src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSw_VdwNone_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     __m128i          ewitab;
  97     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  99     real             *ewtab;
 100     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 101     real             rswitch_scalar,d_scalar;
 102     __m128           dummy_mask,cutoff_mask;
 103     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 104     __m128           one     = _mm_set1_ps(1.0);
 105     __m128           two     = _mm_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_ps(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119
 120     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 121     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 122     beta2            = _mm_mul_ps(beta,beta);
 123     beta3            = _mm_mul_ps(beta,beta2);
 124     ewtab            = fr->ic->tabq_coul_FDV0;
 125     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 126     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 131     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 132     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 133
 134     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 135     rcutoff_scalar   = fr->rcoulomb;
 136     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 137     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 138
 139     rswitch_scalar   = fr->rcoulomb_switch;
 140     rswitch          = _mm_set1_ps(rswitch_scalar);
 141     /* Setup switch parameters */
 142     d_scalar         = rcutoff_scalar-rswitch_scalar;
 143     d                = _mm_set1_ps(d_scalar);
 144     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 145     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 146     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 147     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 148     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 149     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = jnrC = jnrD = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155     j_coord_offsetC = 0;
 156     j_coord_offsetD = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     for(iidx=0;iidx<4*DIM;iidx++)
 162     {
 163         scratch[iidx] = 0.0;
 164     }
 165
 166     /* Start outer loop over neighborlists */
 167     for(iidx=0; iidx<nri; iidx++)
 168     {
 169         /* Load shift vector for this list */
 170         i_shift_offset   = DIM*shiftidx[iidx];
 171
 172         /* Load limits for loop over neighbors */
 173         j_index_start    = jindex[iidx];
 174         j_index_end      = jindex[iidx+1];
 175
 176         /* Get outer coordinate index */
 177         inr              = iinr[iidx];
 178         i_coord_offset   = DIM*inr;
 179
 180         /* Load i particle coords and add shift vector */
 181         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 182                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 183
 184         fix0             = _mm_setzero_ps();
 185         fiy0             = _mm_setzero_ps();
 186         fiz0             = _mm_setzero_ps();
 187         fix1             = _mm_setzero_ps();
 188         fiy1             = _mm_setzero_ps();
 189         fiz1             = _mm_setzero_ps();
 190         fix2             = _mm_setzero_ps();
 191         fiy2             = _mm_setzero_ps();
 192         fiz2             = _mm_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm_setzero_ps();
 196
 197         /* Start inner kernel loop */
 198         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 199         {
 200
 201             /* Get j neighbor index, and coordinate index */
 202             jnrA             = jjnr[jidx];
 203             jnrB             = jjnr[jidx+1];
 204             jnrC             = jjnr[jidx+2];
 205             jnrD             = jjnr[jidx+3];
 206             j_coord_offsetA  = DIM*jnrA;
 207             j_coord_offsetB  = DIM*jnrB;
 208             j_coord_offsetC  = DIM*jnrC;
 209             j_coord_offsetD  = DIM*jnrD;
 210
 211             /* load j atom coordinates */
 212             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 213                                               x+j_coord_offsetC,x+j_coord_offsetD,
 214                                               &jx0,&jy0,&jz0);
 215
 216             /* Calculate displacement vector */
 217             dx00             = _mm_sub_ps(ix0,jx0);
 218             dy00             = _mm_sub_ps(iy0,jy0);
 219             dz00             = _mm_sub_ps(iz0,jz0);
 220             dx10             = _mm_sub_ps(ix1,jx0);
 221             dy10             = _mm_sub_ps(iy1,jy0);
 222             dz10             = _mm_sub_ps(iz1,jz0);
 223             dx20             = _mm_sub_ps(ix2,jx0);
 224             dy20             = _mm_sub_ps(iy2,jy0);
 225             dz20             = _mm_sub_ps(iz2,jz0);
 226
 227             /* Calculate squared distance and things based on it */
 228             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 229             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 230             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 231
 232             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 233             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 234             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 235
 236             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 237             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 238             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 239
 240             /* Load parameters for j particles */
 241             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 242                                                               charge+jnrC+0,charge+jnrD+0);
 243
 244             fjx0             = _mm_setzero_ps();
 245             fjy0             = _mm_setzero_ps();
 246             fjz0             = _mm_setzero_ps();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             if (gmx_mm_any_lt(rsq00,rcutoff2))
 253             {
 254
 255             r00              = _mm_mul_ps(rsq00,rinv00);
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             qq00             = _mm_mul_ps(iq0,jq0);
 259
 260             /* EWALD ELECTROSTATICS */
 261
 262             /* Analytical PME correction */
 263             zeta2            = _mm_mul_ps(beta2,rsq00);
 264             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 265             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 266             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 267             felec            = _mm_mul_ps(qq00,felec);
 268             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 269             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 270             velec            = _mm_mul_ps(qq00,velec);
