src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  89     real             rswitch_scalar,d_scalar;
  90     __m128           dummy_mask,cutoff_mask;
  91     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  92     __m128           one     = _mm_set1_ps(1.0);
  93     __m128           two     = _mm_set1_ps(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm_set1_ps(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107     krf              = _mm_set1_ps(fr->ic->k_rf);
 108     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 109     crf              = _mm_set1_ps(fr->ic->c_rf);
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     /* Setup water-specific parameters */
 115     inr              = nlist->iinr[0];
 116     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 117     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 118     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 119     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 120
 121     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 122     rcutoff_scalar   = fr->rcoulomb;
 123     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 124     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 125
 126     rswitch_scalar   = fr->rvdw_switch;
 127     rswitch          = _mm_set1_ps(rswitch_scalar);
 128     /* Setup switch parameters */
 129     d_scalar         = rcutoff_scalar-rswitch_scalar;
 130     d                = _mm_set1_ps(d_scalar);
 131     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 132     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 133     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 134     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 135     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 136     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 170
 171         fix0             = _mm_setzero_ps();
 172         fiy0             = _mm_setzero_ps();
 173         fiz0             = _mm_setzero_ps();
 174         fix1             = _mm_setzero_ps();
 175         fiy1             = _mm_setzero_ps();
 176         fiz1             = _mm_setzero_ps();
 177         fix2             = _mm_setzero_ps();
 178         fiy2             = _mm_setzero_ps();
 179         fiz2             = _mm_setzero_ps();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm_setzero_ps();
 183         vvdwsum          = _mm_setzero_ps();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             jnrC             = jjnr[jidx+2];
 193             jnrD             = jjnr[jidx+3];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                               x+j_coord_offsetC,x+j_coord_offsetD,
 202                                               &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm_sub_ps(ix0,jx0);
 206             dy00             = _mm_sub_ps(iy0,jy0);
 207             dz00             = _mm_sub_ps(iz0,jz0);
 208             dx10             = _mm_sub_ps(ix1,jx0);
 209             dy10             = _mm_sub_ps(iy1,jy0);
 210             dz10             = _mm_sub_ps(iz1,jz0);
 211             dx20             = _mm_sub_ps(ix2,jx0);
 212             dy20             = _mm_sub_ps(iy2,jy0);
 213             dz20             = _mm_sub_ps(iz2,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 217             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 218             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 219
 220             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 221             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 222             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 223
 224             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 225             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 226             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 230                                                               charge+jnrC+0,charge+jnrD+0);
 231             vdwjidx0A        = 2*vdwtype[jnrA+0];
 232             vdwjidx0B        = 2*vdwtype[jnrB+0];
 233             vdwjidx0C        = 2*vdwtype[jnrC+0];
 234             vdwjidx0D        = 2*vdwtype[jnrD+0];
 235
 236             fjx0             = _mm_setzero_ps();
 237             fjy0             = _mm_setzero_ps();
 238             fjz0             = _mm_setzero_ps();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             if (gmx_mm_any_lt(rsq00,rcutoff2))
 245             {
 246
 247             r00              = _mm_mul_ps(rsq00,rinv00);
 248
 249             /* Compute parameters for interactions between i and j atoms */
 250             qq00             = _mm_mul_ps(iq0,jq0);
 251             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,
 253                                          vdwparam+vdwioffset0+vdwjidx0C,
 254                                          vdwparam+vdwioffset0+vdwjidx0D,
 255                                          &c6_00,&c12_00);
 256
 257             /* REACTION-FIELD ELECTROSTATICS */
 258             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 259             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 260
 261             /* LENNARD-JONES DISPERSION/REPULSION */
 262
 263             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 264             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 265             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 266             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 267             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 268
 269             d                = _mm_sub_ps(r00,rswitch);
 270             d                = _mm_max_ps(d,_mm_setzero_ps());
 271             d2               = _mm_mul_ps(d,d);
 272             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 273
 274             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 275
 276             /* Evaluate switch function */
 277             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 278             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 279             vvdw             = _mm_mul_ps(vvdw,sw);
 280             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velec            = _mm_and_ps(velec,cutoff_mask);
 284             velecsum         = _mm_add_ps(velecsum,velec);
 285             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 286             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 287
 288             fscal            = _mm_add_ps(felec,fvdw);
 289
 290             fscal            = _mm_and_ps(fscal,cutoff_mask);
 291
 292              /* Update vectorial force */
 293             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 294             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 295             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 296
 297             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 298             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 299             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 300
 301             }
