src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwNone_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     __m128i          ewitab;
  97     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  99     real             *ewtab;
 100     __m128           dummy_mask,cutoff_mask;
 101     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 102     __m128           one     = _mm_set1_ps(1.0);
 103     __m128           two     = _mm_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_ps(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117
 118     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 119     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 120     beta2            = _mm_mul_ps(beta,beta);
 121     beta3            = _mm_mul_ps(beta,beta2);
 122     ewtab            = fr->ic->tabq_coul_FDV0;
 123     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 124     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 129     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 130     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 131
 132     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 133     rcutoff_scalar   = fr->rcoulomb;
 134     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 135     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_ps();
 171         fiy0             = _mm_setzero_ps();
 172         fiz0             = _mm_setzero_ps();
 173         fix1             = _mm_setzero_ps();
 174         fiy1             = _mm_setzero_ps();
 175         fiz1             = _mm_setzero_ps();
 176         fix2             = _mm_setzero_ps();
 177         fiy2             = _mm_setzero_ps();
 178         fiz2             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             jnrC             = jjnr[jidx+2];
 191             jnrD             = jjnr[jidx+3];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194             j_coord_offsetC  = DIM*jnrC;
 195             j_coord_offsetD  = DIM*jnrD;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                               x+j_coord_offsetC,x+j_coord_offsetD,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_ps(ix0,jx0);
 204             dy00             = _mm_sub_ps(iy0,jy0);
 205             dz00             = _mm_sub_ps(iz0,jz0);
 206             dx10             = _mm_sub_ps(ix1,jx0);
 207             dy10             = _mm_sub_ps(iy1,jy0);
 208             dz10             = _mm_sub_ps(iz1,jz0);
 209             dx20             = _mm_sub_ps(ix2,jx0);
 210             dy20             = _mm_sub_ps(iy2,jy0);
 211             dz20             = _mm_sub_ps(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 217
 218             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 219             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 220             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 221
 222             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 223             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 224             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 228                                                               charge+jnrC+0,charge+jnrD+0);
 229
 230             fjx0             = _mm_setzero_ps();
 231             fjy0             = _mm_setzero_ps();
 232             fjz0             = _mm_setzero_ps();
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             if (gmx_mm_any_lt(rsq00,rcutoff2))
 239             {
 240
 241             r00              = _mm_mul_ps(rsq00,rinv00);
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             qq00             = _mm_mul_ps(iq0,jq0);
 245
 246             /* EWALD ELECTROSTATICS */
 247
 248             /* Analytical PME correction */
 249             zeta2            = _mm_mul_ps(beta2,rsq00);
 250             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 251             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 252             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 253             felec            = _mm_mul_ps(qq00,felec);
 254             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 255             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 256             velec            = _mm_mul_ps(qq00,velec);
 257
 258             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 259
 260             /* Update potential sum for this i atom from the interaction with this j atom. */
 261             velec            = _mm_and_ps(velec,cutoff_mask);
 262             velecsum         = _mm_add_ps(velecsum,velec);
 263
 264             fscal            = felec;
 265
 266             fscal            = _mm_and_ps(fscal,cutoff_mask);
 267
 268              /* Update vectorial force */
 269             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 270             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 271             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 272
 273             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 274             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 275             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 276
 277             }
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             if (gmx_mm_any_lt(rsq10,rcutoff2))
 284             {
 285
 286             r10              = _mm_mul_ps(rsq10,rinv10);
 287
 288             /* Compute parameters for interactions between i and j atoms */
 289             qq10             = _mm_mul_ps(iq1,jq0);
 290
 291             /* EWALD ELECTROSTATICS */
 292
 293             /* Analytical PME correction */
 294             zeta2            = _mm_mul_ps(beta2,rsq10);
 295             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 296             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 297             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 298             felec            = _mm_mul_ps(qq10,felec);
