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, 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_ElecRFCut_VdwLJSw_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_ElecRFCut_VdwLJSw_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           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 108     real             rswitch_scalar,d_scalar;
 109     __m128           dummy_mask,cutoff_mask;
 110     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 111     __m128           one     = _mm_set1_ps(1.0);
 112     __m128           two     = _mm_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_ps(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     krf              = _mm_set1_ps(fr->ic->k_rf);
 127     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 128     crf              = _mm_set1_ps(fr->ic->c_rf);
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 136     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 137     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->rcoulomb;
 142     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 143     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 144
 145     rswitch_scalar   = fr->rvdw_switch;
 146     rswitch          = _mm_set1_ps(rswitch_scalar);
 147     /* Setup switch parameters */
 148     d_scalar         = rcutoff_scalar-rswitch_scalar;
 149     d                = _mm_set1_ps(d_scalar);
 150     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 151     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 152     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 153     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 154     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 155     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 156
 157     /* Avoid stupid compiler warnings */
 158     jnrA = jnrB = jnrC = jnrD = 0;
 159     j_coord_offsetA = 0;
 160     j_coord_offsetB = 0;
 161     j_coord_offsetC = 0;
 162     j_coord_offsetD = 0;
 163
 164     outeriter        = 0;
 165     inneriter        = 0;
 166
 167     for(iidx=0;iidx<4*DIM;iidx++)
 168     {
 169         scratch[iidx] = 0.0;
 170     }
 171
 172     /* Start outer loop over neighborlists */
 173     for(iidx=0; iidx<nri; iidx++)
 174     {
 175         /* Load shift vector for this list */
 176         i_shift_offset   = DIM*shiftidx[iidx];
 177
 178         /* Load limits for loop over neighbors */
 179         j_index_start    = jindex[iidx];
 180         j_index_end      = jindex[iidx+1];
 181
 182         /* Get outer coordinate index */
 183         inr              = iinr[iidx];
 184         i_coord_offset   = DIM*inr;
 185
 186         /* Load i particle coords and add shift vector */
 187         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 188                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 189
 190         fix0             = _mm_setzero_ps();
 191         fiy0             = _mm_setzero_ps();
 192         fiz0             = _mm_setzero_ps();
 193         fix1             = _mm_setzero_ps();
 194         fiy1             = _mm_setzero_ps();
 195         fiz1             = _mm_setzero_ps();
 196         fix2             = _mm_setzero_ps();
 197         fiy2             = _mm_setzero_ps();
 198         fiz2             = _mm_setzero_ps();
 199         fix3             = _mm_setzero_ps();
 200         fiy3             = _mm_setzero_ps();
 201         fiz3             = _mm_setzero_ps();
 202
 203         /* Reset potential sums */
 204         velecsum         = _mm_setzero_ps();
 205         vvdwsum          = _mm_setzero_ps();
 206
 207         /* Start inner kernel loop */
 208         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 209         {
 210
 211             /* Get j neighbor index, and coordinate index */
 212             jnrA             = jjnr[jidx];
 213             jnrB             = jjnr[jidx+1];
 214             jnrC             = jjnr[jidx+2];
 215             jnrD             = jjnr[jidx+3];
 216             j_coord_offsetA  = DIM*jnrA;
 217             j_coord_offsetB  = DIM*jnrB;
 218             j_coord_offsetC  = DIM*jnrC;
 219             j_coord_offsetD  = DIM*jnrD;
 220
 221             /* load j atom coordinates */
 222             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 223                                               x+j_coord_offsetC,x+j_coord_offsetD,
 224                                               &jx0,&jy0,&jz0);
 225
 226             /* Calculate displacement vector */
 227             dx00             = _mm_sub_ps(ix0,jx0);
 228             dy00             = _mm_sub_ps(iy0,jy0);
 229             dz00             = _mm_sub_ps(iz0,jz0);
 230             dx10             = _mm_sub_ps(ix1,jx0);
 231             dy10             = _mm_sub_ps(iy1,jy0);
 232             dz10             = _mm_sub_ps(iz1,jz0);
 233             dx20             = _mm_sub_ps(ix2,jx0);
 234             dy20             = _mm_sub_ps(iy2,jy0);
 235             dz20             = _mm_sub_ps(iz2,jz0);
 236             dx30             = _mm_sub_ps(ix3,jx0);
 237             dy30             = _mm_sub_ps(iy3,jy0);
 238             dz30             = _mm_sub_ps(iz3,jz0);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 242             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 243             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 244             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 245
 246             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 247             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 248             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 249             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 250
 251             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 252             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 253             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 254             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 255
 256             /* Load parameters for j particles */
 257             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 258                                                               charge+jnrC+0,charge+jnrD+0);
 259             vdwjidx0A        = 2*vdwtype[jnrA+0];
 260             vdwjidx0B        = 2*vdwtype[jnrB+0];
 261             vdwjidx0C        = 2*vdwtype[jnrC+0];
 262             vdwjidx0D        = 2*vdwtype[jnrD+0];
 263
 264             fjx0             = _mm_setzero_ps();
 265             fjy0             = _mm_setzero_ps();
 266             fjz0             = _mm_setzero_ps();
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             if (gmx_mm_any_lt(rsq00,rcutoff2))
