src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSw_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_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSw_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           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 106     real             rswitch_scalar,d_scalar;
 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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 139     rcutoff_scalar   = fr->rcoulomb;
 140     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 141     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 142
 143     rswitch_scalar   = fr->rvdw_switch;
 144     rswitch          = _mm_set1_ps(rswitch_scalar);
 145     /* Setup switch parameters */
 146     d_scalar         = rcutoff_scalar-rswitch_scalar;
 147     d                = _mm_set1_ps(d_scalar);
 148     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 149     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 150     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 151     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 152     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 153     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 154
 155     /* Avoid stupid compiler warnings */
 156     jnrA = jnrB = jnrC = jnrD = 0;
 157     j_coord_offsetA = 0;
 158     j_coord_offsetB = 0;
 159     j_coord_offsetC = 0;
 160     j_coord_offsetD = 0;
 161
 162     outeriter        = 0;
 163     inneriter        = 0;
 164
 165     for(iidx=0;iidx<4*DIM;iidx++)
 166     {
 167         scratch[iidx] = 0.0;
 168     }
 169
 170     /* Start outer loop over neighborlists */
 171     for(iidx=0; iidx<nri; iidx++)
 172     {
 173         /* Load shift vector for this list */
 174         i_shift_offset   = DIM*shiftidx[iidx];
 175
 176         /* Load limits for loop over neighbors */
 177         j_index_start    = jindex[iidx];
 178         j_index_end      = jindex[iidx+1];
 179
 180         /* Get outer coordinate index */
 181         inr              = iinr[iidx];
 182         i_coord_offset   = DIM*inr;
 183
 184         /* Load i particle coords and add shift vector */
 185         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 186                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 187
 188         fix0             = _mm_setzero_ps();
 189         fiy0             = _mm_setzero_ps();
 190         fiz0             = _mm_setzero_ps();
 191         fix1             = _mm_setzero_ps();
 192         fiy1             = _mm_setzero_ps();
 193         fiz1             = _mm_setzero_ps();
 194         fix2             = _mm_setzero_ps();
 195         fiy2             = _mm_setzero_ps();
 196         fiz2             = _mm_setzero_ps();
 197         fix3             = _mm_setzero_ps();
 198         fiy3             = _mm_setzero_ps();
 199         fiz3             = _mm_setzero_ps();
 200
 201         /* Reset potential sums */
 202         velecsum         = _mm_setzero_ps();
 203         vvdwsum          = _mm_setzero_ps();
 204
 205         /* Start inner kernel loop */
 206         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 207         {
 208
 209             /* Get j neighbor index, and coordinate index */
 210             jnrA             = jjnr[jidx];
 211             jnrB             = jjnr[jidx+1];
 212             jnrC             = jjnr[jidx+2];
 213             jnrD             = jjnr[jidx+3];
 214             j_coord_offsetA  = DIM*jnrA;
 215             j_coord_offsetB  = DIM*jnrB;
 216             j_coord_offsetC  = DIM*jnrC;
 217             j_coord_offsetD  = DIM*jnrD;
 218
 219             /* load j atom coordinates */
 220             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 221                                               x+j_coord_offsetC,x+j_coord_offsetD,
 222                                               &jx0,&jy0,&jz0);
 223
 224             /* Calculate displacement vector */
 225             dx00             = _mm_sub_ps(ix0,jx0);
 226             dy00             = _mm_sub_ps(iy0,jy0);
 227             dz00             = _mm_sub_ps(iz0,jz0);
 228             dx10             = _mm_sub_ps(ix1,jx0);
 229             dy10             = _mm_sub_ps(iy1,jy0);
 230             dz10             = _mm_sub_ps(iz1,jz0);
 231             dx20             = _mm_sub_ps(ix2,jx0);
 232             dy20             = _mm_sub_ps(iy2,jy0);
 233             dz20             = _mm_sub_ps(iz2,jz0);
 234             dx30             = _mm_sub_ps(ix3,jx0);
 235             dy30             = _mm_sub_ps(iy3,jy0);
 236             dz30             = _mm_sub_ps(iz3,jz0);
 237
 238             /* Calculate squared distance and things based on it */
 239             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 240             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 241             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 242             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 243
 244             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 245             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 246             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 247             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 248
 249             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 250             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 251             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 252             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 253
 254             /* Load parameters for j particles */
 255             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 256                                                               charge+jnrC+0,charge+jnrD+0);
 257             vdwjidx0A        = 2*vdwtype[jnrA+0];
 258             vdwjidx0B        = 2*vdwtype[jnrB+0];
 259             vdwjidx0C        = 2*vdwtype[jnrC+0];
 260             vdwjidx0D        = 2*vdwtype[jnrD+0];
 261
 262             fjx0             = _mm_setzero_ps();
 263             fjy0             = _mm_setzero_ps();
 264             fjz0             = _mm_setzero_ps();
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             if (gmx_mm_any_lt(rsq00,rcutoff2))
 271             {
 272
 273             r00              = _mm_mul_ps(rsq00,rinv00);
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 277                                          vdwparam+vdwioffset0+vdwjidx0B,
