src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwLJSh_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_VdwLJSh_GeomW4P1_VF_avx_128_fma_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJSh_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              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          ewitab;
 105     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 107     real             *ewtab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->ic->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 130     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 131     beta2            = _mm_mul_ps(beta,beta);
 132     beta3            = _mm_mul_ps(beta,beta2);
 133     ewtab            = fr->ic->tabq_coul_FDV0;
 134     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 135     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 140     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 141     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 142     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 143
 144     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 145     rcutoff_scalar   = fr->ic->rcoulomb;
 146     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 147     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 148
 149     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 150     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = jnrC = jnrD = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156     j_coord_offsetC = 0;
 157     j_coord_offsetD = 0;
 158
 159     outeriter        = 0;
 160     inneriter        = 0;
 161
 162     for(iidx=0;iidx<4*DIM;iidx++)
 163     {
 164         scratch[iidx] = 0.0;
 165     }
 166
 167     /* Start outer loop over neighborlists */
 168     for(iidx=0; iidx<nri; iidx++)
 169     {
 170         /* Load shift vector for this list */
 171         i_shift_offset   = DIM*shiftidx[iidx];
 172
 173         /* Load limits for loop over neighbors */
 174         j_index_start    = jindex[iidx];
 175         j_index_end      = jindex[iidx+1];
 176
 177         /* Get outer coordinate index */
 178         inr              = iinr[iidx];
 179         i_coord_offset   = DIM*inr;
 180
 181         /* Load i particle coords and add shift vector */
 182         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 183                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 184
 185         fix0             = _mm_setzero_ps();
 186         fiy0             = _mm_setzero_ps();
 187         fiz0             = _mm_setzero_ps();
 188         fix1             = _mm_setzero_ps();
 189         fiy1             = _mm_setzero_ps();
 190         fiz1             = _mm_setzero_ps();
 191         fix2             = _mm_setzero_ps();
 192         fiy2             = _mm_setzero_ps();
 193         fiz2             = _mm_setzero_ps();
 194         fix3             = _mm_setzero_ps();
 195         fiy3             = _mm_setzero_ps();
 196         fiz3             = _mm_setzero_ps();
 197
 198         /* Reset potential sums */
 199         velecsum         = _mm_setzero_ps();
 200         vvdwsum          = _mm_setzero_ps();
 201
 202         /* Start inner kernel loop */
 203         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 204         {
 205
 206             /* Get j neighbor index, and coordinate index */
 207             jnrA             = jjnr[jidx];
 208             jnrB             = jjnr[jidx+1];
 209             jnrC             = jjnr[jidx+2];
 210             jnrD             = jjnr[jidx+3];
 211             j_coord_offsetA  = DIM*jnrA;
 212             j_coord_offsetB  = DIM*jnrB;
 213             j_coord_offsetC  = DIM*jnrC;
 214             j_coord_offsetD  = DIM*jnrD;
 215
 216             /* load j atom coordinates */
 217             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 218                                               x+j_coord_offsetC,x+j_coord_offsetD,
 219                                               &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm_sub_ps(ix0,jx0);
 223             dy00             = _mm_sub_ps(iy0,jy0);
 224             dz00             = _mm_sub_ps(iz0,jz0);
 225             dx10             = _mm_sub_ps(ix1,jx0);
 226             dy10             = _mm_sub_ps(iy1,jy0);
 227             dz10             = _mm_sub_ps(iz1,jz0);
 228             dx20             = _mm_sub_ps(ix2,jx0);
 229             dy20             = _mm_sub_ps(iy2,jy0);
 230             dz20             = _mm_sub_ps(iz2,jz0);
 231             dx30             = _mm_sub_ps(ix3,jx0);
 232             dy30             = _mm_sub_ps(iy3,jy0);
 233             dz30             = _mm_sub_ps(iz3,jz0);
 234
 235             /* Calculate squared distance and things based on it */
 236             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 237             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 238             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 239             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 240
 241             rinv10           = avx128fma_invsqrt_f(rsq10);
 242             rinv20           = avx128fma_invsqrt_f(rsq20);
 243             rinv30           = avx128fma_invsqrt_f(rsq30);
 244
 245             rinvsq00         = avx128fma_inv_f(rsq00);
 246             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 247             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 248             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 249
 250             /* Load parameters for j particles */
 251             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 252                                                               charge+jnrC+0,charge+jnrD+0);
 253             vdwjidx0A        = 2*vdwtype[jnrA+0];
 254             vdwjidx0B        = 2*vdwtype[jnrB+0];
 255             vdwjidx0C        = 2*vdwtype[jnrC+0];
 256             vdwjidx0D        = 2*vdwtype[jnrD+0];
 257
 258             fjx0             = _mm_setzero_ps();
 259             fjy0             = _mm_setzero_ps();
 260             fjz0             = _mm_setzero_ps();
 261
 262             /**************************
 263              * CALCULATE INTERACTIONS *
 264              **************************/
 265
 266             if (gmx_mm_any_lt(rsq00,rcutoff2))
 267             {
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 271                                          vdwparam+vdwioffset0+vdwjidx0B,
 272                                          vdwparam+vdwioffset0+vdwjidx0C,
 273                                          vdwparam+vdwioffset0+vdwjidx0D,
 274                                          &c6_00,&c12_00);
 275
 276             /* LENNARD-JONES DISPERSION/REPULSION */
 277
