src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          ewitab;
 105     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 107     real             *ewtab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 130     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 131     beta2            = _mm_mul_ps(beta,beta);
 132     beta3            = _mm_mul_ps(beta,beta2);
 133     ewtab            = fr->ic->tabq_coul_FDV0;
 134     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 135     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 140     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 141     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 142     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 143
 144     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 145     rcutoff_scalar   = fr->rcoulomb;
 146     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 147     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 148
 149     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 150     rvdw             = _mm_set1_ps(fr->rvdw);
 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             /* EWALD ELECTROSTATICS */
 272
 273             /* Analytical PME correction */
 274             zeta2            = _mm_mul_ps(beta2,rsq00);
 275             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 276             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 277             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 278             felec            = _mm_mul_ps(qq00,felec);
 279             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 280             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 281             velec            = _mm_mul_ps(qq00,velec);
 282
 283             /* LENNARD-JONES DISPERSION/REPULSION */
 284
 285             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 286             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 287             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 288             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 289                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 290             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 291
 292             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velec            = _mm_and_ps(velec,cutoff_mask);
 296             velecsum         = _mm_add_ps(velecsum,velec);
 297             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 298             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 299
 300             fscal            = _mm_add_ps(felec,fvdw);
 301
 302             fscal            = _mm_and_ps(fscal,cutoff_mask);
 303
 304              /* Update vectorial force */
 305             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 306             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 307             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 308
 309             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 310             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 311             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 312
 313             }
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             if (gmx_mm_any_lt(rsq10,rcutoff2))
 320             {
 321
 322             r10              = _mm_mul_ps(rsq10,rinv10);
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq10             = _mm_mul_ps(iq1,jq0);
 326
 327             /* EWALD ELECTROSTATICS */
 328
 329             /* Analytical PME correction */
 330             zeta2            = _mm_mul_ps(beta2,rsq10);
 331             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 332             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 333             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 334             felec            = _mm_mul_ps(qq10,felec);
 335             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 336             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 337             velec            = _mm_mul_ps(qq10,velec);
 338
 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,_mm_sub_ps(rinv20,sh_ewald));
 382             velec            = _mm_mul_ps(qq20,velec);
 383
 384             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velec            = _mm_and_ps(velec,cutoff_mask);
 388             velecsum         = _mm_add_ps(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             fscal            = _mm_and_ps(fscal,cutoff_mask);
 393
 394              /* Update vectorial force */
 395             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 396             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 397             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 398
 399             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 400             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 401             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 402
 403             }
 404
 405             fjptrA             = f+j_coord_offsetA;
 406             fjptrB             = f+j_coord_offsetB;
 407             fjptrC             = f+j_coord_offsetC;
 408             fjptrD             = f+j_coord_offsetD;
 409
 410             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 411
 412             /* Inner loop uses 117 flops */
 413         }
 414
 415         if(jidx<j_index_end)
 416         {
 417
 418             /* Get j neighbor index, and coordinate index */
 419             jnrlistA         = jjnr[jidx];
 420             jnrlistB         = jjnr[jidx+1];
 421             jnrlistC         = jjnr[jidx+2];
 422             jnrlistD         = jjnr[jidx+3];
 423             /* Sign of each element will be negative for non-real atoms.
