src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSw_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSw_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     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 103     real             rswitch_scalar,d_scalar;
 104     __m128           dummy_mask,cutoff_mask;
 105     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 106     __m128           one     = _mm_set1_ps(1.0);
 107     __m128           two     = _mm_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     krf              = _mm_set1_ps(fr->ic->k_rf);
 122     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 123     crf              = _mm_set1_ps(fr->ic->c_rf);
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 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     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 136     rcutoff_scalar   = fr->rcoulomb;
 137     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 138     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 139
 140     rswitch_scalar   = fr->rvdw_switch;
 141     rswitch          = _mm_set1_ps(rswitch_scalar);
 142     /* Setup switch parameters */
 143     d_scalar         = rcutoff_scalar-rswitch_scalar;
 144     d                = _mm_set1_ps(d_scalar);
 145     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 146     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 147     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 148     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 149     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 150     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = jnrC = jnrD = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156     j_coord_offsetC = 0;
 157     j_coord_offsetD = 0;
 158
 159     outeriter        = 0;
 160     inneriter        = 0;
 161
 162     for(iidx=0;iidx<4*DIM;iidx++)
 163     {
 164         scratch[iidx] = 0.0;
 165     }
 166
 167     /* Start outer loop over neighborlists */
 168     for(iidx=0; iidx<nri; iidx++)
 169     {
 170         /* Load shift vector for this list */
 171         i_shift_offset   = DIM*shiftidx[iidx];
 172
 173         /* Load limits for loop over neighbors */
 174         j_index_start    = jindex[iidx];
 175         j_index_end      = jindex[iidx+1];
 176
 177         /* Get outer coordinate index */
 178         inr              = iinr[iidx];
 179         i_coord_offset   = DIM*inr;
 180
 181         /* Load i particle coords and add shift vector */
 182         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 183                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 184
 185         fix0             = _mm_setzero_ps();
 186         fiy0             = _mm_setzero_ps();
 187         fiz0             = _mm_setzero_ps();
 188         fix1             = _mm_setzero_ps();
 189         fiy1             = _mm_setzero_ps();
 190         fiz1             = _mm_setzero_ps();
 191         fix2             = _mm_setzero_ps();
 192         fiy2             = _mm_setzero_ps();
 193         fiz2             = _mm_setzero_ps();
 194
 195         /* Reset potential sums */
 196         velecsum         = _mm_setzero_ps();
 197         vvdwsum          = _mm_setzero_ps();
 198
 199         /* Start inner kernel loop */
 200         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 201         {
 202
 203             /* Get j neighbor index, and coordinate index */
 204             jnrA             = jjnr[jidx];
 205             jnrB             = jjnr[jidx+1];
 206             jnrC             = jjnr[jidx+2];
 207             jnrD             = jjnr[jidx+3];
 208             j_coord_offsetA  = DIM*jnrA;
 209             j_coord_offsetB  = DIM*jnrB;
 210             j_coord_offsetC  = DIM*jnrC;
 211             j_coord_offsetD  = DIM*jnrD;
 212
 213             /* load j atom coordinates */
 214             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 215                                               x+j_coord_offsetC,x+j_coord_offsetD,
 216                                               &jx0,&jy0,&jz0);
 217
 218             /* Calculate displacement vector */
 219             dx00             = _mm_sub_ps(ix0,jx0);
 220             dy00             = _mm_sub_ps(iy0,jy0);
 221             dz00             = _mm_sub_ps(iz0,jz0);
 222             dx10             = _mm_sub_ps(ix1,jx0);
 223             dy10             = _mm_sub_ps(iy1,jy0);
 224             dz10             = _mm_sub_ps(iz1,jz0);
 225             dx20             = _mm_sub_ps(ix2,jx0);
 226             dy20             = _mm_sub_ps(iy2,jy0);
 227             dz20             = _mm_sub_ps(iz2,jz0);
 228
 229             /* Calculate squared distance and things based on it */
 230             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 231             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 232             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 233
 234             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 235             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 236             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 237
 238             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 239             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 240             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 241
 242             /* Load parameters for j particles */
 243             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 244                                                               charge+jnrC+0,charge+jnrD+0);
 245             vdwjidx0A        = 2*vdwtype[jnrA+0];
 246             vdwjidx0B        = 2*vdwtype[jnrB+0];
