src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_avx_128_fma_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_128_fma_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_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     __m128           dummy_mask,cutoff_mask;
 105     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 106     __m128           one     = _mm_set1_ps(1.0);
 107     __m128           two     = _mm_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_ps(fr->ic->epsfac);
 120     charge           = mdatoms->chargeA;
 121     krf              = _mm_set1_ps(fr->ic->k_rf);
 122     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 123     crf              = _mm_set1_ps(fr->ic->c_rf);
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 131     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 132     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 133     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 136     rcutoff_scalar   = fr->ic->rcoulomb;
 137     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 138     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 139
 140     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 141     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     for(iidx=0;iidx<4*DIM;iidx++)
 154     {
 155         scratch[iidx] = 0.0;
 156     }
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 175
 176         fix0             = _mm_setzero_ps();
 177         fiy0             = _mm_setzero_ps();
 178         fiz0             = _mm_setzero_ps();
 179         fix1             = _mm_setzero_ps();
 180         fiy1             = _mm_setzero_ps();
 181         fiz1             = _mm_setzero_ps();
 182         fix2             = _mm_setzero_ps();
 183         fiy2             = _mm_setzero_ps();
 184         fiz2             = _mm_setzero_ps();
 185         fix3             = _mm_setzero_ps();
 186         fiy3             = _mm_setzero_ps();
 187         fiz3             = _mm_setzero_ps();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm_setzero_ps();
 191         vvdwsum          = _mm_setzero_ps();
 192
 193         /* Start inner kernel loop */
 194         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 195         {
 196
 197             /* Get j neighbor index, and coordinate index */
 198             jnrA             = jjnr[jidx];
 199             jnrB             = jjnr[jidx+1];
 200             jnrC             = jjnr[jidx+2];
 201             jnrD             = jjnr[jidx+3];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204             j_coord_offsetC  = DIM*jnrC;
 205             j_coord_offsetD  = DIM*jnrD;
 206
 207             /* load j atom coordinates */
 208             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 209                                               x+j_coord_offsetC,x+j_coord_offsetD,
 210                                               &jx0,&jy0,&jz0);
 211
 212             /* Calculate displacement vector */
 213             dx00             = _mm_sub_ps(ix0,jx0);
 214             dy00             = _mm_sub_ps(iy0,jy0);
 215             dz00             = _mm_sub_ps(iz0,jz0);
 216             dx10             = _mm_sub_ps(ix1,jx0);
 217             dy10             = _mm_sub_ps(iy1,jy0);
 218             dz10             = _mm_sub_ps(iz1,jz0);
 219             dx20             = _mm_sub_ps(ix2,jx0);
 220             dy20             = _mm_sub_ps(iy2,jy0);
 221             dz20             = _mm_sub_ps(iz2,jz0);
 222             dx30             = _mm_sub_ps(ix3,jx0);
 223             dy30             = _mm_sub_ps(iy3,jy0);
 224             dz30             = _mm_sub_ps(iz3,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 228             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 229             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 230             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 231
 232             rinv10           = avx128fma_invsqrt_f(rsq10);
 233             rinv20           = avx128fma_invsqrt_f(rsq20);
 234             rinv30           = avx128fma_invsqrt_f(rsq30);
 235
 236             rinvsq00         = avx128fma_inv_f(rsq00);
 237             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 238             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 239             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 240
 241             /* Load parameters for j particles */
 242             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 243                                                               charge+jnrC+0,charge+jnrD+0);
 244             vdwjidx0A        = 2*vdwtype[jnrA+0];
 245             vdwjidx0B        = 2*vdwtype[jnrB+0];
 246             vdwjidx0C        = 2*vdwtype[jnrC+0];
 247             vdwjidx0D        = 2*vdwtype[jnrD+0];
 248
 249             fjx0             = _mm_setzero_ps();
 250             fjy0             = _mm_setzero_ps();
 251             fjz0             = _mm_setzero_ps();
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             if (gmx_mm_any_lt(rsq00,rcutoff2))
 258             {
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 262                                          vdwparam+vdwioffset0+vdwjidx0B,
 263                                          vdwparam+vdwioffset0+vdwjidx0C,
 264                                          vdwparam+vdwioffset0+vdwjidx0D,
 265                                          &c6_00,&c12_00);
 266
 267             /* LENNARD-JONES DISPERSION/REPULSION */
 268
