src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_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_ElecEwSw_VdwLJSw_GeomW4P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_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              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          ewitab;
  92     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  93     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  94     real             *ewtab;
  95     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  96     real             rswitch_scalar,d_scalar;
  97     __m128           dummy_mask,cutoff_mask;
  98     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  99     __m128           one     = _mm_set1_ps(1.0);
 100     __m128           two     = _mm_set1_ps(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_ps(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114     nvdwtype         = fr->ntype;
 115     vdwparam         = fr->nbfp;
 116     vdwtype          = mdatoms->typeA;
 117
 118     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 119     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 120     beta2            = _mm_mul_ps(beta,beta);
 121     beta3            = _mm_mul_ps(beta,beta2);
 122     ewtab            = fr->ic->tabq_coul_FDV0;
 123     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 124     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 129     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 130     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 131     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 132
 133     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 134     rcutoff_scalar   = fr->rcoulomb;
 135     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 136     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 137
 138     rswitch_scalar   = fr->rcoulomb_switch;
 139     rswitch          = _mm_set1_ps(rswitch_scalar);
 140     /* Setup switch parameters */
 141     d_scalar         = rcutoff_scalar-rswitch_scalar;
 142     d                = _mm_set1_ps(d_scalar);
 143     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 144     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 145     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 146     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 147     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 148     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 149
 150     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = jnrC = jnrD = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154     j_coord_offsetC = 0;
 155     j_coord_offsetD = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 182
 183         fix0             = _mm_setzero_ps();
 184         fiy0             = _mm_setzero_ps();
 185         fiz0             = _mm_setzero_ps();
 186         fix1             = _mm_setzero_ps();
 187         fiy1             = _mm_setzero_ps();
 188         fiz1             = _mm_setzero_ps();
 189         fix2             = _mm_setzero_ps();
 190         fiy2             = _mm_setzero_ps();
 191         fiz2             = _mm_setzero_ps();
 192         fix3             = _mm_setzero_ps();
 193         fiy3             = _mm_setzero_ps();
 194         fiz3             = _mm_setzero_ps();
 195
 196         /* Reset potential sums */
 197         velecsum         = _mm_setzero_ps();
 198         vvdwsum          = _mm_setzero_ps();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             jnrC             = jjnr[jidx+2];
 208             jnrD             = jjnr[jidx+3];
 209             j_coord_offsetA  = DIM*jnrA;
 210             j_coord_offsetB  = DIM*jnrB;
 211             j_coord_offsetC  = DIM*jnrC;
 212             j_coord_offsetD  = DIM*jnrD;
 213
 214             /* load j atom coordinates */
 215             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                               x+j_coord_offsetC,x+j_coord_offsetD,
 217                                               &jx0,&jy0,&jz0);
 218
 219             /* Calculate displacement vector */
 220             dx00             = _mm_sub_ps(ix0,jx0);
 221             dy00             = _mm_sub_ps(iy0,jy0);
 222             dz00             = _mm_sub_ps(iz0,jz0);
 223             dx10             = _mm_sub_ps(ix1,jx0);
 224             dy10             = _mm_sub_ps(iy1,jy0);
 225             dz10             = _mm_sub_ps(iz1,jz0);
 226             dx20             = _mm_sub_ps(ix2,jx0);
 227             dy20             = _mm_sub_ps(iy2,jy0);
 228             dz20             = _mm_sub_ps(iz2,jz0);
 229             dx30             = _mm_sub_ps(ix3,jx0);
 230             dy30             = _mm_sub_ps(iy3,jy0);
 231             dz30             = _mm_sub_ps(iz3,jz0);
 232
 233             /* Calculate squared distance and things based on it */
 234             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 235             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 236             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 237             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 238
 239             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 240             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 241             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 242             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 243
 244             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 245             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 246             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 247             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 248
 249             /* Load parameters for j particles */
 250             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 251                                                               charge+jnrC+0,charge+jnrD+0);
 252             vdwjidx0A        = 2*vdwtype[jnrA+0];
 253             vdwjidx0B        = 2*vdwtype[jnrB+0];
 254             vdwjidx0C        = 2*vdwtype[jnrC+0];
 255             vdwjidx0D        = 2*vdwtype[jnrD+0];
 256
 257             fjx0             = _mm_setzero_ps();
 258             fjy0             = _mm_setzero_ps();
 259             fjz0             = _mm_setzero_ps();
 260
 261             /**************************
 262              * CALCULATE INTERACTIONS *
 263              **************************/
 264
 265             if (gmx_mm_any_lt(rsq00,rcutoff2))
 266             {
 267
 268             r00              = _mm_mul_ps(rsq00,rinv00);
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 272                                          vdwparam+vdwioffset0+vdwjidx0B,
 273                                          vdwparam+vdwioffset0+vdwjidx0C,
 274                                          vdwparam+vdwioffset0+vdwjidx0D,
 275                                          &c6_00,&c12_00);
 276
 277             /* LENNARD-JONES DISPERSION/REPULSION */
 278
 279             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 280             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 281             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 282             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 283             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 284
 285             d                = _mm_sub_ps(r00,rswitch);
 286             d                = _mm_max_ps(d,_mm_setzero_ps());
 287             d2               = _mm_mul_ps(d,d);
 288             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 289
 290             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 291
 292             /* Evaluate switch function */
 293             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 294             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 295             vvdw             = _mm_mul_ps(vvdw,sw);
 296             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 300             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 301
