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, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 105     real             rswitch_scalar,d_scalar;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm_set1_ps(fr->ic->k_rf);
 124     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 125     crf              = _mm_set1_ps(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->rcoulomb;
 139     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 140     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 141
 142     rswitch_scalar   = fr->rvdw_switch;
 143     rswitch          = _mm_set1_ps(rswitch_scalar);
 144     /* Setup switch parameters */
 145     d_scalar         = rcutoff_scalar-rswitch_scalar;
 146     d                = _mm_set1_ps(d_scalar);
 147     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 148     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 149     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 150     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 151     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 152     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = jnrC = jnrD = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158     j_coord_offsetC = 0;
 159     j_coord_offsetD = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     for(iidx=0;iidx<4*DIM;iidx++)
 165     {
 166         scratch[iidx] = 0.0;
 167     }
 168
 169     /* Start outer loop over neighborlists */
 170     for(iidx=0; iidx<nri; iidx++)
 171     {
 172         /* Load shift vector for this list */
 173         i_shift_offset   = DIM*shiftidx[iidx];
 174
 175         /* Load limits for loop over neighbors */
 176         j_index_start    = jindex[iidx];
 177         j_index_end      = jindex[iidx+1];
 178
 179         /* Get outer coordinate index */
 180         inr              = iinr[iidx];
 181         i_coord_offset   = DIM*inr;
 182
 183         /* Load i particle coords and add shift vector */
 184         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 185                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 186
 187         fix0             = _mm_setzero_ps();
 188         fiy0             = _mm_setzero_ps();
 189         fiz0             = _mm_setzero_ps();
 190         fix1             = _mm_setzero_ps();
 191         fiy1             = _mm_setzero_ps();
 192         fiz1             = _mm_setzero_ps();
 193         fix2             = _mm_setzero_ps();
 194         fiy2             = _mm_setzero_ps();
 195         fiz2             = _mm_setzero_ps();
 196
 197         /* Reset potential sums */
 198         velecsum         = _mm_setzero_ps();
 199         vvdwsum          = _mm_setzero_ps();
 200
 201         /* Start inner kernel loop */
 202         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 203         {
 204
 205             /* Get j neighbor index, and coordinate index */
 206             jnrA             = jjnr[jidx];
 207             jnrB             = jjnr[jidx+1];
 208             jnrC             = jjnr[jidx+2];
 209             jnrD             = jjnr[jidx+3];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212             j_coord_offsetC  = DIM*jnrC;
 213             j_coord_offsetD  = DIM*jnrD;
 214
 215             /* load j atom coordinates */
 216             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 217                                               x+j_coord_offsetC,x+j_coord_offsetD,
 218                                               &jx0,&jy0,&jz0);
 219
 220             /* Calculate displacement vector */
 221             dx00             = _mm_sub_ps(ix0,jx0);
 222             dy00             = _mm_sub_ps(iy0,jy0);
 223             dz00             = _mm_sub_ps(iz0,jz0);
 224             dx10             = _mm_sub_ps(ix1,jx0);
 225             dy10             = _mm_sub_ps(iy1,jy0);
 226             dz10             = _mm_sub_ps(iz1,jz0);
 227             dx20             = _mm_sub_ps(ix2,jx0);
 228             dy20             = _mm_sub_ps(iy2,jy0);
 229             dz20             = _mm_sub_ps(iz2,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 233             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 234             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 235
 236             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 237             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 238             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 239
 240             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 241             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 242             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 243
 244             /* Load parameters for j particles */
 245             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 246                                                               charge+jnrC+0,charge+jnrD+0);
 247             vdwjidx0A        = 2*vdwtype[jnrA+0];
 248             vdwjidx0B        = 2*vdwtype[jnrB+0];
 249             vdwjidx0C        = 2*vdwtype[jnrC+0];
 250             vdwjidx0D        = 2*vdwtype[jnrD+0];
 251
 252             fjx0             = _mm_setzero_ps();
 253             fjy0             = _mm_setzero_ps();
 254             fjz0             = _mm_setzero_ps();
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             if (gmx_mm_any_lt(rsq00,rcutoff2))
 261             {
 262
 263             r00              = _mm_mul_ps(rsq00,rinv00);
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             qq00             = _mm_mul_ps(iq0,jq0);
 267             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 268                                          vdwparam+vdwioffset0+vdwjidx0B,