 271
 272             d                = _mm_sub_ps(r00,rswitch);
 273             d                = _mm_max_ps(d,_mm_setzero_ps());
 274             d2               = _mm_mul_ps(d,d);
 275             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 276
 277             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 278
 279             /* Evaluate switch function */
 280             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 281             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 282             velec            = _mm_mul_ps(velec,sw);
 283             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 284
 285             /* Update potential sum for this i atom from the interaction with this j atom. */
 286             velec            = _mm_and_ps(velec,cutoff_mask);
 287             velecsum         = _mm_add_ps(velecsum,velec);
 288
 289             fscal            = felec;
 290
 291             fscal            = _mm_and_ps(fscal,cutoff_mask);
 292
 293              /* Update vectorial force */
 294             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 295             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 296             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 297
 298             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 299             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 300             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 301
 302             }
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             if (gmx_mm_any_lt(rsq10,rcutoff2))
 309             {
 310
 311             r10              = _mm_mul_ps(rsq10,rinv10);
 312
 313             /* Compute parameters for interactions between i and j atoms */
 314             qq10             = _mm_mul_ps(iq1,jq0);
 315
 316             /* EWALD ELECTROSTATICS */
 317
 318             /* Analytical PME correction */
 319             zeta2            = _mm_mul_ps(beta2,rsq10);
 320             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 321             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 322             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 323             felec            = _mm_mul_ps(qq10,felec);
 324             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 325             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 326             velec            = _mm_mul_ps(qq10,velec);
 327
 328             d                = _mm_sub_ps(r10,rswitch);
 329             d                = _mm_max_ps(d,_mm_setzero_ps());
 330             d2               = _mm_mul_ps(d,d);
 331             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 332
 333             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 334
 335             /* Evaluate switch function */
 336             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 337             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 338             velec            = _mm_mul_ps(velec,sw);
 339             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velec            = _mm_and_ps(velec,cutoff_mask);
 343             velecsum         = _mm_add_ps(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             fscal            = _mm_and_ps(fscal,cutoff_mask);
 348
 349              /* Update vectorial force */
 350             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 351             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 352             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 353
 354             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 355             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 356             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 357
 358             }
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             if (gmx_mm_any_lt(rsq20,rcutoff2))
 365             {
 366
 367             r20              = _mm_mul_ps(rsq20,rinv20);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm_mul_ps(iq2,jq0);
 371
 372             /* EWALD ELECTROSTATICS */
 373
 374             /* Analytical PME correction */
 375             zeta2            = _mm_mul_ps(beta2,rsq20);
 376             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 377             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 378             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 379             felec            = _mm_mul_ps(qq20,felec);
 380             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 381             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 382             velec            = _mm_mul_ps(qq20,velec);
 383
 384             d                = _mm_sub_ps(r20,rswitch);
 385             d                = _mm_max_ps(d,_mm_setzero_ps());
 386             d2               = _mm_mul_ps(d,d);
 387             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 388
 389             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 390
 391             /* Evaluate switch function */
 392             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 393             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 394             velec            = _mm_mul_ps(velec,sw);
 395             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 396
 397             /* Update potential sum for this i atom from the interaction with this j atom. */
 398             velec            = _mm_and_ps(velec,cutoff_mask);
 399             velecsum         = _mm_add_ps(velecsum,velec);
 400
 401             fscal            = felec;
 402
 403             fscal            = _mm_and_ps(fscal,cutoff_mask);
 404
 405              /* Update vectorial force */
 406             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 407             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 408             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 409
 410             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 411             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 412             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 413
 414             }
 415
 416             fjptrA             = f+j_coord_offsetA;
 417             fjptrB             = f+j_coord_offsetB;
 418             fjptrC             = f+j_coord_offsetC;
 419             fjptrD             = f+j_coord_offsetD;
 420
 421             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 422
 423             /* Inner loop uses 159 flops */
 424         }
 425
 426         if(jidx<j_index_end)
 427         {
 428
 429             /* Get j neighbor index, and coordinate index */
 430             jnrlistA         = jjnr[jidx];
 431             jnrlistB         = jjnr[jidx+1];
 432             jnrlistC         = jjnr[jidx+2];
 433             jnrlistD         = jjnr[jidx+3];
 434             /* Sign of each element will be negative for non-real atoms.