 273             {
 274
 275             r00              = _mm_mul_ps(rsq00,rinv00);
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 279                                          vdwparam+vdwioffset0+vdwjidx0B,
 280                                          vdwparam+vdwioffset0+vdwjidx0C,
 281                                          vdwparam+vdwioffset0+vdwjidx0D,
 282                                          &c6_00,&c12_00);
 283
 284             /* LENNARD-JONES DISPERSION/REPULSION */
 285
 286             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 287             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 288             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 289             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 290             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 291
 292             d                = _mm_sub_ps(r00,rswitch);
 293             d                = _mm_max_ps(d,_mm_setzero_ps());
 294             d2               = _mm_mul_ps(d,d);
 295             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 296
 297             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 298
 299             /* Evaluate switch function */
 300             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 301             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 302             vvdw             = _mm_mul_ps(vvdw,sw);
 303             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 307             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 308
 309             fscal            = fvdw;
 310
 311             fscal            = _mm_and_ps(fscal,cutoff_mask);
 312
 313              /* Update vectorial force */
 314             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 315             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 316             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 317
 318             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 319             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 320             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 321
 322             }
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             if (gmx_mm_any_lt(rsq10,rcutoff2))
 329             {
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq10             = _mm_mul_ps(iq1,jq0);
 333
 334             /* REACTION-FIELD ELECTROSTATICS */
 335             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 336             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 337
 338             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velec            = _mm_and_ps(velec,cutoff_mask);
 342             velecsum         = _mm_add_ps(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             fscal            = _mm_and_ps(fscal,cutoff_mask);
 347
 348              /* Update vectorial force */
 349             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 350             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 351             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 352
 353             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 354             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 355             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 356
 357             }
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             if (gmx_mm_any_lt(rsq20,rcutoff2))
 364             {
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq20             = _mm_mul_ps(iq2,jq0);
 368
 369             /* REACTION-FIELD ELECTROSTATICS */
 370             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 371             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 372
 373             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velec            = _mm_and_ps(velec,cutoff_mask);
 377             velecsum         = _mm_add_ps(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             fscal            = _mm_and_ps(fscal,cutoff_mask);
 382
 383              /* Update vectorial force */
 384             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 385             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 386             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 387
 388             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 389             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 390             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 391
 392             }
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             if (gmx_mm_any_lt(rsq30,rcutoff2))
 399             {
 400
 401             /* Compute parameters for interactions between i and j atoms */
 402             qq30             = _mm_mul_ps(iq3,jq0);
 403
 404             /* REACTION-FIELD ELECTROSTATICS */
 405             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 406             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 407
 408             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 409
 410             /* Update potential sum for this i atom from the interaction with this j atom. */
 411             velec            = _mm_and_ps(velec,cutoff_mask);
 412             velecsum         = _mm_add_ps(velecsum,velec);
 413
 414             fscal            = felec;
 415
 416             fscal            = _mm_and_ps(fscal,cutoff_mask);
 417
 418              /* Update vectorial force */
 419             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 420             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 421             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 422
 423             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 424             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 425             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 426
 427             }
 428
 429             fjptrA             = f+j_coord_offsetA;
 430             fjptrB             = f+j_coord_offsetB;
 431             fjptrC             = f+j_coord_offsetC;
 432             fjptrD             = f+j_coord_offsetD;
 433
 434             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 435
 436             /* Inner loop uses 179 flops */
 437         }
 438
 439         if(jidx<j_index_end)
 440         {
 441
 442             /* Get j neighbor index, and coordinate index */
 443             jnrlistA         = jjnr[jidx];
 444             jnrlistB         = jjnr[jidx+1];
 445             jnrlistC         = jjnr[jidx+2];
 446             jnrlistD         = jjnr[jidx+3];
 447             /* Sign of each element will be negative for non-real atoms.