 278                                          vdwparam+vdwioffset0+vdwjidx0C,
 279                                          vdwparam+vdwioffset0+vdwjidx0D,
 280                                          &c6_00,&c12_00);
 281
 282             /* LENNARD-JONES DISPERSION/REPULSION */
 283
 284             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 285             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 286             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 287             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 288             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 289
 290             d                = _mm_sub_ps(r00,rswitch);
 291             d                = _mm_max_ps(d,_mm_setzero_ps());
 292             d2               = _mm_mul_ps(d,d);
 293             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 294
 295             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 296
 297             /* Evaluate switch function */
 298             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 299             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 300             vvdw             = _mm_mul_ps(vvdw,sw);
 301             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 305             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 306
 307             fscal            = fvdw;
 308
 309             fscal            = _mm_and_ps(fscal,cutoff_mask);
 310
 311              /* Update vectorial force */
 312             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 313             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 314             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 315
 316             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 317             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 318             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 319
 320             }
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             if (gmx_mm_any_lt(rsq10,rcutoff2))
 327             {
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq10             = _mm_mul_ps(iq1,jq0);
 331
 332             /* REACTION-FIELD ELECTROSTATICS */
 333             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 334             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 335
 336             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm_and_ps(velec,cutoff_mask);
 340             velecsum         = _mm_add_ps(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             fscal            = _mm_and_ps(fscal,cutoff_mask);
 345
 346              /* Update vectorial force */
 347             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 348             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 349             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 350
 351             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 352             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 353             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 354
 355             }
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             if (gmx_mm_any_lt(rsq20,rcutoff2))
 362             {
 363
 364             /* Compute parameters for interactions between i and j atoms */
 365             qq20             = _mm_mul_ps(iq2,jq0);
 366
 367             /* REACTION-FIELD ELECTROSTATICS */
 368             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 369             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 370
 371             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velec            = _mm_and_ps(velec,cutoff_mask);
 375             velecsum         = _mm_add_ps(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             fscal            = _mm_and_ps(fscal,cutoff_mask);
 380
 381              /* Update vectorial force */
 382             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 383             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 384             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 385
 386             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 387             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 388             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 389
 390             }
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             if (gmx_mm_any_lt(rsq30,rcutoff2))
 397             {
 398
 399             /* Compute parameters for interactions between i and j atoms */
 400             qq30             = _mm_mul_ps(iq3,jq0);
 401
 402             /* REACTION-FIELD ELECTROSTATICS */
 403             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 404             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 405
 406             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 407
 408             /* Update potential sum for this i atom from the interaction with this j atom. */
 409             velec            = _mm_and_ps(velec,cutoff_mask);
 410             velecsum         = _mm_add_ps(velecsum,velec);
 411
 412             fscal            = felec;
 413
 414             fscal            = _mm_and_ps(fscal,cutoff_mask);
 415
 416              /* Update vectorial force */
 417             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 418             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 419             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 420
 421             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 422             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 423             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 424
 425             }
 426
 427             fjptrA             = f+j_coord_offsetA;
 428             fjptrB             = f+j_coord_offsetB;
 429             fjptrC             = f+j_coord_offsetC;
 430             fjptrD             = f+j_coord_offsetD;
 431
 432             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 433
 434             /* Inner loop uses 179 flops */
 435         }
 436
 437         if(jidx<j_index_end)
 438         {
 439
 440             /* Get j neighbor index, and coordinate index */
 441             jnrlistA         = jjnr[jidx];
 442             jnrlistB         = jjnr[jidx+1];
 443             jnrlistC         = jjnr[jidx+2];
 444             jnrlistD         = jjnr[jidx+3];
 445             /* Sign of each element will be negative for non-real atoms.