 299             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 300             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 301             velec            = _mm_mul_ps(qq10,velec);
 302
 303             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             velec            = _mm_and_ps(velec,cutoff_mask);
 307             velecsum         = _mm_add_ps(velecsum,velec);
 308
 309             fscal            = felec;
 310
 311             fscal            = _mm_and_ps(fscal,cutoff_mask);
 312
 313              /* Update vectorial force */
 314             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 315             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 316             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 317
 318             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 319             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 320             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 321
 322             }
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             if (gmx_mm_any_lt(rsq20,rcutoff2))
 329             {
 330
 331             r20              = _mm_mul_ps(rsq20,rinv20);
 332
 333             /* Compute parameters for interactions between i and j atoms */
 334             qq20             = _mm_mul_ps(iq2,jq0);
 335
 336             /* EWALD ELECTROSTATICS */
 337
 338             /* Analytical PME correction */
 339             zeta2            = _mm_mul_ps(beta2,rsq20);
 340             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 341             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 342             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 343             felec            = _mm_mul_ps(qq20,felec);
 344             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 345             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 346             velec            = _mm_mul_ps(qq20,velec);
 347
 348             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 349
 350             /* Update potential sum for this i atom from the interaction with this j atom. */
 351             velec            = _mm_and_ps(velec,cutoff_mask);
 352             velecsum         = _mm_add_ps(velecsum,velec);
 353
 354             fscal            = felec;
 355
 356             fscal            = _mm_and_ps(fscal,cutoff_mask);
 357
 358              /* Update vectorial force */
 359             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 360             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 361             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 362
 363             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 364             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 365             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 366
 367             }
 368
 369             fjptrA             = f+j_coord_offsetA;
 370             fjptrB             = f+j_coord_offsetB;
 371             fjptrC             = f+j_coord_offsetC;
 372             fjptrD             = f+j_coord_offsetD;
 373
 374             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 375
 376             /* Inner loop uses 99 flops */
 377         }
 378
 379         if(jidx<j_index_end)
 380         {
 381
 382             /* Get j neighbor index, and coordinate index */
 383             jnrlistA         = jjnr[jidx];
 384             jnrlistB         = jjnr[jidx+1];
 385             jnrlistC         = jjnr[jidx+2];
 386             jnrlistD         = jjnr[jidx+3];
 387             /* Sign of each element will be negative for non-real atoms.
 388              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 389              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 390              */
 391             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 392             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 393             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 394             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 395             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 396             j_coord_offsetA  = DIM*jnrA;
 397             j_coord_offsetB  = DIM*jnrB;
 398             j_coord_offsetC  = DIM*jnrC;
 399             j_coord_offsetD  = DIM*jnrD;
 400
 401             /* load j atom coordinates */
 402             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 403                                               x+j_coord_offsetC,x+j_coord_offsetD,
 404                                               &jx0,&jy0,&jz0);
 405
 406             /* Calculate displacement vector */
 407             dx00             = _mm_sub_ps(ix0,jx0);
 408             dy00             = _mm_sub_ps(iy0,jy0);
 409             dz00             = _mm_sub_ps(iz0,jz0);
 410             dx10             = _mm_sub_ps(ix1,jx0);
 411             dy10             = _mm_sub_ps(iy1,jy0);
 412             dz10             = _mm_sub_ps(iz1,jz0);
 413             dx20             = _mm_sub_ps(ix2,jx0);
 414             dy20             = _mm_sub_ps(iy2,jy0);
 415             dz20             = _mm_sub_ps(iz2,jz0);
 416
 417             /* Calculate squared distance and things based on it */
 418             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 419             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 420             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 421
 422             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 423             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 424             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 425
 426             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 427             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 428             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 429
 430             /* Load parameters for j particles */
 431             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 432                                                               charge+jnrC+0,charge+jnrD+0);