 448              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 449              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 450              */
 451             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 452             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 453             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 454             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 455             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 456             j_coord_offsetA  = DIM*jnrA;
 457             j_coord_offsetB  = DIM*jnrB;
 458             j_coord_offsetC  = DIM*jnrC;
 459             j_coord_offsetD  = DIM*jnrD;
 460
 461             /* load j atom coordinates */
 462             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 463                                               x+j_coord_offsetC,x+j_coord_offsetD,
 464                                               &jx0,&jy0,&jz0);
 465
 466             /* Calculate displacement vector */
 467             dx00             = _mm_sub_ps(ix0,jx0);
 468             dy00             = _mm_sub_ps(iy0,jy0);
 469             dz00             = _mm_sub_ps(iz0,jz0);
 470             dx10             = _mm_sub_ps(ix1,jx0);
 471             dy10             = _mm_sub_ps(iy1,jy0);
 472             dz10             = _mm_sub_ps(iz1,jz0);
 473             dx20             = _mm_sub_ps(ix2,jx0);
 474             dy20             = _mm_sub_ps(iy2,jy0);
 475             dz20             = _mm_sub_ps(iz2,jz0);
 476             dx30             = _mm_sub_ps(ix3,jx0);
 477             dy30             = _mm_sub_ps(iy3,jy0);
 478             dz30             = _mm_sub_ps(iz3,jz0);
 479
 480             /* Calculate squared distance and things based on it */
 481             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 482             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 483             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 484             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 485
 486             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 487             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 488             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 489             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 490
 491             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 492             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 493             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 494             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 495
 496             /* Load parameters for j particles */
 497             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 498                                                               charge+jnrC+0,charge+jnrD+0);
 499             vdwjidx0A        = 2*vdwtype[jnrA+0];
 500             vdwjidx0B        = 2*vdwtype[jnrB+0];
 501             vdwjidx0C        = 2*vdwtype[jnrC+0];
 502             vdwjidx0D        = 2*vdwtype[jnrD+0];
 503
 504             fjx0             = _mm_setzero_ps();
 505             fjy0             = _mm_setzero_ps();
 506             fjz0             = _mm_setzero_ps();
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             if (gmx_mm_any_lt(rsq00,rcutoff2))
 513             {
 514
 515             r00              = _mm_mul_ps(rsq00,rinv00);
 516             r00              = _mm_andnot_ps(dummy_mask,r00);
 517
 518             /* Compute parameters for interactions between i and j atoms */
 519             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 520                                          vdwparam+vdwioffset0+vdwjidx0B,
 521                                          vdwparam+vdwioffset0+vdwjidx0C,
 522                                          vdwparam+vdwioffset0+vdwjidx0D,
 523                                          &c6_00,&c12_00);
 524
 525             /* LENNARD-JONES DISPERSION/REPULSION */
 526
 527             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 528             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 529             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 530             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 531             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 532
 533             d                = _mm_sub_ps(r00,rswitch);
 534             d                = _mm_max_ps(d,_mm_setzero_ps());
 535             d2               = _mm_mul_ps(d,d);
 536             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 537
 538             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 539
 540             /* Evaluate switch function */
 541             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 542             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 543             vvdw             = _mm_mul_ps(vvdw,sw);
 544             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 545
 546             /* Update potential sum for this i atom from the interaction with this j atom. */
 547             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 548             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 549             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 550
 551             fscal            = fvdw;
 552
 553             fscal            = _mm_and_ps(fscal,cutoff_mask);
 554
 555             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 556
 557              /* Update vectorial force */
 558             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 559             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 560             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 561
 562             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 563             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 564             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 565
 566             }
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             if (gmx_mm_any_lt(rsq10,rcutoff2))
 573             {
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq10             = _mm_mul_ps(iq1,jq0);
 577
 578             /* REACTION-FIELD ELECTROSTATICS */