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             if (gmx_mm_any_lt(rsq10,rcutoff2))
 308             {
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq10             = _mm_mul_ps(iq1,jq0);
 312
 313             /* REACTION-FIELD ELECTROSTATICS */
 314             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 315             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 316
 317             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velec            = _mm_and_ps(velec,cutoff_mask);
 321             velecsum         = _mm_add_ps(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             fscal            = _mm_and_ps(fscal,cutoff_mask);
 326
 327              /* Update vectorial force */
 328             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 329             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 330             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 331
 332             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 333             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 334             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 335
 336             }
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             if (gmx_mm_any_lt(rsq20,rcutoff2))
 343             {
 344
 345             /* Compute parameters for interactions between i and j atoms */
 346             qq20             = _mm_mul_ps(iq2,jq0);
 347
 348             /* REACTION-FIELD ELECTROSTATICS */
 349             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 350             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 351
 352             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velec            = _mm_and_ps(velec,cutoff_mask);
 356             velecsum         = _mm_add_ps(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             fscal            = _mm_and_ps(fscal,cutoff_mask);
 361
 362              /* Update vectorial force */
 363             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 364             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 365             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 366
 367             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 368             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 369             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 370
 371             }
 372
 373             fjptrA             = f+j_coord_offsetA;
 374             fjptrB             = f+j_coord_offsetB;
 375             fjptrC             = f+j_coord_offsetC;
 376             fjptrD             = f+j_coord_offsetD;
 377
 378             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 379
 380             /* Inner loop uses 151 flops */
 381         }
 382
 383         if(jidx<j_index_end)
 384         {
 385
 386             /* Get j neighbor index, and coordinate index */
 387             jnrlistA         = jjnr[jidx];
 388             jnrlistB         = jjnr[jidx+1];
 389             jnrlistC         = jjnr[jidx+2];
 390             jnrlistD         = jjnr[jidx+3];
 391             /* Sign of each element will be negative for non-real atoms.
 392              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 393              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 394              */
 395             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 396             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 397             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 398             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 399             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 400             j_coord_offsetA  = DIM*jnrA;
 401             j_coord_offsetB  = DIM*jnrB;
 402             j_coord_offsetC  = DIM*jnrC;
 403             j_coord_offsetD  = DIM*jnrD;
 404
 405             /* load j atom coordinates */
 406             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 407                                               x+j_coord_offsetC,x+j_coord_offsetD,
 408                                               &jx0,&jy0,&jz0);
 409
 410             /* Calculate displacement vector */
 411             dx00             = _mm_sub_ps(ix0,jx0);
 412             dy00             = _mm_sub_ps(iy0,jy0);
 413             dz00             = _mm_sub_ps(iz0,jz0);
 414             dx10             = _mm_sub_ps(ix1,jx0);
 415             dy10             = _mm_sub_ps(iy1,jy0);
 416             dz10             = _mm_sub_ps(iz1,jz0);
 417             dx20             = _mm_sub_ps(ix2,jx0);
 418             dy20             = _mm_sub_ps(iy2,jy0);
 419             dz20             = _mm_sub_ps(iz2,jz0);
 420
 421             /* Calculate squared distance and things based on it */
 422             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 423             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 424             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 425
 426             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 427             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 428             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 429
 430             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 431             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 432             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 433
 434             /* Load parameters for j particles */
 435             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 436                                                               charge+jnrC+0,charge+jnrD+0);
 437             vdwjidx0A        = 2*vdwtype[jnrA+0];
 438             vdwjidx0B        = 2*vdwtype[jnrB+0];
 439             vdwjidx0C        = 2*vdwtype[jnrC+0];
 440             vdwjidx0D        = 2*vdwtype[jnrD+0];
 441
 442             fjx0             = _mm_setzero_ps();
 443             fjy0             = _mm_setzero_ps();
 444             fjz0             = _mm_setzero_ps();
 445
 446             /**************************
 447              * CALCULATE INTERACTIONS *
 448              **************************/
 449
 450             if (gmx_mm_any_lt(rsq00,rcutoff2))
 451             {
 452
 453             r00              = _mm_mul_ps(rsq00,rinv00);
 454             r00              = _mm_andnot_ps(dummy_mask,r00);
 455
 456             /* Compute parameters for interactions between i and j atoms */
 457             qq00             = _mm_mul_ps(iq0,jq0);
 458             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 459                                          vdwparam+vdwioffset0+vdwjidx0B,
 460                                          vdwparam+vdwioffset0+vdwjidx0C,
 461                                          vdwparam+vdwioffset0+vdwjidx0D,
 462                                          &c6_00,&c12_00);
 463
 464             /* REACTION-FIELD ELECTROSTATICS */
 465             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 466             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 467
 468             /* LENNARD-JONES DISPERSION/REPULSION */
 469