 302             fscal            = fvdw;
 303
 304             fscal            = _mm_and_ps(fscal,cutoff_mask);
 305
 306              /* Update vectorial force */
 307             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 308             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 309             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 310
 311             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 312             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 313             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 314
 315             }
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             if (gmx_mm_any_lt(rsq10,rcutoff2))
 322             {
 323
 324             r10              = _mm_mul_ps(rsq10,rinv10);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq10             = _mm_mul_ps(iq1,jq0);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Analytical PME correction */
 332             zeta2            = _mm_mul_ps(beta2,rsq10);
 333             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 334             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 335             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 336             felec            = _mm_mul_ps(qq10,felec);
 337             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 338             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 339             velec            = _mm_mul_ps(qq10,velec);
 340
 341             d                = _mm_sub_ps(r10,rswitch);
 342             d                = _mm_max_ps(d,_mm_setzero_ps());
 343             d2               = _mm_mul_ps(d,d);
 344             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 345
 346             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 347
 348             /* Evaluate switch function */
 349             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 350             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 351             velec            = _mm_mul_ps(velec,sw);
 352             cutoff_mask      = _mm_cmplt_ps(rsq10,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             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 364             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 365             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 366
 367             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 368             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 369             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 370
 371             }
 372
 373             /**************************
 374              * CALCULATE INTERACTIONS *
 375              **************************/
 376
 377             if (gmx_mm_any_lt(rsq20,rcutoff2))
 378             {
 379
 380             r20              = _mm_mul_ps(rsq20,rinv20);
 381
 382             /* Compute parameters for interactions between i and j atoms */
 383             qq20             = _mm_mul_ps(iq2,jq0);
 384
 385             /* EWALD ELECTROSTATICS */
 386
 387             /* Analytical PME correction */
 388             zeta2            = _mm_mul_ps(beta2,rsq20);
 389             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 390             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 391             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 392             felec            = _mm_mul_ps(qq20,felec);
 393             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 394             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 395             velec            = _mm_mul_ps(qq20,velec);
 396
 397             d                = _mm_sub_ps(r20,rswitch);
 398             d                = _mm_max_ps(d,_mm_setzero_ps());
 399             d2               = _mm_mul_ps(d,d);
 400             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 401
 402             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 403
 404             /* Evaluate switch function */
 405             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 406             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 407             velec            = _mm_mul_ps(velec,sw);
 408             cutoff_mask      = _mm_cmplt_ps(rsq20,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             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 420             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 421             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 422
 423             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 424             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 425             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 426
 427             }
 428
 429             /**************************
 430              * CALCULATE INTERACTIONS *
 431              **************************/
 432
 433             if (gmx_mm_any_lt(rsq30,rcutoff2))
 434             {
 435
 436             r30              = _mm_mul_ps(rsq30,rinv30);
 437
 438             /* Compute parameters for interactions between i and j atoms */
 439             qq30             = _mm_mul_ps(iq3,jq0);
 440
 441             /* EWALD ELECTROSTATICS */
 442
 443             /* Analytical PME correction */
 444             zeta2            = _mm_mul_ps(beta2,rsq30);
 445             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 446             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 447             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 448             felec            = _mm_mul_ps(qq30,felec);
 449             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 450             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 451             velec            = _mm_mul_ps(qq30,velec);
 452
 453             d                = _mm_sub_ps(r30,rswitch);
 454             d                = _mm_max_ps(d,_mm_setzero_ps());
 455             d2               = _mm_mul_ps(d,d);
 456             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 457
 458             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 459
 460             /* Evaluate switch function */
 461             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 462             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
 463             velec            = _mm_mul_ps(velec,sw);
 464             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 465
 466             /* Update potential sum for this i atom from the interaction with this j atom. */
 467             velec            = _mm_and_ps(velec,cutoff_mask);
 468             velecsum         = _mm_add_ps(velecsum,velec);
 469
 470             fscal            = felec;
 471
 472             fscal            = _mm_and_ps(fscal,cutoff_mask);
 473
 474              /* Update vectorial force */
 475             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 476             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 477             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 478
 479             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 480             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 481             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 482
 483             }
 484
 485             fjptrA             = f+j_coord_offsetA;
 486             fjptrB             = f+j_coord_offsetB;
 487             fjptrC             = f+j_coord_offsetC;
 488             fjptrD             = f+j_coord_offsetD;
 489
 490             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 491
 492             /* Inner loop uses 221 flops */
 493         }
 494
 495         if(jidx<j_index_end)
 496         {
 497
 498             /* Get j neighbor index, and coordinate index */
 499             jnrlistA         = jjnr[jidx];
 500             jnrlistB         = jjnr[jidx+1];
 501             jnrlistC         = jjnr[jidx+2];
 502             jnrlistD         = jjnr[jidx+3];
 503             /* Sign of each element will be negative for non-real atoms.