 435              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 436              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 437              */
 438             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 439             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 440             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 441             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 442             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 443             j_coord_offsetA  = DIM*jnrA;
 444             j_coord_offsetB  = DIM*jnrB;
 445             j_coord_offsetC  = DIM*jnrC;
 446             j_coord_offsetD  = DIM*jnrD;
 447
 448             /* load j atom coordinates */
 449             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 450                                               x+j_coord_offsetC,x+j_coord_offsetD,
 451                                               &jx0,&jy0,&jz0);
 452
 453             /* Calculate displacement vector */
 454             dx00             = _mm_sub_ps(ix0,jx0);
 455             dy00             = _mm_sub_ps(iy0,jy0);
 456             dz00             = _mm_sub_ps(iz0,jz0);
 457             dx10             = _mm_sub_ps(ix1,jx0);
 458             dy10             = _mm_sub_ps(iy1,jy0);
 459             dz10             = _mm_sub_ps(iz1,jz0);
 460             dx20             = _mm_sub_ps(ix2,jx0);
 461             dy20             = _mm_sub_ps(iy2,jy0);
 462             dz20             = _mm_sub_ps(iz2,jz0);
 463
 464             /* Calculate squared distance and things based on it */
 465             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 466             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 467             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 468
 469             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 470             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 471             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 472
 473             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 474             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 475             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 476
 477             /* Load parameters for j particles */
 478             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 479                                                               charge+jnrC+0,charge+jnrD+0);
 480
 481             fjx0             = _mm_setzero_ps();
 482             fjy0             = _mm_setzero_ps();
 483             fjz0             = _mm_setzero_ps();
 484
 485             /**************************
 486              * CALCULATE INTERACTIONS *
 487              **************************/
 488
 489             if (gmx_mm_any_lt(rsq00,rcutoff2))
 490             {
 491
 492             r00              = _mm_mul_ps(rsq00,rinv00);
 493             r00              = _mm_andnot_ps(dummy_mask,r00);
 494
 495             /* Compute parameters for interactions between i and j atoms */
 496             qq00             = _mm_mul_ps(iq0,jq0);
 497
 498             /* EWALD ELECTROSTATICS */
 499
 500             /* Analytical PME correction */
 501             zeta2            = _mm_mul_ps(beta2,rsq00);
 502             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 503             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 504             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 505             felec            = _mm_mul_ps(qq00,felec);
 506             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 507             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 508             velec            = _mm_mul_ps(qq00,velec);
 509
 510             d                = _mm_sub_ps(r00,rswitch);
 511             d                = _mm_max_ps(d,_mm_setzero_ps());
 512             d2               = _mm_mul_ps(d,d);
 513             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 514
 515             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 516
 517             /* Evaluate switch function */
 518             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 519             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 520             velec            = _mm_mul_ps(velec,sw);
 521             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             velec            = _mm_and_ps(velec,cutoff_mask);
 525             velec            = _mm_andnot_ps(dummy_mask,velec);
 526             velecsum         = _mm_add_ps(velecsum,velec);
 527
 528             fscal            = felec;
 529
 530             fscal            = _mm_and_ps(fscal,cutoff_mask);
 531
 532             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 533
 534              /* Update vectorial force */
 535             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 536             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 537             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 538
 539             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 540             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 541             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 542
 543             }
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             if (gmx_mm_any_lt(rsq10,rcutoff2))
 550             {
 551
 552             r10              = _mm_mul_ps(rsq10,rinv10);
 553             r10              = _mm_andnot_ps(dummy_mask,r10);
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             qq10             = _mm_mul_ps(iq1,jq0);
 557
 558             /* EWALD ELECTROSTATICS */
 559
 560             /* Analytical PME correction */
 561             zeta2            = _mm_mul_ps(beta2,rsq10);
 562             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 563             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 564             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 565             felec            = _mm_mul_ps(qq10,felec);
 566             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 567             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 568             velec            = _mm_mul_ps(qq10,velec);
 569
 570             d                = _mm_sub_ps(r10,rswitch);
 571             d                = _mm_max_ps(d,_mm_setzero_ps());
