src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 102     real             rswitch_scalar,d_scalar;
 103     __m128           dummy_mask,cutoff_mask;
 104     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 105     __m128           one     = _mm_set1_ps(1.0);
 106     __m128           two     = _mm_set1_ps(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_ps(fr->ic->epsfac);
 119     charge           = mdatoms->chargeA;
 120     krf              = _mm_set1_ps(fr->ic->k_rf);
 121     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 122     crf              = _mm_set1_ps(fr->ic->c_rf);
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 130     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 131     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 132     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 133
 134     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 135     rcutoff_scalar   = fr->ic->rcoulomb;
 136     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 137     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 138
 139     rswitch_scalar   = fr->ic->rvdw_switch;
 140     rswitch          = _mm_set1_ps(rswitch_scalar);
 141     /* Setup switch parameters */
 142     d_scalar         = rcutoff_scalar-rswitch_scalar;
 143     d                = _mm_set1_ps(d_scalar);
 144     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 145     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 146     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 147     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 148     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 149     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = jnrC = jnrD = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155     j_coord_offsetC = 0;
 156     j_coord_offsetD = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     for(iidx=0;iidx<4*DIM;iidx++)
 162     {
 163         scratch[iidx] = 0.0;
 164     }
 165
 166     /* Start outer loop over neighborlists */
 167     for(iidx=0; iidx<nri; iidx++)
 168     {
 169         /* Load shift vector for this list */
 170         i_shift_offset   = DIM*shiftidx[iidx];
 171
 172         /* Load limits for loop over neighbors */
 173         j_index_start    = jindex[iidx];
 174         j_index_end      = jindex[iidx+1];
 175
 176         /* Get outer coordinate index */
 177         inr              = iinr[iidx];
 178         i_coord_offset   = DIM*inr;
 179
 180         /* Load i particle coords and add shift vector */
 181         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 182                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 183
 184         fix0             = _mm_setzero_ps();
 185         fiy0             = _mm_setzero_ps();
 186         fiz0             = _mm_setzero_ps();
 187         fix1             = _mm_setzero_ps();
 188         fiy1             = _mm_setzero_ps();
 189         fiz1             = _mm_setzero_ps();
 190         fix2             = _mm_setzero_ps();
 191         fiy2             = _mm_setzero_ps();
 192         fiz2             = _mm_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm_setzero_ps();
 196         vvdwsum          = _mm_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211
 212             /* load j atom coordinates */
 213             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                               x+j_coord_offsetC,x+j_coord_offsetD,
 215                                               &jx0,&jy0,&jz0);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm_sub_ps(ix0,jx0);
 219             dy00             = _mm_sub_ps(iy0,jy0);
 220             dz00             = _mm_sub_ps(iz0,jz0);
 221             dx10             = _mm_sub_ps(ix1,jx0);
 222             dy10             = _mm_sub_ps(iy1,jy0);
 223             dz10             = _mm_sub_ps(iz1,jz0);
 224             dx20             = _mm_sub_ps(ix2,jx0);
 225             dy20             = _mm_sub_ps(iy2,jy0);
 226             dz20             = _mm_sub_ps(iz2,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 230             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 231             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 232
 233             rinv00           = avx128fma_invsqrt_f(rsq00);
 234             rinv10           = avx128fma_invsqrt_f(rsq10);
 235             rinv20           = avx128fma_invsqrt_f(rsq20);
 236
 237             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 238             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 239             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 240
 241             /* Load parameters for j particles */
 242             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 243                                                               charge+jnrC+0,charge+jnrD+0);
 244             vdwjidx0A        = 2*vdwtype[jnrA+0];
 245             vdwjidx0B        = 2*vdwtype[jnrB+0];
 246             vdwjidx0C        = 2*vdwtype[jnrC+0];
 247             vdwjidx0D        = 2*vdwtype[jnrD+0];
 248
 249             fjx0             = _mm_setzero_ps();
 250             fjy0             = _mm_setzero_ps();
 251             fjz0             = _mm_setzero_ps();
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             if (gmx_mm_any_lt(rsq00,rcutoff2))
 258             {
 259
 260             r00              = _mm_mul_ps(rsq00,rinv00);
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             qq00             = _mm_mul_ps(iq0,jq0);
