src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwNone_GeomW4P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_VF_avx_128_fma_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_VF_avx_128_fma_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset1;
  83     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128i          ewitab;
  96     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  98     real             *ewtab;
  99     __m128           dummy_mask,cutoff_mask;
 100     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 101     __m128           one     = _mm_set1_ps(1.0);
 102     __m128           two     = _mm_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_ps(fr->ic->epsfac);
 115     charge           = mdatoms->chargeA;
 116
 117     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 118     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 119     beta2            = _mm_mul_ps(beta,beta);
 120     beta3            = _mm_mul_ps(beta,beta2);
 121     ewtab            = fr->ic->tabq_coul_FDV0;
 122     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 123     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 124
 125     /* Setup water-specific parameters */
 126     inr              = nlist->iinr[0];
 127     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 128     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 129     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 130
 131     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 132     rcutoff_scalar   = fr->ic->rcoulomb;
 133     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 134     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     for(iidx=0;iidx<4*DIM;iidx++)
 147     {
 148         scratch[iidx] = 0.0;
 149     }
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 167                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 168
 169         fix1             = _mm_setzero_ps();
 170         fiy1             = _mm_setzero_ps();
 171         fiz1             = _mm_setzero_ps();
 172         fix2             = _mm_setzero_ps();
 173         fiy2             = _mm_setzero_ps();
 174         fiz2             = _mm_setzero_ps();
 175         fix3             = _mm_setzero_ps();
 176         fiy3             = _mm_setzero_ps();
 177         fiz3             = _mm_setzero_ps();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_ps();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             jnrC             = jjnr[jidx+2];
 190             jnrD             = jjnr[jidx+3];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               x+j_coord_offsetC,x+j_coord_offsetD,
 199                                               &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx10             = _mm_sub_ps(ix1,jx0);
 203             dy10             = _mm_sub_ps(iy1,jy0);
 204             dz10             = _mm_sub_ps(iz1,jz0);
 205             dx20             = _mm_sub_ps(ix2,jx0);
 206             dy20             = _mm_sub_ps(iy2,jy0);
 207             dz20             = _mm_sub_ps(iz2,jz0);
 208             dx30             = _mm_sub_ps(ix3,jx0);
 209             dy30             = _mm_sub_ps(iy3,jy0);
 210             dz30             = _mm_sub_ps(iz3,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 215             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 216
 217             rinv10           = avx128fma_invsqrt_f(rsq10);
 218             rinv20           = avx128fma_invsqrt_f(rsq20);
 219             rinv30           = avx128fma_invsqrt_f(rsq30);
 220
 221             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 223             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                               charge+jnrC+0,charge+jnrD+0);
 228
 229             fjx0             = _mm_setzero_ps();
 230             fjy0             = _mm_setzero_ps();
 231             fjz0             = _mm_setzero_ps();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             if (gmx_mm_any_lt(rsq10,rcutoff2))
 238             {
 239
 240             r10              = _mm_mul_ps(rsq10,rinv10);
 241
 242             /* Compute parameters for interactions between i and j atoms */
 243             qq10             = _mm_mul_ps(iq1,jq0);
 244
 245             /* EWALD ELECTROSTATICS */
 246
 247             /* Analytical PME correction */
 248             zeta2            = _mm_mul_ps(beta2,rsq10);
 249             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 250             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 251             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 252             felec            = _mm_mul_ps(qq10,felec);
 253             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 254             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 255             velec            = _mm_mul_ps(qq10,velec);
 256
 257             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 258
 259             /* Update potential sum for this i atom from the interaction with this j atom. */
 260             velec            = _mm_and_ps(velec,cutoff_mask);
 261             velecsum         = _mm_add_ps(velecsum,velec);
 262
 263             fscal            = felec;
 264
 265             fscal            = _mm_and_ps(fscal,cutoff_mask);
 266
 267              /* Update vectorial force */
 268             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 269             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 270             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 271
 272             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 273             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 274             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 275
 276             }
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             if (gmx_mm_any_lt(rsq20,rcutoff2))
 283             {
 284
 285             r20              = _mm_mul_ps(rsq20,rinv20);
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq20             = _mm_mul_ps(iq2,jq0);
