src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwLJ_GeomW4P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  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           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128           dummy_mask,cutoff_mask;
 106     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 107     __m128           one     = _mm_set1_ps(1.0);
 108     __m128           two     = _mm_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_ps(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 129     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 130     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 131     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     for(iidx=0;iidx<4*DIM;iidx++)
 144     {
 145         scratch[iidx] = 0.0;
 146     }
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 164                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 165
 166         fix0             = _mm_setzero_ps();
 167         fiy0             = _mm_setzero_ps();
 168         fiz0             = _mm_setzero_ps();
 169         fix1             = _mm_setzero_ps();
 170         fiy1             = _mm_setzero_ps();
 171         fiz1             = _mm_setzero_ps();
 172         fix2             = _mm_setzero_ps();
 173         fiy2             = _mm_setzero_ps();
 174         fiz2             = _mm_setzero_ps();
 175         fix3             = _mm_setzero_ps();
 176         fiy3             = _mm_setzero_ps();
 177         fiz3             = _mm_setzero_ps();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_ps();
 181         vvdwsum          = _mm_setzero_ps();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             jnrC             = jjnr[jidx+2];
 191             jnrD             = jjnr[jidx+3];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194             j_coord_offsetC  = DIM*jnrC;
 195             j_coord_offsetD  = DIM*jnrD;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                               x+j_coord_offsetC,x+j_coord_offsetD,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_ps(ix0,jx0);
 204             dy00             = _mm_sub_ps(iy0,jy0);
 205             dz00             = _mm_sub_ps(iz0,jz0);
 206             dx10             = _mm_sub_ps(ix1,jx0);
 207             dy10             = _mm_sub_ps(iy1,jy0);
 208             dz10             = _mm_sub_ps(iz1,jz0);
 209             dx20             = _mm_sub_ps(ix2,jx0);
 210             dy20             = _mm_sub_ps(iy2,jy0);
 211             dz20             = _mm_sub_ps(iz2,jz0);
 212             dx30             = _mm_sub_ps(ix3,jx0);
 213             dy30             = _mm_sub_ps(iy3,jy0);
 214             dz30             = _mm_sub_ps(iz3,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 220             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 221
 222             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 223             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 224             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 225
 226             rinvsq00         = gmx_mm_inv_ps(rsq00);
 227             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 228             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 229             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 233                                                               charge+jnrC+0,charge+jnrD+0);
 234             vdwjidx0A        = 2*vdwtype[jnrA+0];
 235             vdwjidx0B        = 2*vdwtype[jnrB+0];
 236             vdwjidx0C        = 2*vdwtype[jnrC+0];
 237             vdwjidx0D        = 2*vdwtype[jnrD+0];
 238
 239             fjx0             = _mm_setzero_ps();
 240             fjy0             = _mm_setzero_ps();
 241             fjz0             = _mm_setzero_ps();
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             /* Compute parameters for interactions between i and j atoms */
 248             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 249                                          vdwparam+vdwioffset0+vdwjidx0B,
 250                                          vdwparam+vdwioffset0+vdwjidx0C,
 251                                          vdwparam+vdwioffset0+vdwjidx0D,
 252                                          &c6_00,&c12_00);
 253
 254             /* LENNARD-JONES DISPERSION/REPULSION */
 255
 256             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 257             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 258             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 259             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 260             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 264
 265             fscal            = fvdw;
 266
 267              /* Update vectorial force */
 268             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 269             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 270             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 271
 272             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 273             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 274             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             qq10             = _mm_mul_ps(iq1,jq0);
 282
 283             /* COULOMB ELECTROSTATICS */
 284             velec            = _mm_mul_ps(qq10,rinv10);
 285             felec            = _mm_mul_ps(velec,rinvsq10);
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             velecsum         = _mm_add_ps(velecsum,velec);
 289
 290             fscal            = felec;
 291
 292              /* Update vectorial force */
 293             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 294             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 295             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 296
 297             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 298             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 299             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq20             = _mm_mul_ps(iq2,jq0);
 307
 308             /* COULOMB ELECTROSTATICS */
 309             velec            = _mm_mul_ps(qq20,rinv20);
 310             felec            = _mm_mul_ps(velec,rinvsq20);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm_add_ps(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317              /* Update vectorial force */
 318             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 319             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 320             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 321
 322             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 323             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 324             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             qq30             = _mm_mul_ps(iq3,jq0);
 332
 333             /* COULOMB ELECTROSTATICS */
 334             velec            = _mm_mul_ps(qq30,rinv30);
 335             felec            = _mm_mul_ps(velec,rinvsq30);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velecsum         = _mm_add_ps(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342              /* Update vectorial force */
 343             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 344             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 345             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 346
 347             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 348             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 349             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 350
 351             fjptrA             = f+j_coord_offsetA;
 352             fjptrB             = f+j_coord_offsetB;
 353             fjptrC             = f+j_coord_offsetC;
 354             fjptrD             = f+j_coord_offsetD;
 355
 356             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 357
 358             /* Inner loop uses 128 flops */
 359         }
 360
 361         if(jidx<j_index_end)
 362         {
 363
 364             /* Get j neighbor index, and coordinate index */
 365             jnrlistA         = jjnr[jidx];
 366             jnrlistB         = jjnr[jidx+1];
 367             jnrlistC         = jjnr[jidx+2];
 368             jnrlistD         = jjnr[jidx+3];
 369             /* Sign of each element will be negative for non-real atoms.