 264             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 265                                          vdwparam+vdwioffset0+vdwjidx0B,
 266                                          vdwparam+vdwioffset0+vdwjidx0C,
 267                                          vdwparam+vdwioffset0+vdwjidx0D,
 268                                          &c6_00,&c12_00);
 269
 270             /* REACTION-FIELD ELECTROSTATICS */
 271             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 272             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 273
 274             /* LENNARD-JONES DISPERSION/REPULSION */
 275
 276             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 277             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 278             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 279             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 280             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 281
 282             d                = _mm_sub_ps(r00,rswitch);
 283             d                = _mm_max_ps(d,_mm_setzero_ps());
 284             d2               = _mm_mul_ps(d,d);
 285             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 286
 287             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 288
 289             /* Evaluate switch function */
 290             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 291             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 292             vvdw             = _mm_mul_ps(vvdw,sw);
 293             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velec            = _mm_and_ps(velec,cutoff_mask);
 297             velecsum         = _mm_add_ps(velecsum,velec);
 298             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 299             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 300
 301             fscal            = _mm_add_ps(felec,fvdw);
 302
 303             fscal            = _mm_and_ps(fscal,cutoff_mask);
 304
 305              /* Update vectorial force */
 306             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 307             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 308             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 309
 310             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 311             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 312             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 313
 314             }
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             if (gmx_mm_any_lt(rsq10,rcutoff2))
 321             {
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm_mul_ps(iq1,jq0);
 325
 326             /* REACTION-FIELD ELECTROSTATICS */
 327             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 328             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 329
 330             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm_and_ps(velec,cutoff_mask);
 334             velecsum         = _mm_add_ps(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             fscal            = _mm_and_ps(fscal,cutoff_mask);
 339
 340              /* Update vectorial force */
 341             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 342             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 343             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 344
 345             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 346             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 347             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 348
 349             }
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             if (gmx_mm_any_lt(rsq20,rcutoff2))
 356             {
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq20             = _mm_mul_ps(iq2,jq0);
 360
 361             /* REACTION-FIELD ELECTROSTATICS */
 362             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 363             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 364
 365             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velec            = _mm_and_ps(velec,cutoff_mask);
 369             velecsum         = _mm_add_ps(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             fscal            = _mm_and_ps(fscal,cutoff_mask);
 374
 375              /* Update vectorial force */
 376             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 377             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 378             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 379
 380             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 381             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 382             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 383
 384             }
 385
 386             fjptrA             = f+j_coord_offsetA;
 387             fjptrB             = f+j_coord_offsetB;
 388             fjptrC             = f+j_coord_offsetC;
 389             fjptrD             = f+j_coord_offsetD;
 390
 391             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 392
 393             /* Inner loop uses 151 flops */
 394         }
 395
 396         if(jidx<j_index_end)
 397         {
 398
 399             /* Get j neighbor index, and coordinate index */
 400             jnrlistA         = jjnr[jidx];
 401             jnrlistB         = jjnr[jidx+1];
 402             jnrlistC         = jjnr[jidx+2];
 403             jnrlistD         = jjnr[jidx+3];
 404             /* Sign of each element will be negative for non-real atoms.
