src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "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_ElecRF_VdwLJ_GeomW4P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwLJ_GeomW4P1_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              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128           dummy_mask,cutoff_mask;
 108     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 109     __m128           one     = _mm_set1_ps(1.0);
 110     __m128           two     = _mm_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_ps(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     krf              = _mm_set1_ps(fr->ic->k_rf);
 125     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 126     crf              = _mm_set1_ps(fr->ic->c_rf);
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[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     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 136     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 170
 171         fix0             = _mm_setzero_ps();
 172         fiy0             = _mm_setzero_ps();
 173         fiz0             = _mm_setzero_ps();
 174         fix1             = _mm_setzero_ps();
 175         fiy1             = _mm_setzero_ps();
 176         fiz1             = _mm_setzero_ps();
 177         fix2             = _mm_setzero_ps();
 178         fiy2             = _mm_setzero_ps();
 179         fiz2             = _mm_setzero_ps();
 180         fix3             = _mm_setzero_ps();
 181         fiy3             = _mm_setzero_ps();
 182         fiz3             = _mm_setzero_ps();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_ps();
 186         vvdwsum          = _mm_setzero_ps();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                               x+j_coord_offsetC,x+j_coord_offsetD,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_ps(ix0,jx0);
 209             dy00             = _mm_sub_ps(iy0,jy0);
 210             dz00             = _mm_sub_ps(iz0,jz0);
 211             dx10             = _mm_sub_ps(ix1,jx0);
 212             dy10             = _mm_sub_ps(iy1,jy0);
 213             dz10             = _mm_sub_ps(iz1,jz0);
 214             dx20             = _mm_sub_ps(ix2,jx0);
 215             dy20             = _mm_sub_ps(iy2,jy0);
 216             dz20             = _mm_sub_ps(iz2,jz0);
 217             dx30             = _mm_sub_ps(ix3,jx0);
 218             dy30             = _mm_sub_ps(iy3,jy0);
 219             dz30             = _mm_sub_ps(iz3,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 223             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 224             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 225             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 226
 227             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 228             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 229             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 230
 231             rinvsq00         = gmx_mm_inv_ps(rsq00);
 232             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 233             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 234             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 235
 236             /* Load parameters for j particles */
 237             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 238                                                               charge+jnrC+0,charge+jnrD+0);
 239             vdwjidx0A        = 2*vdwtype[jnrA+0];
 240             vdwjidx0B        = 2*vdwtype[jnrB+0];
 241             vdwjidx0C        = 2*vdwtype[jnrC+0];
 242             vdwjidx0D        = 2*vdwtype[jnrD+0];
 243
 244             fjx0             = _mm_setzero_ps();
 245             fjy0             = _mm_setzero_ps();
 246             fjz0             = _mm_setzero_ps();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 254                                          vdwparam+vdwioffset0+vdwjidx0B,
 255                                          vdwparam+vdwioffset0+vdwjidx0C,
 256                                          vdwparam+vdwioffset0+vdwjidx0D,
 257                                          &c6_00,&c12_00);
 258
 259             /* LENNARD-JONES DISPERSION/REPULSION */
 260
 261             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 262             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 263             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 264             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 265             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 269
 270             fscal            = fvdw;
 271
 272              /* Update vectorial force */
 273             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 274             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 275             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 276
 277             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 278             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 279             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq10             = _mm_mul_ps(iq1,jq0);
 287
 288             /* REACTION-FIELD ELECTROSTATICS */
 289             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 290             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             velecsum         = _mm_add_ps(velecsum,velec);
 294
 295             fscal            = felec;
 296
 297              /* Update vectorial force */
 298             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 299             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 300             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 301
 302             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 303             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 304             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq20             = _mm_mul_ps(iq2,jq0);
 312
 313             /* REACTION-FIELD ELECTROSTATICS */
 314             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 315             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velecsum         = _mm_add_ps(velecsum,velec);
 319
 320             fscal            = felec;
 321
 322              /* Update vectorial force */
 323             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 324             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 325             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 326
 327             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 328             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 329             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             /* Compute parameters for interactions between i and j atoms */
 336             qq30             = _mm_mul_ps(iq3,jq0);
 337
 338             /* REACTION-FIELD ELECTROSTATICS */
 339             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 340             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velecsum         = _mm_add_ps(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347              /* Update vectorial force */
 348             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 349             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 350             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 351
 352             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 353             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 354             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 355
 356             fjptrA             = f+j_coord_offsetA;
 357             fjptrB             = f+j_coord_offsetB;
 358             fjptrC             = f+j_coord_offsetC;
 359             fjptrD             = f+j_coord_offsetD;
 360
 361             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 362
 363             /* Inner loop uses 140 flops */
 364         }
 365
 366         if(jidx<j_index_end)
 367         {
 368
 369             /* Get j neighbor index, and coordinate index */
 370             jnrlistA         = jjnr[jidx];
 371             jnrlistB         = jjnr[jidx+1];
 372             jnrlistC         = jjnr[jidx+2];
 373             jnrlistD         = jjnr[jidx+3];
 374             /* Sign of each element will be negative for non-real atoms.