 504              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 505              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 506              */
 507             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 508             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 509             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 510             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 511             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 512             j_coord_offsetA  = DIM*jnrA;
 513             j_coord_offsetB  = DIM*jnrB;
 514             j_coord_offsetC  = DIM*jnrC;
 515             j_coord_offsetD  = DIM*jnrD;
 516
 517             /* load j atom coordinates */
 518             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 519                                               x+j_coord_offsetC,x+j_coord_offsetD,
 520                                               &jx0,&jy0,&jz0);
 521
 522             /* Calculate displacement vector */
 523             dx00             = _mm_sub_ps(ix0,jx0);
 524             dy00             = _mm_sub_ps(iy0,jy0);
 525             dz00             = _mm_sub_ps(iz0,jz0);
 526             dx10             = _mm_sub_ps(ix1,jx0);
 527             dy10             = _mm_sub_ps(iy1,jy0);
 528             dz10             = _mm_sub_ps(iz1,jz0);
 529             dx20             = _mm_sub_ps(ix2,jx0);
 530             dy20             = _mm_sub_ps(iy2,jy0);
 531             dz20             = _mm_sub_ps(iz2,jz0);
 532             dx30             = _mm_sub_ps(ix3,jx0);
 533             dy30             = _mm_sub_ps(iy3,jy0);
 534             dz30             = _mm_sub_ps(iz3,jz0);
 535
 536             /* Calculate squared distance and things based on it */
 537             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 538             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 539             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 540             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 541
 542             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 543             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 544             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 545             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 546
 547             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 548             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 549             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 550             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 551
 552             /* Load parameters for j particles */
 553             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 554                                                               charge+jnrC+0,charge+jnrD+0);
 555             vdwjidx0A        = 2*vdwtype[jnrA+0];
 556             vdwjidx0B        = 2*vdwtype[jnrB+0];
 557             vdwjidx0C        = 2*vdwtype[jnrC+0];
 558             vdwjidx0D        = 2*vdwtype[jnrD+0];
 559
 560             fjx0             = _mm_setzero_ps();
 561             fjy0             = _mm_setzero_ps();
 562             fjz0             = _mm_setzero_ps();
 563
 564             /**************************
 565              * CALCULATE INTERACTIONS *
 566              **************************/
 567
 568             if (gmx_mm_any_lt(rsq00,rcutoff2))
 569             {
 570
 571             r00              = _mm_mul_ps(rsq00,rinv00);
 572             r00              = _mm_andnot_ps(dummy_mask,r00);
 573
 574             /* Compute parameters for interactions between i and j atoms */
 575             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 576                                          vdwparam+vdwioffset0+vdwjidx0B,
 577                                          vdwparam+vdwioffset0+vdwjidx0C,
 578                                          vdwparam+vdwioffset0+vdwjidx0D,
 579                                          &c6_00,&c12_00);
 580
 581             /* LENNARD-JONES DISPERSION/REPULSION */
 582
 583             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 584             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 585             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 586             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 587             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 588
 589             d                = _mm_sub_ps(r00,rswitch);
 590             d                = _mm_max_ps(d,_mm_setzero_ps());
 591             d2               = _mm_mul_ps(d,d);
 592             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 593
 594             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 595
 596             /* Evaluate switch function */
 597             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 598             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 599             vvdw             = _mm_mul_ps(vvdw,sw);
 600             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 601
 602             /* Update potential sum for this i atom from the interaction with this j atom. */
 603             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 604             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 605             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 606
 607             fscal            = fvdw;
 608
 609             fscal            = _mm_and_ps(fscal,cutoff_mask);
 610
 611             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 612
 613              /* Update vectorial force */
 614             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 615             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 616             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 617
 618             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 619             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 620             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 621
 622             }
 623
 624             /**************************
 625              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             if (gmx_mm_any_lt(rsq10,rcutoff2))
 629             {
 630
 631             r10              = _mm_mul_ps(rsq10,rinv10);
 632             r10              = _mm_andnot_ps(dummy_mask,r10);
 633
 634             /* Compute parameters for interactions between i and j atoms */
 635             qq10             = _mm_mul_ps(iq1,jq0);
 636
 637             /* EWALD ELECTROSTATICS */
 638
 639             /* Analytical PME correction */
 640             zeta2            = _mm_mul_ps(beta2,rsq10);
 641             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 642             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 643             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 644             felec            = _mm_mul_ps(qq10,felec);
 645             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 646             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 647             velec            = _mm_mul_ps(qq10,velec);
 648
 649             d                = _mm_sub_ps(r10,rswitch);
 650             d                = _mm_max_ps(d,_mm_setzero_ps());
 651             d2               = _mm_mul_ps(d,d);
 652             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 653
 654             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 655
 656             /* Evaluate switch function */
 657             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 658             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 659             velec            = _mm_mul_ps(velec,sw);
 660             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 661
 662             /* Update potential sum for this i atom from the interaction with this j atom. */
 663             velec            = _mm_and_ps(velec,cutoff_mask);
 664             velec            = _mm_andnot_ps(dummy_mask,velec);
 665             velecsum         = _mm_add_ps(velecsum,velec);
 666
 667             fscal            = felec;
 668
 669             fscal            = _mm_and_ps(fscal,cutoff_mask);
 670
 671             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 672
 673              /* Update vectorial force */