 405              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 406              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 407              */
 408             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 409             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 410             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 411             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 412             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 413             j_coord_offsetA  = DIM*jnrA;
 414             j_coord_offsetB  = DIM*jnrB;
 415             j_coord_offsetC  = DIM*jnrC;
 416             j_coord_offsetD  = DIM*jnrD;
 417
 418             /* load j atom coordinates */
 419             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 420                                               x+j_coord_offsetC,x+j_coord_offsetD,
 421                                               &jx0,&jy0,&jz0);
 422
 423             /* Calculate displacement vector */
 424             dx00             = _mm_sub_ps(ix0,jx0);
 425             dy00             = _mm_sub_ps(iy0,jy0);
 426             dz00             = _mm_sub_ps(iz0,jz0);
 427             dx10             = _mm_sub_ps(ix1,jx0);
 428             dy10             = _mm_sub_ps(iy1,jy0);
 429             dz10             = _mm_sub_ps(iz1,jz0);
 430             dx20             = _mm_sub_ps(ix2,jx0);
 431             dy20             = _mm_sub_ps(iy2,jy0);
 432             dz20             = _mm_sub_ps(iz2,jz0);
 433
 434             /* Calculate squared distance and things based on it */
 435             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 436             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 437             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 438
 439             rinv00           = avx128fma_invsqrt_f(rsq00);
 440             rinv10           = avx128fma_invsqrt_f(rsq10);
 441             rinv20           = avx128fma_invsqrt_f(rsq20);
 442
 443             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 444             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 445             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 446
 447             /* Load parameters for j particles */
 448             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 449                                                               charge+jnrC+0,charge+jnrD+0);
 450             vdwjidx0A        = 2*vdwtype[jnrA+0];
 451             vdwjidx0B        = 2*vdwtype[jnrB+0];
 452             vdwjidx0C        = 2*vdwtype[jnrC+0];
 453             vdwjidx0D        = 2*vdwtype[jnrD+0];
 454
 455             fjx0             = _mm_setzero_ps();
 456             fjy0             = _mm_setzero_ps();
 457             fjz0             = _mm_setzero_ps();
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             if (gmx_mm_any_lt(rsq00,rcutoff2))
 464             {
 465
 466             r00              = _mm_mul_ps(rsq00,rinv00);
 467             r00              = _mm_andnot_ps(dummy_mask,r00);
 468
 469             /* Compute parameters for interactions between i and j atoms */
 470             qq00             = _mm_mul_ps(iq0,jq0);
 471             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 472                                          vdwparam+vdwioffset0+vdwjidx0B,
 473                                          vdwparam+vdwioffset0+vdwjidx0C,
 474                                          vdwparam+vdwioffset0+vdwjidx0D,
 475                                          &c6_00,&c12_00);
 476
 477             /* REACTION-FIELD ELECTROSTATICS */
 478             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 479             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 480
 481             /* LENNARD-JONES DISPERSION/REPULSION */
 482
 483             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 484             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 485             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 486             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 487             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 488
 489             d                = _mm_sub_ps(r00,rswitch);
 490             d                = _mm_max_ps(d,_mm_setzero_ps());
 491             d2               = _mm_mul_ps(d,d);
 492             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 493
 494             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 495
 496             /* Evaluate switch function */
 497             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 498             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 499             vvdw             = _mm_mul_ps(vvdw,sw);
 500             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 501
 502             /* Update potential sum for this i atom from the interaction with this j atom. */
 503             velec            = _mm_and_ps(velec,cutoff_mask);
 504             velec            = _mm_andnot_ps(dummy_mask,velec);
 505             velecsum         = _mm_add_ps(velecsum,velec);
 506             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 507             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 508             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 509
 510             fscal            = _mm_add_ps(felec,fvdw);
 511
 512             fscal            = _mm_and_ps(fscal,cutoff_mask);
 513
 514             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 515
 516              /* Update vectorial force */
 517             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 518             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 519             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 520
 521             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 522             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 523             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 524
 525             }
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             if (gmx_mm_any_lt(rsq10,rcutoff2))
 532             {
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq10             = _mm_mul_ps(iq1,jq0);
 536
 537             /* REACTION-FIELD ELECTROSTATICS */
 538             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 539             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 540
 541             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velec            = _mm_and_ps(velec,cutoff_mask);
 545             velec            = _mm_andnot_ps(dummy_mask,velec);
 546             velecsum         = _mm_add_ps(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm_and_ps(fscal,cutoff_mask);