 370              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 371              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 372              */
 373             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 374             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 375             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 376             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 377             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 378             j_coord_offsetA  = DIM*jnrA;
 379             j_coord_offsetB  = DIM*jnrB;
 380             j_coord_offsetC  = DIM*jnrC;
 381             j_coord_offsetD  = DIM*jnrD;
 382
 383             /* load j atom coordinates */
 384             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 385                                               x+j_coord_offsetC,x+j_coord_offsetD,
 386                                               &jx0,&jy0,&jz0);
 387
 388             /* Calculate displacement vector */
 389             dx00             = _mm_sub_ps(ix0,jx0);
 390             dy00             = _mm_sub_ps(iy0,jy0);
 391             dz00             = _mm_sub_ps(iz0,jz0);
 392             dx10             = _mm_sub_ps(ix1,jx0);
 393             dy10             = _mm_sub_ps(iy1,jy0);
 394             dz10             = _mm_sub_ps(iz1,jz0);
 395             dx20             = _mm_sub_ps(ix2,jx0);
 396             dy20             = _mm_sub_ps(iy2,jy0);
 397             dz20             = _mm_sub_ps(iz2,jz0);
 398             dx30             = _mm_sub_ps(ix3,jx0);
 399             dy30             = _mm_sub_ps(iy3,jy0);
 400             dz30             = _mm_sub_ps(iz3,jz0);
 401
 402             /* Calculate squared distance and things based on it */
 403             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 404             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 405             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 406             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 407
 408             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 409             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 410             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 411
 412             rinvsq00         = gmx_mm_inv_ps(rsq00);
 413             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 414             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 415             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 416
 417             /* Load parameters for j particles */
 418             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 419                                                               charge+jnrC+0,charge+jnrD+0);
 420             vdwjidx0A        = 2*vdwtype[jnrA+0];
 421             vdwjidx0B        = 2*vdwtype[jnrB+0];
 422             vdwjidx0C        = 2*vdwtype[jnrC+0];
 423             vdwjidx0D        = 2*vdwtype[jnrD+0];
 424
 425             fjx0             = _mm_setzero_ps();
 426             fjy0             = _mm_setzero_ps();
 427             fjz0             = _mm_setzero_ps();
 428
 429             /**************************
 430              * CALCULATE INTERACTIONS *
 431              **************************/
 432
 433             /* Compute parameters for interactions between i and j atoms */
 434             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 435                                          vdwparam+vdwioffset0+vdwjidx0B,
 436                                          vdwparam+vdwioffset0+vdwjidx0C,
 437                                          vdwparam+vdwioffset0+vdwjidx0D,
 438                                          &c6_00,&c12_00);
 439
 440             /* LENNARD-JONES DISPERSION/REPULSION */
 441
 442             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 443             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 444             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 445             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 446             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 447
 448             /* Update potential sum for this i atom from the interaction with this j atom. */
 449             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 450             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 451
 452             fscal            = fvdw;
 453
 454             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 455
 456              /* Update vectorial force */
 457             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 458             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 459             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 460
 461             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 462             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 463             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 464
 465             /**************************
 466              * CALCULATE INTERACTIONS *
 467              **************************/
 468
 469             /* Compute parameters for interactions between i and j atoms */
 470             qq10             = _mm_mul_ps(iq1,jq0);
 471
 472             /* COULOMB ELECTROSTATICS */
 473             velec            = _mm_mul_ps(qq10,rinv10);
 474             felec            = _mm_mul_ps(velec,rinvsq10);
 475
 476             /* Update potential sum for this i atom from the interaction with this j atom. */
 477             velec            = _mm_andnot_ps(dummy_mask,velec);