 674             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 675             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 676             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 677
 678             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 679             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 680             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 681
 682             }
 683
 684             /**************************
 685              * CALCULATE INTERACTIONS *
 686              **************************/
 687
 688             if (gmx_mm_any_lt(rsq20,rcutoff2))
 689             {
 690
 691             r20              = _mm_mul_ps(rsq20,rinv20);
 692             r20              = _mm_andnot_ps(dummy_mask,r20);
 693
 694             /* Compute parameters for interactions between i and j atoms */
 695             qq20             = _mm_mul_ps(iq2,jq0);
 696
 697             /* EWALD ELECTROSTATICS */
 698
 699             /* Analytical PME correction */
 700             zeta2            = _mm_mul_ps(beta2,rsq20);
 701             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 702             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 703             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 704             felec            = _mm_mul_ps(qq20,felec);
 705             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 706             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 707             velec            = _mm_mul_ps(qq20,velec);
 708
 709             d                = _mm_sub_ps(r20,rswitch);
 710             d                = _mm_max_ps(d,_mm_setzero_ps());
 711             d2               = _mm_mul_ps(d,d);
 712             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 713
 714             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 715
 716             /* Evaluate switch function */
 717             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 718             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 719             velec            = _mm_mul_ps(velec,sw);
 720             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 721
 722             /* Update potential sum for this i atom from the interaction with this j atom. */
 723             velec            = _mm_and_ps(velec,cutoff_mask);
 724             velec            = _mm_andnot_ps(dummy_mask,velec);
 725             velecsum         = _mm_add_ps(velecsum,velec);
 726
 727             fscal            = felec;
 728
 729             fscal            = _mm_and_ps(fscal,cutoff_mask);
 730
 731             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 732
 733              /* Update vectorial force */
 734             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 735             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 736             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 737
 738             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 739             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 740             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 741
 742             }
 743
 744             /**************************
 745              * CALCULATE INTERACTIONS *
 746              **************************/
 747
 748             if (gmx_mm_any_lt(rsq30,rcutoff2))
 749             {
 750
 751             r30              = _mm_mul_ps(rsq30,rinv30);
 752             r30              = _mm_andnot_ps(dummy_mask,r30);
 753
 754             /* Compute parameters for interactions between i and j atoms */
 755             qq30             = _mm_mul_ps(iq3,jq0);
 756
 757             /* EWALD ELECTROSTATICS */
 758
 759             /* Analytical PME correction */
 760             zeta2            = _mm_mul_ps(beta2,rsq30);
 761             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 762             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 763             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 764             felec            = _mm_mul_ps(qq30,felec);
 765             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 766             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 767             velec            = _mm_mul_ps(qq30,velec);
 768
 769             d                = _mm_sub_ps(r30,rswitch);
 770             d                = _mm_max_ps(d,_mm_setzero_ps());
 771             d2               = _mm_mul_ps(d,d);
 772             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 773
 774             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 775
 776             /* Evaluate switch function */
 777             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 778             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
 779             velec            = _mm_mul_ps(velec,sw);
 780             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 781
 782             /* Update potential sum for this i atom from the interaction with this j atom. */
 783             velec            = _mm_and_ps(velec,cutoff_mask);
 784             velec            = _mm_andnot_ps(dummy_mask,velec);
 785             velecsum         = _mm_add_ps(velecsum,velec);
 786
 787             fscal            = felec;
 788
 789             fscal            = _mm_and_ps(fscal,cutoff_mask);
 790
 791             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 792
 793              /* Update vectorial force */
 794             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 795             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 796             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 797
 798             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 799             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 800             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 801
 802             }
 803
 804             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 805             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 806             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 807             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 808
 809             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 810
 811             /* Inner loop uses 225 flops */
 812         }
 813
 814         /* End of innermost loop */
 815
 816         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 817                                               f+i_coord_offset,fshift+i_shift_offset);
 818
 819         ggid                        = gid[iidx];
 820         /* Update potential energies */
 821         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 822         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 823
 824         /* Increment number of inner iterations */
 825         inneriter                  += j_index_end - j_index_start;
 826
 827         /* Outer loop uses 26 flops */
 828     }
 829
 830     /* Increment number of outer iterations */
 831     outeriter        += nri;
 832
 833     /* Update outer/inner flops */
 834
 835     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*225);
 836 }
 837 /*
 838  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_128_fma_single
 839  * Electrostatics interaction: Ewald
 840  * VdW interaction:            LennardJones
 841  * Geometry:                   Water4-Particle
 842  * Calculate force/pot:        Force
 843  */
 844 void
 845 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_128_fma_single
 846                     (t_nblist * gmx_restrict                nlist,
 847                      rvec * gmx_restrict                    xx,
 848                      rvec * gmx_restrict                    ff,
 849                      t_forcerec * gmx_restrict              fr,
 850                      t_mdatoms * gmx_restrict               mdatoms,
 851                      nb_kernel_data_t * gmx_restrict        kernel_data,
 852                      t_nrnb * gmx_restrict                  nrnb)
 853 {
 854     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 855      * just 0 for non-waters.
 856      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 857      * jnr indices corresponding to data put in the four positions in the SIMD register.