 446              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 447              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 448              */
 449             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 450             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 451             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 452             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 453             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 454             j_coord_offsetA  = DIM*jnrA;
 455             j_coord_offsetB  = DIM*jnrB;
 456             j_coord_offsetC  = DIM*jnrC;
 457             j_coord_offsetD  = DIM*jnrD;
 458
 459             /* load j atom coordinates */
 460             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 461                                               x+j_coord_offsetC,x+j_coord_offsetD,
 462                                               &jx0,&jy0,&jz0);
 463
 464             /* Calculate displacement vector */
 465             dx00             = _mm_sub_ps(ix0,jx0);
 466             dy00             = _mm_sub_ps(iy0,jy0);
 467             dz00             = _mm_sub_ps(iz0,jz0);
 468             dx10             = _mm_sub_ps(ix1,jx0);
 469             dy10             = _mm_sub_ps(iy1,jy0);
 470             dz10             = _mm_sub_ps(iz1,jz0);
 471             dx20             = _mm_sub_ps(ix2,jx0);
 472             dy20             = _mm_sub_ps(iy2,jy0);
 473             dz20             = _mm_sub_ps(iz2,jz0);
 474             dx30             = _mm_sub_ps(ix3,jx0);
 475             dy30             = _mm_sub_ps(iy3,jy0);
 476             dz30             = _mm_sub_ps(iz3,jz0);
 477
 478             /* Calculate squared distance and things based on it */
 479             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 480             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 481             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 482             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 483
 484             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 485             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 486             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 487             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 488
 489             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 490             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 491             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 492             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 493
 494             /* Load parameters for j particles */
 495             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 496                                                               charge+jnrC+0,charge+jnrD+0);
 497             vdwjidx0A        = 2*vdwtype[jnrA+0];
 498             vdwjidx0B        = 2*vdwtype[jnrB+0];
 499             vdwjidx0C        = 2*vdwtype[jnrC+0];
 500             vdwjidx0D        = 2*vdwtype[jnrD+0];
 501
 502             fjx0             = _mm_setzero_ps();
 503             fjy0             = _mm_setzero_ps();
 504             fjz0             = _mm_setzero_ps();
 505
 506             /**************************
 507              * CALCULATE INTERACTIONS *
 508              **************************/
 509
 510             if (gmx_mm_any_lt(rsq00,rcutoff2))
 511             {
 512
 513             r00              = _mm_mul_ps(rsq00,rinv00);
 514             r00              = _mm_andnot_ps(dummy_mask,r00);
 515
 516             /* Compute parameters for interactions between i and j atoms */
 517             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 518                                          vdwparam+vdwioffset0+vdwjidx0B,
 519                                          vdwparam+vdwioffset0+vdwjidx0C,
 520                                          vdwparam+vdwioffset0+vdwjidx0D,
 521                                          &c6_00,&c12_00);
 522
 523             /* LENNARD-JONES DISPERSION/REPULSION */
 524
 525             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 526             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 527             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 528             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 529             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 530
 531             d                = _mm_sub_ps(r00,rswitch);
 532             d                = _mm_max_ps(d,_mm_setzero_ps());
 533             d2               = _mm_mul_ps(d,d);
 534             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 535
 536             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 537
 538             /* Evaluate switch function */
 539             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 540             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 541             vvdw             = _mm_mul_ps(vvdw,sw);
 542             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 546             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 547             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 548
 549             fscal            = fvdw;
 550
 551             fscal            = _mm_and_ps(fscal,cutoff_mask);
 552
 553             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 554
 555              /* Update vectorial force */
 556             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 557             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 558             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 559
 560             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 561             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 562             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 563
 564             }
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             if (gmx_mm_any_lt(rsq10,rcutoff2))
 571             {
 572
 573             /* Compute parameters for interactions between i and j atoms */
 574             qq10             = _mm_mul_ps(iq1,jq0);
 575
 576             /* REACTION-FIELD ELECTROSTATICS */
 577             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 578             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 579