 433
 434             fjx0             = _mm_setzero_ps();
 435             fjy0             = _mm_setzero_ps();
 436             fjz0             = _mm_setzero_ps();
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             if (gmx_mm_any_lt(rsq00,rcutoff2))
 443             {
 444
 445             r00              = _mm_mul_ps(rsq00,rinv00);
 446             r00              = _mm_andnot_ps(dummy_mask,r00);
 447
 448             /* Compute parameters for interactions between i and j atoms */
 449             qq00             = _mm_mul_ps(iq0,jq0);
 450
 451             /* EWALD ELECTROSTATICS */
 452
 453             /* Analytical PME correction */
 454             zeta2            = _mm_mul_ps(beta2,rsq00);
 455             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 456             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 457             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 458             felec            = _mm_mul_ps(qq00,felec);
 459             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 460             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 461             velec            = _mm_mul_ps(qq00,velec);
 462
 463             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 464
 465             /* Update potential sum for this i atom from the interaction with this j atom. */
 466             velec            = _mm_and_ps(velec,cutoff_mask);
 467             velec            = _mm_andnot_ps(dummy_mask,velec);
 468             velecsum         = _mm_add_ps(velecsum,velec);
 469
 470             fscal            = felec;
 471
 472             fscal            = _mm_and_ps(fscal,cutoff_mask);
 473
 474             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 475
 476              /* Update vectorial force */
 477             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 478             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 479             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 480
 481             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 482             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 483             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 484
 485             }
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             if (gmx_mm_any_lt(rsq10,rcutoff2))
 492             {
 493
 494             r10              = _mm_mul_ps(rsq10,rinv10);
 495             r10              = _mm_andnot_ps(dummy_mask,r10);
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             qq10             = _mm_mul_ps(iq1,jq0);
 499
 500             /* EWALD ELECTROSTATICS */
 501
 502             /* Analytical PME correction */
 503             zeta2            = _mm_mul_ps(beta2,rsq10);
 504             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 505             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 506             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 507             felec            = _mm_mul_ps(qq10,felec);
 508             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 509             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 510             velec            = _mm_mul_ps(qq10,velec);
 511
 512             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 513
 514             /* Update potential sum for this i atom from the interaction with this j atom. */
 515             velec            = _mm_and_ps(velec,cutoff_mask);
 516             velec            = _mm_andnot_ps(dummy_mask,velec);
 517             velecsum         = _mm_add_ps(velecsum,velec);
 518
 519             fscal            = felec;
 520
 521             fscal            = _mm_and_ps(fscal,cutoff_mask);
 522
 523             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 524
 525              /* Update vectorial force */
 526             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 527             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 528             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 529
 530             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 531             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 532             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 533
 534             }
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 538              **************************/
 539
 540             if (gmx_mm_any_lt(rsq20,rcutoff2))
 541             {
 542
 543             r20              = _mm_mul_ps(rsq20,rinv20);
 544             r20              = _mm_andnot_ps(dummy_mask,r20);
 545
 546             /* Compute parameters for interactions between i and j atoms */
 547             qq20             = _mm_mul_ps(iq2,jq0);
 548
 549             /* EWALD ELECTROSTATICS */
 550
 551             /* Analytical PME correction */
 552             zeta2            = _mm_mul_ps(beta2,rsq20);
 553             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 554             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 555             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 556             felec            = _mm_mul_ps(qq20,felec);
 557             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 558             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 559             velec            = _mm_mul_ps(qq20,velec);
 560
 561             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             velec            = _mm_and_ps(velec,cutoff_mask);
 565             velec            = _mm_andnot_ps(dummy_mask,velec);
 566             velecsum         = _mm_add_ps(velecsum,velec);
 567
 568             fscal            = felec;
 569
 570             fscal            = _mm_and_ps(fscal,cutoff_mask);
 571
 572             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 573
 574              /* Update vectorial force */