 269                                          vdwparam+vdwioffset0+vdwjidx0C,
 270                                          vdwparam+vdwioffset0+vdwjidx0D,
 271                                          &c6_00,&c12_00);
 272
 273             /* REACTION-FIELD ELECTROSTATICS */
 274             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 275             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 276
 277             /* LENNARD-JONES DISPERSION/REPULSION */
 278
 279             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 280             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 281             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 282             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 283             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 284
 285             d                = _mm_sub_ps(r00,rswitch);
 286             d                = _mm_max_ps(d,_mm_setzero_ps());
 287             d2               = _mm_mul_ps(d,d);
 288             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 289
 290             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 291
 292             /* Evaluate switch function */
 293             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 294             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 295             vvdw             = _mm_mul_ps(vvdw,sw);
 296             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velec            = _mm_and_ps(velec,cutoff_mask);
 300             velecsum         = _mm_add_ps(velecsum,velec);
 301             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 302             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 303
 304             fscal            = _mm_add_ps(felec,fvdw);
 305
 306             fscal            = _mm_and_ps(fscal,cutoff_mask);
 307
 308              /* Update vectorial force */
 309             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 310             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 311             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 312
 313             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 314             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 315             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 316
 317             }
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             if (gmx_mm_any_lt(rsq10,rcutoff2))
 324             {
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq10             = _mm_mul_ps(iq1,jq0);
 328
 329             /* REACTION-FIELD ELECTROSTATICS */
 330             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 331             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 332
 333             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _mm_and_ps(velec,cutoff_mask);
 337             velecsum         = _mm_add_ps(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341             fscal            = _mm_and_ps(fscal,cutoff_mask);
 342
 343              /* Update vectorial force */
 344             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 345             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 346             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 347
 348             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 349             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 350             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 351
 352             }
 353
 354             /**************************
 355              * CALCULATE INTERACTIONS *
 356              **************************/
 357
 358             if (gmx_mm_any_lt(rsq20,rcutoff2))
 359             {
 360
 361             /* Compute parameters for interactions between i and j atoms */
 362             qq20             = _mm_mul_ps(iq2,jq0);
 363
 364             /* REACTION-FIELD ELECTROSTATICS */
 365             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 366             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 367
 368             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velec            = _mm_and_ps(velec,cutoff_mask);
 372             velecsum         = _mm_add_ps(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             fscal            = _mm_and_ps(fscal,cutoff_mask);
 377
 378              /* Update vectorial force */
 379             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 380             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 381             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 382
 383             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 384             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 385             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 386
 387             }
 388
 389             fjptrA             = f+j_coord_offsetA;
 390             fjptrB             = f+j_coord_offsetB;
 391             fjptrC             = f+j_coord_offsetC;
 392             fjptrD             = f+j_coord_offsetD;
 393
 394             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 395
 396             /* Inner loop uses 151 flops */
 397         }
 398
 399         if(jidx<j_index_end)
 400         {
 401
 402             /* Get j neighbor index, and coordinate index */
 403             jnrlistA         = jjnr[jidx];
 404             jnrlistB         = jjnr[jidx+1];
 405             jnrlistC         = jjnr[jidx+2];
 406             jnrlistD         = jjnr[jidx+3];
 407             /* Sign of each element will be negative for non-real atoms.