 551
 552             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 553
 554              /* Update vectorial force */
 555             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 556             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 557             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 558
 559             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 560             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 561             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 562
 563             }
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             if (gmx_mm_any_lt(rsq20,rcutoff2))
 570             {
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq20             = _mm_mul_ps(iq2,jq0);
 574
 575             /* REACTION-FIELD ELECTROSTATICS */
 576             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 577             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 578
 579             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm_and_ps(velec,cutoff_mask);
 583             velec            = _mm_andnot_ps(dummy_mask,velec);
 584             velecsum         = _mm_add_ps(velecsum,velec);
 585
 586             fscal            = felec;
 587
 588             fscal            = _mm_and_ps(fscal,cutoff_mask);
 589
 590             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 591
 592              /* Update vectorial force */
 593             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 594             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 595             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 596
 597             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 598             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 599             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 600
 601             }
 602
 603             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 604             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 605             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 606             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 607
 608             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 609
 610             /* Inner loop uses 152 flops */
 611         }
 612
 613         /* End of innermost loop */
 614
 615         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 616                                               f+i_coord_offset,fshift+i_shift_offset);
 617
 618         ggid                        = gid[iidx];
 619         /* Update potential energies */
 620         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 621         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 622
 623         /* Increment number of inner iterations */
 624         inneriter                  += j_index_end - j_index_start;
 625
 626         /* Outer loop uses 20 flops */
 627     }
 628
 629     /* Increment number of outer iterations */
 630     outeriter        += nri;
 631
 632     /* Update outer/inner flops */
 633
 634     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*152);
 635 }
 636 /*
 637  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 638  * Electrostatics interaction: ReactionField
 639  * VdW interaction:            LennardJones
 640  * Geometry:                   Water3-Particle
 641  * Calculate force/pot:        Force
 642  */
 643 void
 644 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 645                     (t_nblist                    * gmx_restrict       nlist,
 646                      rvec                        * gmx_restrict          xx,
 647                      rvec                        * gmx_restrict          ff,
 648                      struct t_forcerec           * gmx_restrict          fr,
 649                      t_mdatoms                   * gmx_restrict     mdatoms,
 650                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 651                      t_nrnb                      * gmx_restrict        nrnb)
 652 {
 653     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 654      * just 0 for non-waters.
 655      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 656      * jnr indices corresponding to data put in the four positions in the SIMD register.
 657      */
 658     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 659     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 660     int              jnrA,jnrB,jnrC,jnrD;
 661     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 662     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 663     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 664     real             rcutoff_scalar;
 665     real             *shiftvec,*fshift,*x,*f;
 666     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 667     real             scratch[4*DIM];
 668     __m128           fscal,rcutoff,rcutoff2,jidxall;
 669     int              vdwioffset0;
 670     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 671     int              vdwioffset1;
 672     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 673     int              vdwioffset2;
 674     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 675     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 676     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 677     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 678     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 679     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 680     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 681     real             *charge;
 682     int              nvdwtype;
 683     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 684     int              *vdwtype;
 685     real             *vdwparam;
 686     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 687     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 688     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 689     real             rswitch_scalar,d_scalar;
 690     __m128           dummy_mask,cutoff_mask;
 691     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 692     __m128           one     = _mm_set1_ps(1.0);
 693     __m128           two     = _mm_set1_ps(2.0);
 694     x                = xx[0];
 695     f                = ff[0];
 696
 697     nri              = nlist->nri;
 698     iinr             = nlist->iinr;
 699     jindex           = nlist->jindex;
 700     jjnr             = nlist->jjnr;
 701     shiftidx         = nlist->shift;
 702     gid              = nlist->gid;
 703     shiftvec         = fr->shift_vec[0];
 704     fshift           = fr->fshift[0];
 705     facel            = _mm_set1_ps(fr->ic->epsfac);
 706     charge           = mdatoms->chargeA;