 269             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 270             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 271             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 272             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 273                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 274             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 275
 276             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 280             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 281
 282             fscal            = fvdw;
 283
 284             fscal            = _mm_and_ps(fscal,cutoff_mask);
 285
 286              /* Update vectorial force */
 287             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 288             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 289             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 290
 291             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 292             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 293             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 294
 295             }
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             if (gmx_mm_any_lt(rsq10,rcutoff2))
 302             {
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq10             = _mm_mul_ps(iq1,jq0);
 306
 307             /* REACTION-FIELD ELECTROSTATICS */
 308             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 309             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 310
 311             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velec            = _mm_and_ps(velec,cutoff_mask);
 315             velecsum         = _mm_add_ps(velecsum,velec);
 316
 317             fscal            = felec;
 318
 319             fscal            = _mm_and_ps(fscal,cutoff_mask);
 320
 321              /* Update vectorial force */
 322             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 323             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 324             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 325
 326             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 327             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 328             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 329
 330             }
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             if (gmx_mm_any_lt(rsq20,rcutoff2))
 337             {
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq20             = _mm_mul_ps(iq2,jq0);
 341
 342             /* REACTION-FIELD ELECTROSTATICS */
 343             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 344             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 345
 346             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velec            = _mm_and_ps(velec,cutoff_mask);
 350             velecsum         = _mm_add_ps(velecsum,velec);
 351
 352             fscal            = felec;
 353
 354             fscal            = _mm_and_ps(fscal,cutoff_mask);
 355
 356              /* Update vectorial force */
 357             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 358             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 359             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 360
 361             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 362             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 363             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 364
 365             }
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             if (gmx_mm_any_lt(rsq30,rcutoff2))
 372             {
 373
 374             /* Compute parameters for interactions between i and j atoms */
 375             qq30             = _mm_mul_ps(iq3,jq0);
 376
 377             /* REACTION-FIELD ELECTROSTATICS */
 378             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 379             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 380
 381             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 382
 383             /* Update potential sum for this i atom from the interaction with this j atom. */
 384             velec            = _mm_and_ps(velec,cutoff_mask);
 385             velecsum         = _mm_add_ps(velecsum,velec);
 386
 387             fscal            = felec;
 388
 389             fscal            = _mm_and_ps(fscal,cutoff_mask);
 390
 391              /* Update vectorial force */
 392             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 393             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 394             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 395
 396             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 397             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 398             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 399
 400             }
 401
 402             fjptrA             = f+j_coord_offsetA;
 403             fjptrB             = f+j_coord_offsetB;
 404             fjptrC             = f+j_coord_offsetC;
 405             fjptrD             = f+j_coord_offsetD;
 406
 407             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 408
 409             /* Inner loop uses 161 flops */
 410         }
 411
 412         if(jidx<j_index_end)
 413         {
 414
 415             /* Get j neighbor index, and coordinate index */
 416             jnrlistA         = jjnr[jidx];
 417             jnrlistB         = jjnr[jidx+1];
 418             jnrlistC         = jjnr[jidx+2];
 419             jnrlistD         = jjnr[jidx+3];
 420             /* Sign of each element will be negative for non-real atoms.