 424              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 425              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 426              */
 427             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 428             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 429             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 430             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 431             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 432             j_coord_offsetA  = DIM*jnrA;
 433             j_coord_offsetB  = DIM*jnrB;
 434             j_coord_offsetC  = DIM*jnrC;
 435             j_coord_offsetD  = DIM*jnrD;
 436
 437             /* load j atom coordinates */
 438             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 439                                               x+j_coord_offsetC,x+j_coord_offsetD,
 440                                               &jx0,&jy0,&jz0);
 441
 442             /* Calculate displacement vector */
 443             dx00             = _mm_sub_ps(ix0,jx0);
 444             dy00             = _mm_sub_ps(iy0,jy0);
 445             dz00             = _mm_sub_ps(iz0,jz0);
 446             dx10             = _mm_sub_ps(ix1,jx0);
 447             dy10             = _mm_sub_ps(iy1,jy0);
 448             dz10             = _mm_sub_ps(iz1,jz0);
 449             dx20             = _mm_sub_ps(ix2,jx0);
 450             dy20             = _mm_sub_ps(iy2,jy0);
 451             dz20             = _mm_sub_ps(iz2,jz0);
 452
 453             /* Calculate squared distance and things based on it */
 454             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 455             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 456             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 457
 458             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 459             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 460             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 461
 462             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 463             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 464             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 465
 466             /* Load parameters for j particles */
 467             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 468                                                               charge+jnrC+0,charge+jnrD+0);
 469             vdwjidx0A        = 2*vdwtype[jnrA+0];
 470             vdwjidx0B        = 2*vdwtype[jnrB+0];
 471             vdwjidx0C        = 2*vdwtype[jnrC+0];
 472             vdwjidx0D        = 2*vdwtype[jnrD+0];
 473
 474             fjx0             = _mm_setzero_ps();
 475             fjy0             = _mm_setzero_ps();
 476             fjz0             = _mm_setzero_ps();
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             if (gmx_mm_any_lt(rsq00,rcutoff2))
 483             {
 484
 485             r00              = _mm_mul_ps(rsq00,rinv00);
 486             r00              = _mm_andnot_ps(dummy_mask,r00);
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             qq00             = _mm_mul_ps(iq0,jq0);
 490             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 491                                          vdwparam+vdwioffset0+vdwjidx0B,
 492                                          vdwparam+vdwioffset0+vdwjidx0C,
 493                                          vdwparam+vdwioffset0+vdwjidx0D,
 494                                          &c6_00,&c12_00);
 495
 496             /* EWALD ELECTROSTATICS */
 497
 498             /* Analytical PME correction */
 499             zeta2            = _mm_mul_ps(beta2,rsq00);
 500             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 501             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 502             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 503             felec            = _mm_mul_ps(qq00,felec);
 504             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 505             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 506             velec            = _mm_mul_ps(qq00,velec);
 507
 508             /* LENNARD-JONES DISPERSION/REPULSION */
 509
 510             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 511             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 512             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 513             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 514                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 515             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 516
 517             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 518
 519             /* Update potential sum for this i atom from the interaction with this j atom. */
 520             velec            = _mm_and_ps(velec,cutoff_mask);
 521             velec            = _mm_andnot_ps(dummy_mask,velec);
 522             velecsum         = _mm_add_ps(velecsum,velec);
 523             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 524             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 525             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 526
 527             fscal            = _mm_add_ps(felec,fvdw);
 528
 529             fscal            = _mm_and_ps(fscal,cutoff_mask);
 530
 531             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 532
 533              /* Update vectorial force */
 534             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 535             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 536             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 537
 538             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 539             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 540             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 541
 542             }
 543
 544             /**************************
 545              * CALCULATE INTERACTIONS *
 546              **************************/
 547
 548             if (gmx_mm_any_lt(rsq10,rcutoff2))
 549             {
 550
 551             r10              = _mm_mul_ps(rsq10,rinv10);
 552             r10              = _mm_andnot_ps(dummy_mask,r10);
 553