 579             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 580             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 581
 582             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm_and_ps(velec,cutoff_mask);
 586             velec            = _mm_andnot_ps(dummy_mask,velec);
 587             velecsum         = _mm_add_ps(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm_and_ps(fscal,cutoff_mask);
 592
 593             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 594
 595              /* Update vectorial force */
 596             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 597             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 598             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 599
 600             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 601             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 602             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 603
 604             }
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             if (gmx_mm_any_lt(rsq20,rcutoff2))
 611             {
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq20             = _mm_mul_ps(iq2,jq0);
 615
 616             /* REACTION-FIELD ELECTROSTATICS */
 617             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 618             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 619
 620             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 621
 622             /* Update potential sum for this i atom from the interaction with this j atom. */
 623             velec            = _mm_and_ps(velec,cutoff_mask);
 624             velec            = _mm_andnot_ps(dummy_mask,velec);
 625             velecsum         = _mm_add_ps(velecsum,velec);
 626
 627             fscal            = felec;
 628
 629             fscal            = _mm_and_ps(fscal,cutoff_mask);
 630
 631             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 632
 633              /* Update vectorial force */
 634             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 635             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 636             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 637
 638             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 639             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 640             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 641
 642             }
 643
 644             /**************************
 645              * CALCULATE INTERACTIONS *
 646              **************************/
 647
 648             if (gmx_mm_any_lt(rsq30,rcutoff2))
 649             {
 650
 651             /* Compute parameters for interactions between i and j atoms */
 652             qq30             = _mm_mul_ps(iq3,jq0);
 653
 654             /* REACTION-FIELD ELECTROSTATICS */
 655             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 656             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 657
 658             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 659
 660             /* Update potential sum for this i atom from the interaction with this j atom. */
 661             velec            = _mm_and_ps(velec,cutoff_mask);
 662             velec            = _mm_andnot_ps(dummy_mask,velec);
 663             velecsum         = _mm_add_ps(velecsum,velec);
 664
 665             fscal            = felec;
 666
 667             fscal            = _mm_and_ps(fscal,cutoff_mask);
 668
 669             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 670
 671              /* Update vectorial force */
 672             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 673             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 674             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 675
 676             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 677             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 678             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 679
 680             }
 681
 682             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 683             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 684             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 685             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 686
 687             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 688
 689             /* Inner loop uses 180 flops */
 690         }
 691
 692         /* End of innermost loop */
 693
 694         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 695                                               f+i_coord_offset,fshift+i_shift_offset);
 696
 697         ggid                        = gid[iidx];
 698         /* Update potential energies */
 699         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 700         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 701
 702         /* Increment number of inner iterations */
 703         inneriter                  += j_index_end - j_index_start;
 704
 705         /* Outer loop uses 26 flops */
 706     }
 707
 708     /* Increment number of outer iterations */
 709     outeriter        += nri;
 710
 711     /* Update outer/inner flops */
 712
 713     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*180);
 714 }
 715 /*
 716  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_128_fma_single
 717  * Electrostatics interaction: ReactionField
 718  * VdW interaction:            LennardJones
 719  * Geometry:                   Water4-Particle
 720  * Calculate force/pot:        Force
 721  */
 722 void
 723 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_128_fma_single
 724                     (t_nblist                    * gmx_restrict       nlist,
 725                      rvec                        * gmx_restrict          xx,
 726                      rvec                        * gmx_restrict          ff,
 727                      t_forcerec                  * gmx_restrict          fr,
 728                      t_mdatoms                   * gmx_restrict     mdatoms,
 729                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 730                      t_nrnb                      * gmx_restrict        nrnb)
 731 {
 732     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 733      * just 0 for non-waters.