 580             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm_and_ps(velec,cutoff_mask);
 584             velec            = _mm_andnot_ps(dummy_mask,velec);
 585             velecsum         = _mm_add_ps(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589             fscal            = _mm_and_ps(fscal,cutoff_mask);
 590
 591             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 592
 593              /* Update vectorial force */
 594             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 595             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 596             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 597
 598             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 599             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 600             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 601
 602             }
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             if (gmx_mm_any_lt(rsq20,rcutoff2))
 609             {
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq20             = _mm_mul_ps(iq2,jq0);
 613
 614             /* REACTION-FIELD ELECTROSTATICS */
 615             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 616             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 617
 618             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 619
 620             /* Update potential sum for this i atom from the interaction with this j atom. */
 621             velec            = _mm_and_ps(velec,cutoff_mask);
 622             velec            = _mm_andnot_ps(dummy_mask,velec);
 623             velecsum         = _mm_add_ps(velecsum,velec);
 624
 625             fscal            = felec;
 626
 627             fscal            = _mm_and_ps(fscal,cutoff_mask);
 628
 629             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 630
 631              /* Update vectorial force */
 632             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 633             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 634             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 635
 636             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 637             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 638             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 639
 640             }
 641
 642             /**************************
 643              * CALCULATE INTERACTIONS *
 644              **************************/
 645
 646             if (gmx_mm_any_lt(rsq30,rcutoff2))
 647             {
 648
 649             /* Compute parameters for interactions between i and j atoms */
 650             qq30             = _mm_mul_ps(iq3,jq0);
 651
 652             /* REACTION-FIELD ELECTROSTATICS */
 653             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 654             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 655
 656             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 657
 658             /* Update potential sum for this i atom from the interaction with this j atom. */
 659             velec            = _mm_and_ps(velec,cutoff_mask);
 660             velec            = _mm_andnot_ps(dummy_mask,velec);
 661             velecsum         = _mm_add_ps(velecsum,velec);
 662
 663             fscal            = felec;
 664
 665             fscal            = _mm_and_ps(fscal,cutoff_mask);
 666
 667             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 668
 669              /* Update vectorial force */
 670             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 671             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 672             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 673
 674             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 675             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 676             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 677
 678             }
 679
 680             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 681             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 682             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 683             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 684
 685             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 686
 687             /* Inner loop uses 180 flops */
 688         }
 689
 690         /* End of innermost loop */
 691
 692         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 693                                               f+i_coord_offset,fshift+i_shift_offset);
 694
 695         ggid                        = gid[iidx];
 696         /* Update potential energies */
 697         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 698         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 699
 700         /* Increment number of inner iterations */
 701         inneriter                  += j_index_end - j_index_start;
 702
 703         /* Outer loop uses 26 flops */
 704     }
 705
 706     /* Increment number of outer iterations */
 707     outeriter        += nri;
 708
 709     /* Update outer/inner flops */
 710
 711     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*180);
 712 }
 713 /*
 714  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_128_fma_single
 715  * Electrostatics interaction: ReactionField
 716  * VdW interaction:            LennardJones
 717  * Geometry:                   Water4-Particle
 718  * Calculate force/pot:        Force
 719  */
 720 void
 721 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_128_fma_single
 722                     (t_nblist                    * gmx_restrict       nlist,
 723                      rvec                        * gmx_restrict          xx,
 724                      rvec                        * gmx_restrict          ff,
 725                      t_forcerec                  * gmx_restrict          fr,
 726                      t_mdatoms                   * gmx_restrict     mdatoms,
 727                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 728                      t_nrnb                      * gmx_restrict        nrnb)
 729 {
 730     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 731      * just 0 for non-waters.
 732      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 733      * jnr indices corresponding to data put in the four positions in the SIMD register.