 408              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 409              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 410              */
 411             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 412             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 413             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 414             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 415             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 416             j_coord_offsetA  = DIM*jnrA;
 417             j_coord_offsetB  = DIM*jnrB;
 418             j_coord_offsetC  = DIM*jnrC;
 419             j_coord_offsetD  = DIM*jnrD;
 420
 421             /* load j atom coordinates */
 422             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 423                                               x+j_coord_offsetC,x+j_coord_offsetD,
 424                                               &jx0,&jy0,&jz0);
 425
 426             /* Calculate displacement vector */
 427             dx00             = _mm_sub_ps(ix0,jx0);
 428             dy00             = _mm_sub_ps(iy0,jy0);
 429             dz00             = _mm_sub_ps(iz0,jz0);
 430             dx10             = _mm_sub_ps(ix1,jx0);
 431             dy10             = _mm_sub_ps(iy1,jy0);
 432             dz10             = _mm_sub_ps(iz1,jz0);
 433             dx20             = _mm_sub_ps(ix2,jx0);
 434             dy20             = _mm_sub_ps(iy2,jy0);
 435             dz20             = _mm_sub_ps(iz2,jz0);
 436
 437             /* Calculate squared distance and things based on it */
 438             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 439             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 440             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 441
 442             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 443             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 444             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 445
 446             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 447             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 448             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 449
 450             /* Load parameters for j particles */
 451             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 452                                                               charge+jnrC+0,charge+jnrD+0);
 453             vdwjidx0A        = 2*vdwtype[jnrA+0];
 454             vdwjidx0B        = 2*vdwtype[jnrB+0];
 455             vdwjidx0C        = 2*vdwtype[jnrC+0];
 456             vdwjidx0D        = 2*vdwtype[jnrD+0];
 457
 458             fjx0             = _mm_setzero_ps();
 459             fjy0             = _mm_setzero_ps();
 460             fjz0             = _mm_setzero_ps();
 461
 462             /**************************
 463              * CALCULATE INTERACTIONS *
 464              **************************/
 465
 466             if (gmx_mm_any_lt(rsq00,rcutoff2))
 467             {
 468
 469             r00              = _mm_mul_ps(rsq00,rinv00);
 470             r00              = _mm_andnot_ps(dummy_mask,r00);
 471
 472             /* Compute parameters for interactions between i and j atoms */
 473             qq00             = _mm_mul_ps(iq0,jq0);
 474             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 475                                          vdwparam+vdwioffset0+vdwjidx0B,
 476                                          vdwparam+vdwioffset0+vdwjidx0C,
 477                                          vdwparam+vdwioffset0+vdwjidx0D,
 478                                          &c6_00,&c12_00);
 479
 480             /* REACTION-FIELD ELECTROSTATICS */
 481             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 482             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 483
 484             /* LENNARD-JONES DISPERSION/REPULSION */
 485
 486             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 487             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 488             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 489             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 490             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 491
 492             d                = _mm_sub_ps(r00,rswitch);
 493             d                = _mm_max_ps(d,_mm_setzero_ps());
 494             d2               = _mm_mul_ps(d,d);
 495             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 496
 497             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 498
 499             /* Evaluate switch function */
 500             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 501             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 502             vvdw             = _mm_mul_ps(vvdw,sw);
 503             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             velec            = _mm_and_ps(velec,cutoff_mask);
 507             velec            = _mm_andnot_ps(dummy_mask,velec);
 508             velecsum         = _mm_add_ps(velecsum,velec);
 509             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 510             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 511             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 512
 513             fscal            = _mm_add_ps(felec,fvdw);
 514
 515             fscal            = _mm_and_ps(fscal,cutoff_mask);
 516
 517             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 518
 519              /* Update vectorial force */
 520             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 521             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 522             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 523
 524             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 525             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 526             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 527
 528             }
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             if (gmx_mm_any_lt(rsq10,rcutoff2))
 535             {
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq10             = _mm_mul_ps(iq1,jq0);
 539
 540             /* REACTION-FIELD ELECTROSTATICS */
 541             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 542             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 543
 544             cutoff_mask      = _mm_cmplt_ps(rsq10,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             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 559             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 560             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 561