 734      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 735      * jnr indices corresponding to data put in the four positions in the SIMD register.
 736      */
 737     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 738     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 739     int              jnrA,jnrB,jnrC,jnrD;
 740     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 741     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 742     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 743     real             rcutoff_scalar;
 744     real             *shiftvec,*fshift,*x,*f;
 745     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 746     real             scratch[4*DIM];
 747     __m128           fscal,rcutoff,rcutoff2,jidxall;
 748     int              vdwioffset0;
 749     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 750     int              vdwioffset1;
 751     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 752     int              vdwioffset2;
 753     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 754     int              vdwioffset3;
 755     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 756     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 757     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 758     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 759     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 760     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 761     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 762     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 763     real             *charge;
 764     int              nvdwtype;
 765     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 766     int              *vdwtype;
 767     real             *vdwparam;
 768     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 769     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 770     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 771     real             rswitch_scalar,d_scalar;
 772     __m128           dummy_mask,cutoff_mask;
 773     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 774     __m128           one     = _mm_set1_ps(1.0);
 775     __m128           two     = _mm_set1_ps(2.0);
 776     x                = xx[0];
 777     f                = ff[0];
 778
 779     nri              = nlist->nri;
 780     iinr             = nlist->iinr;
 781     jindex           = nlist->jindex;
 782     jjnr             = nlist->jjnr;
 783     shiftidx         = nlist->shift;
 784     gid              = nlist->gid;
 785     shiftvec         = fr->shift_vec[0];
 786     fshift           = fr->fshift[0];
 787     facel            = _mm_set1_ps(fr->epsfac);
 788     charge           = mdatoms->chargeA;
 789     krf              = _mm_set1_ps(fr->ic->k_rf);
 790     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 791     crf              = _mm_set1_ps(fr->ic->c_rf);
 792     nvdwtype         = fr->ntype;
 793     vdwparam         = fr->nbfp;
 794     vdwtype          = mdatoms->typeA;
 795
 796     /* Setup water-specific parameters */
 797     inr              = nlist->iinr[0];
 798     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 799     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 800     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 801     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 802
 803     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 804     rcutoff_scalar   = fr->rcoulomb;
 805     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 806     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 807
 808     rswitch_scalar   = fr->rvdw_switch;
 809     rswitch          = _mm_set1_ps(rswitch_scalar);
 810     /* Setup switch parameters */
 811     d_scalar         = rcutoff_scalar-rswitch_scalar;
 812     d                = _mm_set1_ps(d_scalar);
 813     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 814     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 815     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 816     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 817     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 818     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 819
 820     /* Avoid stupid compiler warnings */
 821     jnrA = jnrB = jnrC = jnrD = 0;
 822     j_coord_offsetA = 0;
 823     j_coord_offsetB = 0;
 824     j_coord_offsetC = 0;
 825     j_coord_offsetD = 0;
 826
 827     outeriter        = 0;
 828     inneriter        = 0;
 829
 830     for(iidx=0;iidx<4*DIM;iidx++)
 831     {
 832         scratch[iidx] = 0.0;
 833     }
 834
 835     /* Start outer loop over neighborlists */
 836     for(iidx=0; iidx<nri; iidx++)
 837     {
 838         /* Load shift vector for this list */
 839         i_shift_offset   = DIM*shiftidx[iidx];
 840
 841         /* Load limits for loop over neighbors */
 842         j_index_start    = jindex[iidx];
 843         j_index_end      = jindex[iidx+1];
 844
 845         /* Get outer coordinate index */
 846         inr              = iinr[iidx];
 847         i_coord_offset   = DIM*inr;
 848
 849         /* Load i particle coords and add shift vector */
 850         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 851                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 852
 853         fix0             = _mm_setzero_ps();
 854         fiy0             = _mm_setzero_ps();
 855         fiz0             = _mm_setzero_ps();
 856         fix1             = _mm_setzero_ps();
 857         fiy1             = _mm_setzero_ps();
 858         fiz1             = _mm_setzero_ps();
 859         fix2             = _mm_setzero_ps();
 860         fiy2             = _mm_setzero_ps();
 861         fiz2             = _mm_setzero_ps();
 862         fix3             = _mm_setzero_ps();
 863         fiy3             = _mm_setzero_ps();
 864         fiz3             = _mm_setzero_ps();
 865
 866         /* Start inner kernel loop */
 867         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 868         {
 869
 870             /* Get j neighbor index, and coordinate index */
 871             jnrA             = jjnr[jidx];
 872             jnrB             = jjnr[jidx+1];
 873             jnrC             = jjnr[jidx+2];
 874             jnrD             = jjnr[jidx+3];