 734      */
 735     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 736     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 737     int              jnrA,jnrB,jnrC,jnrD;
 738     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 739     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 740     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 741     real             rcutoff_scalar;
 742     real             *shiftvec,*fshift,*x,*f;
 743     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 744     real             scratch[4*DIM];
 745     __m128           fscal,rcutoff,rcutoff2,jidxall;
 746     int              vdwioffset0;
 747     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 748     int              vdwioffset1;
 749     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 750     int              vdwioffset2;
 751     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 752     int              vdwioffset3;
 753     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 754     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 755     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 756     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 757     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 758     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 759     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 760     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 761     real             *charge;
 762     int              nvdwtype;
 763     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 764     int              *vdwtype;
 765     real             *vdwparam;
 766     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 767     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 768     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 769     real             rswitch_scalar,d_scalar;
 770     __m128           dummy_mask,cutoff_mask;
 771     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 772     __m128           one     = _mm_set1_ps(1.0);
 773     __m128           two     = _mm_set1_ps(2.0);
 774     x                = xx[0];
 775     f                = ff[0];
 776
 777     nri              = nlist->nri;
 778     iinr             = nlist->iinr;
 779     jindex           = nlist->jindex;
 780     jjnr             = nlist->jjnr;
 781     shiftidx         = nlist->shift;
 782     gid              = nlist->gid;
 783     shiftvec         = fr->shift_vec[0];
 784     fshift           = fr->fshift[0];
 785     facel            = _mm_set1_ps(fr->epsfac);
 786     charge           = mdatoms->chargeA;
 787     krf              = _mm_set1_ps(fr->ic->k_rf);
 788     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 789     crf              = _mm_set1_ps(fr->ic->c_rf);
 790     nvdwtype         = fr->ntype;
 791     vdwparam         = fr->nbfp;
 792     vdwtype          = mdatoms->typeA;
 793
 794     /* Setup water-specific parameters */
 795     inr              = nlist->iinr[0];
 796     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 797     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 798     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 799     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 800
 801     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 802     rcutoff_scalar   = fr->rcoulomb;
 803     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 804     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 805
 806     rswitch_scalar   = fr->rvdw_switch;
 807     rswitch          = _mm_set1_ps(rswitch_scalar);
 808     /* Setup switch parameters */
 809     d_scalar         = rcutoff_scalar-rswitch_scalar;
 810     d                = _mm_set1_ps(d_scalar);
 811     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 812     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 813     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 814     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 815     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 816     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 817
 818     /* Avoid stupid compiler warnings */
 819     jnrA = jnrB = jnrC = jnrD = 0;
 820     j_coord_offsetA = 0;
 821     j_coord_offsetB = 0;
 822     j_coord_offsetC = 0;
 823     j_coord_offsetD = 0;
 824
 825     outeriter        = 0;
 826     inneriter        = 0;
 827
 828     for(iidx=0;iidx<4*DIM;iidx++)
 829     {
 830         scratch[iidx] = 0.0;
 831     }
 832
 833     /* Start outer loop over neighborlists */
 834     for(iidx=0; iidx<nri; iidx++)
 835     {
 836         /* Load shift vector for this list */
 837         i_shift_offset   = DIM*shiftidx[iidx];
 838
 839         /* Load limits for loop over neighbors */
 840         j_index_start    = jindex[iidx];
 841         j_index_end      = jindex[iidx+1];
 842
 843         /* Get outer coordinate index */
 844         inr              = iinr[iidx];
 845         i_coord_offset   = DIM*inr;
 846
 847         /* Load i particle coords and add shift vector */
 848         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 849                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 850
 851         fix0             = _mm_setzero_ps();
 852         fiy0             = _mm_setzero_ps();
 853         fiz0             = _mm_setzero_ps();
 854         fix1             = _mm_setzero_ps();
 855         fiy1             = _mm_setzero_ps();
 856         fiz1             = _mm_setzero_ps();
 857         fix2             = _mm_setzero_ps();
 858         fiy2             = _mm_setzero_ps();
 859         fiz2             = _mm_setzero_ps();
 860         fix3             = _mm_setzero_ps();
 861         fiy3             = _mm_setzero_ps();
 862         fiz3             = _mm_setzero_ps();
 863
 864         /* Start inner kernel loop */
 865         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 866         {
 867
 868             /* Get j neighbor index, and coordinate index */
 869             jnrA             = jjnr[jidx];
 870             jnrB             = jjnr[jidx+1];
 871             jnrC             = jjnr[jidx+2];
 872             jnrD             = jjnr[jidx+3];
 873             j_coord_offsetA  = DIM*jnrA;
 874             j_coord_offsetB  = DIM*jnrB;
 875             j_coord_offsetC  = DIM*jnrC;
 876             j_coord_offsetD  = DIM*jnrD;
 877
 878             /* load j atom coordinates */
 879             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 880                                               x+j_coord_offsetC,x+j_coord_offsetD,