 278             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 279             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 280             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 281             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 282                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 283             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 284
 285             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 289             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 290
 291             fscal            = fvdw;
 292
 293             fscal            = _mm_and_ps(fscal,cutoff_mask);
 294
 295              /* Update vectorial force */
 296             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 297             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 298             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 299
 300             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 301             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 302             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 303
 304             }
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             if (gmx_mm_any_lt(rsq10,rcutoff2))
 311             {
 312
 313             r10              = _mm_mul_ps(rsq10,rinv10);
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             qq10             = _mm_mul_ps(iq1,jq0);
 317
 318             /* EWALD ELECTROSTATICS */
 319
 320             /* Analytical PME correction */
 321             zeta2            = _mm_mul_ps(beta2,rsq10);
 322             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 323             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 324             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 325             felec            = _mm_mul_ps(qq10,felec);
 326             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 327             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 328             velec            = _mm_mul_ps(qq10,velec);
 329
 330             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm_and_ps(velec,cutoff_mask);
 334             velecsum         = _mm_add_ps(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             fscal            = _mm_and_ps(fscal,cutoff_mask);
 339
 340              /* Update vectorial force */
 341             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 342             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 343             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 344
 345             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 346             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 347             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 348
 349             }
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             if (gmx_mm_any_lt(rsq20,rcutoff2))
 356             {
 357
 358             r20              = _mm_mul_ps(rsq20,rinv20);
 359
 360             /* Compute parameters for interactions between i and j atoms */
 361             qq20             = _mm_mul_ps(iq2,jq0);
 362
 363             /* EWALD ELECTROSTATICS */
 364
 365             /* Analytical PME correction */
 366             zeta2            = _mm_mul_ps(beta2,rsq20);
 367             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 368             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 369             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 370             felec            = _mm_mul_ps(qq20,felec);
 371             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 372             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 373             velec            = _mm_mul_ps(qq20,velec);
 374
 375             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velec            = _mm_and_ps(velec,cutoff_mask);
 379             velecsum         = _mm_add_ps(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             fscal            = _mm_and_ps(fscal,cutoff_mask);
 384
 385              /* Update vectorial force */
 386             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 387             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 388             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 389
 390             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 391             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 392             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 393
 394             }
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             if (gmx_mm_any_lt(rsq30,rcutoff2))
 401             {
 402
 403             r30              = _mm_mul_ps(rsq30,rinv30);
 404
 405             /* Compute parameters for interactions between i and j atoms */
 406             qq30             = _mm_mul_ps(iq3,jq0);
 407
 408             /* EWALD ELECTROSTATICS */
 409
 410             /* Analytical PME correction */
 411             zeta2            = _mm_mul_ps(beta2,rsq30);
 412             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 413             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 414             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 415             felec            = _mm_mul_ps(qq30,felec);
 416             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 417             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv30,sh_ewald));
 418             velec            = _mm_mul_ps(qq30,velec);
 419
 420             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velec            = _mm_and_ps(velec,cutoff_mask);
 424             velecsum         = _mm_add_ps(velecsum,velec);
 425
 426             fscal            = felec;
 427
 428             fscal            = _mm_and_ps(fscal,cutoff_mask);
 429
 430              /* Update vectorial force */
 431             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 432             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 433             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 434
 435             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 436             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 437             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 438
 439             }
 440
 441             fjptrA             = f+j_coord_offsetA;
 442             fjptrB             = f+j_coord_offsetB;
 443             fjptrC             = f+j_coord_offsetC;
 444             fjptrD             = f+j_coord_offsetD;
 445
 446             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 447
 448             /* Inner loop uses 143 flops */
 449         }
 450
 451         if(jidx<j_index_end)
 452         {
 453
 454             /* Get j neighbor index, and coordinate index */
 455             jnrlistA         = jjnr[jidx];
 456             jnrlistB         = jjnr[jidx+1];
 457             jnrlistC         = jjnr[jidx+2];
 458             jnrlistD         = jjnr[jidx+3];
 459             /* Sign of each element will be negative for non-real atoms.