 562             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 563             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 564             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 565
 566             }
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             if (gmx_mm_any_lt(rsq20,rcutoff2))
 573             {
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq20             = _mm_mul_ps(iq2,jq0);
 577
 578             /* REACTION-FIELD ELECTROSTATICS */
 579             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 580             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 581
 582             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm_and_ps(velec,cutoff_mask);
 586             velec            = _mm_andnot_ps(dummy_mask,velec);
 587             velecsum         = _mm_add_ps(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm_and_ps(fscal,cutoff_mask);
 592
 593             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 594
 595              /* Update vectorial force */
 596             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 597             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 598             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 599
 600             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 601             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 602             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 603
 604             }
 605
 606             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 607             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 608             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 609             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 610
 611             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 612
 613             /* Inner loop uses 152 flops */
 614         }
 615
 616         /* End of innermost loop */
 617
 618         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 619                                               f+i_coord_offset,fshift+i_shift_offset);
 620
 621         ggid                        = gid[iidx];
 622         /* Update potential energies */
 623         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 624         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 625
 626         /* Increment number of inner iterations */
 627         inneriter                  += j_index_end - j_index_start;
 628
 629         /* Outer loop uses 20 flops */
 630     }
 631
 632     /* Increment number of outer iterations */
 633     outeriter        += nri;
 634
 635     /* Update outer/inner flops */
 636
 637     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*152);
 638 }
 639 /*
 640  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 641  * Electrostatics interaction: ReactionField
 642  * VdW interaction:            LennardJones
 643  * Geometry:                   Water3-Particle
 644  * Calculate force/pot:        Force
 645  */
 646 void
 647 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 648                     (t_nblist                    * gmx_restrict       nlist,
 649                      rvec                        * gmx_restrict          xx,
 650                      rvec                        * gmx_restrict          ff,
 651                      t_forcerec                  * gmx_restrict          fr,
 652                      t_mdatoms                   * gmx_restrict     mdatoms,
 653                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 654                      t_nrnb                      * gmx_restrict        nrnb)
 655 {
 656     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 657      * just 0 for non-waters.
 658      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 659      * jnr indices corresponding to data put in the four positions in the SIMD register.
 660      */
 661     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 662     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 663     int              jnrA,jnrB,jnrC,jnrD;
 664     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 665     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 666     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 667     real             rcutoff_scalar;
 668     real             *shiftvec,*fshift,*x,*f;
 669     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 670     real             scratch[4*DIM];
 671     __m128           fscal,rcutoff,rcutoff2,jidxall;
 672     int              vdwioffset0;
 673     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 674     int              vdwioffset1;
 675     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 676     int              vdwioffset2;
 677     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 678     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 679     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 680     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 681     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 682     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 683     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 684     real             *charge;
 685     int              nvdwtype;
 686     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 687     int              *vdwtype;
 688     real             *vdwparam;
 689     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 690     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 691     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 692     real             rswitch_scalar,d_scalar;
 693     __m128           dummy_mask,cutoff_mask;
 694     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 695     __m128           one     = _mm_set1_ps(1.0);
 696     __m128           two     = _mm_set1_ps(2.0);
 697     x                = xx[0];
 698     f                = ff[0];
 699
 700     nri              = nlist->nri;
 701     iinr             = nlist->iinr;
 702     jindex           = nlist->jindex;
 703     jjnr             = nlist->jjnr;
 704     shiftidx         = nlist->shift;
 705     gid              = nlist->gid;
 706     shiftvec         = fr->shift_vec[0];
 707     fshift           = fr->fshift[0];
 708     facel            = _mm_set1_ps(fr->epsfac);
 709     charge           = mdatoms->chargeA;
 710     krf              = _mm_set1_ps(fr->ic->k_rf);
 711     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 712     crf              = _mm_set1_ps(fr->ic->c_rf);
 713     nvdwtype         = fr->ntype;
 714     vdwparam         = fr->nbfp;