 289
 290             /* EWALD ELECTROSTATICS */
 291
 292             /* Analytical PME correction */
 293             zeta2            = _mm_mul_ps(beta2,rsq20);
 294             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 295             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 296             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 297             felec            = _mm_mul_ps(qq20,felec);
 298             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 299             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 300             velec            = _mm_mul_ps(qq20,velec);
 301
 302             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velec            = _mm_and_ps(velec,cutoff_mask);
 306             velecsum         = _mm_add_ps(velecsum,velec);
 307
 308             fscal            = felec;
 309
 310             fscal            = _mm_and_ps(fscal,cutoff_mask);
 311
 312              /* Update vectorial force */
 313             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 314             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 315             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 316
 317             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 318             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 319             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 320
 321             }
 322
 323             /**************************
 324              * CALCULATE INTERACTIONS *
 325              **************************/
 326
 327             if (gmx_mm_any_lt(rsq30,rcutoff2))
 328             {
 329
 330             r30              = _mm_mul_ps(rsq30,rinv30);
 331
 332             /* Compute parameters for interactions between i and j atoms */
 333             qq30             = _mm_mul_ps(iq3,jq0);
 334
 335             /* EWALD ELECTROSTATICS */
 336
 337             /* Analytical PME correction */
 338             zeta2            = _mm_mul_ps(beta2,rsq30);
 339             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 340             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 341             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 342             felec            = _mm_mul_ps(qq30,felec);
 343             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 344             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv30,sh_ewald));
 345             velec            = _mm_mul_ps(qq30,velec);
 346
 347             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velec            = _mm_and_ps(velec,cutoff_mask);
 351             velecsum         = _mm_add_ps(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355             fscal            = _mm_and_ps(fscal,cutoff_mask);
 356
 357              /* Update vectorial force */
 358             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 359             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 360             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 361
 362             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 363             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 364             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 365
 366             }
 367
 368             fjptrA             = f+j_coord_offsetA;
 369             fjptrB             = f+j_coord_offsetB;
 370             fjptrC             = f+j_coord_offsetC;
 371             fjptrD             = f+j_coord_offsetD;
 372
 373             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 374
 375             /* Inner loop uses 99 flops */
 376         }
 377
 378         if(jidx<j_index_end)
 379         {
 380
 381             /* Get j neighbor index, and coordinate index */
 382             jnrlistA         = jjnr[jidx];
 383             jnrlistB         = jjnr[jidx+1];
 384             jnrlistC         = jjnr[jidx+2];
 385             jnrlistD         = jjnr[jidx+3];
 386             /* Sign of each element will be negative for non-real atoms.
 387              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 388              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 389              */
 390             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 391             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 392             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 393             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 394             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 395             j_coord_offsetA  = DIM*jnrA;
 396             j_coord_offsetB  = DIM*jnrB;
 397             j_coord_offsetC  = DIM*jnrC;
 398             j_coord_offsetD  = DIM*jnrD;
 399
 400             /* load j atom coordinates */
 401             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 402                                               x+j_coord_offsetC,x+j_coord_offsetD,
 403                                               &jx0,&jy0,&jz0);
 404
 405             /* Calculate displacement vector */
 406             dx10             = _mm_sub_ps(ix1,jx0);
 407             dy10             = _mm_sub_ps(iy1,jy0);
 408             dz10             = _mm_sub_ps(iz1,jz0);
 409             dx20             = _mm_sub_ps(ix2,jx0);
 410             dy20             = _mm_sub_ps(iy2,jy0);
 411             dz20             = _mm_sub_ps(iz2,jz0);
 412             dx30             = _mm_sub_ps(ix3,jx0);
 413             dy30             = _mm_sub_ps(iy3,jy0);
 414             dz30             = _mm_sub_ps(iz3,jz0);
 415
 416             /* Calculate squared distance and things based on it */
 417             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 418             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 419             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 420
 421             rinv10           = avx128fma_invsqrt_f(rsq10);
 422             rinv20           = avx128fma_invsqrt_f(rsq20);
 423             rinv30           = avx128fma_invsqrt_f(rsq30);