 460              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 461              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 462              */
 463             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 464             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 465             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 466             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 467             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 468             j_coord_offsetA  = DIM*jnrA;
 469             j_coord_offsetB  = DIM*jnrB;
 470             j_coord_offsetC  = DIM*jnrC;
 471             j_coord_offsetD  = DIM*jnrD;
 472
 473             /* load j atom coordinates */
 474             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 475                                               x+j_coord_offsetC,x+j_coord_offsetD,
 476                                               &jx0,&jy0,&jz0);
 477
 478             /* Calculate displacement vector */
 479             dx00             = _mm_sub_ps(ix0,jx0);
 480             dy00             = _mm_sub_ps(iy0,jy0);
 481             dz00             = _mm_sub_ps(iz0,jz0);
 482             dx10             = _mm_sub_ps(ix1,jx0);
 483             dy10             = _mm_sub_ps(iy1,jy0);
 484             dz10             = _mm_sub_ps(iz1,jz0);
 485             dx20             = _mm_sub_ps(ix2,jx0);
 486             dy20             = _mm_sub_ps(iy2,jy0);
 487             dz20             = _mm_sub_ps(iz2,jz0);
 488             dx30             = _mm_sub_ps(ix3,jx0);
 489             dy30             = _mm_sub_ps(iy3,jy0);
 490             dz30             = _mm_sub_ps(iz3,jz0);
 491
 492             /* Calculate squared distance and things based on it */
 493             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 494             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 495             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 496             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 497
 498             rinv10           = avx128fma_invsqrt_f(rsq10);
 499             rinv20           = avx128fma_invsqrt_f(rsq20);
 500             rinv30           = avx128fma_invsqrt_f(rsq30);
 501
 502             rinvsq00         = avx128fma_inv_f(rsq00);
 503             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 504             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 505             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 506
 507             /* Load parameters for j particles */
 508             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 509                                                               charge+jnrC+0,charge+jnrD+0);
 510             vdwjidx0A        = 2*vdwtype[jnrA+0];
 511             vdwjidx0B        = 2*vdwtype[jnrB+0];
 512             vdwjidx0C        = 2*vdwtype[jnrC+0];
 513             vdwjidx0D        = 2*vdwtype[jnrD+0];
 514
 515             fjx0             = _mm_setzero_ps();
 516             fjy0             = _mm_setzero_ps();
 517             fjz0             = _mm_setzero_ps();
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             if (gmx_mm_any_lt(rsq00,rcutoff2))
 524             {
 525
 526             /* Compute parameters for interactions between i and j atoms */
 527             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 528                                          vdwparam+vdwioffset0+vdwjidx0B,
 529                                          vdwparam+vdwioffset0+vdwjidx0C,
 530                                          vdwparam+vdwioffset0+vdwjidx0D,
 531                                          &c6_00,&c12_00);
 532
 533             /* LENNARD-JONES DISPERSION/REPULSION */
 534
 535             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 536             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 537             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 538             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 539                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 540             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 541
 542             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 546             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 547             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 548
 549             fscal            = fvdw;
 550
 551             fscal            = _mm_and_ps(fscal,cutoff_mask);
 552
 553             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 554
 555              /* Update vectorial force */
 556             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 557             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 558             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 559
 560             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 561             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 562             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 563
 564             }
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             if (gmx_mm_any_lt(rsq10,rcutoff2))
 571             {
 572
 573             r10              = _mm_mul_ps(rsq10,rinv10);
 574             r10              = _mm_andnot_ps(dummy_mask,r10);
 575
 576             /* Compute parameters for interactions between i and j atoms */
 577             qq10             = _mm_mul_ps(iq1,jq0);
 578
 579             /* EWALD ELECTROSTATICS */
 580
 581             /* Analytical PME correction */
 582             zeta2            = _mm_mul_ps(beta2,rsq10);
 583             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 584             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 585             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 586             felec            = _mm_mul_ps(qq10,felec);
 587             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 588             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 589             velec            = _mm_mul_ps(qq10,velec);
 590
 591             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velec            = _mm_and_ps(velec,cutoff_mask);
 595             velec            = _mm_andnot_ps(dummy_mask,velec);
 596             velecsum         = _mm_add_ps(velecsum,velec);
 597
 598             fscal            = felec;
 599
 600             fscal            = _mm_and_ps(fscal,cutoff_mask);
 601
 602             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 603
 604              /* Update vectorial force */
 605             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 606             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 607             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 608