 421              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 422              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 423              */
 424             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 425             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 426             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 427             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 428             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 429             j_coord_offsetA  = DIM*jnrA;
 430             j_coord_offsetB  = DIM*jnrB;
 431             j_coord_offsetC  = DIM*jnrC;
 432             j_coord_offsetD  = DIM*jnrD;
 433
 434             /* load j atom coordinates */
 435             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 436                                               x+j_coord_offsetC,x+j_coord_offsetD,
 437                                               &jx0,&jy0,&jz0);
 438
 439             /* Calculate displacement vector */
 440             dx00             = _mm_sub_ps(ix0,jx0);
 441             dy00             = _mm_sub_ps(iy0,jy0);
 442             dz00             = _mm_sub_ps(iz0,jz0);
 443             dx10             = _mm_sub_ps(ix1,jx0);
 444             dy10             = _mm_sub_ps(iy1,jy0);
 445             dz10             = _mm_sub_ps(iz1,jz0);
 446             dx20             = _mm_sub_ps(ix2,jx0);
 447             dy20             = _mm_sub_ps(iy2,jy0);
 448             dz20             = _mm_sub_ps(iz2,jz0);
 449             dx30             = _mm_sub_ps(ix3,jx0);
 450             dy30             = _mm_sub_ps(iy3,jy0);
 451             dz30             = _mm_sub_ps(iz3,jz0);
 452
 453             /* Calculate squared distance and things based on it */
 454             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 455             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 456             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 457             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 458
 459             rinv10           = avx128fma_invsqrt_f(rsq10);
 460             rinv20           = avx128fma_invsqrt_f(rsq20);
 461             rinv30           = avx128fma_invsqrt_f(rsq30);
 462
 463             rinvsq00         = avx128fma_inv_f(rsq00);
 464             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 465             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 466             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 467
 468             /* Load parameters for j particles */
 469             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 470                                                               charge+jnrC+0,charge+jnrD+0);
 471             vdwjidx0A        = 2*vdwtype[jnrA+0];
 472             vdwjidx0B        = 2*vdwtype[jnrB+0];
 473             vdwjidx0C        = 2*vdwtype[jnrC+0];
 474             vdwjidx0D        = 2*vdwtype[jnrD+0];
 475
 476             fjx0             = _mm_setzero_ps();
 477             fjy0             = _mm_setzero_ps();
 478             fjz0             = _mm_setzero_ps();
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             if (gmx_mm_any_lt(rsq00,rcutoff2))
 485             {
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 489                                          vdwparam+vdwioffset0+vdwjidx0B,
 490                                          vdwparam+vdwioffset0+vdwjidx0C,
 491                                          vdwparam+vdwioffset0+vdwjidx0D,
 492                                          &c6_00,&c12_00);
 493
 494             /* LENNARD-JONES DISPERSION/REPULSION */
 495
 496             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 497             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 498             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 499             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 500                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 501             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 502
 503             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 507             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 508             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 509
 510             fscal            = fvdw;
 511
 512             fscal            = _mm_and_ps(fscal,cutoff_mask);
 513
 514             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 515
 516              /* Update vectorial force */
 517             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 518             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 519             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 520
 521             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 522             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 523             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 524
 525             }
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             if (gmx_mm_any_lt(rsq10,rcutoff2))
 532             {
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq10             = _mm_mul_ps(iq1,jq0);
 536
 537             /* REACTION-FIELD ELECTROSTATICS */
 538             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 539             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 540
 541             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velec            = _mm_and_ps(velec,cutoff_mask);
 545             velec            = _mm_andnot_ps(dummy_mask,velec);
 546             velecsum         = _mm_add_ps(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm_and_ps(fscal,cutoff_mask);
 551
 552             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 553
 554              /* Update vectorial force */
 555             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 556             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 557             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 558
 559             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 560             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 561             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 562
 563             }
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             if (gmx_mm_any_lt(rsq20,rcutoff2))
 570             {
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq20             = _mm_mul_ps(iq2,jq0);
 574
 575             /* REACTION-FIELD ELECTROSTATICS */
 576             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 577             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 578
 579             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm_and_ps(velec,cutoff_mask);
 583             velec            = _mm_andnot_ps(dummy_mask,velec);
 584             velecsum         = _mm_add_ps(velecsum,velec);
 585
 586             fscal            = felec;
 587
 588             fscal            = _mm_and_ps(fscal,cutoff_mask);