 858      */
 859     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 860     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 861     int              jnrA,jnrB,jnrC,jnrD;
 862     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 863     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 864     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 865     real             rcutoff_scalar;
 866     real             *shiftvec,*fshift,*x,*f;
 867     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 868     real             scratch[4*DIM];
 869     __m128           fscal,rcutoff,rcutoff2,jidxall;
 870     int              vdwioffset0;
 871     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 872     int              vdwioffset1;
 873     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 874     int              vdwioffset2;
 875     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 876     int              vdwioffset3;
 877     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 878     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 879     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 880     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 881     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 882     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 883     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 884     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 885     real             *charge;
 886     int              nvdwtype;
 887     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 888     int              *vdwtype;
 889     real             *vdwparam;
 890     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 891     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 892     __m128i          ewitab;
 893     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 894     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 895     real             *ewtab;
 896     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 897     real             rswitch_scalar,d_scalar;
 898     __m128           dummy_mask,cutoff_mask;
 899     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 900     __m128           one     = _mm_set1_ps(1.0);
 901     __m128           two     = _mm_set1_ps(2.0);
 902     x                = xx[0];
 903     f                = ff[0];
 904
 905     nri              = nlist->nri;
 906     iinr             = nlist->iinr;
 907     jindex           = nlist->jindex;
 908     jjnr             = nlist->jjnr;
 909     shiftidx         = nlist->shift;
 910     gid              = nlist->gid;
 911     shiftvec         = fr->shift_vec[0];
 912     fshift           = fr->fshift[0];
 913     facel            = _mm_set1_ps(fr->epsfac);
 914     charge           = mdatoms->chargeA;
 915     nvdwtype         = fr->ntype;
 916     vdwparam         = fr->nbfp;
 917     vdwtype          = mdatoms->typeA;
 918
 919     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 920     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 921     beta2            = _mm_mul_ps(beta,beta);
 922     beta3            = _mm_mul_ps(beta,beta2);
 923     ewtab            = fr->ic->tabq_coul_FDV0;
 924     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 925     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 926
 927     /* Setup water-specific parameters */
 928     inr              = nlist->iinr[0];
 929     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 930     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 931     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 932     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 933
 934     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 935     rcutoff_scalar   = fr->rcoulomb;
 936     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 937     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 938
 939     rswitch_scalar   = fr->rcoulomb_switch;
 940     rswitch          = _mm_set1_ps(rswitch_scalar);
 941     /* Setup switch parameters */
 942     d_scalar         = rcutoff_scalar-rswitch_scalar;
 943     d                = _mm_set1_ps(d_scalar);
 944     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 945     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 946     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 947     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 948     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 949     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 950
 951     /* Avoid stupid compiler warnings */
 952     jnrA = jnrB = jnrC = jnrD = 0;
 953     j_coord_offsetA = 0;
 954     j_coord_offsetB = 0;
 955     j_coord_offsetC = 0;
 956     j_coord_offsetD = 0;
 957
 958     outeriter        = 0;
 959     inneriter        = 0;
 960
 961     for(iidx=0;iidx<4*DIM;iidx++)
 962     {
 963         scratch[iidx] = 0.0;
 964     }
 965
 966     /* Start outer loop over neighborlists */
 967     for(iidx=0; iidx<nri; iidx++)
 968     {
 969         /* Load shift vector for this list */
 970         i_shift_offset   = DIM*shiftidx[iidx];
 971
 972         /* Load limits for loop over neighbors */
 973         j_index_start    = jindex[iidx];
 974         j_index_end      = jindex[iidx+1];
 975
 976         /* Get outer coordinate index */
 977         inr              = iinr[iidx];
 978         i_coord_offset   = DIM*inr;
 979
 980         /* Load i particle coords and add shift vector */
 981         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 982                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 983
 984         fix0             = _mm_setzero_ps();
 985         fiy0             = _mm_setzero_ps();
 986         fiz0             = _mm_setzero_ps();
 987         fix1             = _mm_setzero_ps();
 988         fiy1             = _mm_setzero_ps();
 989         fiz1             = _mm_setzero_ps();
 990         fix2             = _mm_setzero_ps();
 991         fiy2             = _mm_setzero_ps();
 992         fiz2             = _mm_setzero_ps();
 993         fix3             = _mm_setzero_ps();
 994         fiy3             = _mm_setzero_ps();
 995         fiz3             = _mm_setzero_ps();
 996
 997         /* Start inner kernel loop */
 998         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 999         {
1000
1001             /* Get j neighbor index, and coordinate index */
1002             jnrA             = jjnr[jidx];
1003             jnrB             = jjnr[jidx+1];
1004             jnrC             = jjnr[jidx+2];
1005             jnrD             = jjnr[jidx+3];
1006             j_coord_offsetA  = DIM*jnrA;
1007             j_coord_offsetB  = DIM*jnrB;
1008             j_coord_offsetC  = DIM*jnrC;
1009             j_coord_offsetD  = DIM*jnrD;
1010
1011             /* load j atom coordinates */
1012             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1013                                               x+j_coord_offsetC,x+j_coord_offsetD,
1014                                               &jx0,&jy0,&jz0);
1015
1016             /* Calculate displacement vector */
1017             dx00             = _mm_sub_ps(ix0,jx0);
1018             dy00             = _mm_sub_ps(iy0,jy0);
1019             dz00             = _mm_sub_ps(iz0,jz0);
1020             dx10             = _mm_sub_ps(ix1,jx0);
1021             dy10             = _mm_sub_ps(iy1,jy0);
1022             dz10             = _mm_sub_ps(iz1,jz0);
1023             dx20             = _mm_sub_ps(ix2,jx0);
1024             dy20             = _mm_sub_ps(iy2,jy0);
1025             dz20             = _mm_sub_ps(iz2,jz0);
1026             dx30             = _mm_sub_ps(ix3,jx0);
1027             dy30             = _mm_sub_ps(iy3,jy0);
1028             dz30             = _mm_sub_ps(iz3,jz0);
1029
1030             /* Calculate squared distance and things based on it */
1031             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1032             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1033             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1034             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1035