 375              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 376              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 377              */
 378             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 379             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 380             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 381             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 382             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 383             j_coord_offsetA  = DIM*jnrA;
 384             j_coord_offsetB  = DIM*jnrB;
 385             j_coord_offsetC  = DIM*jnrC;
 386             j_coord_offsetD  = DIM*jnrD;
 387
 388             /* load j atom coordinates */
 389             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 390                                               x+j_coord_offsetC,x+j_coord_offsetD,
 391                                               &jx0,&jy0,&jz0);
 392
 393             /* Calculate displacement vector */
 394             dx00             = _mm_sub_ps(ix0,jx0);
 395             dy00             = _mm_sub_ps(iy0,jy0);
 396             dz00             = _mm_sub_ps(iz0,jz0);
 397             dx10             = _mm_sub_ps(ix1,jx0);
 398             dy10             = _mm_sub_ps(iy1,jy0);
 399             dz10             = _mm_sub_ps(iz1,jz0);
 400             dx20             = _mm_sub_ps(ix2,jx0);
 401             dy20             = _mm_sub_ps(iy2,jy0);
 402             dz20             = _mm_sub_ps(iz2,jz0);
 403             dx30             = _mm_sub_ps(ix3,jx0);
 404             dy30             = _mm_sub_ps(iy3,jy0);
 405             dz30             = _mm_sub_ps(iz3,jz0);
 406
 407             /* Calculate squared distance and things based on it */
 408             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 409             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 410             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 411             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 412
 413             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 414             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 415             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 416
 417             rinvsq00         = gmx_mm_inv_ps(rsq00);
 418             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 419             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 420             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 421
 422             /* Load parameters for j particles */
 423             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 424                                                               charge+jnrC+0,charge+jnrD+0);
 425             vdwjidx0A        = 2*vdwtype[jnrA+0];
 426             vdwjidx0B        = 2*vdwtype[jnrB+0];
 427             vdwjidx0C        = 2*vdwtype[jnrC+0];
 428             vdwjidx0D        = 2*vdwtype[jnrD+0];
 429
 430             fjx0             = _mm_setzero_ps();
 431             fjy0             = _mm_setzero_ps();
 432             fjz0             = _mm_setzero_ps();
 433
 434             /**************************
 435              * CALCULATE INTERACTIONS *
 436              **************************/
 437
 438             /* Compute parameters for interactions between i and j atoms */
 439             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 440                                          vdwparam+vdwioffset0+vdwjidx0B,
 441                                          vdwparam+vdwioffset0+vdwjidx0C,
 442                                          vdwparam+vdwioffset0+vdwjidx0D,
 443                                          &c6_00,&c12_00);
 444
 445             /* LENNARD-JONES DISPERSION/REPULSION */
 446
 447             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 448             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 449             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 450             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 451             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 452
 453             /* Update potential sum for this i atom from the interaction with this j atom. */
 454             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 455             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 456
 457             fscal            = fvdw;
 458
 459             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 460
 461              /* Update vectorial force */
 462             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 463             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 464             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 465
 466             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 467             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 468             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 469
 470             /**************************
 471              * CALCULATE INTERACTIONS *
 472              **************************/
 473
 474             /* Compute parameters for interactions between i and j atoms */
 475             qq10             = _mm_mul_ps(iq1,jq0);
 476
 477             /* REACTION-FIELD ELECTROSTATICS */