 589
 590             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 591
 592              /* Update vectorial force */
 593             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 594             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 595             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 596
 597             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 598             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 599             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 600
 601             }
 602
 603             /**************************
 604              * CALCULATE INTERACTIONS *
 605              **************************/
 606
 607             if (gmx_mm_any_lt(rsq30,rcutoff2))
 608             {
 609
 610             /* Compute parameters for interactions between i and j atoms */
 611             qq30             = _mm_mul_ps(iq3,jq0);
 612
 613             /* REACTION-FIELD ELECTROSTATICS */
 614             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 615             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 616
 617             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 618
 619             /* Update potential sum for this i atom from the interaction with this j atom. */
 620             velec            = _mm_and_ps(velec,cutoff_mask);
 621             velec            = _mm_andnot_ps(dummy_mask,velec);
 622             velecsum         = _mm_add_ps(velecsum,velec);
 623
 624             fscal            = felec;
 625
 626             fscal            = _mm_and_ps(fscal,cutoff_mask);
 627
 628             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 629
 630              /* Update vectorial force */
 631             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 632             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 633             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 634
 635             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 636             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 637             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 638
 639             }
 640
 641             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 642             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 643             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 644             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 645
 646             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 647
 648             /* Inner loop uses 161 flops */
 649         }
 650
 651         /* End of innermost loop */
 652
 653         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 654                                               f+i_coord_offset,fshift+i_shift_offset);
 655
 656         ggid                        = gid[iidx];
 657         /* Update potential energies */
 658         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 659         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 660
 661         /* Increment number of inner iterations */
 662         inneriter                  += j_index_end - j_index_start;
 663
 664         /* Outer loop uses 26 flops */
 665     }
 666
 667     /* Increment number of outer iterations */
 668     outeriter        += nri;
 669
 670     /* Update outer/inner flops */
 671
 672     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*161);
 673 }
 674 /*
 675  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_128_fma_single
 676  * Electrostatics interaction: ReactionField
 677  * VdW interaction:            LennardJones
 678  * Geometry:                   Water4-Particle
 679  * Calculate force/pot:        Force
 680  */
 681 void
 682 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_128_fma_single
 683                     (t_nblist                    * gmx_restrict       nlist,
 684                      rvec                        * gmx_restrict          xx,
 685                      rvec                        * gmx_restrict          ff,
 686                      struct t_forcerec           * gmx_restrict          fr,
 687                      t_mdatoms                   * gmx_restrict     mdatoms,
 688                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 689                      t_nrnb                      * gmx_restrict        nrnb)
 690 {
 691     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 692      * just 0 for non-waters.
 693      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 694      * jnr indices corresponding to data put in the four positions in the SIMD register.
 695      */
 696     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 697     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 698     int              jnrA,jnrB,jnrC,jnrD;
 699     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 700     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 701     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 702     real             rcutoff_scalar;
 703     real             *shiftvec,*fshift,*x,*f;
 704     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 705     real             scratch[4*DIM];
 706     __m128           fscal,rcutoff,rcutoff2,jidxall;
 707     int              vdwioffset0;
 708     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 709     int              vdwioffset1;
 710     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 711     int              vdwioffset2;
 712     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 713     int              vdwioffset3;
 714     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 715     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 716     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 717     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 718     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 719     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 720     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 721     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 722     real             *charge;
 723     int              nvdwtype;
 724     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 725     int              *vdwtype;
 726     real             *vdwparam;
 727     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 728     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 729     __m128           dummy_mask,cutoff_mask;
 730     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 731     __m128           one     = _mm_set1_ps(1.0);
 732     __m128           two     = _mm_set1_ps(2.0);
 733     x                = xx[0];
 734     f                = ff[0];
 735
 736     nri              = nlist->nri;
 737     iinr             = nlist->iinr;
 738     jindex           = nlist->jindex;
 739     jjnr             = nlist->jjnr;
 740     shiftidx         = nlist->shift;
 741     gid              = nlist->gid;
 742     shiftvec         = fr->shift_vec[0];