 554             /* Compute parameters for interactions between i and j atoms */
 555             qq10             = _mm_mul_ps(iq1,jq0);
 556
 557             /* EWALD ELECTROSTATICS */
 558
 559             /* Analytical PME correction */
 560             zeta2            = _mm_mul_ps(beta2,rsq10);
 561             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 562             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 563             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 564             felec            = _mm_mul_ps(qq10,felec);
 565             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 566             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 567             velec            = _mm_mul_ps(qq10,velec);
 568
 569             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 570
 571             /* Update potential sum for this i atom from the interaction with this j atom. */
 572             velec            = _mm_and_ps(velec,cutoff_mask);
 573             velec            = _mm_andnot_ps(dummy_mask,velec);
 574             velecsum         = _mm_add_ps(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm_and_ps(fscal,cutoff_mask);
 579
 580             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 581
 582              /* Update vectorial force */
 583             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 584             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 585             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 586
 587             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 588             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 589             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 590
 591             }
 592
 593             /**************************
 594              * CALCULATE INTERACTIONS *
 595              **************************/
 596
 597             if (gmx_mm_any_lt(rsq20,rcutoff2))
 598             {
 599
 600             r20              = _mm_mul_ps(rsq20,rinv20);
 601             r20              = _mm_andnot_ps(dummy_mask,r20);
 602
 603             /* Compute parameters for interactions between i and j atoms */
 604             qq20             = _mm_mul_ps(iq2,jq0);
 605
 606             /* EWALD ELECTROSTATICS */
 607
 608             /* Analytical PME correction */
 609             zeta2            = _mm_mul_ps(beta2,rsq20);
 610             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 611             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 612             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 613             felec            = _mm_mul_ps(qq20,felec);
 614             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 615             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 616             velec            = _mm_mul_ps(qq20,velec);
 617
 618             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 619
 620             /* Update potential sum for this i atom from the interaction with this j atom. */
 621             velec            = _mm_and_ps(velec,cutoff_mask);
 622             velec            = _mm_andnot_ps(dummy_mask,velec);
 623             velecsum         = _mm_add_ps(velecsum,velec);
 624
 625             fscal            = felec;
 626
 627             fscal            = _mm_and_ps(fscal,cutoff_mask);
 628
 629             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 630
 631              /* Update vectorial force */
 632             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 633             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 634             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 635
 636             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 637             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 638             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 639
 640             }
 641
 642             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 643             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 644             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 645             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 646
 647             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 648
 649             /* Inner loop uses 120 flops */
 650         }
 651
 652         /* End of innermost loop */
 653
 654         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 655                                               f+i_coord_offset,fshift+i_shift_offset);
 656
 657         ggid                        = gid[iidx];
 658         /* Update potential energies */
 659         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 660         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 661
 662         /* Increment number of inner iterations */
 663         inneriter                  += j_index_end - j_index_start;
 664
 665         /* Outer loop uses 20 flops */
 666     }
 667
 668     /* Increment number of outer iterations */
 669     outeriter        += nri;
 670
 671     /* Update outer/inner flops */
 672
 673     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 674 }
 675 /*
 676  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_single
 677  * Electrostatics interaction: Ewald
 678  * VdW interaction:            LennardJones
 679  * Geometry:                   Water3-Particle
 680  * Calculate force/pot:        Force
 681  */
 682 void
 683 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_single
 684                     (t_nblist                    * gmx_restrict       nlist,
 685                      rvec                        * gmx_restrict          xx,
 686                      rvec                        * gmx_restrict          ff,
 687                      t_forcerec                  * gmx_restrict          fr,
 688                      t_mdatoms                   * gmx_restrict     mdatoms,
 689                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 690                      t_nrnb                      * gmx_restrict        nrnb)
 691 {
 692     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 693      * just 0 for non-waters.
 694      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 695      * jnr indices corresponding to data put in the four positions in the SIMD register.