 247             vdwjidx0C        = 2*vdwtype[jnrC+0];
 248             vdwjidx0D        = 2*vdwtype[jnrD+0];
 249
 250             fjx0             = _mm_setzero_ps();
 251             fjy0             = _mm_setzero_ps();
 252             fjz0             = _mm_setzero_ps();
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             if (gmx_mm_any_lt(rsq00,rcutoff2))
 259             {
 260
 261             r00              = _mm_mul_ps(rsq00,rinv00);
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             qq00             = _mm_mul_ps(iq0,jq0);
 265             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 266                                          vdwparam+vdwioffset0+vdwjidx0B,
 267                                          vdwparam+vdwioffset0+vdwjidx0C,
 268                                          vdwparam+vdwioffset0+vdwjidx0D,
 269                                          &c6_00,&c12_00);
 270
 271             /* REACTION-FIELD ELECTROSTATICS */
 272             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 273             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 274
 275             /* LENNARD-JONES DISPERSION/REPULSION */
 276
 277             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 278             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 279             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 280             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 281             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 282
 283             d                = _mm_sub_ps(r00,rswitch);
 284             d                = _mm_max_ps(d,_mm_setzero_ps());
 285             d2               = _mm_mul_ps(d,d);
 286             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 287
 288             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 289
 290             /* Evaluate switch function */
 291             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 292             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 293             vvdw             = _mm_mul_ps(vvdw,sw);
 294             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velec            = _mm_and_ps(velec,cutoff_mask);
 298             velecsum         = _mm_add_ps(velecsum,velec);
 299             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 300             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 301
 302             fscal            = _mm_add_ps(felec,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             /* Compute parameters for interactions between i and j atoms */
 325             qq10             = _mm_mul_ps(iq1,jq0);
 326
 327             /* REACTION-FIELD ELECTROSTATICS */
 328             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 329             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 330
 331             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velec            = _mm_and_ps(velec,cutoff_mask);
 335             velecsum         = _mm_add_ps(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             fscal            = _mm_and_ps(fscal,cutoff_mask);
 340
 341              /* Update vectorial force */
 342             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 343             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 344             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 345
 346             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 347             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 348             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 349
 350             }
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             if (gmx_mm_any_lt(rsq20,rcutoff2))
 357             {
 358
 359             /* Compute parameters for interactions between i and j atoms */
 360             qq20             = _mm_mul_ps(iq2,jq0);
 361
 362             /* REACTION-FIELD ELECTROSTATICS */
 363             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 364             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 365
 366             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velec            = _mm_and_ps(velec,cutoff_mask);
 370             velecsum         = _mm_add_ps(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             fscal            = _mm_and_ps(fscal,cutoff_mask);
 375
 376              /* Update vectorial force */
 377             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 378             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 379             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 380
 381             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 382             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 383             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 384
 385             }
 386
 387             fjptrA             = f+j_coord_offsetA;
 388             fjptrB             = f+j_coord_offsetB;
 389             fjptrC             = f+j_coord_offsetC;
 390             fjptrD             = f+j_coord_offsetD;
 391
 392             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 393
 394             /* Inner loop uses 151 flops */
 395         }
 396
 397         if(jidx<j_index_end)
 398         {
 399
 400             /* Get j neighbor index, and coordinate index */
 401             jnrlistA         = jjnr[jidx];
 402             jnrlistB         = jjnr[jidx+1];
 403             jnrlistC         = jjnr[jidx+2];
 404             jnrlistD         = jjnr[jidx+3];
 405             /* Sign of each element will be negative for non-real atoms.