 743     fshift           = fr->fshift[0];
 744     facel            = _mm_set1_ps(fr->ic->epsfac);
 745     charge           = mdatoms->chargeA;
 746     krf              = _mm_set1_ps(fr->ic->k_rf);
 747     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 748     crf              = _mm_set1_ps(fr->ic->c_rf);
 749     nvdwtype         = fr->ntype;
 750     vdwparam         = fr->nbfp;
 751     vdwtype          = mdatoms->typeA;
 752
 753     /* Setup water-specific parameters */
 754     inr              = nlist->iinr[0];
 755     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 756     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 757     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 758     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 759
 760     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 761     rcutoff_scalar   = fr->ic->rcoulomb;
 762     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 763     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 764
 765     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 766     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 767
 768     /* Avoid stupid compiler warnings */
 769     jnrA = jnrB = jnrC = jnrD = 0;
 770     j_coord_offsetA = 0;
 771     j_coord_offsetB = 0;
 772     j_coord_offsetC = 0;
 773     j_coord_offsetD = 0;
 774
 775     outeriter        = 0;
 776     inneriter        = 0;
 777
 778     for(iidx=0;iidx<4*DIM;iidx++)
 779     {
 780         scratch[iidx] = 0.0;
 781     }
 782
 783     /* Start outer loop over neighborlists */
 784     for(iidx=0; iidx<nri; iidx++)
 785     {
 786         /* Load shift vector for this list */
 787         i_shift_offset   = DIM*shiftidx[iidx];
 788
 789         /* Load limits for loop over neighbors */
 790         j_index_start    = jindex[iidx];
 791         j_index_end      = jindex[iidx+1];
 792
 793         /* Get outer coordinate index */
 794         inr              = iinr[iidx];
 795         i_coord_offset   = DIM*inr;
 796
 797         /* Load i particle coords and add shift vector */
 798         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 799                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 800
 801         fix0             = _mm_setzero_ps();
 802         fiy0             = _mm_setzero_ps();
 803         fiz0             = _mm_setzero_ps();
 804         fix1             = _mm_setzero_ps();
 805         fiy1             = _mm_setzero_ps();
 806         fiz1             = _mm_setzero_ps();
 807         fix2             = _mm_setzero_ps();
 808         fiy2             = _mm_setzero_ps();
 809         fiz2             = _mm_setzero_ps();
 810         fix3             = _mm_setzero_ps();
 811         fiy3             = _mm_setzero_ps();
 812         fiz3             = _mm_setzero_ps();
 813
 814         /* Start inner kernel loop */
 815         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 816         {
 817
 818             /* Get j neighbor index, and coordinate index */
 819             jnrA             = jjnr[jidx];
 820             jnrB             = jjnr[jidx+1];
 821             jnrC             = jjnr[jidx+2];
 822             jnrD             = jjnr[jidx+3];
 823             j_coord_offsetA  = DIM*jnrA;
 824             j_coord_offsetB  = DIM*jnrB;
 825             j_coord_offsetC  = DIM*jnrC;
 826             j_coord_offsetD  = DIM*jnrD;
 827
 828             /* load j atom coordinates */
 829             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 830                                               x+j_coord_offsetC,x+j_coord_offsetD,
 831                                               &jx0,&jy0,&jz0);
 832
 833             /* Calculate displacement vector */
 834             dx00             = _mm_sub_ps(ix0,jx0);
 835             dy00             = _mm_sub_ps(iy0,jy0);
 836             dz00             = _mm_sub_ps(iz0,jz0);
 837             dx10             = _mm_sub_ps(ix1,jx0);
 838             dy10             = _mm_sub_ps(iy1,jy0);
 839             dz10             = _mm_sub_ps(iz1,jz0);
 840             dx20             = _mm_sub_ps(ix2,jx0);
 841             dy20             = _mm_sub_ps(iy2,jy0);
 842             dz20             = _mm_sub_ps(iz2,jz0);
 843             dx30             = _mm_sub_ps(ix3,jx0);
 844             dy30             = _mm_sub_ps(iy3,jy0);
 845             dz30             = _mm_sub_ps(iz3,jz0);
 846
 847             /* Calculate squared distance and things based on it */
 848             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 849             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 850             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 851             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 852
 853             rinv10           = avx128fma_invsqrt_f(rsq10);
 854             rinv20           = avx128fma_invsqrt_f(rsq20);
 855             rinv30           = avx128fma_invsqrt_f(rsq30);
 856
 857             rinvsq00         = avx128fma_inv_f(rsq00);
 858             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 859             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 860             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 861
 862             /* Load parameters for j particles */
 863             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 864                                                               charge+jnrC+0,charge+jnrD+0);
 865             vdwjidx0A        = 2*vdwtype[jnrA+0];
 866             vdwjidx0B        = 2*vdwtype[jnrB+0];
 867             vdwjidx0C        = 2*vdwtype[jnrC+0];
 868             vdwjidx0D        = 2*vdwtype[jnrD+0];
 869
 870             fjx0             = _mm_setzero_ps();
 871             fjy0             = _mm_setzero_ps();
 872             fjz0             = _mm_setzero_ps();
 873
 874             /**************************
 875              * CALCULATE INTERACTIONS *
 876              **************************/
 877
 878             if (gmx_mm_any_lt(rsq00,rcutoff2))
 879             {
 880
 881             /* Compute parameters for interactions between i and j atoms */
 882             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 883                                          vdwparam+vdwioffset0+vdwjidx0B,
 884                                          vdwparam+vdwioffset0+vdwjidx0C,
 885                                          vdwparam+vdwioffset0+vdwjidx0D,
 886                                          &c6_00,&c12_00);
 887
 888             /* LENNARD-JONES DISPERSION/REPULSION */
 889
 890             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 891             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 892
 893             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 894
 895             fscal            = fvdw;
 896
 897             fscal            = _mm_and_ps(fscal,cutoff_mask);
 898
 899              /* Update vectorial force */