 696      */
 697     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 698     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 699     int              jnrA,jnrB,jnrC,jnrD;
 700     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 701     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 702     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 703     real             rcutoff_scalar;
 704     real             *shiftvec,*fshift,*x,*f;
 705     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 706     real             scratch[4*DIM];
 707     __m128           fscal,rcutoff,rcutoff2,jidxall;
 708     int              vdwioffset0;
 709     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 710     int              vdwioffset1;
 711     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 712     int              vdwioffset2;
 713     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 714     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 715     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 716     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 717     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 718     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 719     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 720     real             *charge;
 721     int              nvdwtype;
 722     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 723     int              *vdwtype;
 724     real             *vdwparam;
 725     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 726     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 727     __m128i          ewitab;
 728     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 729     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 730     real             *ewtab;
 731     __m128           dummy_mask,cutoff_mask;
 732     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 733     __m128           one     = _mm_set1_ps(1.0);
 734     __m128           two     = _mm_set1_ps(2.0);
 735     x                = xx[0];
 736     f                = ff[0];
 737
 738     nri              = nlist->nri;
 739     iinr             = nlist->iinr;
 740     jindex           = nlist->jindex;
 741     jjnr             = nlist->jjnr;
 742     shiftidx         = nlist->shift;
 743     gid              = nlist->gid;
 744     shiftvec         = fr->shift_vec[0];
 745     fshift           = fr->fshift[0];
 746     facel            = _mm_set1_ps(fr->epsfac);
 747     charge           = mdatoms->chargeA;
 748     nvdwtype         = fr->ntype;
 749     vdwparam         = fr->nbfp;
 750     vdwtype          = mdatoms->typeA;
 751
 752     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 753     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 754     beta2            = _mm_mul_ps(beta,beta);
 755     beta3            = _mm_mul_ps(beta,beta2);
 756     ewtab            = fr->ic->tabq_coul_F;
 757     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 758     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 759
 760     /* Setup water-specific parameters */
 761     inr              = nlist->iinr[0];
 762     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 763     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 764     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 765     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 766
 767     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 768     rcutoff_scalar   = fr->rcoulomb;
 769     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 770     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 771
 772     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 773     rvdw             = _mm_set1_ps(fr->rvdw);
 774
 775     /* Avoid stupid compiler warnings */
 776     jnrA = jnrB = jnrC = jnrD = 0;
 777     j_coord_offsetA = 0;
 778     j_coord_offsetB = 0;
 779     j_coord_offsetC = 0;
 780     j_coord_offsetD = 0;
 781
 782     outeriter        = 0;
 783     inneriter        = 0;
 784
 785     for(iidx=0;iidx<4*DIM;iidx++)
 786     {
 787         scratch[iidx] = 0.0;
 788     }
 789
 790     /* Start outer loop over neighborlists */
 791     for(iidx=0; iidx<nri; iidx++)
 792     {
 793         /* Load shift vector for this list */
 794         i_shift_offset   = DIM*shiftidx[iidx];
 795
 796         /* Load limits for loop over neighbors */
 797         j_index_start    = jindex[iidx];
 798         j_index_end      = jindex[iidx+1];
 799
 800         /* Get outer coordinate index */
 801         inr              = iinr[iidx];
 802         i_coord_offset   = DIM*inr;
 803
 804         /* Load i particle coords and add shift vector */
 805         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 806                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 807
 808         fix0             = _mm_setzero_ps();
 809         fiy0             = _mm_setzero_ps();
 810         fiz0             = _mm_setzero_ps();
 811         fix1             = _mm_setzero_ps();
 812         fiy1             = _mm_setzero_ps();
 813         fiz1             = _mm_setzero_ps();
 814         fix2             = _mm_setzero_ps();
 815         fiy2             = _mm_setzero_ps();
 816         fiz2             = _mm_setzero_ps();
 817
 818         /* Start inner kernel loop */
 819         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 820         {
 821
 822             /* Get j neighbor index, and coordinate index */
 823             jnrA             = jjnr[jidx];
 824             jnrB             = jjnr[jidx+1];
 825             jnrC             = jjnr[jidx+2];
 826             jnrD             = jjnr[jidx+3];
 827             j_coord_offsetA  = DIM*jnrA;
 828             j_coord_offsetB  = DIM*jnrB;
 829             j_coord_offsetC  = DIM*jnrC;
 830             j_coord_offsetD  = DIM*jnrD;
 831
 832             /* load j atom coordinates */