 572             d2               = _mm_mul_ps(d,d);
 573             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 574
 575             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 576
 577             /* Evaluate switch function */
 578             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 579             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 580             velec            = _mm_mul_ps(velec,sw);
 581             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 582
 583             /* Update potential sum for this i atom from the interaction with this j atom. */
 584             velec            = _mm_and_ps(velec,cutoff_mask);
 585             velec            = _mm_andnot_ps(dummy_mask,velec);
 586             velecsum         = _mm_add_ps(velecsum,velec);
 587
 588             fscal            = felec;
 589
 590             fscal            = _mm_and_ps(fscal,cutoff_mask);
 591
 592             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 593
 594              /* Update vectorial force */
 595             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 596             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 597             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 598
 599             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 600             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 601             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 602
 603             }
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             if (gmx_mm_any_lt(rsq20,rcutoff2))
 610             {
 611
 612             r20              = _mm_mul_ps(rsq20,rinv20);
 613             r20              = _mm_andnot_ps(dummy_mask,r20);
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             qq20             = _mm_mul_ps(iq2,jq0);
 617
 618             /* EWALD ELECTROSTATICS */
 619
 620             /* Analytical PME correction */
 621             zeta2            = _mm_mul_ps(beta2,rsq20);
 622             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 623             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 624             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 625             felec            = _mm_mul_ps(qq20,felec);
 626             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 627             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 628             velec            = _mm_mul_ps(qq20,velec);
 629
 630             d                = _mm_sub_ps(r20,rswitch);
 631             d                = _mm_max_ps(d,_mm_setzero_ps());
 632             d2               = _mm_mul_ps(d,d);
 633             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 634
 635             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 636
 637             /* Evaluate switch function */
 638             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 639             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 640             velec            = _mm_mul_ps(velec,sw);
 641             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 642
 643             /* Update potential sum for this i atom from the interaction with this j atom. */
 644             velec            = _mm_and_ps(velec,cutoff_mask);
 645             velec            = _mm_andnot_ps(dummy_mask,velec);
 646             velecsum         = _mm_add_ps(velecsum,velec);
 647
 648             fscal            = felec;
 649
 650             fscal            = _mm_and_ps(fscal,cutoff_mask);
 651
 652             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 653
 654              /* Update vectorial force */
 655             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 656             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 657             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 658
 659             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 660             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 661             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 662
 663             }
 664
 665             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 666             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 667             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 668             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 669
 670             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 671
 672             /* Inner loop uses 162 flops */
 673         }
 674
 675         /* End of innermost loop */
 676
 677         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 678                                               f+i_coord_offset,fshift+i_shift_offset);
 679
 680         ggid                        = gid[iidx];
 681         /* Update potential energies */
 682         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 683
 684         /* Increment number of inner iterations */
 685         inneriter                  += j_index_end - j_index_start;
 686
 687         /* Outer loop uses 19 flops */
 688     }
 689
 690     /* Increment number of outer iterations */
 691     outeriter        += nri;
 692
 693     /* Update outer/inner flops */
 694
 695     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*162);
 696 }
 697 /*
 698  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_avx_128_fma_single
 699  * Electrostatics interaction: Ewald
 700  * VdW interaction:            None
 701  * Geometry:                   Water3-Particle
 702  * Calculate force/pot:        Force
 703  */
 704 void
 705 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_avx_128_fma_single
 706                     (t_nblist                    * gmx_restrict       nlist,
 707                      rvec                        * gmx_restrict          xx,
 708                      rvec                        * gmx_restrict          ff,
 709                      t_forcerec                  * gmx_restrict          fr,
 710                      t_mdatoms                   * gmx_restrict     mdatoms,
 711                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 712                      t_nrnb                      * gmx_restrict        nrnb)
 713 {
 714     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 715      * just 0 for non-waters.