 478             velecsum         = _mm_add_ps(velecsum,velec);
 479
 480             fscal            = felec;
 481
 482             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 483
 484              /* Update vectorial force */
 485             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 486             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 487             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 488
 489             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 490             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 491             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 492
 493             /**************************
 494              * CALCULATE INTERACTIONS *
 495              **************************/
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             qq20             = _mm_mul_ps(iq2,jq0);
 499
 500             /* COULOMB ELECTROSTATICS */
 501             velec            = _mm_mul_ps(qq20,rinv20);
 502             felec            = _mm_mul_ps(velec,rinvsq20);
 503
 504             /* Update potential sum for this i atom from the interaction with this j atom. */
 505             velec            = _mm_andnot_ps(dummy_mask,velec);
 506             velecsum         = _mm_add_ps(velecsum,velec);
 507
 508             fscal            = felec;
 509
 510             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 511
 512              /* Update vectorial force */
 513             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 514             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 515             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 516
 517             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 518             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 519             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 520
 521             /**************************
 522              * CALCULATE INTERACTIONS *
 523              **************************/
 524
 525             /* Compute parameters for interactions between i and j atoms */
 526             qq30             = _mm_mul_ps(iq3,jq0);
 527
 528             /* COULOMB ELECTROSTATICS */
 529             velec            = _mm_mul_ps(qq30,rinv30);
 530             felec            = _mm_mul_ps(velec,rinvsq30);
 531
 532             /* Update potential sum for this i atom from the interaction with this j atom. */
 533             velec            = _mm_andnot_ps(dummy_mask,velec);
 534             velecsum         = _mm_add_ps(velecsum,velec);
 535
 536             fscal            = felec;
 537
 538             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 539
 540              /* Update vectorial force */
 541             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 542             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 543             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 544
 545             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 546             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 547             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 548
 549             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 550             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 551             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 552             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 553
 554             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 555
 556             /* Inner loop uses 128 flops */
 557         }
 558
 559         /* End of innermost loop */
 560
 561         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 562                                               f+i_coord_offset,fshift+i_shift_offset);
 563
 564         ggid                        = gid[iidx];
 565         /* Update potential energies */
 566         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 567         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 568
 569         /* Increment number of inner iterations */
 570         inneriter                  += j_index_end - j_index_start;
 571
 572         /* Outer loop uses 26 flops */
 573     }
 574
 575     /* Increment number of outer iterations */
 576     outeriter        += nri;
 577
 578     /* Update outer/inner flops */
 579
 580     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*128);
 581 }
 582 /*
 583  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_128_fma_single
 584  * Electrostatics interaction: Coulomb
 585  * VdW interaction:            LennardJones
 586  * Geometry:                   Water4-Particle
 587  * Calculate force/pot:        Force
 588  */
 589 void
 590 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_avx_128_fma_single
 591                     (t_nblist                    * gmx_restrict       nlist,
 592                      rvec                        * gmx_restrict          xx,
 593                      rvec                        * gmx_restrict          ff,
 594                      t_forcerec                  * gmx_restrict          fr,
 595                      t_mdatoms                   * gmx_restrict     mdatoms,
 596                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 597                      t_nrnb                      * gmx_restrict        nrnb)
 598 {
 599     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 600      * just 0 for non-waters.
 601      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 602      * jnr indices corresponding to data put in the four positions in the SIMD register.