 470             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 471             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 472             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 473             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 474             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 475
 476             d                = _mm_sub_ps(r00,rswitch);
 477             d                = _mm_max_ps(d,_mm_setzero_ps());
 478             d2               = _mm_mul_ps(d,d);
 479             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 480
 481             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 482
 483             /* Evaluate switch function */
 484             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 485             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 486             vvdw             = _mm_mul_ps(vvdw,sw);
 487             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 488
 489             /* Update potential sum for this i atom from the interaction with this j atom. */
 490             velec            = _mm_and_ps(velec,cutoff_mask);
 491             velec            = _mm_andnot_ps(dummy_mask,velec);
 492             velecsum         = _mm_add_ps(velecsum,velec);
 493             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 494             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 495             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 496
 497             fscal            = _mm_add_ps(felec,fvdw);
 498
 499             fscal            = _mm_and_ps(fscal,cutoff_mask);
 500
 501             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 502
 503              /* Update vectorial force */
 504             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 505             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 506             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 507
 508             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 509             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 510             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 511
 512             }
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             if (gmx_mm_any_lt(rsq10,rcutoff2))
 519             {
 520
 521             /* Compute parameters for interactions between i and j atoms */
 522             qq10             = _mm_mul_ps(iq1,jq0);
 523
 524             /* REACTION-FIELD ELECTROSTATICS */
 525             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 526             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 527
 528             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm_and_ps(velec,cutoff_mask);
 532             velec            = _mm_andnot_ps(dummy_mask,velec);
 533             velecsum         = _mm_add_ps(velecsum,velec);
 534
 535             fscal            = felec;
 536
 537             fscal            = _mm_and_ps(fscal,cutoff_mask);
 538
 539             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 540
 541              /* Update vectorial force */
 542             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 543             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 544             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 545
 546             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 547             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 548             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 549
 550             }
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             if (gmx_mm_any_lt(rsq20,rcutoff2))
 557             {
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             qq20             = _mm_mul_ps(iq2,jq0);
 561
 562             /* REACTION-FIELD ELECTROSTATICS */
 563             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 564             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 565
 566             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velec            = _mm_and_ps(velec,cutoff_mask);
 570             velec            = _mm_andnot_ps(dummy_mask,velec);
 571             velecsum         = _mm_add_ps(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             fscal            = _mm_and_ps(fscal,cutoff_mask);
 576
 577             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 578
 579              /* Update vectorial force */
 580             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 581             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 582             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 583
 584             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 585             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 586             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 587
 588             }
 589
 590             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 591             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 592             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 593             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 594
 595             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 596
 597             /* Inner loop uses 152 flops */
 598         }
 599
 600         /* End of innermost loop */
 601
 602         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 603                                               f+i_coord_offset,fshift+i_shift_offset);
 604
 605         ggid                        = gid[iidx];
 606         /* Update potential energies */
 607         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 608         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 609
 610         /* Increment number of inner iterations */
 611         inneriter                  += j_index_end - j_index_start;
 612
 613         /* Outer loop uses 20 flops */
 614     }
 615
 616     /* Increment number of outer iterations */
 617     outeriter        += nri;
 618
 619     /* Update outer/inner flops */
 620
 621     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*152);
 622 }
 623 /*
 624  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 625  * Electrostatics interaction: ReactionField
 626  * VdW interaction:            LennardJones
 627  * Geometry:                   Water3-Particle
 628  * Calculate force/pot:        Force
 629  */
 630 void
 631 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 632                     (t_nblist * gmx_restrict                nlist,
 633                      rvec * gmx_restrict                    xx,
 634                      rvec * gmx_restrict                    ff,
 635                      t_forcerec * gmx_restrict              fr,
 636                      t_mdatoms * gmx_restrict               mdatoms,
 637                      nb_kernel_data_t * gmx_restrict        kernel_data,
 638                      t_nrnb * gmx_restrict                  nrnb)
 639 {
 640     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 641      * just 0 for non-waters.