 716      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 717      * jnr indices corresponding to data put in the four positions in the SIMD register.
 718      */
 719     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 720     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 721     int              jnrA,jnrB,jnrC,jnrD;
 722     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 723     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 724     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 725     real             rcutoff_scalar;
 726     real             *shiftvec,*fshift,*x,*f;
 727     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 728     real             scratch[4*DIM];
 729     __m128           fscal,rcutoff,rcutoff2,jidxall;
 730     int              vdwioffset0;
 731     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 732     int              vdwioffset1;
 733     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 734     int              vdwioffset2;
 735     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 736     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 737     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 738     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 739     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 740     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 741     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 742     real             *charge;
 743     __m128i          ewitab;
 744     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 745     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 746     real             *ewtab;
 747     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 748     real             rswitch_scalar,d_scalar;
 749     __m128           dummy_mask,cutoff_mask;
 750     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 751     __m128           one     = _mm_set1_ps(1.0);
 752     __m128           two     = _mm_set1_ps(2.0);
 753     x                = xx[0];
 754     f                = ff[0];
 755
 756     nri              = nlist->nri;
 757     iinr             = nlist->iinr;
 758     jindex           = nlist->jindex;
 759     jjnr             = nlist->jjnr;
 760     shiftidx         = nlist->shift;
 761     gid              = nlist->gid;
 762     shiftvec         = fr->shift_vec[0];
 763     fshift           = fr->fshift[0];
 764     facel            = _mm_set1_ps(fr->epsfac);
 765     charge           = mdatoms->chargeA;
 766
 767     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 768     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 769     beta2            = _mm_mul_ps(beta,beta);
 770     beta3            = _mm_mul_ps(beta,beta2);
 771     ewtab            = fr->ic->tabq_coul_FDV0;
 772     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 773     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 774
 775     /* Setup water-specific parameters */
 776     inr              = nlist->iinr[0];
 777     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 778     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 779     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 780
 781     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 782     rcutoff_scalar   = fr->rcoulomb;
 783     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 784     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 785
 786     rswitch_scalar   = fr->rcoulomb_switch;
 787     rswitch          = _mm_set1_ps(rswitch_scalar);
 788     /* Setup switch parameters */
 789     d_scalar         = rcutoff_scalar-rswitch_scalar;
 790     d                = _mm_set1_ps(d_scalar);
 791     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 792     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 793     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 794     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 795     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 796     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 797
 798     /* Avoid stupid compiler warnings */
 799     jnrA = jnrB = jnrC = jnrD = 0;
 800     j_coord_offsetA = 0;
 801     j_coord_offsetB = 0;
 802     j_coord_offsetC = 0;
 803     j_coord_offsetD = 0;
 804
 805     outeriter        = 0;
 806     inneriter        = 0;
 807
 808     for(iidx=0;iidx<4*DIM;iidx++)
 809     {
 810         scratch[iidx] = 0.0;
 811     }
 812
 813     /* Start outer loop over neighborlists */
 814     for(iidx=0; iidx<nri; iidx++)
 815     {
 816         /* Load shift vector for this list */
 817         i_shift_offset   = DIM*shiftidx[iidx];
 818
 819         /* Load limits for loop over neighbors */
 820         j_index_start    = jindex[iidx];
 821         j_index_end      = jindex[iidx+1];
 822
 823         /* Get outer coordinate index */
 824         inr              = iinr[iidx];
 825         i_coord_offset   = DIM*inr;
 826
 827         /* Load i particle coords and add shift vector */
 828         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 829                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 830
 831         fix0             = _mm_setzero_ps();
 832         fiy0             = _mm_setzero_ps();
 833         fiz0             = _mm_setzero_ps();
 834         fix1             = _mm_setzero_ps();
 835         fiy1             = _mm_setzero_ps();
 836         fiz1             = _mm_setzero_ps();
 837         fix2             = _mm_setzero_ps();
 838         fiy2             = _mm_setzero_ps();
 839         fiz2             = _mm_setzero_ps();
 840
 841         /* Start inner kernel loop */
 842         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 843         {
 844
 845             /* Get j neighbor index, and coordinate index */
 846             jnrA             = jjnr[jidx];
 847             jnrB             = jjnr[jidx+1];
 848             jnrC             = jjnr[jidx+2];
 849             jnrD             = jjnr[jidx+3];
 850             j_coord_offsetA  = DIM*jnrA;
 851             j_coord_offsetB  = DIM*jnrB;
 852             j_coord_offsetC  = DIM*jnrC;
 853             j_coord_offsetD  = DIM*jnrD;
 854
 855             /* load j atom coordinates */
 856             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 857                                               x+j_coord_offsetC,x+j_coord_offsetD,
 858                                               &jx0,&jy0,&jz0);
 859
 860             /* Calculate displacement vector */
 861             dx00             = _mm_sub_ps(ix0,jx0);
 862             dy00             = _mm_sub_ps(iy0,jy0);