 875             j_coord_offsetA  = DIM*jnrA;
 876             j_coord_offsetB  = DIM*jnrB;
 877             j_coord_offsetC  = DIM*jnrC;
 878             j_coord_offsetD  = DIM*jnrD;
 879
 880             /* load j atom coordinates */
 881             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 882                                               x+j_coord_offsetC,x+j_coord_offsetD,
 883                                               &jx0,&jy0,&jz0);
 884
 885             /* Calculate displacement vector */
 886             dx00             = _mm_sub_ps(ix0,jx0);
 887             dy00             = _mm_sub_ps(iy0,jy0);
 888             dz00             = _mm_sub_ps(iz0,jz0);
 889             dx10             = _mm_sub_ps(ix1,jx0);
 890             dy10             = _mm_sub_ps(iy1,jy0);
 891             dz10             = _mm_sub_ps(iz1,jz0);
 892             dx20             = _mm_sub_ps(ix2,jx0);
 893             dy20             = _mm_sub_ps(iy2,jy0);
 894             dz20             = _mm_sub_ps(iz2,jz0);
 895             dx30             = _mm_sub_ps(ix3,jx0);
 896             dy30             = _mm_sub_ps(iy3,jy0);
 897             dz30             = _mm_sub_ps(iz3,jz0);
 898
 899             /* Calculate squared distance and things based on it */
 900             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 901             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 902             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 903             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 904
 905             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 906             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 907             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 908             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 909
 910             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 911             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 912             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 913             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 914
 915             /* Load parameters for j particles */
 916             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 917                                                               charge+jnrC+0,charge+jnrD+0);
 918             vdwjidx0A        = 2*vdwtype[jnrA+0];
 919             vdwjidx0B        = 2*vdwtype[jnrB+0];
 920             vdwjidx0C        = 2*vdwtype[jnrC+0];
 921             vdwjidx0D        = 2*vdwtype[jnrD+0];
 922
 923             fjx0             = _mm_setzero_ps();
 924             fjy0             = _mm_setzero_ps();
 925             fjz0             = _mm_setzero_ps();
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             if (gmx_mm_any_lt(rsq00,rcutoff2))
 932             {
 933
 934             r00              = _mm_mul_ps(rsq00,rinv00);
 935
 936             /* Compute parameters for interactions between i and j atoms */
 937             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 938                                          vdwparam+vdwioffset0+vdwjidx0B,
 939                                          vdwparam+vdwioffset0+vdwjidx0C,
 940                                          vdwparam+vdwioffset0+vdwjidx0D,
 941                                          &c6_00,&c12_00);
 942
 943             /* LENNARD-JONES DISPERSION/REPULSION */
 944
 945             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 946             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 947             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 948             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 949             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 950
 951             d                = _mm_sub_ps(r00,rswitch);
 952             d                = _mm_max_ps(d,_mm_setzero_ps());
 953             d2               = _mm_mul_ps(d,d);
 954             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 955
 956             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 957
 958             /* Evaluate switch function */
 959             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 960             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 961             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 962
 963             fscal            = fvdw;
 964
 965             fscal            = _mm_and_ps(fscal,cutoff_mask);
 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
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             if (gmx_mm_any_lt(rsq10,rcutoff2))
 983             {
 984
 985             /* Compute parameters for interactions between i and j atoms */
 986             qq10             = _mm_mul_ps(iq1,jq0);
 987
 988             /* REACTION-FIELD ELECTROSTATICS */
 989             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 990
 991             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 992
 993             fscal            = felec;
 994
 995             fscal            = _mm_and_ps(fscal,cutoff_mask);
 996
 997              /* Update vectorial force */
 998             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 999             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1000             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1001
1002             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1003             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1004             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1005
1006             }
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             if (gmx_mm_any_lt(rsq20,rcutoff2))
1013             {
1014
1015             /* Compute parameters for interactions between i and j atoms */
1016             qq20             = _mm_mul_ps(iq2,jq0);
1017
1018             /* REACTION-FIELD ELECTROSTATICS */
1019             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1020
1021             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1022
1023             fscal            = felec;
1024
1025             fscal            = _mm_and_ps(fscal,cutoff_mask);
1026
1027              /* Update vectorial force */