 833             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 834                                               x+j_coord_offsetC,x+j_coord_offsetD,
 835                                               &jx0,&jy0,&jz0);
 836
 837             /* Calculate displacement vector */
 838             dx00             = _mm_sub_ps(ix0,jx0);
 839             dy00             = _mm_sub_ps(iy0,jy0);
 840             dz00             = _mm_sub_ps(iz0,jz0);
 841             dx10             = _mm_sub_ps(ix1,jx0);
 842             dy10             = _mm_sub_ps(iy1,jy0);
 843             dz10             = _mm_sub_ps(iz1,jz0);
 844             dx20             = _mm_sub_ps(ix2,jx0);
 845             dy20             = _mm_sub_ps(iy2,jy0);
 846             dz20             = _mm_sub_ps(iz2,jz0);
 847
 848             /* Calculate squared distance and things based on it */
 849             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 850             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 851             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 852
 853             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 854             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 855             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 856
 857             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 858             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 859             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 860
 861             /* Load parameters for j particles */
 862             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 863                                                               charge+jnrC+0,charge+jnrD+0);
 864             vdwjidx0A        = 2*vdwtype[jnrA+0];
 865             vdwjidx0B        = 2*vdwtype[jnrB+0];
 866             vdwjidx0C        = 2*vdwtype[jnrC+0];
 867             vdwjidx0D        = 2*vdwtype[jnrD+0];
 868
 869             fjx0             = _mm_setzero_ps();
 870             fjy0             = _mm_setzero_ps();
 871             fjz0             = _mm_setzero_ps();
 872
 873             /**************************
 874              * CALCULATE INTERACTIONS *
 875              **************************/
 876
 877             if (gmx_mm_any_lt(rsq00,rcutoff2))
 878             {
 879
 880             r00              = _mm_mul_ps(rsq00,rinv00);
 881
 882             /* Compute parameters for interactions between i and j atoms */
 883             qq00             = _mm_mul_ps(iq0,jq0);
 884             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 885                                          vdwparam+vdwioffset0+vdwjidx0B,
 886                                          vdwparam+vdwioffset0+vdwjidx0C,
 887                                          vdwparam+vdwioffset0+vdwjidx0D,
 888                                          &c6_00,&c12_00);
 889
 890             /* EWALD ELECTROSTATICS */
 891
 892             /* Analytical PME correction */
 893             zeta2            = _mm_mul_ps(beta2,rsq00);
 894             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 895             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 896             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 897             felec            = _mm_mul_ps(qq00,felec);
 898
 899             /* LENNARD-JONES DISPERSION/REPULSION */
 900
 901             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 902             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 903
 904             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 905
 906             fscal            = _mm_add_ps(felec,fvdw);
 907
 908             fscal            = _mm_and_ps(fscal,cutoff_mask);
 909
 910              /* Update vectorial force */
 911             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 912             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 913             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 914
 915             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 916             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 917             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 918
 919             }
 920
 921             /**************************
 922              * CALCULATE INTERACTIONS *
 923              **************************/
 924
 925             if (gmx_mm_any_lt(rsq10,rcutoff2))
 926             {
 927
 928             r10              = _mm_mul_ps(rsq10,rinv10);
 929
 930             /* Compute parameters for interactions between i and j atoms */
 931             qq10             = _mm_mul_ps(iq1,jq0);
 932
 933             /* EWALD ELECTROSTATICS */
 934
 935             /* Analytical PME correction */
 936             zeta2            = _mm_mul_ps(beta2,rsq10);
 937             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 938             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 939             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 940             felec            = _mm_mul_ps(qq10,felec);
 941
 942             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 943
 944             fscal            = felec;
 945
 946             fscal            = _mm_and_ps(fscal,cutoff_mask);
 947
 948              /* Update vectorial force */
 949             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 950             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 951             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 952
 953             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 954             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 955             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 956
 957             }
 958
 959             /**************************
 960              * CALCULATE INTERACTIONS *