 707     krf              = _mm_set1_ps(fr->ic->k_rf);
 708     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 709     crf              = _mm_set1_ps(fr->ic->c_rf);
 710     nvdwtype         = fr->ntype;
 711     vdwparam         = fr->nbfp;
 712     vdwtype          = mdatoms->typeA;
 713
 714     /* Setup water-specific parameters */
 715     inr              = nlist->iinr[0];
 716     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 717     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 718     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 719     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 720
 721     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 722     rcutoff_scalar   = fr->ic->rcoulomb;
 723     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 724     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 725
 726     rswitch_scalar   = fr->ic->rvdw_switch;
 727     rswitch          = _mm_set1_ps(rswitch_scalar);
 728     /* Setup switch parameters */
 729     d_scalar         = rcutoff_scalar-rswitch_scalar;
 730     d                = _mm_set1_ps(d_scalar);
 731     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 732     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 733     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 734     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 735     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 736     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 737
 738     /* Avoid stupid compiler warnings */
 739     jnrA = jnrB = jnrC = jnrD = 0;
 740     j_coord_offsetA = 0;
 741     j_coord_offsetB = 0;
 742     j_coord_offsetC = 0;
 743     j_coord_offsetD = 0;
 744
 745     outeriter        = 0;
 746     inneriter        = 0;
 747
 748     for(iidx=0;iidx<4*DIM;iidx++)
 749     {
 750         scratch[iidx] = 0.0;
 751     }
 752
 753     /* Start outer loop over neighborlists */
 754     for(iidx=0; iidx<nri; iidx++)
 755     {
 756         /* Load shift vector for this list */
 757         i_shift_offset   = DIM*shiftidx[iidx];
 758
 759         /* Load limits for loop over neighbors */
 760         j_index_start    = jindex[iidx];
 761         j_index_end      = jindex[iidx+1];
 762
 763         /* Get outer coordinate index */
 764         inr              = iinr[iidx];
 765         i_coord_offset   = DIM*inr;
 766
 767         /* Load i particle coords and add shift vector */
 768         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 769                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 770
 771         fix0             = _mm_setzero_ps();
 772         fiy0             = _mm_setzero_ps();
 773         fiz0             = _mm_setzero_ps();
 774         fix1             = _mm_setzero_ps();
 775         fiy1             = _mm_setzero_ps();
 776         fiz1             = _mm_setzero_ps();
 777         fix2             = _mm_setzero_ps();
 778         fiy2             = _mm_setzero_ps();
 779         fiz2             = _mm_setzero_ps();
 780
 781         /* Start inner kernel loop */
 782         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 783         {
 784
 785             /* Get j neighbor index, and coordinate index */
 786             jnrA             = jjnr[jidx];
 787             jnrB             = jjnr[jidx+1];
 788             jnrC             = jjnr[jidx+2];
 789             jnrD             = jjnr[jidx+3];
 790             j_coord_offsetA  = DIM*jnrA;
 791             j_coord_offsetB  = DIM*jnrB;
 792             j_coord_offsetC  = DIM*jnrC;
 793             j_coord_offsetD  = DIM*jnrD;
 794
 795             /* load j atom coordinates */
 796             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 797                                               x+j_coord_offsetC,x+j_coord_offsetD,
 798                                               &jx0,&jy0,&jz0);
 799
 800             /* Calculate displacement vector */
 801             dx00             = _mm_sub_ps(ix0,jx0);
 802             dy00             = _mm_sub_ps(iy0,jy0);
 803             dz00             = _mm_sub_ps(iz0,jz0);
 804             dx10             = _mm_sub_ps(ix1,jx0);
 805             dy10             = _mm_sub_ps(iy1,jy0);
 806             dz10             = _mm_sub_ps(iz1,jz0);
 807             dx20             = _mm_sub_ps(ix2,jx0);
 808             dy20             = _mm_sub_ps(iy2,jy0);
 809             dz20             = _mm_sub_ps(iz2,jz0);
 810
 811             /* Calculate squared distance and things based on it */
 812             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 813             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 814             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 815
 816             rinv00           = avx128fma_invsqrt_f(rsq00);
 817             rinv10           = avx128fma_invsqrt_f(rsq10);
 818             rinv20           = avx128fma_invsqrt_f(rsq20);
 819
 820             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 821             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 822             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 823
 824             /* Load parameters for j particles */
 825             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 826                                                               charge+jnrC+0,charge+jnrD+0);
 827             vdwjidx0A        = 2*vdwtype[jnrA+0];
 828             vdwjidx0B        = 2*vdwtype[jnrB+0];
 829             vdwjidx0C        = 2*vdwtype[jnrC+0];
 830             vdwjidx0D        = 2*vdwtype[jnrD+0];
 831
 832             fjx0             = _mm_setzero_ps();
 833             fjy0             = _mm_setzero_ps();
 834             fjz0             = _mm_setzero_ps();
 835
 836             /**************************
 837              * CALCULATE INTERACTIONS *
 838              **************************/
 839
 840             if (gmx_mm_any_lt(rsq00,rcutoff2))
 841             {
 842
 843             r00              = _mm_mul_ps(rsq00,rinv00);
 844
 845             /* Compute parameters for interactions between i and j atoms */
 846             qq00             = _mm_mul_ps(iq0,jq0);
 847             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 848                                          vdwparam+vdwioffset0+vdwjidx0B,
 849                                          vdwparam+vdwioffset0+vdwjidx0C,
 850                                          vdwparam+vdwioffset0+vdwjidx0D,
 851                                          &c6_00,&c12_00);
 852
 853             /* REACTION-FIELD ELECTROSTATICS */
 854             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 855
 856             /* LENNARD-JONES DISPERSION/REPULSION */
 857