 424
 425             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 426             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 427             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 428
 429             /* Load parameters for j particles */
 430             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 431                                                               charge+jnrC+0,charge+jnrD+0);
 432
 433             fjx0             = _mm_setzero_ps();
 434             fjy0             = _mm_setzero_ps();
 435             fjz0             = _mm_setzero_ps();
 436
 437             /**************************
 438              * CALCULATE INTERACTIONS *
 439              **************************/
 440
 441             if (gmx_mm_any_lt(rsq10,rcutoff2))
 442             {
 443
 444             r10              = _mm_mul_ps(rsq10,rinv10);
 445             r10              = _mm_andnot_ps(dummy_mask,r10);
 446
 447             /* Compute parameters for interactions between i and j atoms */
 448             qq10             = _mm_mul_ps(iq1,jq0);
 449
 450             /* EWALD ELECTROSTATICS */
 451
 452             /* Analytical PME correction */
 453             zeta2            = _mm_mul_ps(beta2,rsq10);
 454             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 455             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 456             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 457             felec            = _mm_mul_ps(qq10,felec);
 458             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 459             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 460             velec            = _mm_mul_ps(qq10,velec);
 461
 462             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 463
 464             /* Update potential sum for this i atom from the interaction with this j atom. */
 465             velec            = _mm_and_ps(velec,cutoff_mask);
 466             velec            = _mm_andnot_ps(dummy_mask,velec);
 467             velecsum         = _mm_add_ps(velecsum,velec);
 468
 469             fscal            = felec;
 470
 471             fscal            = _mm_and_ps(fscal,cutoff_mask);
 472
 473             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 474
 475              /* Update vectorial force */
 476             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 477             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 478             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 479
 480             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 481             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 482             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 483
 484             }
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             if (gmx_mm_any_lt(rsq20,rcutoff2))
 491             {
 492
 493             r20              = _mm_mul_ps(rsq20,rinv20);
 494             r20              = _mm_andnot_ps(dummy_mask,r20);
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             qq20             = _mm_mul_ps(iq2,jq0);
 498
 499             /* EWALD ELECTROSTATICS */
 500
 501             /* Analytical PME correction */
 502             zeta2            = _mm_mul_ps(beta2,rsq20);
 503             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 504             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 505             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 506             felec            = _mm_mul_ps(qq20,felec);
 507             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 508             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 509             velec            = _mm_mul_ps(qq20,velec);
 510
 511             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 512
 513             /* Update potential sum for this i atom from the interaction with this j atom. */
 514             velec            = _mm_and_ps(velec,cutoff_mask);
 515             velec            = _mm_andnot_ps(dummy_mask,velec);
 516             velecsum         = _mm_add_ps(velecsum,velec);
 517
 518             fscal            = felec;
 519
 520             fscal            = _mm_and_ps(fscal,cutoff_mask);
 521
 522             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 523
 524              /* Update vectorial force */
 525             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 526             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 527             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 528
 529             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 530             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 531             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 532
 533             }
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             if (gmx_mm_any_lt(rsq30,rcutoff2))
 540             {
 541
 542             r30              = _mm_mul_ps(rsq30,rinv30);
 543             r30              = _mm_andnot_ps(dummy_mask,r30);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             qq30             = _mm_mul_ps(iq3,jq0);
 547
 548             /* EWALD ELECTROSTATICS */
 549
 550             /* Analytical PME correction */
 551             zeta2            = _mm_mul_ps(beta2,rsq30);
 552             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 553             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 554             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 555             felec            = _mm_mul_ps(qq30,felec);
 556             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 557             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv30,sh_ewald));
 558             velec            = _mm_mul_ps(qq30,velec);
 559
 560             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             velec            = _mm_and_ps(velec,cutoff_mask);
 564             velec            = _mm_andnot_ps(dummy_mask,velec);