 603      */
 604     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 605     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 606     int              jnrA,jnrB,jnrC,jnrD;
 607     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 608     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 609     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 610     real             rcutoff_scalar;
 611     real             *shiftvec,*fshift,*x,*f;
 612     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 613     real             scratch[4*DIM];
 614     __m128           fscal,rcutoff,rcutoff2,jidxall;
 615     int              vdwioffset0;
 616     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 617     int              vdwioffset1;
 618     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 619     int              vdwioffset2;
 620     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 621     int              vdwioffset3;
 622     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 623     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 624     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 625     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 626     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 627     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 628     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 629     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 630     real             *charge;
 631     int              nvdwtype;
 632     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 633     int              *vdwtype;
 634     real             *vdwparam;
 635     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 636     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 637     __m128           dummy_mask,cutoff_mask;
 638     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 639     __m128           one     = _mm_set1_ps(1.0);
 640     __m128           two     = _mm_set1_ps(2.0);
 641     x                = xx[0];
 642     f                = ff[0];
 643
 644     nri              = nlist->nri;
 645     iinr             = nlist->iinr;
 646     jindex           = nlist->jindex;
 647     jjnr             = nlist->jjnr;
 648     shiftidx         = nlist->shift;
 649     gid              = nlist->gid;
 650     shiftvec         = fr->shift_vec[0];
 651     fshift           = fr->fshift[0];
 652     facel            = _mm_set1_ps(fr->epsfac);
 653     charge           = mdatoms->chargeA;
 654     nvdwtype         = fr->ntype;
 655     vdwparam         = fr->nbfp;
 656     vdwtype          = mdatoms->typeA;
 657
 658     /* Setup water-specific parameters */
 659     inr              = nlist->iinr[0];
 660     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 661     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 662     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 663     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 664
 665     /* Avoid stupid compiler warnings */
 666     jnrA = jnrB = jnrC = jnrD = 0;
 667     j_coord_offsetA = 0;
 668     j_coord_offsetB = 0;
 669     j_coord_offsetC = 0;
 670     j_coord_offsetD = 0;
 671
 672     outeriter        = 0;
 673     inneriter        = 0;
 674
 675     for(iidx=0;iidx<4*DIM;iidx++)
 676     {
 677         scratch[iidx] = 0.0;
 678     }
 679
 680     /* Start outer loop over neighborlists */
 681     for(iidx=0; iidx<nri; iidx++)
 682     {
 683         /* Load shift vector for this list */
 684         i_shift_offset   = DIM*shiftidx[iidx];
 685
 686         /* Load limits for loop over neighbors */
 687         j_index_start    = jindex[iidx];
 688         j_index_end      = jindex[iidx+1];
 689
 690         /* Get outer coordinate index */
 691         inr              = iinr[iidx];
 692         i_coord_offset   = DIM*inr;
 693
 694         /* Load i particle coords and add shift vector */
 695         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 696                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 697
 698         fix0             = _mm_setzero_ps();
 699         fiy0             = _mm_setzero_ps();
 700         fiz0             = _mm_setzero_ps();
 701         fix1             = _mm_setzero_ps();
 702         fiy1             = _mm_setzero_ps();
 703         fiz1             = _mm_setzero_ps();
 704         fix2             = _mm_setzero_ps();
 705         fiy2             = _mm_setzero_ps();
 706         fiz2             = _mm_setzero_ps();
 707         fix3             = _mm_setzero_ps();
 708         fiy3             = _mm_setzero_ps();
 709         fiz3             = _mm_setzero_ps();
 710
 711         /* Start inner kernel loop */
 712         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 713         {
 714
 715             /* Get j neighbor index, and coordinate index */
 716             jnrA             = jjnr[jidx];
 717             jnrB             = jjnr[jidx+1];
 718             jnrC             = jjnr[jidx+2];
 719             jnrD             = jjnr[jidx+3];
 720             j_coord_offsetA  = DIM*jnrA;
 721             j_coord_offsetB  = DIM*jnrB;
 722             j_coord_offsetC  = DIM*jnrC;
 723             j_coord_offsetD  = DIM*jnrD;
 724
 725             /* load j atom coordinates */
 726             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 727                                               x+j_coord_offsetC,x+j_coord_offsetD,
 728                                               &jx0,&jy0,&jz0);
 729
 730             /* Calculate displacement vector */
 731             dx00             = _mm_sub_ps(ix0,jx0);
 732             dy00             = _mm_sub_ps(iy0,jy0);
 733             dz00             = _mm_sub_ps(iz0,jz0);
 734             dx10             = _mm_sub_ps(ix1,jx0);
 735             dy10             = _mm_sub_ps(iy1,jy0);
 736             dz10             = _mm_sub_ps(iz1,jz0);
 737             dx20             = _mm_sub_ps(ix2,jx0);
 738             dy20             = _mm_sub_ps(iy2,jy0);
 739             dz20             = _mm_sub_ps(iz2,jz0);
 740             dx30             = _mm_sub_ps(ix3,jx0);
 741             dy30             = _mm_sub_ps(iy3,jy0);
 742             dz30             = _mm_sub_ps(iz3,jz0);
 743
 744             /* Calculate squared distance and things based on it */