1028             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1029             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1030             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1031
1032             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1033             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1034             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1035
1036             }
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             if (gmx_mm_any_lt(rsq30,rcutoff2))
1043             {
1044
1045             /* Compute parameters for interactions between i and j atoms */
1046             qq30             = _mm_mul_ps(iq3,jq0);
1047
1048             /* REACTION-FIELD ELECTROSTATICS */
1049             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1050
1051             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1052
1053             fscal            = felec;
1054
1055             fscal            = _mm_and_ps(fscal,cutoff_mask);
1056
1057              /* Update vectorial force */
1058             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1059             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1060             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1061
1062             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1063             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1064             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1065
1066             }
1067
1068             fjptrA             = f+j_coord_offsetA;
1069             fjptrB             = f+j_coord_offsetB;
1070             fjptrC             = f+j_coord_offsetC;
1071             fjptrD             = f+j_coord_offsetD;
1072
1073             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1074
1075             /* Inner loop uses 158 flops */
1076         }
1077
1078         if(jidx<j_index_end)
1079         {
1080
1081             /* Get j neighbor index, and coordinate index */
1082             jnrlistA         = jjnr[jidx];
1083             jnrlistB         = jjnr[jidx+1];
1084             jnrlistC         = jjnr[jidx+2];
1085             jnrlistD         = jjnr[jidx+3];
1086             /* Sign of each element will be negative for non-real atoms.
1087              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1088              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1089              */
1090             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1091             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1092             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1093             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1094             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1095             j_coord_offsetA  = DIM*jnrA;
1096             j_coord_offsetB  = DIM*jnrB;
1097             j_coord_offsetC  = DIM*jnrC;
1098             j_coord_offsetD  = DIM*jnrD;
1099
1100             /* load j atom coordinates */
1101             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1102                                               x+j_coord_offsetC,x+j_coord_offsetD,
1103                                               &jx0,&jy0,&jz0);
1104
1105             /* Calculate displacement vector */
1106             dx00             = _mm_sub_ps(ix0,jx0);
1107             dy00             = _mm_sub_ps(iy0,jy0);
1108             dz00             = _mm_sub_ps(iz0,jz0);
1109             dx10             = _mm_sub_ps(ix1,jx0);
1110             dy10             = _mm_sub_ps(iy1,jy0);
1111             dz10             = _mm_sub_ps(iz1,jz0);
1112             dx20             = _mm_sub_ps(ix2,jx0);
1113             dy20             = _mm_sub_ps(iy2,jy0);
1114             dz20             = _mm_sub_ps(iz2,jz0);
1115             dx30             = _mm_sub_ps(ix3,jx0);
1116             dy30             = _mm_sub_ps(iy3,jy0);
1117             dz30             = _mm_sub_ps(iz3,jz0);
1118
1119             /* Calculate squared distance and things based on it */
1120             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1121             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1122             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1123             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1124
1125             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1126             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1127             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1128             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1129
1130             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1131             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1132             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1133             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1134
1135             /* Load parameters for j particles */
1136             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1137                                                               charge+jnrC+0,charge+jnrD+0);
1138             vdwjidx0A        = 2*vdwtype[jnrA+0];
1139             vdwjidx0B        = 2*vdwtype[jnrB+0];
1140             vdwjidx0C        = 2*vdwtype[jnrC+0];
1141             vdwjidx0D        = 2*vdwtype[jnrD+0];
1142
1143             fjx0             = _mm_setzero_ps();
1144             fjy0             = _mm_setzero_ps();
1145             fjz0             = _mm_setzero_ps();
1146
1147             /**************************
1148              * CALCULATE INTERACTIONS *
1149              **************************/
1150
1151             if (gmx_mm_any_lt(rsq00,rcutoff2))
1152             {
1153
1154             r00              = _mm_mul_ps(rsq00,rinv00);
1155             r00              = _mm_andnot_ps(dummy_mask,r00);
1156
1157             /* Compute parameters for interactions between i and j atoms */
1158             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1159                                          vdwparam+vdwioffset0+vdwjidx0B,
1160                                          vdwparam+vdwioffset0+vdwjidx0C,
1161                                          vdwparam+vdwioffset0+vdwjidx0D,
1162                                          &c6_00,&c12_00);
1163
1164             /* LENNARD-JONES DISPERSION/REPULSION */