 575             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 576             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 577             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 578
 579             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 580             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 581             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 582
 583             }
 584
 585             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 586             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 587             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 588             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 589
 590             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 591
 592             /* Inner loop uses 102 flops */
 593         }
 594
 595         /* End of innermost loop */
 596
 597         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 598                                               f+i_coord_offset,fshift+i_shift_offset);
 599
 600         ggid                        = gid[iidx];
 601         /* Update potential energies */
 602         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 603
 604         /* Increment number of inner iterations */
 605         inneriter                  += j_index_end - j_index_start;
 606
 607         /* Outer loop uses 19 flops */
 608     }
 609
 610     /* Increment number of outer iterations */
 611     outeriter        += nri;
 612
 613     /* Update outer/inner flops */
 614
 615     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*102);
 616 }
 617 /*
 618  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_avx_128_fma_single
 619  * Electrostatics interaction: Ewald
 620  * VdW interaction:            None
 621  * Geometry:                   Water3-Particle
 622  * Calculate force/pot:        Force
 623  */
 624 void
 625 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_avx_128_fma_single
 626                     (t_nblist                    * gmx_restrict       nlist,
 627                      rvec                        * gmx_restrict          xx,
 628                      rvec                        * gmx_restrict          ff,
 629                      t_forcerec                  * gmx_restrict          fr,
 630                      t_mdatoms                   * gmx_restrict     mdatoms,
 631                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 632                      t_nrnb                      * gmx_restrict        nrnb)
 633 {
 634     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 635      * just 0 for non-waters.
 636      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 637      * jnr indices corresponding to data put in the four positions in the SIMD register.
 638      */
 639     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 640     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 641     int              jnrA,jnrB,jnrC,jnrD;
 642     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 643     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 644     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 645     real             rcutoff_scalar;
 646     real             *shiftvec,*fshift,*x,*f;
 647     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 648     real             scratch[4*DIM];
 649     __m128           fscal,rcutoff,rcutoff2,jidxall;
 650     int              vdwioffset0;
 651     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 652     int              vdwioffset1;
 653     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 654     int              vdwioffset2;
 655     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 656     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 657     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 658     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 659     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 660     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 661     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 662     real             *charge;
 663     __m128i          ewitab;
 664     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 665     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 666     real             *ewtab;
 667     __m128           dummy_mask,cutoff_mask;
 668     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 669     __m128           one     = _mm_set1_ps(1.0);
 670     __m128           two     = _mm_set1_ps(2.0);
 671     x                = xx[0];
 672     f                = ff[0];
 673
 674     nri              = nlist->nri;
 675     iinr             = nlist->iinr;
 676     jindex           = nlist->jindex;
 677     jjnr             = nlist->jjnr;
 678     shiftidx         = nlist->shift;
 679     gid              = nlist->gid;
 680     shiftvec         = fr->shift_vec[0];
 681     fshift           = fr->fshift[0];
 682     facel            = _mm_set1_ps(fr->epsfac);
 683     charge           = mdatoms->chargeA;
 684
 685     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 686     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 687     beta2            = _mm_mul_ps(beta,beta);
 688     beta3            = _mm_mul_ps(beta,beta2);
 689     ewtab            = fr->ic->tabq_coul_F;
 690     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 691     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 692
 693     /* Setup water-specific parameters */
 694     inr              = nlist->iinr[0];
 695     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 696     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 697     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 698
 699     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 700     rcutoff_scalar   = fr->rcoulomb;
 701     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 702     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 703
 704     /* Avoid stupid compiler warnings */
 705     jnrA = jnrB = jnrC = jnrD = 0;