 478             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 479             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 480
 481             /* Update potential sum for this i atom from the interaction with this j atom. */
 482             velec            = _mm_andnot_ps(dummy_mask,velec);
 483             velecsum         = _mm_add_ps(velecsum,velec);
 484
 485             fscal            = felec;
 486
 487             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 488
 489              /* Update vectorial force */
 490             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 491             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 492             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 493
 494             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 495             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 496             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             qq20             = _mm_mul_ps(iq2,jq0);
 504
 505             /* REACTION-FIELD ELECTROSTATICS */
 506             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 507             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 508
 509             /* Update potential sum for this i atom from the interaction with this j atom. */
 510             velec            = _mm_andnot_ps(dummy_mask,velec);
 511             velecsum         = _mm_add_ps(velecsum,velec);
 512
 513             fscal            = felec;
 514
 515             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 516
 517              /* Update vectorial force */
 518             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 519             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 520             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 521
 522             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 523             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 524             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq30             = _mm_mul_ps(iq3,jq0);
 532
 533             /* REACTION-FIELD ELECTROSTATICS */
 534             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 535             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velec            = _mm_andnot_ps(dummy_mask,velec);
 539             velecsum         = _mm_add_ps(velecsum,velec);
 540
 541             fscal            = felec;
 542
 543             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 544
 545              /* Update vectorial force */
 546             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 547             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 548             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 549
 550             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 551             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 552             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 553
 554             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 555             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 556             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 557             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 558
 559             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 560
 561             /* Inner loop uses 140 flops */
 562         }
 563
 564         /* End of innermost loop */
 565
 566         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 567                                               f+i_coord_offset,fshift+i_shift_offset);
 568
 569         ggid                        = gid[iidx];
 570         /* Update potential energies */
 571         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 572         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 573
 574         /* Increment number of inner iterations */
 575         inneriter                  += j_index_end - j_index_start;
 576
 577         /* Outer loop uses 26 flops */
 578     }
 579
 580     /* Increment number of outer iterations */
 581     outeriter        += nri;
 582
 583     /* Update outer/inner flops */
 584
 585     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*140);
 586 }
 587 /*
 588  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_128_fma_single
 589  * Electrostatics interaction: ReactionField
 590  * VdW interaction:            LennardJones
 591  * Geometry:                   Water4-Particle
 592  * Calculate force/pot:        Force
 593  */
 594 void
 595 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_128_fma_single
 596                     (t_nblist                    * gmx_restrict       nlist,
 597                      rvec                        * gmx_restrict          xx,
 598                      rvec                        * gmx_restrict          ff,
 599                      t_forcerec                  * gmx_restrict          fr,
 600                      t_mdatoms                   * gmx_restrict     mdatoms,
 601                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 602                      t_nrnb                      * gmx_restrict        nrnb)
 603 {
 604     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 605      * just 0 for non-waters.
 606      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 607      * jnr indices corresponding to data put in the four positions in the SIMD register.