 609             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 610             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 611             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 612
 613             }
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             if (gmx_mm_any_lt(rsq20,rcutoff2))
 620             {
 621
 622             r20              = _mm_mul_ps(rsq20,rinv20);
 623             r20              = _mm_andnot_ps(dummy_mask,r20);
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq20             = _mm_mul_ps(iq2,jq0);
 627
 628             /* EWALD ELECTROSTATICS */
 629
 630             /* Analytical PME correction */
 631             zeta2            = _mm_mul_ps(beta2,rsq20);
 632             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 633             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 634             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 635             felec            = _mm_mul_ps(qq20,felec);
 636             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 637             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 638             velec            = _mm_mul_ps(qq20,velec);
 639
 640             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 641
 642             /* Update potential sum for this i atom from the interaction with this j atom. */
 643             velec            = _mm_and_ps(velec,cutoff_mask);
 644             velec            = _mm_andnot_ps(dummy_mask,velec);
 645             velecsum         = _mm_add_ps(velecsum,velec);
 646
 647             fscal            = felec;
 648
 649             fscal            = _mm_and_ps(fscal,cutoff_mask);
 650
 651             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 652
 653              /* Update vectorial force */
 654             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 655             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 656             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 657
 658             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 659             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 660             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 661
 662             }
 663
 664             /**************************
 665              * CALCULATE INTERACTIONS *
 666              **************************/
 667
 668             if (gmx_mm_any_lt(rsq30,rcutoff2))
 669             {
 670
 671             r30              = _mm_mul_ps(rsq30,rinv30);
 672             r30              = _mm_andnot_ps(dummy_mask,r30);
 673
 674             /* Compute parameters for interactions between i and j atoms */
 675             qq30             = _mm_mul_ps(iq3,jq0);
 676
 677             /* EWALD ELECTROSTATICS */
 678
 679             /* Analytical PME correction */
 680             zeta2            = _mm_mul_ps(beta2,rsq30);
 681             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 682             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
 683             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 684             felec            = _mm_mul_ps(qq30,felec);
 685             pmecorrV         = avx128fma_pmecorrV_f(zeta2);
 686             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv30,sh_ewald));
 687             velec            = _mm_mul_ps(qq30,velec);
 688
 689             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 690
 691             /* Update potential sum for this i atom from the interaction with this j atom. */
 692             velec            = _mm_and_ps(velec,cutoff_mask);
 693             velec            = _mm_andnot_ps(dummy_mask,velec);
 694             velecsum         = _mm_add_ps(velecsum,velec);
 695
 696             fscal            = felec;
 697
 698             fscal            = _mm_and_ps(fscal,cutoff_mask);
 699
 700             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 701
 702              /* Update vectorial force */
 703             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 704             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 705             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 706
 707             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 708             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 709             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 710
 711             }
 712
 713             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 714             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 715             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 716             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 717
 718             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 719
 720             /* Inner loop uses 146 flops */
 721         }
 722
 723         /* End of innermost loop */
 724
 725         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 726                                               f+i_coord_offset,fshift+i_shift_offset);
 727
 728         ggid                        = gid[iidx];
 729         /* Update potential energies */
 730         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 731         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 732
 733         /* Increment number of inner iterations */
 734         inneriter                  += j_index_end - j_index_start;
 735
 736         /* Outer loop uses 26 flops */
 737     }
 738
 739     /* Increment number of outer iterations */
 740     outeriter        += nri;
 741
 742     /* Update outer/inner flops */
 743
 744     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*146);
 745 }
 746 /*
 747  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_128_fma_single
 748  * Electrostatics interaction: Ewald
 749  * VdW interaction:            LennardJones
 750  * Geometry:                   Water4-Particle
 751  * Calculate force/pot:        Force
 752  */
 753 void
 754 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_128_fma_single
 755                     (t_nblist                    * gmx_restrict       nlist,
 756                      rvec                        * gmx_restrict          xx,
 757                      rvec                        * gmx_restrict          ff,
 758                      struct t_forcerec           * gmx_restrict          fr,
 759                      t_mdatoms                   * gmx_restrict     mdatoms,
 760                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 761                      t_nrnb                      * gmx_restrict        nrnb)
 762 {
 763     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 764      * just 0 for non-waters.