 406              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 407              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 408              */
 409             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 410             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 411             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 412             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 413             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 414             j_coord_offsetA  = DIM*jnrA;
 415             j_coord_offsetB  = DIM*jnrB;
 416             j_coord_offsetC  = DIM*jnrC;
 417             j_coord_offsetD  = DIM*jnrD;
 418
 419             /* load j atom coordinates */
 420             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 421                                               x+j_coord_offsetC,x+j_coord_offsetD,
 422                                               &jx0,&jy0,&jz0);
 423
 424             /* Calculate displacement vector */
 425             dx00             = _mm_sub_ps(ix0,jx0);
 426             dy00             = _mm_sub_ps(iy0,jy0);
 427             dz00             = _mm_sub_ps(iz0,jz0);
 428             dx10             = _mm_sub_ps(ix1,jx0);
 429             dy10             = _mm_sub_ps(iy1,jy0);
 430             dz10             = _mm_sub_ps(iz1,jz0);
 431             dx20             = _mm_sub_ps(ix2,jx0);
 432             dy20             = _mm_sub_ps(iy2,jy0);
 433             dz20             = _mm_sub_ps(iz2,jz0);
 434
 435             /* Calculate squared distance and things based on it */
 436             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 437             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 438             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 439
 440             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 441             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 442             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 443
 444             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 445             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 446             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 447
 448             /* Load parameters for j particles */
 449             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 450                                                               charge+jnrC+0,charge+jnrD+0);
 451             vdwjidx0A        = 2*vdwtype[jnrA+0];
 452             vdwjidx0B        = 2*vdwtype[jnrB+0];
 453             vdwjidx0C        = 2*vdwtype[jnrC+0];
 454             vdwjidx0D        = 2*vdwtype[jnrD+0];
 455
 456             fjx0             = _mm_setzero_ps();
 457             fjy0             = _mm_setzero_ps();
 458             fjz0             = _mm_setzero_ps();
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             if (gmx_mm_any_lt(rsq00,rcutoff2))
 465             {
 466
 467             r00              = _mm_mul_ps(rsq00,rinv00);
 468             r00              = _mm_andnot_ps(dummy_mask,r00);
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             qq00             = _mm_mul_ps(iq0,jq0);
 472             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 473                                          vdwparam+vdwioffset0+vdwjidx0B,
 474                                          vdwparam+vdwioffset0+vdwjidx0C,
 475                                          vdwparam+vdwioffset0+vdwjidx0D,
 476                                          &c6_00,&c12_00);
 477
 478             /* REACTION-FIELD ELECTROSTATICS */
 479             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 480             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 481
 482             /* LENNARD-JONES DISPERSION/REPULSION */
 483
 484             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 485             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 486             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 487             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 488             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 489
 490             d                = _mm_sub_ps(r00,rswitch);
 491             d                = _mm_max_ps(d,_mm_setzero_ps());
 492             d2               = _mm_mul_ps(d,d);
 493             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 494
 495             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 496
 497             /* Evaluate switch function */
 498             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 499             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 500             vvdw             = _mm_mul_ps(vvdw,sw);
 501             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velec            = _mm_and_ps(velec,cutoff_mask);
 505             velec            = _mm_andnot_ps(dummy_mask,velec);
 506             velecsum         = _mm_add_ps(velecsum,velec);
 507             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 508             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 509             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 510
 511             fscal            = _mm_add_ps(felec,fvdw);
 512
 513             fscal            = _mm_and_ps(fscal,cutoff_mask);
 514
 515             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 516
 517              /* Update vectorial force */
 518             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 519             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 520             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 521
 522             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 523             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 524             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 525
 526             }
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             if (gmx_mm_any_lt(rsq10,rcutoff2))
 533             {
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq10             = _mm_mul_ps(iq1,jq0);
 537
 538             /* REACTION-FIELD ELECTROSTATICS */
 539             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 540             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 541
 542             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             velec            = _mm_and_ps(velec,cutoff_mask);
 546             velec            = _mm_andnot_ps(dummy_mask,velec);
 547             velecsum         = _mm_add_ps(velecsum,velec);
 548
 549             fscal            = felec;
 550
 551             fscal            = _mm_and_ps(fscal,cutoff_mask);
 552
 553             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 554