 863             dz00             = _mm_sub_ps(iz0,jz0);
 864             dx10             = _mm_sub_ps(ix1,jx0);
 865             dy10             = _mm_sub_ps(iy1,jy0);
 866             dz10             = _mm_sub_ps(iz1,jz0);
 867             dx20             = _mm_sub_ps(ix2,jx0);
 868             dy20             = _mm_sub_ps(iy2,jy0);
 869             dz20             = _mm_sub_ps(iz2,jz0);
 870
 871             /* Calculate squared distance and things based on it */
 872             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 873             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 874             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 875
 876             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 877             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 878             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 879
 880             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 881             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 882             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 883
 884             /* Load parameters for j particles */
 885             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 886                                                               charge+jnrC+0,charge+jnrD+0);
 887
 888             fjx0             = _mm_setzero_ps();
 889             fjy0             = _mm_setzero_ps();
 890             fjz0             = _mm_setzero_ps();
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             if (gmx_mm_any_lt(rsq00,rcutoff2))
 897             {
 898
 899             r00              = _mm_mul_ps(rsq00,rinv00);
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             qq00             = _mm_mul_ps(iq0,jq0);
 903
 904             /* EWALD ELECTROSTATICS */
 905
 906             /* Analytical PME correction */
 907             zeta2            = _mm_mul_ps(beta2,rsq00);
 908             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 909             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 910             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 911             felec            = _mm_mul_ps(qq00,felec);
 912             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 913             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 914             velec            = _mm_mul_ps(qq00,velec);
 915
 916             d                = _mm_sub_ps(r00,rswitch);
 917             d                = _mm_max_ps(d,_mm_setzero_ps());
 918             d2               = _mm_mul_ps(d,d);
 919             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 920
 921             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 922
 923             /* Evaluate switch function */
 924             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 925             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 926             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 927
 928             fscal            = felec;
 929
 930             fscal            = _mm_and_ps(fscal,cutoff_mask);
 931
 932              /* Update vectorial force */
 933             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 934             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 935             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 936
 937             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 938             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 939             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 940
 941             }
 942
 943             /**************************
 944              * CALCULATE INTERACTIONS *
 945              **************************/
 946
 947             if (gmx_mm_any_lt(rsq10,rcutoff2))
 948             {
 949
 950             r10              = _mm_mul_ps(rsq10,rinv10);
 951
 952             /* Compute parameters for interactions between i and j atoms */
 953             qq10             = _mm_mul_ps(iq1,jq0);
 954
 955             /* EWALD ELECTROSTATICS */
 956
 957             /* Analytical PME correction */
 958             zeta2            = _mm_mul_ps(beta2,rsq10);
 959             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 960             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 961             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 962             felec            = _mm_mul_ps(qq10,felec);
 963             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 964             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 965             velec            = _mm_mul_ps(qq10,velec);
 966
 967             d                = _mm_sub_ps(r10,rswitch);
 968             d                = _mm_max_ps(d,_mm_setzero_ps());
 969             d2               = _mm_mul_ps(d,d);
 970             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 971
 972             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 973
 974             /* Evaluate switch function */
 975             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 976             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 977             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 978
 979             fscal            = felec;
 980
 981             fscal            = _mm_and_ps(fscal,cutoff_mask);
 982
 983              /* Update vectorial force */
 984             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 985             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 986             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 987
 988             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 989             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 990             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 991
 992             }
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             if (gmx_mm_any_lt(rsq20,rcutoff2))
 999             {
1000
1001             r20              = _mm_mul_ps(rsq20,rinv20);
1002
1003             /* Compute parameters for interactions between i and j atoms */
1004             qq20             = _mm_mul_ps(iq2,jq0);
1005
1006             /* EWALD ELECTROSTATICS */
1007
1008             /* Analytical PME correction */
1009             zeta2            = _mm_mul_ps(beta2,rsq20);
1010             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1011             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1012             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1013             felec            = _mm_mul_ps(qq20,felec);