 765      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 766      * jnr indices corresponding to data put in the four positions in the SIMD register.
 767      */
 768     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 769     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 770     int              jnrA,jnrB,jnrC,jnrD;
 771     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 772     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 773     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 774     real             rcutoff_scalar;
 775     real             *shiftvec,*fshift,*x,*f;
 776     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 777     real             scratch[4*DIM];
 778     __m128           fscal,rcutoff,rcutoff2,jidxall;
 779     int              vdwioffset0;
 780     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 781     int              vdwioffset1;
 782     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 783     int              vdwioffset2;
 784     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 785     int              vdwioffset3;
 786     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 787     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 788     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 789     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 790     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 791     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 792     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 793     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 794     real             *charge;
 795     int              nvdwtype;
 796     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 797     int              *vdwtype;
 798     real             *vdwparam;
 799     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 800     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 801     __m128i          ewitab;
 802     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 803     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 804     real             *ewtab;
 805     __m128           dummy_mask,cutoff_mask;
 806     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 807     __m128           one     = _mm_set1_ps(1.0);
 808     __m128           two     = _mm_set1_ps(2.0);
 809     x                = xx[0];
 810     f                = ff[0];
 811
 812     nri              = nlist->nri;
 813     iinr             = nlist->iinr;
 814     jindex           = nlist->jindex;
 815     jjnr             = nlist->jjnr;
 816     shiftidx         = nlist->shift;
 817     gid              = nlist->gid;
 818     shiftvec         = fr->shift_vec[0];
 819     fshift           = fr->fshift[0];
 820     facel            = _mm_set1_ps(fr->ic->epsfac);
 821     charge           = mdatoms->chargeA;
 822     nvdwtype         = fr->ntype;
 823     vdwparam         = fr->nbfp;
 824     vdwtype          = mdatoms->typeA;
 825
 826     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 827     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 828     beta2            = _mm_mul_ps(beta,beta);
 829     beta3            = _mm_mul_ps(beta,beta2);
 830     ewtab            = fr->ic->tabq_coul_F;
 831     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 832     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 833
 834     /* Setup water-specific parameters */
 835     inr              = nlist->iinr[0];
 836     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 837     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 838     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 839     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 840
 841     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 842     rcutoff_scalar   = fr->ic->rcoulomb;
 843     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 844     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 845
 846     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 847     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 848
 849     /* Avoid stupid compiler warnings */
 850     jnrA = jnrB = jnrC = jnrD = 0;
 851     j_coord_offsetA = 0;
 852     j_coord_offsetB = 0;
 853     j_coord_offsetC = 0;
 854     j_coord_offsetD = 0;
 855
 856     outeriter        = 0;
 857     inneriter        = 0;
 858
 859     for(iidx=0;iidx<4*DIM;iidx++)
 860     {
 861         scratch[iidx] = 0.0;
 862     }
 863
 864     /* Start outer loop over neighborlists */
 865     for(iidx=0; iidx<nri; iidx++)
 866     {
 867         /* Load shift vector for this list */
 868         i_shift_offset   = DIM*shiftidx[iidx];
 869
 870         /* Load limits for loop over neighbors */
 871         j_index_start    = jindex[iidx];
 872         j_index_end      = jindex[iidx+1];
 873
 874         /* Get outer coordinate index */
 875         inr              = iinr[iidx];
 876         i_coord_offset   = DIM*inr;
 877
 878         /* Load i particle coords and add shift vector */
 879         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 880                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 881
 882         fix0             = _mm_setzero_ps();
 883         fiy0             = _mm_setzero_ps();
 884         fiz0             = _mm_setzero_ps();
 885         fix1             = _mm_setzero_ps();
 886         fiy1             = _mm_setzero_ps();
 887         fiz1             = _mm_setzero_ps();
 888         fix2             = _mm_setzero_ps();
 889         fiy2             = _mm_setzero_ps();
 890         fiz2             = _mm_setzero_ps();
 891         fix3             = _mm_setzero_ps();
 892         fiy3             = _mm_setzero_ps();
 893         fiz3             = _mm_setzero_ps();
 894
 895         /* Start inner kernel loop */
 896         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 897         {
 898
 899             /* Get j neighbor index, and coordinate index */
 900             jnrA             = jjnr[jidx];
 901             jnrB             = jjnr[jidx+1];
 902             jnrC             = jjnr[jidx+2];
 903             jnrD             = jjnr[jidx+3];
 904             j_coord_offsetA  = DIM*jnrA;
 905             j_coord_offsetB  = DIM*jnrB;
 906             j_coord_offsetC  = DIM*jnrC;
 907             j_coord_offsetD  = DIM*jnrD;
 908
 909             /* load j atom coordinates */
 910             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 911                                               x+j_coord_offsetC,x+j_coord_offsetD,