 565             velecsum         = _mm_add_ps(velecsum,velec);
 566
 567             fscal            = felec;
 568
 569             fscal            = _mm_and_ps(fscal,cutoff_mask);
 570
 571             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 572
 573              /* Update vectorial force */
 574             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 575             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 576             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 577
 578             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 579             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 580             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 581
 582             }
 583
 584             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 585             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 586             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 587             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 588
 589             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 590
 591             /* Inner loop uses 102 flops */
 592         }
 593
 594         /* End of innermost loop */
 595
 596         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 597                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 598
 599         ggid                        = gid[iidx];
 600         /* Update potential energies */
 601         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 602
 603         /* Increment number of inner iterations */
 604         inneriter                  += j_index_end - j_index_start;
 605
 606         /* Outer loop uses 19 flops */
 607     }
 608
 609     /* Increment number of outer iterations */
 610     outeriter        += nri;
 611
 612     /* Update outer/inner flops */
 613
 614     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*102);
 615 }
 616 /*
 617  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_avx_128_fma_single
 618  * Electrostatics interaction: Ewald
 619  * VdW interaction:            None
 620  * Geometry:                   Water4-Particle
 621  * Calculate force/pot:        Force
 622  */
 623 void
 624 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_avx_128_fma_single
 625                     (t_nblist                    * gmx_restrict       nlist,
 626                      rvec                        * gmx_restrict          xx,
 627                      rvec                        * gmx_restrict          ff,
 628                      struct t_forcerec           * gmx_restrict          fr,
 629                      t_mdatoms                   * gmx_restrict     mdatoms,
 630                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 631                      t_nrnb                      * gmx_restrict        nrnb)
 632 {
 633     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 634      * just 0 for non-waters.
 635      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 636      * jnr indices corresponding to data put in the four positions in the SIMD register.
 637      */
 638     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 639     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 640     int              jnrA,jnrB,jnrC,jnrD;
 641     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 642     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 643     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 644     real             rcutoff_scalar;
 645     real             *shiftvec,*fshift,*x,*f;
 646     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 647     real             scratch[4*DIM];
 648     __m128           fscal,rcutoff,rcutoff2,jidxall;
 649     int              vdwioffset1;
 650     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 651     int              vdwioffset2;
 652     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 653     int              vdwioffset3;
 654     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 655     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 656     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 657     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 658     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 659     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 660     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 661     real             *charge;
 662     __m128i          ewitab;
 663     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 664     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 665     real             *ewtab;
 666     __m128           dummy_mask,cutoff_mask;
 667     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 668     __m128           one     = _mm_set1_ps(1.0);
 669     __m128           two     = _mm_set1_ps(2.0);
 670     x                = xx[0];
 671     f                = ff[0];
 672
 673     nri              = nlist->nri;
 674     iinr             = nlist->iinr;
 675     jindex           = nlist->jindex;
 676     jjnr             = nlist->jjnr;
 677     shiftidx         = nlist->shift;
 678     gid              = nlist->gid;
 679     shiftvec         = fr->shift_vec[0];
 680     fshift           = fr->fshift[0];
 681     facel            = _mm_set1_ps(fr->ic->epsfac);
 682     charge           = mdatoms->chargeA;
 683
 684     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 685     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 686     beta2            = _mm_mul_ps(beta,beta);
 687     beta3            = _mm_mul_ps(beta,beta2);
 688     ewtab            = fr->ic->tabq_coul_F;
 689     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 690     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 691
 692     /* Setup water-specific parameters */
 693     inr              = nlist->iinr[0];
 694     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 695     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 696     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 697