 706     j_coord_offsetA = 0;
 707     j_coord_offsetB = 0;
 708     j_coord_offsetC = 0;
 709     j_coord_offsetD = 0;
 710
 711     outeriter        = 0;
 712     inneriter        = 0;
 713
 714     for(iidx=0;iidx<4*DIM;iidx++)
 715     {
 716         scratch[iidx] = 0.0;
 717     }
 718
 719     /* Start outer loop over neighborlists */
 720     for(iidx=0; iidx<nri; iidx++)
 721     {
 722         /* Load shift vector for this list */
 723         i_shift_offset   = DIM*shiftidx[iidx];
 724
 725         /* Load limits for loop over neighbors */
 726         j_index_start    = jindex[iidx];
 727         j_index_end      = jindex[iidx+1];
 728
 729         /* Get outer coordinate index */
 730         inr              = iinr[iidx];
 731         i_coord_offset   = DIM*inr;
 732
 733         /* Load i particle coords and add shift vector */
 734         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 735                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 736
 737         fix0             = _mm_setzero_ps();
 738         fiy0             = _mm_setzero_ps();
 739         fiz0             = _mm_setzero_ps();
 740         fix1             = _mm_setzero_ps();
 741         fiy1             = _mm_setzero_ps();
 742         fiz1             = _mm_setzero_ps();
 743         fix2             = _mm_setzero_ps();
 744         fiy2             = _mm_setzero_ps();
 745         fiz2             = _mm_setzero_ps();
 746
 747         /* Start inner kernel loop */
 748         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 749         {
 750
 751             /* Get j neighbor index, and coordinate index */
 752             jnrA             = jjnr[jidx];
 753             jnrB             = jjnr[jidx+1];
 754             jnrC             = jjnr[jidx+2];
 755             jnrD             = jjnr[jidx+3];
 756             j_coord_offsetA  = DIM*jnrA;
 757             j_coord_offsetB  = DIM*jnrB;
 758             j_coord_offsetC  = DIM*jnrC;
 759             j_coord_offsetD  = DIM*jnrD;
 760
 761             /* load j atom coordinates */
 762             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 763                                               x+j_coord_offsetC,x+j_coord_offsetD,
 764                                               &jx0,&jy0,&jz0);
 765
 766             /* Calculate displacement vector */
 767             dx00             = _mm_sub_ps(ix0,jx0);
 768             dy00             = _mm_sub_ps(iy0,jy0);
 769             dz00             = _mm_sub_ps(iz0,jz0);
 770             dx10             = _mm_sub_ps(ix1,jx0);
 771             dy10             = _mm_sub_ps(iy1,jy0);
 772             dz10             = _mm_sub_ps(iz1,jz0);
 773             dx20             = _mm_sub_ps(ix2,jx0);
 774             dy20             = _mm_sub_ps(iy2,jy0);
 775             dz20             = _mm_sub_ps(iz2,jz0);
 776
 777             /* Calculate squared distance and things based on it */
 778             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 779             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 780             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 781
 782             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 783             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 784             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 785
 786             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 787             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 788             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 789
 790             /* Load parameters for j particles */
 791             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 792                                                               charge+jnrC+0,charge+jnrD+0);
 793
 794             fjx0             = _mm_setzero_ps();
 795             fjy0             = _mm_setzero_ps();
 796             fjz0             = _mm_setzero_ps();
 797
 798             /**************************
 799              * CALCULATE INTERACTIONS *
 800              **************************/
 801
 802             if (gmx_mm_any_lt(rsq00,rcutoff2))
 803             {
 804
 805             r00              = _mm_mul_ps(rsq00,rinv00);
 806
 807             /* Compute parameters for interactions between i and j atoms */
 808             qq00             = _mm_mul_ps(iq0,jq0);
 809
 810             /* EWALD ELECTROSTATICS */
 811
 812             /* Analytical PME correction */
 813             zeta2            = _mm_mul_ps(beta2,rsq00);
 814             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 815             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 816             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 817             felec            = _mm_mul_ps(qq00,felec);
 818
 819             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 820
 821             fscal            = felec;
 822
 823             fscal            = _mm_and_ps(fscal,cutoff_mask);
 824
 825              /* Update vectorial force */
 826             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 827             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 828             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 829
 830             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 831             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 832             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 833
 834             }
 835
 836             /**************************
 837              * CALCULATE INTERACTIONS *
 838              **************************/
 839
 840             if (gmx_mm_any_lt(rsq10,rcutoff2))
 841             {
 842
 843             r10              = _mm_mul_ps(rsq10,rinv10);
 844
 845             /* Compute parameters for interactions between i and j atoms */
 846             qq10             = _mm_mul_ps(iq1,jq0);
 847
 848             /* EWALD ELECTROSTATICS */
 849
 850             /* Analytical PME correction */