 608      */
 609     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 610     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 611     int              jnrA,jnrB,jnrC,jnrD;
 612     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 613     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 614     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 615     real             rcutoff_scalar;
 616     real             *shiftvec,*fshift,*x,*f;
 617     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 618     real             scratch[4*DIM];
 619     __m128           fscal,rcutoff,rcutoff2,jidxall;
 620     int              vdwioffset0;
 621     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 622     int              vdwioffset1;
 623     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 624     int              vdwioffset2;
 625     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 626     int              vdwioffset3;
 627     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 628     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 629     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 630     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 631     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 632     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 633     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 634     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 635     real             *charge;
 636     int              nvdwtype;
 637     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 638     int              *vdwtype;
 639     real             *vdwparam;
 640     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 641     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 642     __m128           dummy_mask,cutoff_mask;
 643     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 644     __m128           one     = _mm_set1_ps(1.0);
 645     __m128           two     = _mm_set1_ps(2.0);
 646     x                = xx[0];
 647     f                = ff[0];
 648
 649     nri              = nlist->nri;
 650     iinr             = nlist->iinr;
 651     jindex           = nlist->jindex;
 652     jjnr             = nlist->jjnr;
 653     shiftidx         = nlist->shift;
 654     gid              = nlist->gid;
 655     shiftvec         = fr->shift_vec[0];
 656     fshift           = fr->fshift[0];
 657     facel            = _mm_set1_ps(fr->epsfac);
 658     charge           = mdatoms->chargeA;
 659     krf              = _mm_set1_ps(fr->ic->k_rf);
 660     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 661     crf              = _mm_set1_ps(fr->ic->c_rf);
 662     nvdwtype         = fr->ntype;
 663     vdwparam         = fr->nbfp;
 664     vdwtype          = mdatoms->typeA;
 665
 666     /* Setup water-specific parameters */
 667     inr              = nlist->iinr[0];
 668     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 669     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 670     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 671     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 672
 673     /* Avoid stupid compiler warnings */
 674     jnrA = jnrB = jnrC = jnrD = 0;
 675     j_coord_offsetA = 0;
 676     j_coord_offsetB = 0;
 677     j_coord_offsetC = 0;
 678     j_coord_offsetD = 0;
 679
 680     outeriter        = 0;
 681     inneriter        = 0;
 682
 683     for(iidx=0;iidx<4*DIM;iidx++)
 684     {
 685         scratch[iidx] = 0.0;
 686     }
 687
 688     /* Start outer loop over neighborlists */
 689     for(iidx=0; iidx<nri; iidx++)
 690     {
 691         /* Load shift vector for this list */
 692         i_shift_offset   = DIM*shiftidx[iidx];
 693
 694         /* Load limits for loop over neighbors */
 695         j_index_start    = jindex[iidx];
 696         j_index_end      = jindex[iidx+1];
 697
 698         /* Get outer coordinate index */
 699         inr              = iinr[iidx];
 700         i_coord_offset   = DIM*inr;
 701
 702         /* Load i particle coords and add shift vector */
 703         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 704                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 705
 706         fix0             = _mm_setzero_ps();
 707         fiy0             = _mm_setzero_ps();
 708         fiz0             = _mm_setzero_ps();
 709         fix1             = _mm_setzero_ps();
 710         fiy1             = _mm_setzero_ps();
 711         fiz1             = _mm_setzero_ps();
 712         fix2             = _mm_setzero_ps();
 713         fiy2             = _mm_setzero_ps();
 714         fiz2             = _mm_setzero_ps();
 715         fix3             = _mm_setzero_ps();
 716         fiy3             = _mm_setzero_ps();
 717         fiz3             = _mm_setzero_ps();
 718
 719         /* Start inner kernel loop */
 720         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 721         {
 722
 723             /* Get j neighbor index, and coordinate index */
 724             jnrA             = jjnr[jidx];
 725             jnrB             = jjnr[jidx+1];
 726             jnrC             = jjnr[jidx+2];
 727             jnrD             = jjnr[jidx+3];
 728             j_coord_offsetA  = DIM*jnrA;
 729             j_coord_offsetB  = DIM*jnrB;
 730             j_coord_offsetC  = DIM*jnrC;
 731             j_coord_offsetD  = DIM*jnrD;
 732
 733             /* load j atom coordinates */
 734             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 735                                               x+j_coord_offsetC,x+j_coord_offsetD,
 736                                               &jx0,&jy0,&jz0);
 737
 738             /* Calculate displacement vector */
 739             dx00             = _mm_sub_ps(ix0,jx0);
 740             dy00             = _mm_sub_ps(iy0,jy0);
 741             dz00             = _mm_sub_ps(iz0,jz0);
 742             dx10             = _mm_sub_ps(ix1,jx0);
 743             dy10             = _mm_sub_ps(iy1,jy0);
 744             dz10             = _mm_sub_ps(iz1,jz0);
 745             dx20             = _mm_sub_ps(ix2,jx0);
 746             dy20             = _mm_sub_ps(iy2,jy0);
 747             dz20             = _mm_sub_ps(iz2,jz0);
 748             dx30             = _mm_sub_ps(ix3,jx0);