 715     vdwtype          = mdatoms->typeA;
 716
 717     /* Setup water-specific parameters */
 718     inr              = nlist->iinr[0];
 719     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 720     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 721     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 722     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 723
 724     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 725     rcutoff_scalar   = fr->rcoulomb;
 726     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 727     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 728
 729     rswitch_scalar   = fr->rvdw_switch;
 730     rswitch          = _mm_set1_ps(rswitch_scalar);
 731     /* Setup switch parameters */
 732     d_scalar         = rcutoff_scalar-rswitch_scalar;
 733     d                = _mm_set1_ps(d_scalar);
 734     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 735     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 736     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 737     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 738     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 739     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 740
 741     /* Avoid stupid compiler warnings */
 742     jnrA = jnrB = jnrC = jnrD = 0;
 743     j_coord_offsetA = 0;
 744     j_coord_offsetB = 0;
 745     j_coord_offsetC = 0;
 746     j_coord_offsetD = 0;
 747
 748     outeriter        = 0;
 749     inneriter        = 0;
 750
 751     for(iidx=0;iidx<4*DIM;iidx++)
 752     {
 753         scratch[iidx] = 0.0;
 754     }
 755
 756     /* Start outer loop over neighborlists */
 757     for(iidx=0; iidx<nri; iidx++)
 758     {
 759         /* Load shift vector for this list */
 760         i_shift_offset   = DIM*shiftidx[iidx];
 761
 762         /* Load limits for loop over neighbors */
 763         j_index_start    = jindex[iidx];
 764         j_index_end      = jindex[iidx+1];
 765
 766         /* Get outer coordinate index */
 767         inr              = iinr[iidx];
 768         i_coord_offset   = DIM*inr;
 769
 770         /* Load i particle coords and add shift vector */
 771         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 772                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 773
 774         fix0             = _mm_setzero_ps();
 775         fiy0             = _mm_setzero_ps();
 776         fiz0             = _mm_setzero_ps();
 777         fix1             = _mm_setzero_ps();
 778         fiy1             = _mm_setzero_ps();
 779         fiz1             = _mm_setzero_ps();
 780         fix2             = _mm_setzero_ps();
 781         fiy2             = _mm_setzero_ps();
 782         fiz2             = _mm_setzero_ps();
 783
 784         /* Start inner kernel loop */
 785         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 786         {
 787
 788             /* Get j neighbor index, and coordinate index */
 789             jnrA             = jjnr[jidx];
 790             jnrB             = jjnr[jidx+1];
 791             jnrC             = jjnr[jidx+2];
 792             jnrD             = jjnr[jidx+3];
 793             j_coord_offsetA  = DIM*jnrA;
 794             j_coord_offsetB  = DIM*jnrB;
 795             j_coord_offsetC  = DIM*jnrC;
 796             j_coord_offsetD  = DIM*jnrD;
 797
 798             /* load j atom coordinates */
 799             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 800                                               x+j_coord_offsetC,x+j_coord_offsetD,
 801                                               &jx0,&jy0,&jz0);
 802
 803             /* Calculate displacement vector */
 804             dx00             = _mm_sub_ps(ix0,jx0);
 805             dy00             = _mm_sub_ps(iy0,jy0);
 806             dz00             = _mm_sub_ps(iz0,jz0);
 807             dx10             = _mm_sub_ps(ix1,jx0);
 808             dy10             = _mm_sub_ps(iy1,jy0);
 809             dz10             = _mm_sub_ps(iz1,jz0);
 810             dx20             = _mm_sub_ps(ix2,jx0);
 811             dy20             = _mm_sub_ps(iy2,jy0);
 812             dz20             = _mm_sub_ps(iz2,jz0);
 813
 814             /* Calculate squared distance and things based on it */
 815             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 816             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 817             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 818
 819             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 820             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 821             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 822
 823             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 824             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 825             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 826
 827             /* Load parameters for j particles */
 828             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 829                                                               charge+jnrC+0,charge+jnrD+0);
 830             vdwjidx0A        = 2*vdwtype[jnrA+0];
 831             vdwjidx0B        = 2*vdwtype[jnrB+0];
 832             vdwjidx0C        = 2*vdwtype[jnrC+0];
 833             vdwjidx0D        = 2*vdwtype[jnrD+0];
 834
 835             fjx0             = _mm_setzero_ps();
 836             fjy0             = _mm_setzero_ps();
 837             fjz0             = _mm_setzero_ps();
 838
 839             /**************************
 840              * CALCULATE INTERACTIONS *
 841              **************************/
 842
 843             if (gmx_mm_any_lt(rsq00,rcutoff2))
 844             {
 845
 846             r00              = _mm_mul_ps(rsq00,rinv00);
 847
 848             /* Compute parameters for interactions between i and j atoms */
 849             qq00             = _mm_mul_ps(iq0,jq0);
 850             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 851                                          vdwparam+vdwioffset0+vdwjidx0B,
 852                                          vdwparam+vdwioffset0+vdwjidx0C,
 853                                          vdwparam+vdwioffset0+vdwjidx0D,
 854                                          &c6_00,&c12_00);
 855
 856             /* REACTION-FIELD ELECTROSTATICS */
 857             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 858
 859             /* LENNARD-JONES DISPERSION/REPULSION */
 860
 861             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 862             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 863             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 864             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 865             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 866