 851             zeta2            = _mm_mul_ps(beta2,rsq10);
 852             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 853             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 854             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 855             felec            = _mm_mul_ps(qq10,felec);
 856
 857             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 858
 859             fscal            = felec;
 860
 861             fscal            = _mm_and_ps(fscal,cutoff_mask);
 862
 863              /* Update vectorial force */
 864             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 865             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 866             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 867
 868             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 869             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 870             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 871
 872             }
 873
 874             /**************************
 875              * CALCULATE INTERACTIONS *
 876              **************************/
 877
 878             if (gmx_mm_any_lt(rsq20,rcutoff2))
 879             {
 880
 881             r20              = _mm_mul_ps(rsq20,rinv20);
 882
 883             /* Compute parameters for interactions between i and j atoms */
 884             qq20             = _mm_mul_ps(iq2,jq0);
 885
 886             /* EWALD ELECTROSTATICS */
 887
 888             /* Analytical PME correction */
 889             zeta2            = _mm_mul_ps(beta2,rsq20);
 890             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 891             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 892             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 893             felec            = _mm_mul_ps(qq20,felec);
 894
 895             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 896
 897             fscal            = felec;
 898
 899             fscal            = _mm_and_ps(fscal,cutoff_mask);
 900
 901              /* Update vectorial force */
 902             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 903             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 904             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 905
 906             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 907             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 908             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 909
 910             }
 911
 912             fjptrA             = f+j_coord_offsetA;
 913             fjptrB             = f+j_coord_offsetB;
 914             fjptrC             = f+j_coord_offsetC;
 915             fjptrD             = f+j_coord_offsetD;
 916
 917             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 918
 919             /* Inner loop uses 93 flops */
 920         }
 921
 922         if(jidx<j_index_end)
 923         {
 924
 925             /* Get j neighbor index, and coordinate index */
 926             jnrlistA         = jjnr[jidx];
 927             jnrlistB         = jjnr[jidx+1];
 928             jnrlistC         = jjnr[jidx+2];
 929             jnrlistD         = jjnr[jidx+3];
 930             /* Sign of each element will be negative for non-real atoms.
 931              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 932              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 933              */
 934             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 935             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 936             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 937             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 938             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 939             j_coord_offsetA  = DIM*jnrA;
 940             j_coord_offsetB  = DIM*jnrB;
 941             j_coord_offsetC  = DIM*jnrC;
 942             j_coord_offsetD  = DIM*jnrD;
 943
 944             /* load j atom coordinates */
 945             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 946                                               x+j_coord_offsetC,x+j_coord_offsetD,
 947                                               &jx0,&jy0,&jz0);
 948
 949             /* Calculate displacement vector */
 950             dx00             = _mm_sub_ps(ix0,jx0);
 951             dy00             = _mm_sub_ps(iy0,jy0);
 952             dz00             = _mm_sub_ps(iz0,jz0);
 953             dx10             = _mm_sub_ps(ix1,jx0);
 954             dy10             = _mm_sub_ps(iy1,jy0);
 955             dz10             = _mm_sub_ps(iz1,jz0);
 956             dx20             = _mm_sub_ps(ix2,jx0);
 957             dy20             = _mm_sub_ps(iy2,jy0);
 958             dz20             = _mm_sub_ps(iz2,jz0);
 959
 960             /* Calculate squared distance and things based on it */
 961             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 962             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 963             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 964
 965             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 966             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 967             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 968
 969             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 970             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 971             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 972
 973             /* Load parameters for j particles */
 974             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 975                                                               charge+jnrC+0,charge+jnrD+0);
 976
 977             fjx0             = _mm_setzero_ps();
 978             fjy0             = _mm_setzero_ps();
 979             fjz0             = _mm_setzero_ps();
 980
 981             /**************************