 912                                               &jx0,&jy0,&jz0);
 913
 914             /* Calculate displacement vector */
 915             dx00             = _mm_sub_ps(ix0,jx0);
 916             dy00             = _mm_sub_ps(iy0,jy0);
 917             dz00             = _mm_sub_ps(iz0,jz0);
 918             dx10             = _mm_sub_ps(ix1,jx0);
 919             dy10             = _mm_sub_ps(iy1,jy0);
 920             dz10             = _mm_sub_ps(iz1,jz0);
 921             dx20             = _mm_sub_ps(ix2,jx0);
 922             dy20             = _mm_sub_ps(iy2,jy0);
 923             dz20             = _mm_sub_ps(iz2,jz0);
 924             dx30             = _mm_sub_ps(ix3,jx0);
 925             dy30             = _mm_sub_ps(iy3,jy0);
 926             dz30             = _mm_sub_ps(iz3,jz0);
 927
 928             /* Calculate squared distance and things based on it */
 929             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 930             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 931             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 932             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 933
 934             rinv10           = avx128fma_invsqrt_f(rsq10);
 935             rinv20           = avx128fma_invsqrt_f(rsq20);
 936             rinv30           = avx128fma_invsqrt_f(rsq30);
 937
 938             rinvsq00         = avx128fma_inv_f(rsq00);
 939             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 940             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 941             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 942
 943             /* Load parameters for j particles */
 944             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 945                                                               charge+jnrC+0,charge+jnrD+0);
 946             vdwjidx0A        = 2*vdwtype[jnrA+0];
 947             vdwjidx0B        = 2*vdwtype[jnrB+0];
 948             vdwjidx0C        = 2*vdwtype[jnrC+0];
 949             vdwjidx0D        = 2*vdwtype[jnrD+0];
 950
 951             fjx0             = _mm_setzero_ps();
 952             fjy0             = _mm_setzero_ps();
 953             fjz0             = _mm_setzero_ps();
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             if (gmx_mm_any_lt(rsq00,rcutoff2))
 960             {
 961
 962             /* Compute parameters for interactions between i and j atoms */
 963             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 964                                          vdwparam+vdwioffset0+vdwjidx0B,
 965                                          vdwparam+vdwioffset0+vdwjidx0C,
 966                                          vdwparam+vdwioffset0+vdwjidx0D,
 967                                          &c6_00,&c12_00);
 968
 969             /* LENNARD-JONES DISPERSION/REPULSION */
 970
 971             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 972             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 973
 974             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 975
 976             fscal            = fvdw;
 977
 978             fscal            = _mm_and_ps(fscal,cutoff_mask);
 979
 980              /* Update vectorial force */
 981             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 982             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 983             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 984
 985             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 986             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 987             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 988
 989             }
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             if (gmx_mm_any_lt(rsq10,rcutoff2))
 996             {
 997
 998             r10              = _mm_mul_ps(rsq10,rinv10);
 999
1000             /* Compute parameters for interactions between i and j atoms */
1001             qq10             = _mm_mul_ps(iq1,jq0);
1002
1003             /* EWALD ELECTROSTATICS */
1004
1005             /* Analytical PME correction */
1006             zeta2            = _mm_mul_ps(beta2,rsq10);
1007             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1008             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1009             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1010             felec            = _mm_mul_ps(qq10,felec);
1011
1012             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1013
1014             fscal            = felec;
1015
1016             fscal            = _mm_and_ps(fscal,cutoff_mask);
1017
1018              /* Update vectorial force */
1019             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1020             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1021             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1022
1023             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1024             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1025             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1026
1027             }
1028
1029             /**************************
1030              * CALCULATE INTERACTIONS *
1031              **************************/
1032
1033             if (gmx_mm_any_lt(rsq20,rcutoff2))
1034             {
1035
1036             r20              = _mm_mul_ps(rsq20,rinv20);
1037
1038             /* Compute parameters for interactions between i and j atoms */
1039             qq20             = _mm_mul_ps(iq2,jq0);
1040
1041             /* EWALD ELECTROSTATICS */
1042
1043             /* Analytical PME correction */
1044             zeta2            = _mm_mul_ps(beta2,rsq20);
1045             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1046             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1047             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1048             felec            = _mm_mul_ps(qq20,felec);
1049
1050             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1051
1052             fscal            = felec;
1053
1054             fscal            = _mm_and_ps(fscal,cutoff_mask);
1055
1056              /* Update vectorial force */
1057             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1058             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1059             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1060