 745             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 746             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 747             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 748             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 749
 750             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 751             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 752             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 753
 754             rinvsq00         = gmx_mm_inv_ps(rsq00);
 755             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 756             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 757             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 758
 759             /* Load parameters for j particles */
 760             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 761                                                               charge+jnrC+0,charge+jnrD+0);
 762             vdwjidx0A        = 2*vdwtype[jnrA+0];
 763             vdwjidx0B        = 2*vdwtype[jnrB+0];
 764             vdwjidx0C        = 2*vdwtype[jnrC+0];
 765             vdwjidx0D        = 2*vdwtype[jnrD+0];
 766
 767             fjx0             = _mm_setzero_ps();
 768             fjy0             = _mm_setzero_ps();
 769             fjz0             = _mm_setzero_ps();
 770
 771             /**************************
 772              * CALCULATE INTERACTIONS *
 773              **************************/
 774
 775             /* Compute parameters for interactions between i and j atoms */
 776             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 777                                          vdwparam+vdwioffset0+vdwjidx0B,
 778                                          vdwparam+vdwioffset0+vdwjidx0C,
 779                                          vdwparam+vdwioffset0+vdwjidx0D,
 780                                          &c6_00,&c12_00);
 781
 782             /* LENNARD-JONES DISPERSION/REPULSION */
 783
 784             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 785             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 786
 787             fscal            = fvdw;
 788
 789              /* Update vectorial force */
 790             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 791             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 792             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 793
 794             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 795             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 796             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 797
 798             /**************************
 799              * CALCULATE INTERACTIONS *
 800              **************************/
 801
 802             /* Compute parameters for interactions between i and j atoms */
 803             qq10             = _mm_mul_ps(iq1,jq0);
 804
 805             /* COULOMB ELECTROSTATICS */
 806             velec            = _mm_mul_ps(qq10,rinv10);
 807             felec            = _mm_mul_ps(velec,rinvsq10);
 808
 809             fscal            = felec;
 810
 811              /* Update vectorial force */
 812             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 813             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 814             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 815
 816             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 817             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 818             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 819
 820             /**************************
 821              * CALCULATE INTERACTIONS *
 822              **************************/
 823
 824             /* Compute parameters for interactions between i and j atoms */
 825             qq20             = _mm_mul_ps(iq2,jq0);
 826
 827             /* COULOMB ELECTROSTATICS */
 828             velec            = _mm_mul_ps(qq20,rinv20);
 829             felec            = _mm_mul_ps(velec,rinvsq20);
 830
 831             fscal            = felec;
 832
 833              /* Update vectorial force */
 834             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 835             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 836             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 837
 838             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 839             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 840             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 841
 842             /**************************
 843              * CALCULATE INTERACTIONS *
 844              **************************/
 845
 846             /* Compute parameters for interactions between i and j atoms */
 847             qq30             = _mm_mul_ps(iq3,jq0);
 848
 849             /* COULOMB ELECTROSTATICS */
 850             velec            = _mm_mul_ps(qq30,rinv30);
 851             felec            = _mm_mul_ps(velec,rinvsq30);
 852
 853             fscal            = felec;
 854
 855              /* Update vectorial force */
 856             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 857             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 858             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 859
 860             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 861             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 862             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 863
 864             fjptrA             = f+j_coord_offsetA;
 865             fjptrB             = f+j_coord_offsetB;
 866             fjptrC             = f+j_coord_offsetC;
 867             fjptrD             = f+j_coord_offsetD;
 868
 869             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 870
 871             /* Inner loop uses 120 flops */
 872         }
 873
 874         if(jidx<j_index_end)
 875         {
 876
 877             /* Get j neighbor index, and coordinate index */
 878             jnrlistA         = jjnr[jidx];
 879             jnrlistB         = jjnr[jidx+1];
 880             jnrlistC         = jjnr[jidx+2];
 881             jnrlistD         = jjnr[jidx+3];
 882             /* Sign of each element will be negative for non-real atoms.