 698     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 699     rcutoff_scalar   = fr->ic->rcoulomb;
 700     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 701     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 702
 703     /* Avoid stupid compiler warnings */
 704     jnrA = jnrB = jnrC = jnrD = 0;
 705     j_coord_offsetA = 0;
 706     j_coord_offsetB = 0;
 707     j_coord_offsetC = 0;
 708     j_coord_offsetD = 0;
 709
 710     outeriter        = 0;
 711     inneriter        = 0;
 712
 713     for(iidx=0;iidx<4*DIM;iidx++)
 714     {
 715         scratch[iidx] = 0.0;
 716     }
 717
 718     /* Start outer loop over neighborlists */
 719     for(iidx=0; iidx<nri; iidx++)
 720     {
 721         /* Load shift vector for this list */
 722         i_shift_offset   = DIM*shiftidx[iidx];
 723
 724         /* Load limits for loop over neighbors */
 725         j_index_start    = jindex[iidx];
 726         j_index_end      = jindex[iidx+1];
 727
 728         /* Get outer coordinate index */
 729         inr              = iinr[iidx];
 730         i_coord_offset   = DIM*inr;
 731
 732         /* Load i particle coords and add shift vector */
 733         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 734                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 735
 736         fix1             = _mm_setzero_ps();
 737         fiy1             = _mm_setzero_ps();
 738         fiz1             = _mm_setzero_ps();
 739         fix2             = _mm_setzero_ps();
 740         fiy2             = _mm_setzero_ps();
 741         fiz2             = _mm_setzero_ps();
 742         fix3             = _mm_setzero_ps();
 743         fiy3             = _mm_setzero_ps();
 744         fiz3             = _mm_setzero_ps();
 745
 746         /* Start inner kernel loop */
 747         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 748         {
 749
 750             /* Get j neighbor index, and coordinate index */
 751             jnrA             = jjnr[jidx];
 752             jnrB             = jjnr[jidx+1];
 753             jnrC             = jjnr[jidx+2];
 754             jnrD             = jjnr[jidx+3];
 755             j_coord_offsetA  = DIM*jnrA;
 756             j_coord_offsetB  = DIM*jnrB;
 757             j_coord_offsetC  = DIM*jnrC;
 758             j_coord_offsetD  = DIM*jnrD;
 759
 760             /* load j atom coordinates */
 761             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 762                                               x+j_coord_offsetC,x+j_coord_offsetD,
 763                                               &jx0,&jy0,&jz0);
 764
 765             /* Calculate displacement vector */
 766             dx10             = _mm_sub_ps(ix1,jx0);
 767             dy10             = _mm_sub_ps(iy1,jy0);
 768             dz10             = _mm_sub_ps(iz1,jz0);
 769             dx20             = _mm_sub_ps(ix2,jx0);
 770             dy20             = _mm_sub_ps(iy2,jy0);
 771             dz20             = _mm_sub_ps(iz2,jz0);
 772             dx30             = _mm_sub_ps(ix3,jx0);
 773             dy30             = _mm_sub_ps(iy3,jy0);
 774             dz30             = _mm_sub_ps(iz3,jz0);
 775
 776             /* Calculate squared distance and things based on it */
 777             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 778             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 779             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 780
 781             rinv10           = avx128fma_invsqrt_f(rsq10);
 782             rinv20           = avx128fma_invsqrt_f(rsq20);
 783             rinv30           = avx128fma_invsqrt_f(rsq30);
 784
 785             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 786             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 787             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 788
 789             /* Load parameters for j particles */
 790             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 791                                                               charge+jnrC+0,charge+jnrD+0);
 792
 793             fjx0             = _mm_setzero_ps();
 794             fjy0             = _mm_setzero_ps();
 795             fjz0             = _mm_setzero_ps();
 796
 797             /**************************
 798              * CALCULATE INTERACTIONS *
 799              **************************/
 800
 801             if (gmx_mm_any_lt(rsq10,rcutoff2))
 802             {
 803
 804             r10              = _mm_mul_ps(rsq10,rinv10);
 805
 806             /* Compute parameters for interactions between i and j atoms */
 807             qq10             = _mm_mul_ps(iq1,jq0);
 808
 809             /* EWALD ELECTROSTATICS */
 810
 811             /* Analytical PME correction */
 812             zeta2            = _mm_mul_ps(beta2,rsq10);
 813             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 814             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 815             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 816             felec            = _mm_mul_ps(qq10,felec);
 817
 818             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 819
 820             fscal            = felec;
 821
 822             fscal            = _mm_and_ps(fscal,cutoff_mask);
 823
 824              /* Update vectorial force */
 825             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 826             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 827             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 828
 829             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 830             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 831             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 832
 833             }
 834
 835             /**************************
 836              * CALCULATE INTERACTIONS *
 837              **************************/
 838
 839             if (gmx_mm_any_lt(rsq20,rcutoff2))
 840             {
 841
 842             r20              = _mm_mul_ps(rsq20,rinv20);