1061             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1062             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1063             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1064
1065             }
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             if (gmx_mm_any_lt(rsq30,rcutoff2))
1072             {
1073
1074             r30              = _mm_mul_ps(rsq30,rinv30);
1075
1076             /* Compute parameters for interactions between i and j atoms */
1077             qq30             = _mm_mul_ps(iq3,jq0);
1078
1079             /* EWALD ELECTROSTATICS */
1080
1081             /* Analytical PME correction */
1082             zeta2            = _mm_mul_ps(beta2,rsq30);
1083             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1084             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1085             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1086             felec            = _mm_mul_ps(qq30,felec);
1087
1088             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1089
1090             fscal            = felec;
1091
1092             fscal            = _mm_and_ps(fscal,cutoff_mask);
1093
1094              /* Update vectorial force */
1095             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1096             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1097             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1098
1099             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1100             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1101             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1102
1103             }
1104
1105             fjptrA             = f+j_coord_offsetA;
1106             fjptrB             = f+j_coord_offsetB;
1107             fjptrC             = f+j_coord_offsetC;
1108             fjptrD             = f+j_coord_offsetD;
1109
1110             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1111
1112             /* Inner loop uses 126 flops */
1113         }
1114
1115         if(jidx<j_index_end)
1116         {
1117
1118             /* Get j neighbor index, and coordinate index */
1119             jnrlistA         = jjnr[jidx];
1120             jnrlistB         = jjnr[jidx+1];
1121             jnrlistC         = jjnr[jidx+2];
1122             jnrlistD         = jjnr[jidx+3];
1123             /* Sign of each element will be negative for non-real atoms.
1124              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1125              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1126              */
1127             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1128             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1129             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1130             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1131             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1132             j_coord_offsetA  = DIM*jnrA;
1133             j_coord_offsetB  = DIM*jnrB;
1134             j_coord_offsetC  = DIM*jnrC;
1135             j_coord_offsetD  = DIM*jnrD;
1136
1137             /* load j atom coordinates */
1138             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1139                                               x+j_coord_offsetC,x+j_coord_offsetD,
1140                                               &jx0,&jy0,&jz0);
1141
1142             /* Calculate displacement vector */
1143             dx00             = _mm_sub_ps(ix0,jx0);
1144             dy00             = _mm_sub_ps(iy0,jy0);
1145             dz00             = _mm_sub_ps(iz0,jz0);
1146             dx10             = _mm_sub_ps(ix1,jx0);
1147             dy10             = _mm_sub_ps(iy1,jy0);
1148             dz10             = _mm_sub_ps(iz1,jz0);
1149             dx20             = _mm_sub_ps(ix2,jx0);
1150             dy20             = _mm_sub_ps(iy2,jy0);
1151             dz20             = _mm_sub_ps(iz2,jz0);
1152             dx30             = _mm_sub_ps(ix3,jx0);
1153             dy30             = _mm_sub_ps(iy3,jy0);
1154             dz30             = _mm_sub_ps(iz3,jz0);
1155
1156             /* Calculate squared distance and things based on it */
1157             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1158             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1159             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1160             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1161
1162             rinv10           = avx128fma_invsqrt_f(rsq10);
1163             rinv20           = avx128fma_invsqrt_f(rsq20);
1164             rinv30           = avx128fma_invsqrt_f(rsq30);
1165
1166             rinvsq00         = avx128fma_inv_f(rsq00);
1167             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1168             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1169             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1170
1171             /* Load parameters for j particles */
1172             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1173                                                               charge+jnrC+0,charge+jnrD+0);
1174             vdwjidx0A        = 2*vdwtype[jnrA+0];
1175             vdwjidx0B        = 2*vdwtype[jnrB+0];
1176             vdwjidx0C        = 2*vdwtype[jnrC+0];
1177             vdwjidx0D        = 2*vdwtype[jnrD+0];
1178
1179             fjx0             = _mm_setzero_ps();
1180             fjy0             = _mm_setzero_ps();
1181             fjz0             = _mm_setzero_ps();
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             if (gmx_mm_any_lt(rsq00,rcutoff2))
1188             {
1189
1190             /* Compute parameters for interactions between i and j atoms */
1191             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1192                                          vdwparam+vdwioffset0+vdwjidx0B,
1193                                          vdwparam+vdwioffset0+vdwjidx0C,
1194                                          vdwparam+vdwioffset0+vdwjidx0D,
1195                                          &c6_00,&c12_00);
1196
1197             /* LENNARD-JONES DISPERSION/REPULSION */
1198
1199             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1200             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1201
1202             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1203
1204             fscal            = fvdw;
1205
1206             fscal            = _mm_and_ps(fscal,cutoff_mask);
1207