1036             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1037             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1038             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1039             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1040
1041             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1042             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1043             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1044             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1045
1046             /* Load parameters for j particles */
1047             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1048                                                               charge+jnrC+0,charge+jnrD+0);
1049             vdwjidx0A        = 2*vdwtype[jnrA+0];
1050             vdwjidx0B        = 2*vdwtype[jnrB+0];
1051             vdwjidx0C        = 2*vdwtype[jnrC+0];
1052             vdwjidx0D        = 2*vdwtype[jnrD+0];
1053
1054             fjx0             = _mm_setzero_ps();
1055             fjy0             = _mm_setzero_ps();
1056             fjz0             = _mm_setzero_ps();
1057
1058             /**************************
1059              * CALCULATE INTERACTIONS *
1060              **************************/
1061
1062             if (gmx_mm_any_lt(rsq00,rcutoff2))
1063             {
1064
1065             r00              = _mm_mul_ps(rsq00,rinv00);
1066
1067             /* Compute parameters for interactions between i and j atoms */
1068             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1069                                          vdwparam+vdwioffset0+vdwjidx0B,
1070                                          vdwparam+vdwioffset0+vdwjidx0C,
1071                                          vdwparam+vdwioffset0+vdwjidx0D,
1072                                          &c6_00,&c12_00);
1073
1074             /* LENNARD-JONES DISPERSION/REPULSION */
1075
1076             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1077             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1078             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1079             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1080             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1081
1082             d                = _mm_sub_ps(r00,rswitch);
1083             d                = _mm_max_ps(d,_mm_setzero_ps());
1084             d2               = _mm_mul_ps(d,d);
1085             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1086
1087             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1088
1089             /* Evaluate switch function */
1090             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1091             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1092             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1093
1094             fscal            = fvdw;
1095
1096             fscal            = _mm_and_ps(fscal,cutoff_mask);
1097
1098              /* Update vectorial force */
1099             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1100             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1101             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1102
1103             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1104             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1105             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1106
1107             }
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             if (gmx_mm_any_lt(rsq10,rcutoff2))
1114             {
1115
1116             r10              = _mm_mul_ps(rsq10,rinv10);
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq10             = _mm_mul_ps(iq1,jq0);
1120
1121             /* EWALD ELECTROSTATICS */
1122
1123             /* Analytical PME correction */
1124             zeta2            = _mm_mul_ps(beta2,rsq10);
1125             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1126             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1127             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1128             felec            = _mm_mul_ps(qq10,felec);
1129             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1130             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
1131             velec            = _mm_mul_ps(qq10,velec);
1132
1133             d                = _mm_sub_ps(r10,rswitch);
1134             d                = _mm_max_ps(d,_mm_setzero_ps());
1135             d2               = _mm_mul_ps(d,d);
1136             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1137
1138             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1139
1140             /* Evaluate switch function */
1141             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1142             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1143             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1144
1145             fscal            = felec;
1146
1147             fscal            = _mm_and_ps(fscal,cutoff_mask);
1148
1149              /* Update vectorial force */
1150             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1151             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1152             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1153
1154             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1155             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1156             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1157
1158             }
1159
1160             /**************************
1161              * CALCULATE INTERACTIONS *
1162              **************************/
1163
1164             if (gmx_mm_any_lt(rsq20,rcutoff2))
1165             {
1166
1167             r20              = _mm_mul_ps(rsq20,rinv20);
1168
1169             /* Compute parameters for interactions between i and j atoms */
1170             qq20             = _mm_mul_ps(iq2,jq0);
1171
1172             /* EWALD ELECTROSTATICS */
1173
1174             /* Analytical PME correction */
1175             zeta2            = _mm_mul_ps(beta2,rsq20);
1176             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1177             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1178             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1179             felec            = _mm_mul_ps(qq20,felec);
1180             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1181             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1182             velec            = _mm_mul_ps(qq20,velec);
1183
1184             d                = _mm_sub_ps(r20,rswitch);
1185             d                = _mm_max_ps(d,_mm_setzero_ps());
1186             d2               = _mm_mul_ps(d,d);
1187             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1188
1189             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1190
1191             /* Evaluate switch function */
1192             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1193             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1194             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1195
1196             fscal            = felec;
1197
1198             fscal            = _mm_and_ps(fscal,cutoff_mask);
1199
1200              /* Update vectorial force */
1201             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1202             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1203             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1204
1205             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1206             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1207             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1208
1209             }
1210
1211             /**************************
1212              * CALCULATE INTERACTIONS *
1213              **************************/
1214
1215             if (gmx_mm_any_lt(rsq30,rcutoff2))
1216             {
1217
1218             r30              = _mm_mul_ps(rsq30,rinv30);
1219
1220             /* Compute parameters for interactions between i and j atoms */
1221             qq30             = _mm_mul_ps(iq3,jq0);
1222
1223             /* EWALD ELECTROSTATICS */