 858             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 859             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 860             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 861             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 862             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 863
 864             d                = _mm_sub_ps(r00,rswitch);
 865             d                = _mm_max_ps(d,_mm_setzero_ps());
 866             d2               = _mm_mul_ps(d,d);
 867             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 868
 869             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 870
 871             /* Evaluate switch function */
 872             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 873             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 874             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 875
 876             fscal            = _mm_add_ps(felec,fvdw);
 877
 878             fscal            = _mm_and_ps(fscal,cutoff_mask);
 879
 880              /* Update vectorial force */
 881             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 882             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 883             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 884
 885             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 886             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 887             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 888
 889             }
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             if (gmx_mm_any_lt(rsq10,rcutoff2))
 896             {
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             qq10             = _mm_mul_ps(iq1,jq0);
 900
 901             /* REACTION-FIELD ELECTROSTATICS */
 902             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 903
 904             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 905
 906             fscal            = felec;
 907
 908             fscal            = _mm_and_ps(fscal,cutoff_mask);
 909
 910              /* Update vectorial force */
 911             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 912             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 913             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 914
 915             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 916             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 917             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 918
 919             }
 920
 921             /**************************
 922              * CALCULATE INTERACTIONS *
 923              **************************/
 924
 925             if (gmx_mm_any_lt(rsq20,rcutoff2))
 926             {
 927
 928             /* Compute parameters for interactions between i and j atoms */
 929             qq20             = _mm_mul_ps(iq2,jq0);
 930
 931             /* REACTION-FIELD ELECTROSTATICS */
 932             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 933
 934             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 935
 936             fscal            = felec;
 937
 938             fscal            = _mm_and_ps(fscal,cutoff_mask);
 939
 940              /* Update vectorial force */
 941             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 942             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 943             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 944
 945             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 946             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 947             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 948
 949             }
 950
 951             fjptrA             = f+j_coord_offsetA;
 952             fjptrB             = f+j_coord_offsetB;
 953             fjptrC             = f+j_coord_offsetC;
 954             fjptrD             = f+j_coord_offsetD;
 955
 956             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 957
 958             /* Inner loop uses 130 flops */
 959         }
 960
 961         if(jidx<j_index_end)
 962         {
 963
 964             /* Get j neighbor index, and coordinate index */
 965             jnrlistA         = jjnr[jidx];
 966             jnrlistB         = jjnr[jidx+1];
 967             jnrlistC         = jjnr[jidx+2];
 968             jnrlistD         = jjnr[jidx+3];
 969             /* Sign of each element will be negative for non-real atoms.
 970              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 971              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 972              */
 973             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 974             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 975             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 976             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 977             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 978             j_coord_offsetA  = DIM*jnrA;
 979             j_coord_offsetB  = DIM*jnrB;
 980             j_coord_offsetC  = DIM*jnrC;
 981             j_coord_offsetD  = DIM*jnrD;
 982
 983             /* load j atom coordinates */
 984             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 985                                               x+j_coord_offsetC,x+j_coord_offsetD,
 986                                               &jx0,&jy0,&jz0);
 987
 988             /* Calculate displacement vector */
 989             dx00             = _mm_sub_ps(ix0,jx0);
 990             dy00             = _mm_sub_ps(iy0,jy0);
 991             dz00             = _mm_sub_ps(iz0,jz0);
 992             dx10             = _mm_sub_ps(ix1,jx0);
 993             dy10             = _mm_sub_ps(iy1,jy0);
 994             dz10             = _mm_sub_ps(iz1,jz0);
 995             dx20             = _mm_sub_ps(ix2,jx0);
 996             dy20             = _mm_sub_ps(iy2,jy0);
 997             dz20             = _mm_sub_ps(iz2,jz0);
 998
 999             /* Calculate squared distance and things based on it */
1000             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1001             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1002             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1003
1004             rinv00           = avx128fma_invsqrt_f(rsq00);
1005             rinv10           = avx128fma_invsqrt_f(rsq10);
1006             rinv20           = avx128fma_invsqrt_f(rsq20);
1007
1008             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1009             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1010             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1011
1012             /* Load parameters for j particles */