 642      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 643      * jnr indices corresponding to data put in the four positions in the SIMD register.
 644      */
 645     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 646     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 647     int              jnrA,jnrB,jnrC,jnrD;
 648     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 649     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 650     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 651     real             rcutoff_scalar;
 652     real             *shiftvec,*fshift,*x,*f;
 653     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 654     real             scratch[4*DIM];
 655     __m128           fscal,rcutoff,rcutoff2,jidxall;
 656     int              vdwioffset0;
 657     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 658     int              vdwioffset1;
 659     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 660     int              vdwioffset2;
 661     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 662     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 663     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 664     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 665     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 666     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 667     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 668     real             *charge;
 669     int              nvdwtype;
 670     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 671     int              *vdwtype;
 672     real             *vdwparam;
 673     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 674     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 675     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 676     real             rswitch_scalar,d_scalar;
 677     __m128           dummy_mask,cutoff_mask;
 678     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 679     __m128           one     = _mm_set1_ps(1.0);
 680     __m128           two     = _mm_set1_ps(2.0);
 681     x                = xx[0];
 682     f                = ff[0];
 683
 684     nri              = nlist->nri;
 685     iinr             = nlist->iinr;
 686     jindex           = nlist->jindex;
 687     jjnr             = nlist->jjnr;
 688     shiftidx         = nlist->shift;
 689     gid              = nlist->gid;
 690     shiftvec         = fr->shift_vec[0];
 691     fshift           = fr->fshift[0];
 692     facel            = _mm_set1_ps(fr->epsfac);
 693     charge           = mdatoms->chargeA;
 694     krf              = _mm_set1_ps(fr->ic->k_rf);
 695     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 696     crf              = _mm_set1_ps(fr->ic->c_rf);
 697     nvdwtype         = fr->ntype;
 698     vdwparam         = fr->nbfp;
 699     vdwtype          = mdatoms->typeA;
 700
 701     /* Setup water-specific parameters */
 702     inr              = nlist->iinr[0];
 703     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 704     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 705     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 706     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 707
 708     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 709     rcutoff_scalar   = fr->rcoulomb;
 710     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 711     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 712
 713     rswitch_scalar   = fr->rvdw_switch;
 714     rswitch          = _mm_set1_ps(rswitch_scalar);
 715     /* Setup switch parameters */
 716     d_scalar         = rcutoff_scalar-rswitch_scalar;
 717     d                = _mm_set1_ps(d_scalar);
 718     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 719     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 720     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 721     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 722     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 723     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 724
 725     /* Avoid stupid compiler warnings */
 726     jnrA = jnrB = jnrC = jnrD = 0;
 727     j_coord_offsetA = 0;
 728     j_coord_offsetB = 0;
 729     j_coord_offsetC = 0;
 730     j_coord_offsetD = 0;
 731
 732     outeriter        = 0;
 733     inneriter        = 0;
 734
 735     for(iidx=0;iidx<4*DIM;iidx++)
 736     {
 737         scratch[iidx] = 0.0;
 738     }
 739
 740     /* Start outer loop over neighborlists */
 741     for(iidx=0; iidx<nri; iidx++)
 742     {
 743         /* Load shift vector for this list */
 744         i_shift_offset   = DIM*shiftidx[iidx];
 745
 746         /* Load limits for loop over neighbors */
 747         j_index_start    = jindex[iidx];
 748         j_index_end      = jindex[iidx+1];
 749
 750         /* Get outer coordinate index */
 751         inr              = iinr[iidx];
 752         i_coord_offset   = DIM*inr;
 753
 754         /* Load i particle coords and add shift vector */
 755         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 756                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 757
 758         fix0             = _mm_setzero_ps();
 759         fiy0             = _mm_setzero_ps();
 760         fiz0             = _mm_setzero_ps();
 761         fix1             = _mm_setzero_ps();
 762         fiy1             = _mm_setzero_ps();
 763         fiz1             = _mm_setzero_ps();
 764         fix2             = _mm_setzero_ps();
 765         fiy2             = _mm_setzero_ps();
 766         fiz2             = _mm_setzero_ps();
 767
 768         /* Start inner kernel loop */
 769         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 770         {