1014             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1015             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1016             velec            = _mm_mul_ps(qq20,velec);
1017
1018             d                = _mm_sub_ps(r20,rswitch);
1019             d                = _mm_max_ps(d,_mm_setzero_ps());
1020             d2               = _mm_mul_ps(d,d);
1021             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1022
1023             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1024
1025             /* Evaluate switch function */
1026             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1027             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1028             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1029
1030             fscal            = felec;
1031
1032             fscal            = _mm_and_ps(fscal,cutoff_mask);
1033
1034              /* Update vectorial force */
1035             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1036             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1037             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1038
1039             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1040             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1041             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1042
1043             }
1044
1045             fjptrA             = f+j_coord_offsetA;
1046             fjptrB             = f+j_coord_offsetB;
1047             fjptrC             = f+j_coord_offsetC;
1048             fjptrD             = f+j_coord_offsetD;
1049
1050             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1051
1052             /* Inner loop uses 150 flops */
1053         }
1054
1055         if(jidx<j_index_end)
1056         {
1057
1058             /* Get j neighbor index, and coordinate index */
1059             jnrlistA         = jjnr[jidx];
1060             jnrlistB         = jjnr[jidx+1];
1061             jnrlistC         = jjnr[jidx+2];
1062             jnrlistD         = jjnr[jidx+3];
1063             /* Sign of each element will be negative for non-real atoms.
1064              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1065              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1066              */
1067             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1068             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1069             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1070             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1071             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1072             j_coord_offsetA  = DIM*jnrA;
1073             j_coord_offsetB  = DIM*jnrB;
1074             j_coord_offsetC  = DIM*jnrC;
1075             j_coord_offsetD  = DIM*jnrD;
1076
1077             /* load j atom coordinates */
1078             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1079                                               x+j_coord_offsetC,x+j_coord_offsetD,
1080                                               &jx0,&jy0,&jz0);
1081
1082             /* Calculate displacement vector */
1083             dx00             = _mm_sub_ps(ix0,jx0);
1084             dy00             = _mm_sub_ps(iy0,jy0);
1085             dz00             = _mm_sub_ps(iz0,jz0);
1086             dx10             = _mm_sub_ps(ix1,jx0);
1087             dy10             = _mm_sub_ps(iy1,jy0);
1088             dz10             = _mm_sub_ps(iz1,jz0);
1089             dx20             = _mm_sub_ps(ix2,jx0);
1090             dy20             = _mm_sub_ps(iy2,jy0);
1091             dz20             = _mm_sub_ps(iz2,jz0);
1092
1093             /* Calculate squared distance and things based on it */
1094             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1095             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1096             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1097
1098             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1099             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1100             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1101
1102             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1103             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1104             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1105
1106             /* Load parameters for j particles */
1107             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1108                                                               charge+jnrC+0,charge+jnrD+0);
1109
1110             fjx0             = _mm_setzero_ps();
1111             fjy0             = _mm_setzero_ps();
1112             fjz0             = _mm_setzero_ps();
1113
1114             /**************************
1115              * CALCULATE INTERACTIONS *
1116              **************************/
1117
1118             if (gmx_mm_any_lt(rsq00,rcutoff2))
1119             {
1120
1121             r00              = _mm_mul_ps(rsq00,rinv00);
1122             r00              = _mm_andnot_ps(dummy_mask,r00);
1123
1124             /* Compute parameters for interactions between i and j atoms */
1125             qq00             = _mm_mul_ps(iq0,jq0);
1126
1127             /* EWALD ELECTROSTATICS */
1128
1129             /* Analytical PME correction */
1130             zeta2            = _mm_mul_ps(beta2,rsq00);
1131             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1132             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1133             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1134             felec            = _mm_mul_ps(qq00,felec);
1135             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1136             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
1137             velec            = _mm_mul_ps(qq00,velec);
1138
1139             d                = _mm_sub_ps(r00,rswitch);
1140             d                = _mm_max_ps(d,_mm_setzero_ps());
1141             d2               = _mm_mul_ps(d,d);
1142             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1143
1144             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1145
1146             /* Evaluate switch function */
1147             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1148             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1149             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1150