 555              /* Update vectorial force */
 556             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 557             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 558             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 559
 560             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 561             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 562             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 563
 564             }
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             if (gmx_mm_any_lt(rsq20,rcutoff2))
 571             {
 572
 573             /* Compute parameters for interactions between i and j atoms */
 574             qq20             = _mm_mul_ps(iq2,jq0);
 575
 576             /* REACTION-FIELD ELECTROSTATICS */
 577             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 578             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 579
 580             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm_and_ps(velec,cutoff_mask);
 584             velec            = _mm_andnot_ps(dummy_mask,velec);
 585             velecsum         = _mm_add_ps(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589             fscal            = _mm_and_ps(fscal,cutoff_mask);
 590
 591             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 592
 593              /* Update vectorial force */
 594             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 595             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 596             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 597
 598             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 599             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 600             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 601
 602             }
 603
 604             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 605             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 606             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 607             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 608
 609             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 610
 611             /* Inner loop uses 152 flops */
 612         }
 613
 614         /* End of innermost loop */
 615
 616         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 617                                               f+i_coord_offset,fshift+i_shift_offset);
 618
 619         ggid                        = gid[iidx];
 620         /* Update potential energies */
 621         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 622         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 623
 624         /* Increment number of inner iterations */
 625         inneriter                  += j_index_end - j_index_start;
 626
 627         /* Outer loop uses 20 flops */
 628     }
 629
 630     /* Increment number of outer iterations */
 631     outeriter        += nri;
 632
 633     /* Update outer/inner flops */
 634
 635     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*152);
 636 }
 637 /*
 638  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 639  * Electrostatics interaction: ReactionField
 640  * VdW interaction:            LennardJones
 641  * Geometry:                   Water3-Particle
 642  * Calculate force/pot:        Force
 643  */
 644 void
 645 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_128_fma_single
 646                     (t_nblist                    * gmx_restrict       nlist,
 647                      rvec                        * gmx_restrict          xx,
 648                      rvec                        * gmx_restrict          ff,
 649                      t_forcerec                  * gmx_restrict          fr,
 650                      t_mdatoms                   * gmx_restrict     mdatoms,
 651                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 652                      t_nrnb                      * gmx_restrict        nrnb)
 653 {
 654     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 655      * just 0 for non-waters.
 656      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 657      * jnr indices corresponding to data put in the four positions in the SIMD register.
 658      */
 659     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 660     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 661     int              jnrA,jnrB,jnrC,jnrD;
 662     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 663     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 664     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 665     real             rcutoff_scalar;
 666     real             *shiftvec,*fshift,*x,*f;
 667     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 668     real             scratch[4*DIM];
 669     __m128           fscal,rcutoff,rcutoff2,jidxall;
 670     int              vdwioffset0;
 671     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 672     int              vdwioffset1;
 673     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 674     int              vdwioffset2;
 675     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 676     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 677     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 678     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 679     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 680     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 681     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 682     real             *charge;
 683     int              nvdwtype;
 684     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 685     int              *vdwtype;
 686     real             *vdwparam;
 687     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 688     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 689     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 690     real             rswitch_scalar,d_scalar;
 691     __m128           dummy_mask,cutoff_mask;
 692     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 693     __m128           one     = _mm_set1_ps(1.0);
 694     __m128           two     = _mm_set1_ps(2.0);
 695     x                = xx[0];
 696     f                = ff[0];
 697
 698     nri              = nlist->nri;
 699     iinr             = nlist->iinr;
 700     jindex           = nlist->jindex;
 701     jjnr             = nlist->jjnr;
 702     shiftidx         = nlist->shift;
 703     gid              = nlist->gid;
 704     shiftvec         = fr->shift_vec[0];
 705     fshift           = fr->fshift[0];
 706     facel            = _mm_set1_ps(fr->epsfac);
 707     charge           = mdatoms->chargeA;
 708     krf              = _mm_set1_ps(fr->ic->k_rf);