 881                                               &jx0,&jy0,&jz0);
 882
 883             /* Calculate displacement vector */
 884             dx00             = _mm_sub_ps(ix0,jx0);
 885             dy00             = _mm_sub_ps(iy0,jy0);
 886             dz00             = _mm_sub_ps(iz0,jz0);
 887             dx10             = _mm_sub_ps(ix1,jx0);
 888             dy10             = _mm_sub_ps(iy1,jy0);
 889             dz10             = _mm_sub_ps(iz1,jz0);
 890             dx20             = _mm_sub_ps(ix2,jx0);
 891             dy20             = _mm_sub_ps(iy2,jy0);
 892             dz20             = _mm_sub_ps(iz2,jz0);
 893             dx30             = _mm_sub_ps(ix3,jx0);
 894             dy30             = _mm_sub_ps(iy3,jy0);
 895             dz30             = _mm_sub_ps(iz3,jz0);
 896
 897             /* Calculate squared distance and things based on it */
 898             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 899             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 900             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 901             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 902
 903             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 904             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 905             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 906             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 907
 908             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 909             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 910             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 911             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 912
 913             /* Load parameters for j particles */
 914             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 915                                                               charge+jnrC+0,charge+jnrD+0);
 916             vdwjidx0A        = 2*vdwtype[jnrA+0];
 917             vdwjidx0B        = 2*vdwtype[jnrB+0];
 918             vdwjidx0C        = 2*vdwtype[jnrC+0];
 919             vdwjidx0D        = 2*vdwtype[jnrD+0];
 920
 921             fjx0             = _mm_setzero_ps();
 922             fjy0             = _mm_setzero_ps();
 923             fjz0             = _mm_setzero_ps();
 924
 925             /**************************
 926              * CALCULATE INTERACTIONS *
 927              **************************/
 928
 929             if (gmx_mm_any_lt(rsq00,rcutoff2))
 930             {
 931
 932             r00              = _mm_mul_ps(rsq00,rinv00);
 933
 934             /* Compute parameters for interactions between i and j atoms */
 935             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 936                                          vdwparam+vdwioffset0+vdwjidx0B,
 937                                          vdwparam+vdwioffset0+vdwjidx0C,
 938                                          vdwparam+vdwioffset0+vdwjidx0D,
 939                                          &c6_00,&c12_00);
 940
 941             /* LENNARD-JONES DISPERSION/REPULSION */
 942
 943             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 944             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 945             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 946             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 947             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 948
 949             d                = _mm_sub_ps(r00,rswitch);
 950             d                = _mm_max_ps(d,_mm_setzero_ps());
 951             d2               = _mm_mul_ps(d,d);
 952             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 953
 954             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 955
 956             /* Evaluate switch function */
 957             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 958             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 959             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 960
 961             fscal            = fvdw;
 962
 963             fscal            = _mm_and_ps(fscal,cutoff_mask);
 964
 965              /* Update vectorial force */
 966             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 967             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 968             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 969
 970             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 971             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 972             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 973
 974             }
 975
 976             /**************************
 977              * CALCULATE INTERACTIONS *
 978              **************************/
 979
 980             if (gmx_mm_any_lt(rsq10,rcutoff2))
 981             {
 982
 983             /* Compute parameters for interactions between i and j atoms */
 984             qq10             = _mm_mul_ps(iq1,jq0);
 985
 986             /* REACTION-FIELD ELECTROSTATICS */
 987             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 988
 989             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 990
 991             fscal            = felec;
 992
 993             fscal            = _mm_and_ps(fscal,cutoff_mask);
 994
 995              /* Update vectorial force */
 996             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 997             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 998             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 999
1000             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1001             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1002             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1003
1004             }
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             if (gmx_mm_any_lt(rsq20,rcutoff2))
1011             {
1012
1013             /* Compute parameters for interactions between i and j atoms */
1014             qq20             = _mm_mul_ps(iq2,jq0);
1015
1016             /* REACTION-FIELD ELECTROSTATICS */
1017             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1018
1019             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1020
1021             fscal            = felec;
1022
1023             fscal            = _mm_and_ps(fscal,cutoff_mask);