1151             fscal            = felec;
1152
1153             fscal            = _mm_and_ps(fscal,cutoff_mask);
1154
1155             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1156
1157              /* Update vectorial force */
1158             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1159             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1160             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1161
1162             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1163             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1164             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1165
1166             }
1167
1168             /**************************
1169              * CALCULATE INTERACTIONS *
1170              **************************/
1171
1172             if (gmx_mm_any_lt(rsq10,rcutoff2))
1173             {
1174
1175             r10              = _mm_mul_ps(rsq10,rinv10);
1176             r10              = _mm_andnot_ps(dummy_mask,r10);
1177
1178             /* Compute parameters for interactions between i and j atoms */
1179             qq10             = _mm_mul_ps(iq1,jq0);
1180
1181             /* EWALD ELECTROSTATICS */
1182
1183             /* Analytical PME correction */
1184             zeta2            = _mm_mul_ps(beta2,rsq10);
1185             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1186             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1187             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1188             felec            = _mm_mul_ps(qq10,felec);
1189             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1190             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
1191             velec            = _mm_mul_ps(qq10,velec);
1192
1193             d                = _mm_sub_ps(r10,rswitch);
1194             d                = _mm_max_ps(d,_mm_setzero_ps());
1195             d2               = _mm_mul_ps(d,d);
1196             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1197
1198             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1199
1200             /* Evaluate switch function */
1201             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1202             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1203             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1204
1205             fscal            = felec;
1206
1207             fscal            = _mm_and_ps(fscal,cutoff_mask);
1208
1209             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1210
1211              /* Update vectorial force */
1212             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1213             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1214             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1215
1216             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1217             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1218             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1219
1220             }
1221
1222             /**************************
1223              * CALCULATE INTERACTIONS *
1224              **************************/
1225
1226             if (gmx_mm_any_lt(rsq20,rcutoff2))
1227             {
1228
1229             r20              = _mm_mul_ps(rsq20,rinv20);
1230             r20              = _mm_andnot_ps(dummy_mask,r20);
1231
1232             /* Compute parameters for interactions between i and j atoms */
1233             qq20             = _mm_mul_ps(iq2,jq0);
1234
1235             /* EWALD ELECTROSTATICS */
1236
1237             /* Analytical PME correction */
1238             zeta2            = _mm_mul_ps(beta2,rsq20);
1239             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1240             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1241             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1242             felec            = _mm_mul_ps(qq20,felec);
1243             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1244             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1245             velec            = _mm_mul_ps(qq20,velec);
1246
1247             d                = _mm_sub_ps(r20,rswitch);
1248             d                = _mm_max_ps(d,_mm_setzero_ps());
1249             d2               = _mm_mul_ps(d,d);
1250             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1251
1252             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1253
1254             /* Evaluate switch function */
1255             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1256             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1257             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1258
1259             fscal            = felec;
1260
1261             fscal            = _mm_and_ps(fscal,cutoff_mask);
1262
1263             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1264
1265              /* Update vectorial force */
1266             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1267             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1268             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1269
1270             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1271             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1272             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1273
1274             }
1275
1276             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1277             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1278             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1279             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1280
1281             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1282
1283             /* Inner loop uses 153 flops */
1284         }
1285
1286         /* End of innermost loop */
1287
1288         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1289                                               f+i_coord_offset,fshift+i_shift_offset);
1290
1291         /* Increment number of inner iterations */
1292         inneriter                  += j_index_end - j_index_start;
1293
1294         /* Outer loop uses 18 flops */
1295     }
1296
1297     /* Increment number of outer iterations */
1298     outeriter        += nri;
1299
1300     /* Update outer/inner flops */
1301
1302     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*153);
1303 }