 961              **************************/
 962
 963             if (gmx_mm_any_lt(rsq20,rcutoff2))
 964             {
 965
 966             r20              = _mm_mul_ps(rsq20,rinv20);
 967
 968             /* Compute parameters for interactions between i and j atoms */
 969             qq20             = _mm_mul_ps(iq2,jq0);
 970
 971             /* EWALD ELECTROSTATICS */
 972
 973             /* Analytical PME correction */
 974             zeta2            = _mm_mul_ps(beta2,rsq20);
 975             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 976             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 977             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 978             felec            = _mm_mul_ps(qq20,felec);
 979
 980             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 981
 982             fscal            = felec;
 983
 984             fscal            = _mm_and_ps(fscal,cutoff_mask);
 985
 986              /* Update vectorial force */
 987             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 988             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 989             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 990
 991             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 992             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 993             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 994
 995             }
 996
 997             fjptrA             = f+j_coord_offsetA;
 998             fjptrB             = f+j_coord_offsetB;
 999             fjptrC             = f+j_coord_offsetC;
1000             fjptrD             = f+j_coord_offsetD;
1001
1002             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1003
1004             /* Inner loop uses 100 flops */
1005         }
1006
1007         if(jidx<j_index_end)
1008         {
1009
1010             /* Get j neighbor index, and coordinate index */
1011             jnrlistA         = jjnr[jidx];
1012             jnrlistB         = jjnr[jidx+1];
1013             jnrlistC         = jjnr[jidx+2];
1014             jnrlistD         = jjnr[jidx+3];
1015             /* Sign of each element will be negative for non-real atoms.
1016              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1017              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1018              */
1019             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1020             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1021             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1022             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1023             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1024             j_coord_offsetA  = DIM*jnrA;
1025             j_coord_offsetB  = DIM*jnrB;
1026             j_coord_offsetC  = DIM*jnrC;
1027             j_coord_offsetD  = DIM*jnrD;
1028
1029             /* load j atom coordinates */
1030             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1031                                               x+j_coord_offsetC,x+j_coord_offsetD,
1032                                               &jx0,&jy0,&jz0);
1033
1034             /* Calculate displacement vector */
1035             dx00             = _mm_sub_ps(ix0,jx0);
1036             dy00             = _mm_sub_ps(iy0,jy0);
1037             dz00             = _mm_sub_ps(iz0,jz0);
1038             dx10             = _mm_sub_ps(ix1,jx0);
1039             dy10             = _mm_sub_ps(iy1,jy0);
1040             dz10             = _mm_sub_ps(iz1,jz0);
1041             dx20             = _mm_sub_ps(ix2,jx0);
1042             dy20             = _mm_sub_ps(iy2,jy0);
1043             dz20             = _mm_sub_ps(iz2,jz0);
1044
1045             /* Calculate squared distance and things based on it */
1046             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1047             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1048             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1049
1050             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1051             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1052             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1053
1054             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1055             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1056             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1057
1058             /* Load parameters for j particles */
1059             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1060                                                               charge+jnrC+0,charge+jnrD+0);
1061             vdwjidx0A        = 2*vdwtype[jnrA+0];
1062             vdwjidx0B        = 2*vdwtype[jnrB+0];
1063             vdwjidx0C        = 2*vdwtype[jnrC+0];
1064             vdwjidx0D        = 2*vdwtype[jnrD+0];
1065
1066             fjx0             = _mm_setzero_ps();
1067             fjy0             = _mm_setzero_ps();
1068             fjz0             = _mm_setzero_ps();
1069
1070             /**************************
1071              * CALCULATE INTERACTIONS *
1072              **************************/
1073
1074             if (gmx_mm_any_lt(rsq00,rcutoff2))
1075             {
1076
1077             r00              = _mm_mul_ps(rsq00,rinv00);
1078             r00              = _mm_andnot_ps(dummy_mask,r00);
1079
1080             /* Compute parameters for interactions between i and j atoms */
1081             qq00             = _mm_mul_ps(iq0,jq0);
1082             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1083                                          vdwparam+vdwioffset0+vdwjidx0B,
1084                                          vdwparam+vdwioffset0+vdwjidx0C,
1085                                          vdwparam+vdwioffset0+vdwjidx0D,