 709     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 710     crf              = _mm_set1_ps(fr->ic->c_rf);
 711     nvdwtype         = fr->ntype;
 712     vdwparam         = fr->nbfp;
 713     vdwtype          = mdatoms->typeA;
 714
 715     /* Setup water-specific parameters */
 716     inr              = nlist->iinr[0];
 717     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 718     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 719     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 720     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 721
 722     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 723     rcutoff_scalar   = fr->rcoulomb;
 724     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 725     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 726
 727     rswitch_scalar   = fr->rvdw_switch;
 728     rswitch          = _mm_set1_ps(rswitch_scalar);
 729     /* Setup switch parameters */
 730     d_scalar         = rcutoff_scalar-rswitch_scalar;
 731     d                = _mm_set1_ps(d_scalar);
 732     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 733     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 734     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 735     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 736     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 737     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 738
 739     /* Avoid stupid compiler warnings */
 740     jnrA = jnrB = jnrC = jnrD = 0;
 741     j_coord_offsetA = 0;
 742     j_coord_offsetB = 0;
 743     j_coord_offsetC = 0;
 744     j_coord_offsetD = 0;
 745
 746     outeriter        = 0;
 747     inneriter        = 0;
 748
 749     for(iidx=0;iidx<4*DIM;iidx++)
 750     {
 751         scratch[iidx] = 0.0;
 752     }
 753
 754     /* Start outer loop over neighborlists */
 755     for(iidx=0; iidx<nri; iidx++)
 756     {
 757         /* Load shift vector for this list */
 758         i_shift_offset   = DIM*shiftidx[iidx];
 759
 760         /* Load limits for loop over neighbors */
 761         j_index_start    = jindex[iidx];
 762         j_index_end      = jindex[iidx+1];
 763
 764         /* Get outer coordinate index */
 765         inr              = iinr[iidx];
 766         i_coord_offset   = DIM*inr;
 767
 768         /* Load i particle coords and add shift vector */
 769         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 770                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 771
 772         fix0             = _mm_setzero_ps();
 773         fiy0             = _mm_setzero_ps();
 774         fiz0             = _mm_setzero_ps();
 775         fix1             = _mm_setzero_ps();
 776         fiy1             = _mm_setzero_ps();
 777         fiz1             = _mm_setzero_ps();
 778         fix2             = _mm_setzero_ps();
 779         fiy2             = _mm_setzero_ps();
 780         fiz2             = _mm_setzero_ps();
 781
 782         /* Start inner kernel loop */
 783         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 784         {
 785
 786             /* Get j neighbor index, and coordinate index */
 787             jnrA             = jjnr[jidx];
 788             jnrB             = jjnr[jidx+1];
 789             jnrC             = jjnr[jidx+2];
 790             jnrD             = jjnr[jidx+3];
 791             j_coord_offsetA  = DIM*jnrA;
 792             j_coord_offsetB  = DIM*jnrB;
 793             j_coord_offsetC  = DIM*jnrC;
 794             j_coord_offsetD  = DIM*jnrD;
 795
 796             /* load j atom coordinates */
 797             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 798                                               x+j_coord_offsetC,x+j_coord_offsetD,
 799                                               &jx0,&jy0,&jz0);
 800
 801             /* Calculate displacement vector */
 802             dx00             = _mm_sub_ps(ix0,jx0);
 803             dy00             = _mm_sub_ps(iy0,jy0);
 804             dz00             = _mm_sub_ps(iz0,jz0);
 805             dx10             = _mm_sub_ps(ix1,jx0);
 806             dy10             = _mm_sub_ps(iy1,jy0);
 807             dz10             = _mm_sub_ps(iz1,jz0);
 808             dx20             = _mm_sub_ps(ix2,jx0);
 809             dy20             = _mm_sub_ps(iy2,jy0);
 810             dz20             = _mm_sub_ps(iz2,jz0);
 811
 812             /* Calculate squared distance and things based on it */
 813             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 814             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 815             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 816
 817             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 818             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 819             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 820
 821             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 822             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 823             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 824
 825             /* Load parameters for j particles */
 826             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 827                                                               charge+jnrC+0,charge+jnrD+0);
 828             vdwjidx0A        = 2*vdwtype[jnrA+0];
 829             vdwjidx0B        = 2*vdwtype[jnrB+0];
 830             vdwjidx0C        = 2*vdwtype[jnrC+0];
 831             vdwjidx0D        = 2*vdwtype[jnrD+0];
 832
 833             fjx0             = _mm_setzero_ps();
 834             fjy0             = _mm_setzero_ps();
 835             fjz0             = _mm_setzero_ps();
 836
 837             /**************************
 838              * CALCULATE INTERACTIONS *
 839              **************************/
 840
 841             if (gmx_mm_any_lt(rsq00,rcutoff2))
 842             {
 843
 844             r00              = _mm_mul_ps(rsq00,rinv00);
 845
 846             /* Compute parameters for interactions between i and j atoms */
 847             qq00             = _mm_mul_ps(iq0,jq0);
 848             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 849                                          vdwparam+vdwioffset0+vdwjidx0B,
 850                                          vdwparam+vdwioffset0+vdwjidx0C,
 851                                          vdwparam+vdwioffset0+vdwjidx0D,
 852                                          &c6_00,&c12_00);
 853
 854             /* REACTION-FIELD ELECTROSTATICS */
 855             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 856
 857             /* LENNARD-JONES DISPERSION/REPULSION */
 858