 843
 844             /* Compute parameters for interactions between i and j atoms */
 845             qq20             = _mm_mul_ps(iq2,jq0);
 846
 847             /* EWALD ELECTROSTATICS */
 848
 849             /* Analytical PME correction */
 850             zeta2            = _mm_mul_ps(beta2,rsq20);
 851             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 852             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 853             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 854             felec            = _mm_mul_ps(qq20,felec);
 855
 856             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 857
 858             fscal            = felec;
 859
 860             fscal            = _mm_and_ps(fscal,cutoff_mask);
 861
 862              /* Update vectorial force */
 863             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 864             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 865             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 866
 867             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 868             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 869             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 870
 871             }
 872
 873             /**************************
 874              * CALCULATE INTERACTIONS *
 875              **************************/
 876
 877             if (gmx_mm_any_lt(rsq30,rcutoff2))
 878             {
 879
 880             r30              = _mm_mul_ps(rsq30,rinv30);
 881
 882             /* Compute parameters for interactions between i and j atoms */
 883             qq30             = _mm_mul_ps(iq3,jq0);
 884
 885             /* EWALD ELECTROSTATICS */
 886
 887             /* Analytical PME correction */
 888             zeta2            = _mm_mul_ps(beta2,rsq30);
 889             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 890             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 891             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 892             felec            = _mm_mul_ps(qq30,felec);
 893
 894             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 895
 896             fscal            = felec;
 897
 898             fscal            = _mm_and_ps(fscal,cutoff_mask);
 899
 900              /* Update vectorial force */
 901             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 902             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 903             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 904
 905             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 906             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 907             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 908
 909             }
 910
 911             fjptrA             = f+j_coord_offsetA;
 912             fjptrB             = f+j_coord_offsetB;
 913             fjptrC             = f+j_coord_offsetC;
 914             fjptrD             = f+j_coord_offsetD;
 915
 916             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 917
 918             /* Inner loop uses 93 flops */
 919         }
 920
 921         if(jidx<j_index_end)
 922         {
 923
 924             /* Get j neighbor index, and coordinate index */
 925             jnrlistA         = jjnr[jidx];
 926             jnrlistB         = jjnr[jidx+1];
 927             jnrlistC         = jjnr[jidx+2];
 928             jnrlistD         = jjnr[jidx+3];
 929             /* Sign of each element will be negative for non-real atoms.
 930              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 931              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 932              */
 933             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 934             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 935             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 936             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 937             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 938             j_coord_offsetA  = DIM*jnrA;
 939             j_coord_offsetB  = DIM*jnrB;
 940             j_coord_offsetC  = DIM*jnrC;
 941             j_coord_offsetD  = DIM*jnrD;
 942
 943             /* load j atom coordinates */
 944             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 945                                               x+j_coord_offsetC,x+j_coord_offsetD,
 946                                               &jx0,&jy0,&jz0);
 947
 948             /* Calculate displacement vector */
 949             dx10             = _mm_sub_ps(ix1,jx0);
 950             dy10             = _mm_sub_ps(iy1,jy0);
 951             dz10             = _mm_sub_ps(iz1,jz0);
 952             dx20             = _mm_sub_ps(ix2,jx0);
 953             dy20             = _mm_sub_ps(iy2,jy0);
 954             dz20             = _mm_sub_ps(iz2,jz0);
 955             dx30             = _mm_sub_ps(ix3,jx0);
 956             dy30             = _mm_sub_ps(iy3,jy0);
 957             dz30             = _mm_sub_ps(iz3,jz0);
 958
 959             /* Calculate squared distance and things based on it */
 960             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 961             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 962             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 963
 964             rinv10           = avx128fma_invsqrt_f(rsq10);
 965             rinv20           = avx128fma_invsqrt_f(rsq20);
 966             rinv30           = avx128fma_invsqrt_f(rsq30);
 967
 968             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 969             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 970             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 971
 972             /* Load parameters for j particles */
 973             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 974                                                               charge+jnrC+0,charge+jnrD+0);
 975
 976             fjx0             = _mm_setzero_ps();