 900             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 901             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 902             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 903
 904             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 905             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 906             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 907
 908             }
 909
 910             /**************************
 911              * CALCULATE INTERACTIONS *
 912              **************************/
 913
 914             if (gmx_mm_any_lt(rsq10,rcutoff2))
 915             {
 916
 917             /* Compute parameters for interactions between i and j atoms */
 918             qq10             = _mm_mul_ps(iq1,jq0);
 919
 920             /* REACTION-FIELD ELECTROSTATICS */
 921             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 922
 923             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 924
 925             fscal            = felec;
 926
 927             fscal            = _mm_and_ps(fscal,cutoff_mask);
 928
 929              /* Update vectorial force */
 930             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 931             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 932             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 933
 934             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 935             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 936             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 937
 938             }
 939
 940             /**************************
 941              * CALCULATE INTERACTIONS *
 942              **************************/
 943
 944             if (gmx_mm_any_lt(rsq20,rcutoff2))
 945             {
 946
 947             /* Compute parameters for interactions between i and j atoms */
 948             qq20             = _mm_mul_ps(iq2,jq0);
 949
 950             /* REACTION-FIELD ELECTROSTATICS */
 951             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 952
 953             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 954
 955             fscal            = felec;
 956
 957             fscal            = _mm_and_ps(fscal,cutoff_mask);
 958
 959              /* Update vectorial force */
 960             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 961             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 962             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 963
 964             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 965             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 966             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 967
 968             }
 969
 970             /**************************
 971              * CALCULATE INTERACTIONS *
 972              **************************/
 973
 974             if (gmx_mm_any_lt(rsq30,rcutoff2))
 975             {
 976
 977             /* Compute parameters for interactions between i and j atoms */
 978             qq30             = _mm_mul_ps(iq3,jq0);
 979
 980             /* REACTION-FIELD ELECTROSTATICS */
 981             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 982
 983             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 984
 985             fscal            = felec;
 986
 987             fscal            = _mm_and_ps(fscal,cutoff_mask);
 988
 989              /* Update vectorial force */
 990             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 991             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 992             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 993
 994             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 995             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 996             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 997
 998             }
 999
1000             fjptrA             = f+j_coord_offsetA;
1001             fjptrB             = f+j_coord_offsetB;
1002             fjptrC             = f+j_coord_offsetC;
1003             fjptrD             = f+j_coord_offsetD;
1004
1005             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1006
1007             /* Inner loop uses 132 flops */
1008         }
1009
1010         if(jidx<j_index_end)
1011         {
1012
1013             /* Get j neighbor index, and coordinate index */
1014             jnrlistA         = jjnr[jidx];
1015             jnrlistB         = jjnr[jidx+1];
1016             jnrlistC         = jjnr[jidx+2];
1017             jnrlistD         = jjnr[jidx+3];
1018             /* Sign of each element will be negative for non-real atoms.
1019              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1020              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1021              */
1022             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1023             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1024             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1025             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1026             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1027             j_coord_offsetA  = DIM*jnrA;
1028             j_coord_offsetB  = DIM*jnrB;
1029             j_coord_offsetC  = DIM*jnrC;
1030             j_coord_offsetD  = DIM*jnrD;
1031
1032             /* load j atom coordinates */
1033             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1034                                               x+j_coord_offsetC,x+j_coord_offsetD,
1035                                               &jx0,&jy0,&jz0);
1036
1037             /* Calculate displacement vector */
1038             dx00             = _mm_sub_ps(ix0,jx0);
1039             dy00             = _mm_sub_ps(iy0,jy0);
1040             dz00             = _mm_sub_ps(iz0,jz0);
1041             dx10             = _mm_sub_ps(ix1,jx0);
1042             dy10             = _mm_sub_ps(iy1,jy0);
1043             dz10             = _mm_sub_ps(iz1,jz0);
1044             dx20             = _mm_sub_ps(ix2,jx0);
1045             dy20             = _mm_sub_ps(iy2,jy0);
1046             dz20             = _mm_sub_ps(iz2,jz0);
1047             dx30             = _mm_sub_ps(ix3,jx0);
1048             dy30             = _mm_sub_ps(iy3,jy0);
1049             dz30             = _mm_sub_ps(iz3,jz0);
1050
1051             /* Calculate squared distance and things based on it */
1052             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1053             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1054             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1055             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1056
1057             rinv10           = avx128fma_invsqrt_f(rsq10);
1058             rinv20           = avx128fma_invsqrt_f(rsq20);
1059             rinv30           = avx128fma_invsqrt_f(rsq30);
1060
1061             rinvsq00         = avx128fma_inv_f(rsq00);