 859             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 860             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 861             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 862             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 863             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 864
 865             d                = _mm_sub_ps(r00,rswitch);
 866             d                = _mm_max_ps(d,_mm_setzero_ps());
 867             d2               = _mm_mul_ps(d,d);
 868             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 869
 870             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 871
 872             /* Evaluate switch function */
 873             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 874             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 875             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 876
 877             fscal            = _mm_add_ps(felec,fvdw);
 878
 879             fscal            = _mm_and_ps(fscal,cutoff_mask);
 880
 881              /* Update vectorial force */
 882             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 883             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 884             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 885
 886             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 887             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 888             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 889
 890             }
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             if (gmx_mm_any_lt(rsq10,rcutoff2))
 897             {
 898
 899             /* Compute parameters for interactions between i and j atoms */
 900             qq10             = _mm_mul_ps(iq1,jq0);
 901
 902             /* REACTION-FIELD ELECTROSTATICS */
 903             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 904
 905             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 906
 907             fscal            = felec;
 908
 909             fscal            = _mm_and_ps(fscal,cutoff_mask);
 910
 911              /* Update vectorial force */
 912             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 913             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 914             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 915
 916             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 917             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 918             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 919
 920             }
 921
 922             /**************************
 923              * CALCULATE INTERACTIONS *
 924              **************************/
 925
 926             if (gmx_mm_any_lt(rsq20,rcutoff2))
 927             {
 928
 929             /* Compute parameters for interactions between i and j atoms */
 930             qq20             = _mm_mul_ps(iq2,jq0);
 931
 932             /* REACTION-FIELD ELECTROSTATICS */
 933             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 934
 935             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 936
 937             fscal            = felec;
 938
 939             fscal            = _mm_and_ps(fscal,cutoff_mask);
 940
 941              /* Update vectorial force */
 942             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 943             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 944             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 945
 946             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 947             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 948             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 949
 950             }
 951
 952             fjptrA             = f+j_coord_offsetA;
 953             fjptrB             = f+j_coord_offsetB;
 954             fjptrC             = f+j_coord_offsetC;
 955             fjptrD             = f+j_coord_offsetD;
 956
 957             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 958
 959             /* Inner loop uses 130 flops */
 960         }
 961
 962         if(jidx<j_index_end)
 963         {
 964
 965             /* Get j neighbor index, and coordinate index */
 966             jnrlistA         = jjnr[jidx];
 967             jnrlistB         = jjnr[jidx+1];
 968             jnrlistC         = jjnr[jidx+2];
 969             jnrlistD         = jjnr[jidx+3];
 970             /* Sign of each element will be negative for non-real atoms.
 971              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 972              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 973              */
 974             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 975             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 976             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 977             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 978             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 979             j_coord_offsetA  = DIM*jnrA;
 980             j_coord_offsetB  = DIM*jnrB;
 981             j_coord_offsetC  = DIM*jnrC;
 982             j_coord_offsetD  = DIM*jnrD;
 983
 984             /* load j atom coordinates */
 985             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 986                                               x+j_coord_offsetC,x+j_coord_offsetD,
 987                                               &jx0,&jy0,&jz0);
 988
 989             /* Calculate displacement vector */
 990             dx00             = _mm_sub_ps(ix0,jx0);
 991             dy00             = _mm_sub_ps(iy0,jy0);
 992             dz00             = _mm_sub_ps(iz0,jz0);
 993             dx10             = _mm_sub_ps(ix1,jx0);
 994             dy10             = _mm_sub_ps(iy1,jy0);
 995             dz10             = _mm_sub_ps(iz1,jz0);
 996             dx20             = _mm_sub_ps(ix2,jx0);
 997             dy20             = _mm_sub_ps(iy2,jy0);
 998             dz20             = _mm_sub_ps(iz2,jz0);
 999
1000             /* Calculate squared distance and things based on it */
1001             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1002             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1003             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1004
1005             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1006             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1007             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1008
1009             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1010             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1011             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1012
1013             /* Load parameters for j particles */