 867             d                = _mm_sub_ps(r00,rswitch);
 868             d                = _mm_max_ps(d,_mm_setzero_ps());
 869             d2               = _mm_mul_ps(d,d);
 870             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 871
 872             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 873
 874             /* Evaluate switch function */
 875             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 876             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 877             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 878
 879             fscal            = _mm_add_ps(felec,fvdw);
 880
 881             fscal            = _mm_and_ps(fscal,cutoff_mask);
 882
 883              /* Update vectorial force */
 884             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 885             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 886             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 887
 888             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 889             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 890             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 891
 892             }
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             if (gmx_mm_any_lt(rsq10,rcutoff2))
 899             {
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             qq10             = _mm_mul_ps(iq1,jq0);
 903
 904             /* REACTION-FIELD ELECTROSTATICS */
 905             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 906
 907             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 908
 909             fscal            = felec;
 910
 911             fscal            = _mm_and_ps(fscal,cutoff_mask);
 912
 913              /* Update vectorial force */
 914             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 915             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 916             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 917
 918             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 919             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 920             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 921
 922             }
 923
 924             /**************************
 925              * CALCULATE INTERACTIONS *
 926              **************************/
 927
 928             if (gmx_mm_any_lt(rsq20,rcutoff2))
 929             {
 930
 931             /* Compute parameters for interactions between i and j atoms */
 932             qq20             = _mm_mul_ps(iq2,jq0);
 933
 934             /* REACTION-FIELD ELECTROSTATICS */
 935             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 936
 937             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 938
 939             fscal            = felec;
 940
 941             fscal            = _mm_and_ps(fscal,cutoff_mask);
 942
 943              /* Update vectorial force */
 944             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 945             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 946             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 947
 948             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 949             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 950             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 951
 952             }
 953
 954             fjptrA             = f+j_coord_offsetA;
 955             fjptrB             = f+j_coord_offsetB;
 956             fjptrC             = f+j_coord_offsetC;
 957             fjptrD             = f+j_coord_offsetD;
 958
 959             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 960
 961             /* Inner loop uses 130 flops */
 962         }
 963
 964         if(jidx<j_index_end)
 965         {
 966
 967             /* Get j neighbor index, and coordinate index */
 968             jnrlistA         = jjnr[jidx];
 969             jnrlistB         = jjnr[jidx+1];
 970             jnrlistC         = jjnr[jidx+2];
 971             jnrlistD         = jjnr[jidx+3];
 972             /* Sign of each element will be negative for non-real atoms.
 973              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 974              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 975              */
 976             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 977             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 978             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 979             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 980             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 981             j_coord_offsetA  = DIM*jnrA;
 982             j_coord_offsetB  = DIM*jnrB;
 983             j_coord_offsetC  = DIM*jnrC;
 984             j_coord_offsetD  = DIM*jnrD;
 985
 986             /* load j atom coordinates */
 987             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 988                                               x+j_coord_offsetC,x+j_coord_offsetD,
 989                                               &jx0,&jy0,&jz0);
 990
 991             /* Calculate displacement vector */
 992             dx00             = _mm_sub_ps(ix0,jx0);
 993             dy00             = _mm_sub_ps(iy0,jy0);
 994             dz00             = _mm_sub_ps(iz0,jz0);
 995             dx10             = _mm_sub_ps(ix1,jx0);
 996             dy10             = _mm_sub_ps(iy1,jy0);
 997             dz10             = _mm_sub_ps(iz1,jz0);
 998             dx20             = _mm_sub_ps(ix2,jx0);
 999             dy20             = _mm_sub_ps(iy2,jy0);
1000             dz20             = _mm_sub_ps(iz2,jz0);
1001
1002             /* Calculate squared distance and things based on it */
1003             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1004             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1005             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1006
1007             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1008             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1009             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1010
1011             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1012             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1013             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1014
1015             /* Load parameters for j particles */