 771
 772             /* Get j neighbor index, and coordinate index */
 773             jnrA             = jjnr[jidx];
 774             jnrB             = jjnr[jidx+1];
 775             jnrC             = jjnr[jidx+2];
 776             jnrD             = jjnr[jidx+3];
 777             j_coord_offsetA  = DIM*jnrA;
 778             j_coord_offsetB  = DIM*jnrB;
 779             j_coord_offsetC  = DIM*jnrC;
 780             j_coord_offsetD  = DIM*jnrD;
 781
 782             /* load j atom coordinates */
 783             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 784                                               x+j_coord_offsetC,x+j_coord_offsetD,
 785                                               &jx0,&jy0,&jz0);
 786
 787             /* Calculate displacement vector */
 788             dx00             = _mm_sub_ps(ix0,jx0);
 789             dy00             = _mm_sub_ps(iy0,jy0);
 790             dz00             = _mm_sub_ps(iz0,jz0);
 791             dx10             = _mm_sub_ps(ix1,jx0);
 792             dy10             = _mm_sub_ps(iy1,jy0);
 793             dz10             = _mm_sub_ps(iz1,jz0);
 794             dx20             = _mm_sub_ps(ix2,jx0);
 795             dy20             = _mm_sub_ps(iy2,jy0);
 796             dz20             = _mm_sub_ps(iz2,jz0);
 797
 798             /* Calculate squared distance and things based on it */
 799             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 800             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 801             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 802
 803             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 804             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 805             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 806
 807             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 808             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 809             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 810
 811             /* Load parameters for j particles */
 812             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 813                                                               charge+jnrC+0,charge+jnrD+0);
 814             vdwjidx0A        = 2*vdwtype[jnrA+0];
 815             vdwjidx0B        = 2*vdwtype[jnrB+0];
 816             vdwjidx0C        = 2*vdwtype[jnrC+0];
 817             vdwjidx0D        = 2*vdwtype[jnrD+0];
 818
 819             fjx0             = _mm_setzero_ps();
 820             fjy0             = _mm_setzero_ps();
 821             fjz0             = _mm_setzero_ps();
 822
 823             /**************************
 824              * CALCULATE INTERACTIONS *
 825              **************************/
 826
 827             if (gmx_mm_any_lt(rsq00,rcutoff2))
 828             {
 829
 830             r00              = _mm_mul_ps(rsq00,rinv00);
 831
 832             /* Compute parameters for interactions between i and j atoms */
 833             qq00             = _mm_mul_ps(iq0,jq0);
 834             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 835                                          vdwparam+vdwioffset0+vdwjidx0B,
 836                                          vdwparam+vdwioffset0+vdwjidx0C,
 837                                          vdwparam+vdwioffset0+vdwjidx0D,
 838                                          &c6_00,&c12_00);
 839
 840             /* REACTION-FIELD ELECTROSTATICS */
 841             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 842
 843             /* LENNARD-JONES DISPERSION/REPULSION */
 844
 845             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 846             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 847             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 848             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 849             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 850
 851             d                = _mm_sub_ps(r00,rswitch);
 852             d                = _mm_max_ps(d,_mm_setzero_ps());
 853             d2               = _mm_mul_ps(d,d);
 854             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 855
 856             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 857
 858             /* Evaluate switch function */
 859             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 860             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 861             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 862
 863             fscal            = _mm_add_ps(felec,fvdw);
 864
 865             fscal            = _mm_and_ps(fscal,cutoff_mask);
 866
 867              /* Update vectorial force */
 868             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 869             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 870             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 871
 872             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 873             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 874             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 875
 876             }
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             if (gmx_mm_any_lt(rsq10,rcutoff2))
 883             {
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             qq10             = _mm_mul_ps(iq1,jq0);
 887
 888             /* REACTION-FIELD ELECTROSTATICS */
 889             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 890
 891             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 892
 893             fscal            = felec;
 894
 895             fscal            = _mm_and_ps(fscal,cutoff_mask);
 896
 897              /* Update vectorial force */