 982              * CALCULATE INTERACTIONS *
 983              **************************/
 984
 985             if (gmx_mm_any_lt(rsq00,rcutoff2))
 986             {
 987
 988             r00              = _mm_mul_ps(rsq00,rinv00);
 989             r00              = _mm_andnot_ps(dummy_mask,r00);
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             qq00             = _mm_mul_ps(iq0,jq0);
 993
 994             /* EWALD ELECTROSTATICS */
 995
 996             /* Analytical PME correction */
 997             zeta2            = _mm_mul_ps(beta2,rsq00);
 998             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 999             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1000             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1001             felec            = _mm_mul_ps(qq00,felec);
1002
1003             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1004
1005             fscal            = felec;
1006
1007             fscal            = _mm_and_ps(fscal,cutoff_mask);
1008
1009             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1010
1011              /* Update vectorial force */
1012             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1013             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1014             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1015
1016             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1017             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1018             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1019
1020             }
1021
1022             /**************************
1023              * CALCULATE INTERACTIONS *
1024              **************************/
1025
1026             if (gmx_mm_any_lt(rsq10,rcutoff2))
1027             {
1028
1029             r10              = _mm_mul_ps(rsq10,rinv10);
1030             r10              = _mm_andnot_ps(dummy_mask,r10);
1031
1032             /* Compute parameters for interactions between i and j atoms */
1033             qq10             = _mm_mul_ps(iq1,jq0);
1034
1035             /* EWALD ELECTROSTATICS */
1036
1037             /* Analytical PME correction */
1038             zeta2            = _mm_mul_ps(beta2,rsq10);
1039             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1040             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1041             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1042             felec            = _mm_mul_ps(qq10,felec);
1043
1044             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1045
1046             fscal            = felec;
1047
1048             fscal            = _mm_and_ps(fscal,cutoff_mask);
1049
1050             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1051
1052              /* Update vectorial force */
1053             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1054             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1055             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1056
1057             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1058             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1059             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1060
1061             }
1062
1063             /**************************
1064              * CALCULATE INTERACTIONS *
1065              **************************/
1066
1067             if (gmx_mm_any_lt(rsq20,rcutoff2))
1068             {
1069
1070             r20              = _mm_mul_ps(rsq20,rinv20);
1071             r20              = _mm_andnot_ps(dummy_mask,r20);
1072
1073             /* Compute parameters for interactions between i and j atoms */
1074             qq20             = _mm_mul_ps(iq2,jq0);
1075
1076             /* EWALD ELECTROSTATICS */
1077
1078             /* Analytical PME correction */
1079             zeta2            = _mm_mul_ps(beta2,rsq20);
1080             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1081             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1082             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1083             felec            = _mm_mul_ps(qq20,felec);
1084
1085             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1086
1087             fscal            = felec;
1088
1089             fscal            = _mm_and_ps(fscal,cutoff_mask);
1090
1091             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1092
1093              /* Update vectorial force */
1094             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1095             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1096             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1097
1098             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1099             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1100             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1101
1102             }
1103
1104             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1105             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1106             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1107             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1108
1109             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1110
1111             /* Inner loop uses 96 flops */
1112         }
1113
1114         /* End of innermost loop */
1115
1116         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1117                                               f+i_coord_offset,fshift+i_shift_offset);
1118
1119         /* Increment number of inner iterations */
1120         inneriter                  += j_index_end - j_index_start;
1121
1122         /* Outer loop uses 18 flops */
1123     }
1124
1125     /* Increment number of outer iterations */
1126     outeriter        += nri;
1127
1128     /* Update outer/inner flops */
1129
1130     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*96);
1131 }