1014             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1015                                                               charge+jnrC+0,charge+jnrD+0);
1016             vdwjidx0A        = 2*vdwtype[jnrA+0];
1017             vdwjidx0B        = 2*vdwtype[jnrB+0];
1018             vdwjidx0C        = 2*vdwtype[jnrC+0];
1019             vdwjidx0D        = 2*vdwtype[jnrD+0];
1020
1021             fjx0             = _mm_setzero_ps();
1022             fjy0             = _mm_setzero_ps();
1023             fjz0             = _mm_setzero_ps();
1024
1025             /**************************
1026              * CALCULATE INTERACTIONS *
1027              **************************/
1028
1029             if (gmx_mm_any_lt(rsq00,rcutoff2))
1030             {
1031
1032             r00              = _mm_mul_ps(rsq00,rinv00);
1033             r00              = _mm_andnot_ps(dummy_mask,r00);
1034
1035             /* Compute parameters for interactions between i and j atoms */
1036             qq00             = _mm_mul_ps(iq0,jq0);
1037             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1038                                          vdwparam+vdwioffset0+vdwjidx0B,
1039                                          vdwparam+vdwioffset0+vdwjidx0C,
1040                                          vdwparam+vdwioffset0+vdwjidx0D,
1041                                          &c6_00,&c12_00);
1042
1043             /* REACTION-FIELD ELECTROSTATICS */
1044             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1045
1046             /* LENNARD-JONES DISPERSION/REPULSION */
1047
1048             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1049             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1050             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1051             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
1052             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1053
1054             d                = _mm_sub_ps(r00,rswitch);
1055             d                = _mm_max_ps(d,_mm_setzero_ps());
1056             d2               = _mm_mul_ps(d,d);
1057             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
1058
1059             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
1060
1061             /* Evaluate switch function */
1062             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1063             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1064             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1065
1066             fscal            = _mm_add_ps(felec,fvdw);
1067
1068             fscal            = _mm_and_ps(fscal,cutoff_mask);
1069
1070             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1071
1072              /* Update vectorial force */
1073             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1074             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1075             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1076
1077             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1078             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1079             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1080
1081             }
1082
1083             /**************************
1084              * CALCULATE INTERACTIONS *
1085              **************************/
1086
1087             if (gmx_mm_any_lt(rsq10,rcutoff2))
1088             {
1089
1090             /* Compute parameters for interactions between i and j atoms */
1091             qq10             = _mm_mul_ps(iq1,jq0);
1092
1093             /* REACTION-FIELD ELECTROSTATICS */
1094             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1095
1096             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1097
1098             fscal            = felec;
1099
1100             fscal            = _mm_and_ps(fscal,cutoff_mask);
1101
1102             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1103
1104              /* Update vectorial force */
1105             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1106             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1107             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1108
1109             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1110             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1111             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1112
1113             }
1114
1115             /**************************
1116              * CALCULATE INTERACTIONS *
1117              **************************/
1118
1119             if (gmx_mm_any_lt(rsq20,rcutoff2))
1120             {
1121
1122             /* Compute parameters for interactions between i and j atoms */
1123             qq20             = _mm_mul_ps(iq2,jq0);
1124
1125             /* REACTION-FIELD ELECTROSTATICS */
1126             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1127
1128             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1129
1130             fscal            = felec;
1131
1132             fscal            = _mm_and_ps(fscal,cutoff_mask);
1133
1134             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1135
1136              /* Update vectorial force */
1137             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1138             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1139             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1140
1141             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1142             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1143             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1144
1145             }
1146
1147             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1148             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1149             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1150             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1151
1152             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1153
1154             /* Inner loop uses 131 flops */
1155         }
1156
1157         /* End of innermost loop */
1158
1159         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1160                                               f+i_coord_offset,fshift+i_shift_offset);
1161
1162         /* Increment number of inner iterations */
1163         inneriter                  += j_index_end - j_index_start;
1164
1165         /* Outer loop uses 18 flops */
1166     }
1167
1168     /* Increment number of outer iterations */
1169     outeriter        += nri;
1170
1171     /* Update outer/inner flops */
1172
1173     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*131);
1174 }