1062             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1063             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1064             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1065
1066             /* Load parameters for j particles */
1067             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1068                                                               charge+jnrC+0,charge+jnrD+0);
1069             vdwjidx0A        = 2*vdwtype[jnrA+0];
1070             vdwjidx0B        = 2*vdwtype[jnrB+0];
1071             vdwjidx0C        = 2*vdwtype[jnrC+0];
1072             vdwjidx0D        = 2*vdwtype[jnrD+0];
1073
1074             fjx0             = _mm_setzero_ps();
1075             fjy0             = _mm_setzero_ps();
1076             fjz0             = _mm_setzero_ps();
1077
1078             /**************************
1079              * CALCULATE INTERACTIONS *
1080              **************************/
1081
1082             if (gmx_mm_any_lt(rsq00,rcutoff2))
1083             {
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1087                                          vdwparam+vdwioffset0+vdwjidx0B,
1088                                          vdwparam+vdwioffset0+vdwjidx0C,
1089                                          vdwparam+vdwioffset0+vdwjidx0D,
1090                                          &c6_00,&c12_00);
1091
1092             /* LENNARD-JONES DISPERSION/REPULSION */
1093
1094             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1095             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1096
1097             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1098
1099             fscal            = fvdw;
1100
1101             fscal            = _mm_and_ps(fscal,cutoff_mask);
1102
1103             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1104
1105              /* Update vectorial force */
1106             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1107             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1108             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1109
1110             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1111             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1112             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1113
1114             }
1115
1116             /**************************
1117              * CALCULATE INTERACTIONS *
1118              **************************/
1119
1120             if (gmx_mm_any_lt(rsq10,rcutoff2))
1121             {
1122
1123             /* Compute parameters for interactions between i and j atoms */
1124             qq10             = _mm_mul_ps(iq1,jq0);
1125
1126             /* REACTION-FIELD ELECTROSTATICS */
1127             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1128
1129             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1130
1131             fscal            = felec;
1132
1133             fscal            = _mm_and_ps(fscal,cutoff_mask);
1134
1135             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1136
1137              /* Update vectorial force */
1138             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1139             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1140             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1141
1142             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1143             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1144             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1145
1146             }
1147
1148             /**************************
1149              * CALCULATE INTERACTIONS *
1150              **************************/
1151
1152             if (gmx_mm_any_lt(rsq20,rcutoff2))
1153             {
1154
1155             /* Compute parameters for interactions between i and j atoms */
1156             qq20             = _mm_mul_ps(iq2,jq0);
1157
1158             /* REACTION-FIELD ELECTROSTATICS */
1159             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1160
1161             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm_and_ps(fscal,cutoff_mask);
1166
1167             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1168
1169              /* Update vectorial force */
1170             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1171             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1172             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1173
1174             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1175             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1176             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1177
1178             }
1179
1180             /**************************
1181              * CALCULATE INTERACTIONS *
1182              **************************/
1183
1184             if (gmx_mm_any_lt(rsq30,rcutoff2))
1185             {
1186
1187             /* Compute parameters for interactions between i and j atoms */
1188             qq30             = _mm_mul_ps(iq3,jq0);
1189
1190             /* REACTION-FIELD ELECTROSTATICS */
1191             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1192
1193             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1194
1195             fscal            = felec;
1196
1197             fscal            = _mm_and_ps(fscal,cutoff_mask);
1198
1199             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1200
1201              /* Update vectorial force */
1202             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1203             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1204             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1205
1206             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1207             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1208             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1209
1210             }
1211
1212             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1213             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1214             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1215             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1216
1217             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1218
1219             /* Inner loop uses 132 flops */
1220         }
1221
1222         /* End of innermost loop */
1223
1224         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1225                                               f+i_coord_offset,fshift+i_shift_offset);
1226
1227         /* Increment number of inner iterations */
1228         inneriter                  += j_index_end - j_index_start;
1229
1230         /* Outer loop uses 24 flops */
1231     }
1232
1233     /* Increment number of outer iterations */
1234     outeriter        += nri;
1235
1236     /* Update outer/inner flops */
1237
1238     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*132);
1239 }