1024
1025              /* Update vectorial force */
1026             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1027             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1028             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1029
1030             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1031             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1032             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1033
1034             }
1035
1036             /**************************
1037              * CALCULATE INTERACTIONS *
1038              **************************/
1039
1040             if (gmx_mm_any_lt(rsq30,rcutoff2))
1041             {
1042
1043             /* Compute parameters for interactions between i and j atoms */
1044             qq30             = _mm_mul_ps(iq3,jq0);
1045
1046             /* REACTION-FIELD ELECTROSTATICS */
1047             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1048
1049             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1050
1051             fscal            = felec;
1052
1053             fscal            = _mm_and_ps(fscal,cutoff_mask);
1054
1055              /* Update vectorial force */
1056             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1057             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1058             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1059
1060             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1061             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1062             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1063
1064             }
1065
1066             fjptrA             = f+j_coord_offsetA;
1067             fjptrB             = f+j_coord_offsetB;
1068             fjptrC             = f+j_coord_offsetC;
1069             fjptrD             = f+j_coord_offsetD;
1070
1071             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1072
1073             /* Inner loop uses 158 flops */
1074         }
1075
1076         if(jidx<j_index_end)
1077         {
1078
1079             /* Get j neighbor index, and coordinate index */
1080             jnrlistA         = jjnr[jidx];
1081             jnrlistB         = jjnr[jidx+1];
1082             jnrlistC         = jjnr[jidx+2];
1083             jnrlistD         = jjnr[jidx+3];
1084             /* Sign of each element will be negative for non-real atoms.
1085              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1086              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1087              */
1088             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1089             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1090             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1091             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1092             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1093             j_coord_offsetA  = DIM*jnrA;
1094             j_coord_offsetB  = DIM*jnrB;
1095             j_coord_offsetC  = DIM*jnrC;
1096             j_coord_offsetD  = DIM*jnrD;
1097
1098             /* load j atom coordinates */
1099             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1100                                               x+j_coord_offsetC,x+j_coord_offsetD,
1101                                               &jx0,&jy0,&jz0);
1102
1103             /* Calculate displacement vector */
1104             dx00             = _mm_sub_ps(ix0,jx0);
1105             dy00             = _mm_sub_ps(iy0,jy0);
1106             dz00             = _mm_sub_ps(iz0,jz0);
1107             dx10             = _mm_sub_ps(ix1,jx0);
1108             dy10             = _mm_sub_ps(iy1,jy0);
1109             dz10             = _mm_sub_ps(iz1,jz0);
1110             dx20             = _mm_sub_ps(ix2,jx0);
1111             dy20             = _mm_sub_ps(iy2,jy0);
1112             dz20             = _mm_sub_ps(iz2,jz0);
1113             dx30             = _mm_sub_ps(ix3,jx0);
1114             dy30             = _mm_sub_ps(iy3,jy0);
1115             dz30             = _mm_sub_ps(iz3,jz0);
1116
1117             /* Calculate squared distance and things based on it */
1118             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1119             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1120             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1121             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1122
1123             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1124             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1125             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1126             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1127
1128             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1129             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1130             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1131             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1132
1133             /* Load parameters for j particles */
1134             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1135                                                               charge+jnrC+0,charge+jnrD+0);
1136             vdwjidx0A        = 2*vdwtype[jnrA+0];
1137             vdwjidx0B        = 2*vdwtype[jnrB+0];
1138             vdwjidx0C        = 2*vdwtype[jnrC+0];
1139             vdwjidx0D        = 2*vdwtype[jnrD+0];
1140
1141             fjx0             = _mm_setzero_ps();
1142             fjy0             = _mm_setzero_ps();
1143             fjz0             = _mm_setzero_ps();
1144
1145             /**************************
1146              * CALCULATE INTERACTIONS *
1147              **************************/
1148
1149             if (gmx_mm_any_lt(rsq00,rcutoff2))
1150             {
1151
1152             r00              = _mm_mul_ps(rsq00,rinv00);
1153             r00              = _mm_andnot_ps(dummy_mask,r00);
1154
1155             /* Compute parameters for interactions between i and j atoms */
1156             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1157                                          vdwparam+vdwioffset0+vdwjidx0B,
1158                                          vdwparam+vdwioffset0+vdwjidx0C,
1159                                          vdwparam+vdwioffset0+vdwjidx0D,
1160                                          &c6_00,&c12_00);