 749             dy30             = _mm_sub_ps(iy3,jy0);
 750             dz30             = _mm_sub_ps(iz3,jz0);
 751
 752             /* Calculate squared distance and things based on it */
 753             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 754             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 755             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 756             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 757
 758             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 759             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 760             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 761
 762             rinvsq00         = gmx_mm_inv_ps(rsq00);
 763             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 764             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 765             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 766
 767             /* Load parameters for j particles */
 768             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 769                                                               charge+jnrC+0,charge+jnrD+0);
 770             vdwjidx0A        = 2*vdwtype[jnrA+0];
 771             vdwjidx0B        = 2*vdwtype[jnrB+0];
 772             vdwjidx0C        = 2*vdwtype[jnrC+0];
 773             vdwjidx0D        = 2*vdwtype[jnrD+0];
 774
 775             fjx0             = _mm_setzero_ps();
 776             fjy0             = _mm_setzero_ps();
 777             fjz0             = _mm_setzero_ps();
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             /* Compute parameters for interactions between i and j atoms */
 784             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 785                                          vdwparam+vdwioffset0+vdwjidx0B,
 786                                          vdwparam+vdwioffset0+vdwjidx0C,
 787                                          vdwparam+vdwioffset0+vdwjidx0D,
 788                                          &c6_00,&c12_00);
 789
 790             /* LENNARD-JONES DISPERSION/REPULSION */
 791
 792             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 793             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 794
 795             fscal            = fvdw;
 796
 797              /* Update vectorial force */
 798             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 799             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 800             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 801
 802             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 803             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 804             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 805
 806             /**************************
 807              * CALCULATE INTERACTIONS *
 808              **************************/
 809
 810             /* Compute parameters for interactions between i and j atoms */
 811             qq10             = _mm_mul_ps(iq1,jq0);
 812
 813             /* REACTION-FIELD ELECTROSTATICS */
 814             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 815
 816             fscal            = felec;
 817
 818              /* Update vectorial force */
 819             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 820             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 821             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 822
 823             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 824             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 825             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 826
 827             /**************************
 828              * CALCULATE INTERACTIONS *
 829              **************************/
 830
 831             /* Compute parameters for interactions between i and j atoms */
 832             qq20             = _mm_mul_ps(iq2,jq0);
 833
 834             /* REACTION-FIELD ELECTROSTATICS */
 835             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 836
 837             fscal            = felec;
 838
 839              /* Update vectorial force */
 840             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 841             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 842             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 843
 844             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 845             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 846             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 847
 848             /**************************
 849              * CALCULATE INTERACTIONS *
 850              **************************/
 851
 852             /* Compute parameters for interactions between i and j atoms */
 853             qq30             = _mm_mul_ps(iq3,jq0);
 854
 855             /* REACTION-FIELD ELECTROSTATICS */
 856             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 857
 858             fscal            = felec;
 859
 860              /* Update vectorial force */
 861             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 862             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 863             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 864
 865             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 866             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 867             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 868
 869             fjptrA             = f+j_coord_offsetA;
 870             fjptrB             = f+j_coord_offsetB;
 871             fjptrC             = f+j_coord_offsetC;
 872             fjptrD             = f+j_coord_offsetD;
 873
 874             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 875
 876             /* Inner loop uses 120 flops */
 877         }
 878
 879         if(jidx<j_index_end)
 880         {
 881
 882             /* Get j neighbor index, and coordinate index */
 883             jnrlistA         = jjnr[jidx];
 884             jnrlistB         = jjnr[jidx+1];
 885             jnrlistC         = jjnr[jidx+2];
 886             jnrlistD         = jjnr[jidx+3];
 887             /* Sign of each element will be negative for non-real atoms.