1165
1166             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1167             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1168             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1169             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1170             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1171
1172             d                = _mm_sub_ps(r00,rswitch);
1173             d                = _mm_max_ps(d,_mm_setzero_ps());
1174             d2               = _mm_mul_ps(d,d);
1175             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1176
1177             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1178
1179             /* Evaluate switch function */
1180             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1181             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1182             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1183
1184             fscal            = fvdw;
1185
1186             fscal            = _mm_and_ps(fscal,cutoff_mask);
1187
1188             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1189
1190              /* Update vectorial force */
1191             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1192             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1193             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1194
1195             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1196             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1197             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1198
1199             }
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             if (gmx_mm_any_lt(rsq10,rcutoff2))
1206             {
1207
1208             /* Compute parameters for interactions between i and j atoms */
1209             qq10             = _mm_mul_ps(iq1,jq0);
1210
1211             /* REACTION-FIELD ELECTROSTATICS */
1212             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1213
1214             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1215
1216             fscal            = felec;
1217
1218             fscal            = _mm_and_ps(fscal,cutoff_mask);
1219
1220             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1221
1222              /* Update vectorial force */
1223             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1224             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1225             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1226
1227             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1228             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1229             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1230
1231             }
1232
1233             /**************************
1234              * CALCULATE INTERACTIONS *
1235              **************************/
1236
1237             if (gmx_mm_any_lt(rsq20,rcutoff2))
1238             {
1239
1240             /* Compute parameters for interactions between i and j atoms */
1241             qq20             = _mm_mul_ps(iq2,jq0);
1242
1243             /* REACTION-FIELD ELECTROSTATICS */
1244             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1245
1246             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1247
1248             fscal            = felec;
1249
1250             fscal            = _mm_and_ps(fscal,cutoff_mask);
1251
1252             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1253
1254              /* Update vectorial force */
1255             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1256             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1257             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1258
1259             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1260             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1261             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1262
1263             }
1264
1265             /**************************
1266              * CALCULATE INTERACTIONS *
1267              **************************/
1268
1269             if (gmx_mm_any_lt(rsq30,rcutoff2))
1270             {
1271
1272             /* Compute parameters for interactions between i and j atoms */
1273             qq30             = _mm_mul_ps(iq3,jq0);
1274
1275             /* REACTION-FIELD ELECTROSTATICS */
1276             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1277
1278             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1279
1280             fscal            = felec;
1281
1282             fscal            = _mm_and_ps(fscal,cutoff_mask);
1283
1284             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1285
1286              /* Update vectorial force */
1287             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1288             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1289             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1290
1291             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1292             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1293             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1294
1295             }
1296
1297             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1298             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1299             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1300             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1301
1302             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1303
1304             /* Inner loop uses 159 flops */
1305         }
1306
1307         /* End of innermost loop */
1308
1309         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1310                                               f+i_coord_offset,fshift+i_shift_offset);
1311
1312         /* Increment number of inner iterations */
1313         inneriter                  += j_index_end - j_index_start;
1314
1315         /* Outer loop uses 24 flops */
1316     }
1317
1318     /* Increment number of outer iterations */
1319     outeriter        += nri;
1320
1321     /* Update outer/inner flops */
1322
1323     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*159);
1324 }