1161
1162             /* LENNARD-JONES DISPERSION/REPULSION */
1163
1164             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1165             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1166             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1167             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1168             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1169
1170             d                = _mm_sub_ps(r00,rswitch);
1171             d                = _mm_max_ps(d,_mm_setzero_ps());
1172             d2               = _mm_mul_ps(d,d);
1173             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1174
1175             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1176
1177             /* Evaluate switch function */
1178             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1179             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1180             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1181
1182             fscal            = fvdw;
1183
1184             fscal            = _mm_and_ps(fscal,cutoff_mask);
1185
1186             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1187
1188              /* Update vectorial force */
1189             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1190             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1191             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1192
1193             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1194             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1195             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1196
1197             }
1198
1199             /**************************
1200              * CALCULATE INTERACTIONS *
1201              **************************/
1202
1203             if (gmx_mm_any_lt(rsq10,rcutoff2))
1204             {
1205
1206             /* Compute parameters for interactions between i and j atoms */
1207             qq10             = _mm_mul_ps(iq1,jq0);
1208
1209             /* REACTION-FIELD ELECTROSTATICS */
1210             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1211
1212             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1213
1214             fscal            = felec;
1215
1216             fscal            = _mm_and_ps(fscal,cutoff_mask);
1217
1218             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1219
1220              /* Update vectorial force */
1221             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1222             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1223             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1224
1225             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1226             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1227             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1228
1229             }
1230
1231             /**************************
1232              * CALCULATE INTERACTIONS *
1233              **************************/
1234
1235             if (gmx_mm_any_lt(rsq20,rcutoff2))
1236             {
1237
1238             /* Compute parameters for interactions between i and j atoms */
1239             qq20             = _mm_mul_ps(iq2,jq0);
1240
1241             /* REACTION-FIELD ELECTROSTATICS */
1242             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1243
1244             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1245
1246             fscal            = felec;
1247
1248             fscal            = _mm_and_ps(fscal,cutoff_mask);
1249
1250             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1251
1252              /* Update vectorial force */
1253             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1254             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1255             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1256
1257             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1258             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1259             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1260
1261             }
1262
1263             /**************************
1264              * CALCULATE INTERACTIONS *
1265              **************************/
1266
1267             if (gmx_mm_any_lt(rsq30,rcutoff2))
1268             {
1269
1270             /* Compute parameters for interactions between i and j atoms */
1271             qq30             = _mm_mul_ps(iq3,jq0);
1272
1273             /* REACTION-FIELD ELECTROSTATICS */
1274             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1275
1276             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1277
1278             fscal            = felec;
1279
1280             fscal            = _mm_and_ps(fscal,cutoff_mask);
1281
1282             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1283
1284              /* Update vectorial force */
1285             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1286             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1287             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1288
1289             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1290             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1291             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1292
1293             }
1294
1295             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1296             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1297             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1298             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1299
1300             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1301
1302             /* Inner loop uses 159 flops */
1303         }
1304
1305         /* End of innermost loop */
1306
1307         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1308                                               f+i_coord_offset,fshift+i_shift_offset);
1309
1310         /* Increment number of inner iterations */
1311         inneriter                  += j_index_end - j_index_start;
1312
1313         /* Outer loop uses 24 flops */
1314     }
1315
1316     /* Increment number of outer iterations */
1317     outeriter        += nri;
1318
1319     /* Update outer/inner flops */
1320
1321     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*159);
1322 }