1013             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1014                                                               charge+jnrC+0,charge+jnrD+0);
1015             vdwjidx0A        = 2*vdwtype[jnrA+0];
1016             vdwjidx0B        = 2*vdwtype[jnrB+0];
1017             vdwjidx0C        = 2*vdwtype[jnrC+0];
1018             vdwjidx0D        = 2*vdwtype[jnrD+0];
1019
1020             fjx0             = _mm_setzero_ps();
1021             fjy0             = _mm_setzero_ps();
1022             fjz0             = _mm_setzero_ps();
1023
1024             /**************************
1025              * CALCULATE INTERACTIONS *
1026              **************************/
1027
1028             if (gmx_mm_any_lt(rsq00,rcutoff2))
1029             {
1030
1031             r00              = _mm_mul_ps(rsq00,rinv00);
1032             r00              = _mm_andnot_ps(dummy_mask,r00);
1033
1034             /* Compute parameters for interactions between i and j atoms */
1035             qq00             = _mm_mul_ps(iq0,jq0);
1036             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1037                                          vdwparam+vdwioffset0+vdwjidx0B,
1038                                          vdwparam+vdwioffset0+vdwjidx0C,
1039                                          vdwparam+vdwioffset0+vdwjidx0D,
1040                                          &c6_00,&c12_00);
1041
1042             /* REACTION-FIELD ELECTROSTATICS */
1043             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1044
1045             /* LENNARD-JONES DISPERSION/REPULSION */
1046
1047             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1048             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1049             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1050             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1051             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1052
1053             d                = _mm_sub_ps(r00,rswitch);
1054             d                = _mm_max_ps(d,_mm_setzero_ps());
1055             d2               = _mm_mul_ps(d,d);
1056             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1057
1058             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1059
1060             /* Evaluate switch function */
1061             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1062             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1063             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1064
1065             fscal            = _mm_add_ps(felec,fvdw);
1066
1067             fscal            = _mm_and_ps(fscal,cutoff_mask);
1068
1069             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1070
1071              /* Update vectorial force */
1072             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1073             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1074             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1075
1076             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1077             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1078             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1079
1080             }
1081
1082             /**************************
1083              * CALCULATE INTERACTIONS *
1084              **************************/
1085
1086             if (gmx_mm_any_lt(rsq10,rcutoff2))
1087             {
1088
1089             /* Compute parameters for interactions between i and j atoms */
1090             qq10             = _mm_mul_ps(iq1,jq0);
1091
1092             /* REACTION-FIELD ELECTROSTATICS */
1093             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1094
1095             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1096
1097             fscal            = felec;
1098
1099             fscal            = _mm_and_ps(fscal,cutoff_mask);
1100
1101             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1102
1103              /* Update vectorial force */
1104             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1105             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1106             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1107
1108             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1109             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1110             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1111
1112             }
1113
1114             /**************************
1115              * CALCULATE INTERACTIONS *
1116              **************************/
1117
1118             if (gmx_mm_any_lt(rsq20,rcutoff2))
1119             {
1120
1121             /* Compute parameters for interactions between i and j atoms */
1122             qq20             = _mm_mul_ps(iq2,jq0);
1123
1124             /* REACTION-FIELD ELECTROSTATICS */
1125             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1126
1127             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1128
1129             fscal            = felec;
1130
1131             fscal            = _mm_and_ps(fscal,cutoff_mask);
1132
1133             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1134
1135              /* Update vectorial force */
1136             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1137             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1138             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1139
1140             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1141             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1142             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1143
1144             }
1145
1146             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1147             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1148             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1149             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1150
1151             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1152
1153             /* Inner loop uses 131 flops */
1154         }
1155
1156         /* End of innermost loop */
1157
1158         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1159                                               f+i_coord_offset,fshift+i_shift_offset);
1160
1161         /* Increment number of inner iterations */
1162         inneriter                  += j_index_end - j_index_start;
1163
1164         /* Outer loop uses 18 flops */
1165     }
1166
1167     /* Increment number of outer iterations */
1168     outeriter        += nri;
1169
1170     /* Update outer/inner flops */
1171
1172     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*131);
1173 }