1086                                          &c6_00,&c12_00);
1087
1088             /* EWALD ELECTROSTATICS */
1089
1090             /* Analytical PME correction */
1091             zeta2            = _mm_mul_ps(beta2,rsq00);
1092             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1093             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1094             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1095             felec            = _mm_mul_ps(qq00,felec);
1096
1097             /* LENNARD-JONES DISPERSION/REPULSION */
1098
1099             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1100             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1101
1102             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1103
1104             fscal            = _mm_add_ps(felec,fvdw);
1105
1106             fscal            = _mm_and_ps(fscal,cutoff_mask);
1107
1108             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1109
1110              /* Update vectorial force */
1111             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1112             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1113             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1114
1115             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1116             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1117             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1118
1119             }
1120
1121             /**************************
1122              * CALCULATE INTERACTIONS *
1123              **************************/
1124
1125             if (gmx_mm_any_lt(rsq10,rcutoff2))
1126             {
1127
1128             r10              = _mm_mul_ps(rsq10,rinv10);
1129             r10              = _mm_andnot_ps(dummy_mask,r10);
1130
1131             /* Compute parameters for interactions between i and j atoms */
1132             qq10             = _mm_mul_ps(iq1,jq0);
1133
1134             /* EWALD ELECTROSTATICS */
1135
1136             /* Analytical PME correction */
1137             zeta2            = _mm_mul_ps(beta2,rsq10);
1138             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1139             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1140             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1141             felec            = _mm_mul_ps(qq10,felec);
1142
1143             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1144
1145             fscal            = felec;
1146
1147             fscal            = _mm_and_ps(fscal,cutoff_mask);
1148
1149             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1150
1151              /* Update vectorial force */
1152             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1153             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1154             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1155
1156             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1157             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1158             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1159
1160             }
1161
1162             /**************************
1163              * CALCULATE INTERACTIONS *
1164              **************************/
1165
1166             if (gmx_mm_any_lt(rsq20,rcutoff2))
1167             {
1168
1169             r20              = _mm_mul_ps(rsq20,rinv20);
1170             r20              = _mm_andnot_ps(dummy_mask,r20);
1171
1172             /* Compute parameters for interactions between i and j atoms */
1173             qq20             = _mm_mul_ps(iq2,jq0);
1174
1175             /* EWALD ELECTROSTATICS */
1176
1177             /* Analytical PME correction */
1178             zeta2            = _mm_mul_ps(beta2,rsq20);
1179             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1180             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1181             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1182             felec            = _mm_mul_ps(qq20,felec);
1183
1184             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1185
1186             fscal            = felec;
1187
1188             fscal            = _mm_and_ps(fscal,cutoff_mask);
1189
1190             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1191
1192              /* Update vectorial force */
1193             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1194             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1195             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1196
1197             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1198             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1199             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1200
1201             }
1202
1203             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1204             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1205             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1206             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1207
1208             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1209
1210             /* Inner loop uses 103 flops */
1211         }
1212
1213         /* End of innermost loop */
1214
1215         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1216                                               f+i_coord_offset,fshift+i_shift_offset);
1217
1218         /* Increment number of inner iterations */
1219         inneriter                  += j_index_end - j_index_start;
1220
1221         /* Outer loop uses 18 flops */
1222     }
1223
1224     /* Increment number of outer iterations */
1225     outeriter        += nri;
1226
1227     /* Update outer/inner flops */
1228
1229     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*103);
1230 }