1208             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1209
1210              /* Update vectorial force */
1211             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1212             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1213             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1214
1215             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1216             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1217             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1218
1219             }
1220
1221             /**************************
1222              * CALCULATE INTERACTIONS *
1223              **************************/
1224
1225             if (gmx_mm_any_lt(rsq10,rcutoff2))
1226             {
1227
1228             r10              = _mm_mul_ps(rsq10,rinv10);
1229             r10              = _mm_andnot_ps(dummy_mask,r10);
1230
1231             /* Compute parameters for interactions between i and j atoms */
1232             qq10             = _mm_mul_ps(iq1,jq0);
1233
1234             /* EWALD ELECTROSTATICS */
1235
1236             /* Analytical PME correction */
1237             zeta2            = _mm_mul_ps(beta2,rsq10);
1238             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1239             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1240             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1241             felec            = _mm_mul_ps(qq10,felec);
1242
1243             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1244
1245             fscal            = felec;
1246
1247             fscal            = _mm_and_ps(fscal,cutoff_mask);
1248
1249             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1250
1251              /* Update vectorial force */
1252             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1253             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1254             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1255
1256             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1257             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1258             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1259
1260             }
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             if (gmx_mm_any_lt(rsq20,rcutoff2))
1267             {
1268
1269             r20              = _mm_mul_ps(rsq20,rinv20);
1270             r20              = _mm_andnot_ps(dummy_mask,r20);
1271
1272             /* Compute parameters for interactions between i and j atoms */
1273             qq20             = _mm_mul_ps(iq2,jq0);
1274
1275             /* EWALD ELECTROSTATICS */
1276
1277             /* Analytical PME correction */
1278             zeta2            = _mm_mul_ps(beta2,rsq20);
1279             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1280             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1281             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1282             felec            = _mm_mul_ps(qq20,felec);
1283
1284             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1285
1286             fscal            = felec;
1287
1288             fscal            = _mm_and_ps(fscal,cutoff_mask);
1289
1290             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1291
1292              /* Update vectorial force */
1293             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1294             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1295             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1296
1297             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1298             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1299             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1300
1301             }
1302
1303             /**************************
1304              * CALCULATE INTERACTIONS *
1305              **************************/
1306
1307             if (gmx_mm_any_lt(rsq30,rcutoff2))
1308             {
1309
1310             r30              = _mm_mul_ps(rsq30,rinv30);
1311             r30              = _mm_andnot_ps(dummy_mask,r30);
1312
1313             /* Compute parameters for interactions between i and j atoms */
1314             qq30             = _mm_mul_ps(iq3,jq0);
1315
1316             /* EWALD ELECTROSTATICS */
1317
1318             /* Analytical PME correction */
1319             zeta2            = _mm_mul_ps(beta2,rsq30);
1320             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1321             pmecorrF         = avx128fma_pmecorrF_f(zeta2);
1322             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1323             felec            = _mm_mul_ps(qq30,felec);
1324
1325             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1326
1327             fscal            = felec;
1328
1329             fscal            = _mm_and_ps(fscal,cutoff_mask);
1330
1331             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1332
1333              /* Update vectorial force */
1334             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1335             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1336             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1337
1338             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1339             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1340             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1341
1342             }
1343
1344             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1345             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1346             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1347             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1348
1349             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1350
1351             /* Inner loop uses 129 flops */
1352         }
1353
1354         /* End of innermost loop */
1355
1356         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1357                                               f+i_coord_offset,fshift+i_shift_offset);
1358
1359         /* Increment number of inner iterations */
1360         inneriter                  += j_index_end - j_index_start;
1361
1362         /* Outer loop uses 24 flops */
1363     }
1364
1365     /* Increment number of outer iterations */
1366     outeriter        += nri;
1367
1368     /* Update outer/inner flops */
1369
1370     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*129);
1371 }