 977             fjy0             = _mm_setzero_ps();
 978             fjz0             = _mm_setzero_ps();
 979
 980             /**************************
 981              * CALCULATE INTERACTIONS *
 982              **************************/
 983
 984             if (gmx_mm_any_lt(rsq10,rcutoff2))
 985             {
 986
 987             r10              = _mm_mul_ps(rsq10,rinv10);
 988             r10              = _mm_andnot_ps(dummy_mask,r10);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq10             = _mm_mul_ps(iq1,jq0);
 992
 993             /* EWALD ELECTROSTATICS */
 994
 995             /* Analytical PME correction */
 996             zeta2            = _mm_mul_ps(beta2,rsq10);
 997             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 998             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 999             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1000             felec            = _mm_mul_ps(qq10,felec);
1001
1002             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1003
1004             fscal            = felec;
1005
1006             fscal            = _mm_and_ps(fscal,cutoff_mask);
1007
1008             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1009
1010              /* Update vectorial force */
1011             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1012             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1013             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1014
1015             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1016             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1017             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1018
1019             }
1020
1021             /**************************
1022              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             if (gmx_mm_any_lt(rsq20,rcutoff2))
1026             {
1027
1028             r20              = _mm_mul_ps(rsq20,rinv20);
1029             r20              = _mm_andnot_ps(dummy_mask,r20);
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq20             = _mm_mul_ps(iq2,jq0);
1033
1034             /* EWALD ELECTROSTATICS */
1035
1036             /* Analytical PME correction */
1037             zeta2            = _mm_mul_ps(beta2,rsq20);
1038             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1039             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1040             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1041             felec            = _mm_mul_ps(qq20,felec);
1042
1043             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1044
1045             fscal            = felec;
1046
1047             fscal            = _mm_and_ps(fscal,cutoff_mask);
1048
1049             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1050
1051              /* Update vectorial force */
1052             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1053             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1054             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1055
1056             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1057             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1058             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1059
1060             }
1061
1062             /**************************
1063              * CALCULATE INTERACTIONS *
1064              **************************/
1065
1066             if (gmx_mm_any_lt(rsq30,rcutoff2))
1067             {
1068
1069             r30              = _mm_mul_ps(rsq30,rinv30);
1070             r30              = _mm_andnot_ps(dummy_mask,r30);
1071
1072             /* Compute parameters for interactions between i and j atoms */
1073             qq30             = _mm_mul_ps(iq3,jq0);
1074
1075             /* EWALD ELECTROSTATICS */
1076
1077             /* Analytical PME correction */
1078             zeta2            = _mm_mul_ps(beta2,rsq30);
1079             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1080             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1081             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1082             felec            = _mm_mul_ps(qq30,felec);
1083
1084             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1085
1086             fscal            = felec;
1087
1088             fscal            = _mm_and_ps(fscal,cutoff_mask);
1089
1090             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1091
1092              /* Update vectorial force */
1093             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1094             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1095             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1096
1097             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1098             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1099             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1100
1101             }
1102
1103             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1104             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1105             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1106             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1107
1108             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1109
1110             /* Inner loop uses 96 flops */
1111         }
1112
1113         /* End of innermost loop */
1114
1115         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1116                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1117
1118         /* Increment number of inner iterations */
1119         inneriter                  += j_index_end - j_index_start;
1120
1121         /* Outer loop uses 18 flops */
1122     }
1123
1124     /* Increment number of outer iterations */
1125     outeriter        += nri;
1126
1127     /* Update outer/inner flops */
1128
1129     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*96);
1130 }