 898             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 899             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 900             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 901
 902             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 903             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 904             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 905
 906             }
 907
 908             /**************************
 909              * CALCULATE INTERACTIONS *
 910              **************************/
 911
 912             if (gmx_mm_any_lt(rsq20,rcutoff2))
 913             {
 914
 915             /* Compute parameters for interactions between i and j atoms */
 916             qq20             = _mm_mul_ps(iq2,jq0);
 917
 918             /* REACTION-FIELD ELECTROSTATICS */
 919             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 920
 921             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 922
 923             fscal            = felec;
 924
 925             fscal            = _mm_and_ps(fscal,cutoff_mask);
 926
 927              /* Update vectorial force */
 928             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 929             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 930             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 931
 932             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 933             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 934             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 935
 936             }
 937
 938             fjptrA             = f+j_coord_offsetA;
 939             fjptrB             = f+j_coord_offsetB;
 940             fjptrC             = f+j_coord_offsetC;
 941             fjptrD             = f+j_coord_offsetD;
 942
 943             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 944
 945             /* Inner loop uses 130 flops */
 946         }
 947
 948         if(jidx<j_index_end)
 949         {
 950
 951             /* Get j neighbor index, and coordinate index */
 952             jnrlistA         = jjnr[jidx];
 953             jnrlistB         = jjnr[jidx+1];
 954             jnrlistC         = jjnr[jidx+2];
 955             jnrlistD         = jjnr[jidx+3];
 956             /* Sign of each element will be negative for non-real atoms.
 957              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 958              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 959              */
 960             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 961             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 962             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 963             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 964             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 965             j_coord_offsetA  = DIM*jnrA;
 966             j_coord_offsetB  = DIM*jnrB;
 967             j_coord_offsetC  = DIM*jnrC;
 968             j_coord_offsetD  = DIM*jnrD;
 969
 970             /* load j atom coordinates */
 971             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 972                                               x+j_coord_offsetC,x+j_coord_offsetD,
 973                                               &jx0,&jy0,&jz0);
 974
 975             /* Calculate displacement vector */
 976             dx00             = _mm_sub_ps(ix0,jx0);
 977             dy00             = _mm_sub_ps(iy0,jy0);
 978             dz00             = _mm_sub_ps(iz0,jz0);
 979             dx10             = _mm_sub_ps(ix1,jx0);
 980             dy10             = _mm_sub_ps(iy1,jy0);
 981             dz10             = _mm_sub_ps(iz1,jz0);
 982             dx20             = _mm_sub_ps(ix2,jx0);
 983             dy20             = _mm_sub_ps(iy2,jy0);
 984             dz20             = _mm_sub_ps(iz2,jz0);
 985
 986             /* Calculate squared distance and things based on it */
 987             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 988             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 989             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 990
 991             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 992             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 993             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 994
 995             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 996             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 997             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 998
 999             /* Load parameters for j particles */
1000             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1001                                                               charge+jnrC+0,charge+jnrD+0);
1002             vdwjidx0A        = 2*vdwtype[jnrA+0];
1003             vdwjidx0B        = 2*vdwtype[jnrB+0];
1004             vdwjidx0C        = 2*vdwtype[jnrC+0];
1005             vdwjidx0D        = 2*vdwtype[jnrD+0];
1006
1007             fjx0             = _mm_setzero_ps();
1008             fjy0             = _mm_setzero_ps();
1009             fjz0             = _mm_setzero_ps();
1010
1011             /**************************
1012              * CALCULATE INTERACTIONS *
1013              **************************/
1014
1015             if (gmx_mm_any_lt(rsq00,rcutoff2))
1016             {
1017
1018             r00              = _mm_mul_ps(rsq00,rinv00);
1019             r00              = _mm_andnot_ps(dummy_mask,r00);
1020
1021             /* Compute parameters for interactions between i and j atoms */
1022             qq00             = _mm_mul_ps(iq0,jq0);