1224
1225             /* Analytical PME correction */
1226             zeta2            = _mm_mul_ps(beta2,rsq30);
1227             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1228             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1229             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1230             felec            = _mm_mul_ps(qq30,felec);
1231             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1232             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
1233             velec            = _mm_mul_ps(qq30,velec);
1234
1235             d                = _mm_sub_ps(r30,rswitch);
1236             d                = _mm_max_ps(d,_mm_setzero_ps());
1237             d2               = _mm_mul_ps(d,d);
1238             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1239
1240             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1241
1242             /* Evaluate switch function */
1243             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1244             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1245             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1246
1247             fscal            = felec;
1248
1249             fscal            = _mm_and_ps(fscal,cutoff_mask);
1250
1251              /* Update vectorial force */
1252             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1253             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1254             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1255
1256             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1257             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1258             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1259
1260             }
1261
1262             fjptrA             = f+j_coord_offsetA;
1263             fjptrB             = f+j_coord_offsetB;
1264             fjptrC             = f+j_coord_offsetC;
1265             fjptrD             = f+j_coord_offsetD;
1266
1267             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1268
1269             /* Inner loop uses 209 flops */
1270         }
1271
1272         if(jidx<j_index_end)
1273         {
1274
1275             /* Get j neighbor index, and coordinate index */
1276             jnrlistA         = jjnr[jidx];
1277             jnrlistB         = jjnr[jidx+1];
1278             jnrlistC         = jjnr[jidx+2];
1279             jnrlistD         = jjnr[jidx+3];
1280             /* Sign of each element will be negative for non-real atoms.
1281              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1282              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1283              */
1284             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1285             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1286             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1287             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1288             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1289             j_coord_offsetA  = DIM*jnrA;
1290             j_coord_offsetB  = DIM*jnrB;
1291             j_coord_offsetC  = DIM*jnrC;
1292             j_coord_offsetD  = DIM*jnrD;
1293
1294             /* load j atom coordinates */
1295             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1296                                               x+j_coord_offsetC,x+j_coord_offsetD,
1297                                               &jx0,&jy0,&jz0);
1298
1299             /* Calculate displacement vector */
1300             dx00             = _mm_sub_ps(ix0,jx0);
1301             dy00             = _mm_sub_ps(iy0,jy0);
1302             dz00             = _mm_sub_ps(iz0,jz0);
1303             dx10             = _mm_sub_ps(ix1,jx0);
1304             dy10             = _mm_sub_ps(iy1,jy0);
1305             dz10             = _mm_sub_ps(iz1,jz0);
1306             dx20             = _mm_sub_ps(ix2,jx0);
1307             dy20             = _mm_sub_ps(iy2,jy0);
1308             dz20             = _mm_sub_ps(iz2,jz0);
1309             dx30             = _mm_sub_ps(ix3,jx0);
1310             dy30             = _mm_sub_ps(iy3,jy0);
1311             dz30             = _mm_sub_ps(iz3,jz0);
1312
1313             /* Calculate squared distance and things based on it */
1314             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1315             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1316             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1317             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1318
1319             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1320             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1321             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1322             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1323
1324             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1325             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1326             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1327             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1328
1329             /* Load parameters for j particles */
1330             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1331                                                               charge+jnrC+0,charge+jnrD+0);
1332             vdwjidx0A        = 2*vdwtype[jnrA+0];
1333             vdwjidx0B        = 2*vdwtype[jnrB+0];
1334             vdwjidx0C        = 2*vdwtype[jnrC+0];
1335             vdwjidx0D        = 2*vdwtype[jnrD+0];
1336
1337             fjx0             = _mm_setzero_ps();
1338             fjy0             = _mm_setzero_ps();
1339             fjz0             = _mm_setzero_ps();
1340
1341             /**************************
1342              * CALCULATE INTERACTIONS *
1343              **************************/
1344
1345             if (gmx_mm_any_lt(rsq00,rcutoff2))
1346             {
1347
1348             r00              = _mm_mul_ps(rsq00,rinv00);
1349             r00              = _mm_andnot_ps(dummy_mask,r00);
1350
1351             /* Compute parameters for interactions between i and j atoms */
1352             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1353                                          vdwparam+vdwioffset0+vdwjidx0B,
1354                                          vdwparam+vdwioffset0+vdwjidx0C,
1355                                          vdwparam+vdwioffset0+vdwjidx0D,
1356                                          &c6_00,&c12_00);
1357
1358             /* LENNARD-JONES DISPERSION/REPULSION */
1359
1360             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1361             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1362             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1363             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1364             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1365
1366             d                = _mm_sub_ps(r00,rswitch);
1367             d                = _mm_max_ps(d,_mm_setzero_ps());
1368             d2               = _mm_mul_ps(d,d);
1369             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1370
1371             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1372
1373             /* Evaluate switch function */
1374             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1375             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1376             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1377
1378             fscal            = fvdw;
1379
1380             fscal            = _mm_and_ps(fscal,cutoff_mask);
1381
1382             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1383
1384              /* Update vectorial force */
1385             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1386             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1387             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1388
1389             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1390             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1391             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1392