 883              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 884              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 885              */
 886             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 887             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 888             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 889             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 890             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 891             j_coord_offsetA  = DIM*jnrA;
 892             j_coord_offsetB  = DIM*jnrB;
 893             j_coord_offsetC  = DIM*jnrC;
 894             j_coord_offsetD  = DIM*jnrD;
 895
 896             /* load j atom coordinates */
 897             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 898                                               x+j_coord_offsetC,x+j_coord_offsetD,
 899                                               &jx0,&jy0,&jz0);
 900
 901             /* Calculate displacement vector */
 902             dx00             = _mm_sub_ps(ix0,jx0);
 903             dy00             = _mm_sub_ps(iy0,jy0);
 904             dz00             = _mm_sub_ps(iz0,jz0);
 905             dx10             = _mm_sub_ps(ix1,jx0);
 906             dy10             = _mm_sub_ps(iy1,jy0);
 907             dz10             = _mm_sub_ps(iz1,jz0);
 908             dx20             = _mm_sub_ps(ix2,jx0);
 909             dy20             = _mm_sub_ps(iy2,jy0);
 910             dz20             = _mm_sub_ps(iz2,jz0);
 911             dx30             = _mm_sub_ps(ix3,jx0);
 912             dy30             = _mm_sub_ps(iy3,jy0);
 913             dz30             = _mm_sub_ps(iz3,jz0);
 914
 915             /* Calculate squared distance and things based on it */
 916             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 917             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 918             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 919             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 920
 921             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 922             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 923             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 924
 925             rinvsq00         = gmx_mm_inv_ps(rsq00);
 926             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 927             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 928             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 929
 930             /* Load parameters for j particles */
 931             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 932                                                               charge+jnrC+0,charge+jnrD+0);
 933             vdwjidx0A        = 2*vdwtype[jnrA+0];
 934             vdwjidx0B        = 2*vdwtype[jnrB+0];
 935             vdwjidx0C        = 2*vdwtype[jnrC+0];
 936             vdwjidx0D        = 2*vdwtype[jnrD+0];
 937
 938             fjx0             = _mm_setzero_ps();
 939             fjy0             = _mm_setzero_ps();
 940             fjz0             = _mm_setzero_ps();
 941
 942             /**************************
 943              * CALCULATE INTERACTIONS *
 944              **************************/
 945
 946             /* Compute parameters for interactions between i and j atoms */
 947             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 948                                          vdwparam+vdwioffset0+vdwjidx0B,
 949                                          vdwparam+vdwioffset0+vdwjidx0C,
 950                                          vdwparam+vdwioffset0+vdwjidx0D,
 951                                          &c6_00,&c12_00);
 952
 953             /* LENNARD-JONES DISPERSION/REPULSION */
 954
 955             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 956             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 957
 958             fscal            = fvdw;
 959
 960             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 961
 962              /* Update vectorial force */
 963             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 964             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 965             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 966
 967             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 968             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 969             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq10             = _mm_mul_ps(iq1,jq0);
 977
 978             /* COULOMB ELECTROSTATICS */
 979             velec            = _mm_mul_ps(qq10,rinv10);
 980             felec            = _mm_mul_ps(velec,rinvsq10);
 981
 982             fscal            = felec;
 983
 984             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 985
 986              /* Update vectorial force */
 987             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 988             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 989             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 990
 991             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 992             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 993             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 994
 995             /**************************
 996              * CALCULATE INTERACTIONS *
 997              **************************/
 998
 999             /* Compute parameters for interactions between i and j atoms */
1000             qq20             = _mm_mul_ps(iq2,jq0);
1001
1002             /* COULOMB ELECTROSTATICS */
1003             velec            = _mm_mul_ps(qq20,rinv20);
1004             felec            = _mm_mul_ps(velec,rinvsq20);
1005
1006             fscal            = felec;
1007
1008             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1009
1010              /* Update vectorial force */
1011             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1012             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1013             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1014
1015             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1016             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1017             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1018
1019             /**************************
1020              * CALCULATE INTERACTIONS *
1021              **************************/
1022
1023             /* Compute parameters for interactions between i and j atoms */
1024             qq30             = _mm_mul_ps(iq3,jq0);
1025
1026             /* COULOMB ELECTROSTATICS */
1027             velec            = _mm_mul_ps(qq30,rinv30);
1028             felec            = _mm_mul_ps(velec,rinvsq30);
1029
1030             fscal            = felec;
1031
1032             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1033
1034              /* Update vectorial force */
1035             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1036             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1037             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1038
1039             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1040             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1041             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1042
1043             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1044             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1045             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1046             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1047
1048             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1049
1050             /* Inner loop uses 120 flops */
1051         }
1052
1053         /* End of innermost loop */
1054
1055         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1056                                               f+i_coord_offset,fshift+i_shift_offset);
1057
1058         /* Increment number of inner iterations */
1059         inneriter                  += j_index_end - j_index_start;
1060
1061         /* Outer loop uses 24 flops */
1062     }
1063
1064     /* Increment number of outer iterations */
1065     outeriter        += nri;
1066
1067     /* Update outer/inner flops */
1068
1069     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*120);
1070 }