 888              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 889              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 890              */
 891             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 892             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 893             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 894             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 895             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 896             j_coord_offsetA  = DIM*jnrA;
 897             j_coord_offsetB  = DIM*jnrB;
 898             j_coord_offsetC  = DIM*jnrC;
 899             j_coord_offsetD  = DIM*jnrD;
 900
 901             /* load j atom coordinates */
 902             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 903                                               x+j_coord_offsetC,x+j_coord_offsetD,
 904                                               &jx0,&jy0,&jz0);
 905
 906             /* Calculate displacement vector */
 907             dx00             = _mm_sub_ps(ix0,jx0);
 908             dy00             = _mm_sub_ps(iy0,jy0);
 909             dz00             = _mm_sub_ps(iz0,jz0);
 910             dx10             = _mm_sub_ps(ix1,jx0);
 911             dy10             = _mm_sub_ps(iy1,jy0);
 912             dz10             = _mm_sub_ps(iz1,jz0);
 913             dx20             = _mm_sub_ps(ix2,jx0);
 914             dy20             = _mm_sub_ps(iy2,jy0);
 915             dz20             = _mm_sub_ps(iz2,jz0);
 916             dx30             = _mm_sub_ps(ix3,jx0);
 917             dy30             = _mm_sub_ps(iy3,jy0);
 918             dz30             = _mm_sub_ps(iz3,jz0);
 919
 920             /* Calculate squared distance and things based on it */
 921             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 922             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 923             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 924             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 925
 926             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 927             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 928             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 929
 930             rinvsq00         = gmx_mm_inv_ps(rsq00);
 931             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 932             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 933             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 934
 935             /* Load parameters for j particles */
 936             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 937                                                               charge+jnrC+0,charge+jnrD+0);
 938             vdwjidx0A        = 2*vdwtype[jnrA+0];
 939             vdwjidx0B        = 2*vdwtype[jnrB+0];
 940             vdwjidx0C        = 2*vdwtype[jnrC+0];
 941             vdwjidx0D        = 2*vdwtype[jnrD+0];
 942
 943             fjx0             = _mm_setzero_ps();
 944             fjy0             = _mm_setzero_ps();
 945             fjz0             = _mm_setzero_ps();
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             /* Compute parameters for interactions between i and j atoms */
 952             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 953                                          vdwparam+vdwioffset0+vdwjidx0B,
 954                                          vdwparam+vdwioffset0+vdwjidx0C,
 955                                          vdwparam+vdwioffset0+vdwjidx0D,
 956                                          &c6_00,&c12_00);
 957
 958             /* LENNARD-JONES DISPERSION/REPULSION */
 959
 960             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 961             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 962
 963             fscal            = fvdw;
 964
 965             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 966
 967              /* Update vectorial force */
 968             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 969             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 970             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 971
 972             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 973             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 974             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 975
 976             /**************************
 977              * CALCULATE INTERACTIONS *
 978              **************************/
 979
 980             /* Compute parameters for interactions between i and j atoms */
 981             qq10             = _mm_mul_ps(iq1,jq0);
 982
 983             /* REACTION-FIELD ELECTROSTATICS */
 984             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 985
 986             fscal            = felec;
 987
 988             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 989
 990              /* Update vectorial force */
 991             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 992             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 993             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 994
 995             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 996             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 997             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             /* Compute parameters for interactions between i and j atoms */
1004             qq20             = _mm_mul_ps(iq2,jq0);
1005
1006             /* REACTION-FIELD ELECTROSTATICS */
1007             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1008
1009             fscal            = felec;
1010
1011             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1012
1013              /* Update vectorial force */
1014             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1015             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1016             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1017
1018             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1019             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1020             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1021
1022             /**************************
1023              * CALCULATE INTERACTIONS *
1024              **************************/
1025
1026             /* Compute parameters for interactions between i and j atoms */
1027             qq30             = _mm_mul_ps(iq3,jq0);
1028
1029             /* REACTION-FIELD ELECTROSTATICS */
1030             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1031
1032             fscal            = felec;
1033
1034             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1035
1036              /* Update vectorial force */
1037             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1038             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1039             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1040
1041             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1042             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1043             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1044
1045             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1046             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1047             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1048             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1049
1050             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1051
1052             /* Inner loop uses 120 flops */
1053         }
1054
1055         /* End of innermost loop */
1056
1057         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1058                                               f+i_coord_offset,fshift+i_shift_offset);
1059
1060         /* Increment number of inner iterations */
1061         inneriter                  += j_index_end - j_index_start;
1062
1063         /* Outer loop uses 24 flops */
1064     }
1065
1066     /* Increment number of outer iterations */
1067     outeriter        += nri;
1068
1069     /* Update outer/inner flops */
1070
1071     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*120);
1072 }