1393             }
1394
1395             /**************************
1396              * CALCULATE INTERACTIONS *
1397              **************************/
1398
1399             if (gmx_mm_any_lt(rsq10,rcutoff2))
1400             {
1401
1402             r10              = _mm_mul_ps(rsq10,rinv10);
1403             r10              = _mm_andnot_ps(dummy_mask,r10);
1404
1405             /* Compute parameters for interactions between i and j atoms */
1406             qq10             = _mm_mul_ps(iq1,jq0);
1407
1408             /* EWALD ELECTROSTATICS */
1409
1410             /* Analytical PME correction */
1411             zeta2            = _mm_mul_ps(beta2,rsq10);
1412             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1413             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1414             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1415             felec            = _mm_mul_ps(qq10,felec);
1416             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1417             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
1418             velec            = _mm_mul_ps(qq10,velec);
1419
1420             d                = _mm_sub_ps(r10,rswitch);
1421             d                = _mm_max_ps(d,_mm_setzero_ps());
1422             d2               = _mm_mul_ps(d,d);
1423             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1424
1425             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1426
1427             /* Evaluate switch function */
1428             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1429             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1430             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1431
1432             fscal            = felec;
1433
1434             fscal            = _mm_and_ps(fscal,cutoff_mask);
1435
1436             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1437
1438              /* Update vectorial force */
1439             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1440             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1441             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1442
1443             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1444             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1445             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1446
1447             }
1448
1449             /**************************
1450              * CALCULATE INTERACTIONS *
1451              **************************/
1452
1453             if (gmx_mm_any_lt(rsq20,rcutoff2))
1454             {
1455
1456             r20              = _mm_mul_ps(rsq20,rinv20);
1457             r20              = _mm_andnot_ps(dummy_mask,r20);
1458
1459             /* Compute parameters for interactions between i and j atoms */
1460             qq20             = _mm_mul_ps(iq2,jq0);
1461
1462             /* EWALD ELECTROSTATICS */
1463
1464             /* Analytical PME correction */
1465             zeta2            = _mm_mul_ps(beta2,rsq20);
1466             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1467             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1468             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1469             felec            = _mm_mul_ps(qq20,felec);
1470             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1471             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1472             velec            = _mm_mul_ps(qq20,velec);
1473
1474             d                = _mm_sub_ps(r20,rswitch);
1475             d                = _mm_max_ps(d,_mm_setzero_ps());
1476             d2               = _mm_mul_ps(d,d);
1477             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1478
1479             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1480
1481             /* Evaluate switch function */
1482             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1483             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1484             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1485
1486             fscal            = felec;
1487
1488             fscal            = _mm_and_ps(fscal,cutoff_mask);
1489
1490             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1491
1492              /* Update vectorial force */
1493             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1494             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1495             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1496
1497             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1498             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1499             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1500
1501             }
1502
1503             /**************************
1504              * CALCULATE INTERACTIONS *
1505              **************************/
1506
1507             if (gmx_mm_any_lt(rsq30,rcutoff2))
1508             {
1509
1510             r30              = _mm_mul_ps(rsq30,rinv30);
1511             r30              = _mm_andnot_ps(dummy_mask,r30);
1512
1513             /* Compute parameters for interactions between i and j atoms */
1514             qq30             = _mm_mul_ps(iq3,jq0);
1515
1516             /* EWALD ELECTROSTATICS */
1517
1518             /* Analytical PME correction */
1519             zeta2            = _mm_mul_ps(beta2,rsq30);
1520             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1521             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1522             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1523             felec            = _mm_mul_ps(qq30,felec);
1524             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1525             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
1526             velec            = _mm_mul_ps(qq30,velec);
1527
1528             d                = _mm_sub_ps(r30,rswitch);
1529             d                = _mm_max_ps(d,_mm_setzero_ps());
1530             d2               = _mm_mul_ps(d,d);
1531             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1532
1533             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1534
1535             /* Evaluate switch function */
1536             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1537             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1538             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1539
1540             fscal            = felec;
1541
1542             fscal            = _mm_and_ps(fscal,cutoff_mask);
1543
1544             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1545
1546              /* Update vectorial force */
1547             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1548             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1549             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1550
1551             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1552             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1553             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1554
1555             }
1556
1557             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1558             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1559             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1560             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1561
1562             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1563
1564             /* Inner loop uses 213 flops */
1565         }
1566
1567         /* End of innermost loop */
1568
1569         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1570                                               f+i_coord_offset,fshift+i_shift_offset);
1571
1572         /* Increment number of inner iterations */
1573         inneriter                  += j_index_end - j_index_start;
1574
1575         /* Outer loop uses 24 flops */
1576     }
1577
1578     /* Increment number of outer iterations */
1579     outeriter        += nri;
1580
1581     /* Update outer/inner flops */
1582
1583     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*213);
1584 }