1023             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1024                                          vdwparam+vdwioffset0+vdwjidx0B,
1025                                          vdwparam+vdwioffset0+vdwjidx0C,
1026                                          vdwparam+vdwioffset0+vdwjidx0D,
1027                                          &c6_00,&c12_00);
1028
1029             /* REACTION-FIELD ELECTROSTATICS */
1030             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1031
1032             /* LENNARD-JONES DISPERSION/REPULSION */
1033
1034             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1035             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1036             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1037             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1038             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1039
1040             d                = _mm_sub_ps(r00,rswitch);
1041             d                = _mm_max_ps(d,_mm_setzero_ps());
1042             d2               = _mm_mul_ps(d,d);
1043             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1044
1045             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1046
1047             /* Evaluate switch function */
1048             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1049             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1050             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1051
1052             fscal            = _mm_add_ps(felec,fvdw);
1053
1054             fscal            = _mm_and_ps(fscal,cutoff_mask);
1055
1056             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1057
1058              /* Update vectorial force */
1059             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1060             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1061             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1062
1063             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1064             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1065             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1066
1067             }
1068
1069             /**************************
1070              * CALCULATE INTERACTIONS *
1071              **************************/
1072
1073             if (gmx_mm_any_lt(rsq10,rcutoff2))
1074             {
1075
1076             /* Compute parameters for interactions between i and j atoms */
1077             qq10             = _mm_mul_ps(iq1,jq0);
1078
1079             /* REACTION-FIELD ELECTROSTATICS */
1080             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1081
1082             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1083
1084             fscal            = felec;
1085
1086             fscal            = _mm_and_ps(fscal,cutoff_mask);
1087
1088             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1089
1090              /* Update vectorial force */
1091             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1092             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1093             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1094
1095             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1096             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1097             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1098
1099             }
1100
1101             /**************************
1102              * CALCULATE INTERACTIONS *
1103              **************************/
1104
1105             if (gmx_mm_any_lt(rsq20,rcutoff2))
1106             {
1107
1108             /* Compute parameters for interactions between i and j atoms */
1109             qq20             = _mm_mul_ps(iq2,jq0);
1110
1111             /* REACTION-FIELD ELECTROSTATICS */
1112             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1113
1114             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1115
1116             fscal            = felec;
1117
1118             fscal            = _mm_and_ps(fscal,cutoff_mask);
1119
1120             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1121
1122              /* Update vectorial force */
1123             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1124             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1125             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1126
1127             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1128             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1129             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1130
1131             }
1132
1133             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1134             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1135             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1136             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1137
1138             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1139
1140             /* Inner loop uses 131 flops */
1141         }
1142
1143         /* End of innermost loop */
1144
1145         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1146                                               f+i_coord_offset,fshift+i_shift_offset);
1147
1148         /* Increment number of inner iterations */
1149         inneriter                  += j_index_end - j_index_start;
1150
1151         /* Outer loop uses 18 flops */
1152     }
1153
1154     /* Increment number of outer iterations */
1155     outeriter        += nri;
1156
1157     /* Update outer/inner flops */
1158
1159     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*131);
1160 }