1016             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1017                                                               charge+jnrC+0,charge+jnrD+0);
1018             vdwjidx0A        = 2*vdwtype[jnrA+0];
1019             vdwjidx0B        = 2*vdwtype[jnrB+0];
1020             vdwjidx0C        = 2*vdwtype[jnrC+0];
1021             vdwjidx0D        = 2*vdwtype[jnrD+0];
1022
1023             fjx0             = _mm_setzero_ps();
1024             fjy0             = _mm_setzero_ps();
1025             fjz0             = _mm_setzero_ps();
1026
1027             /**************************
1028              * CALCULATE INTERACTIONS *
1029              **************************/
1030
1031             if (gmx_mm_any_lt(rsq00,rcutoff2))
1032             {
1033
1034             r00              = _mm_mul_ps(rsq00,rinv00);
1035             r00              = _mm_andnot_ps(dummy_mask,r00);
1036
1037             /* Compute parameters for interactions between i and j atoms */
1038             qq00             = _mm_mul_ps(iq0,jq0);
1039             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1040                                          vdwparam+vdwioffset0+vdwjidx0B,
1041                                          vdwparam+vdwioffset0+vdwjidx0C,
1042                                          vdwparam+vdwioffset0+vdwjidx0D,
1043                                          &c6_00,&c12_00);
1044
1045             /* REACTION-FIELD ELECTROSTATICS */
1046             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1047
1048             /* LENNARD-JONES DISPERSION/REPULSION */
1049
1050             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1051             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1052             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1053             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1054             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1055
1056             d                = _mm_sub_ps(r00,rswitch);
1057             d                = _mm_max_ps(d,_mm_setzero_ps());
1058             d2               = _mm_mul_ps(d,d);
1059             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1060
1061             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1062
1063             /* Evaluate switch function */
1064             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1065             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1066             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1067
1068             fscal            = _mm_add_ps(felec,fvdw);
1069
1070             fscal            = _mm_and_ps(fscal,cutoff_mask);
1071
1072             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1073
1074              /* Update vectorial force */
1075             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1076             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1077             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1078
1079             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1080             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1081             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1082
1083             }
1084
1085             /**************************
1086              * CALCULATE INTERACTIONS *
1087              **************************/
1088
1089             if (gmx_mm_any_lt(rsq10,rcutoff2))
1090             {
1091
1092             /* Compute parameters for interactions between i and j atoms */
1093             qq10             = _mm_mul_ps(iq1,jq0);
1094
1095             /* REACTION-FIELD ELECTROSTATICS */
1096             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1097
1098             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1099
1100             fscal            = felec;
1101
1102             fscal            = _mm_and_ps(fscal,cutoff_mask);
1103
1104             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1105
1106              /* Update vectorial force */
1107             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1108             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1109             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1110
1111             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1112             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1113             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1114
1115             }
1116
1117             /**************************
1118              * CALCULATE INTERACTIONS *
1119              **************************/
1120
1121             if (gmx_mm_any_lt(rsq20,rcutoff2))
1122             {
1123
1124             /* Compute parameters for interactions between i and j atoms */
1125             qq20             = _mm_mul_ps(iq2,jq0);
1126
1127             /* REACTION-FIELD ELECTROSTATICS */
1128             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1129
1130             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1131
1132             fscal            = felec;
1133
1134             fscal            = _mm_and_ps(fscal,cutoff_mask);
1135
1136             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1137
1138              /* Update vectorial force */
1139             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1140             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1141             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1142
1143             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1144             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1145             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1146
1147             }
1148
1149             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1150             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1151             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1152             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1153
1154             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1155
1156             /* Inner loop uses 131 flops */
1157         }
1158
1159         /* End of innermost loop */
1160
1161         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1162                                               f+i_coord_offset,fshift+i_shift_offset);
1163
1164         /* Increment number of inner iterations */
1165         inneriter                  += j_index_end - j_index_start;
1166
1167         /* Outer loop uses 18 flops */
1168     }
1169
1170     /* Increment number of outer iterations */
1171     outeriter        += nri;
1172
1173     /* Update outer/inner flops */
1174
1175     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*131);
1176 }