src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128i          ewitab;
  97     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  99     real             *ewtab;
 100     __m128           dummy_mask,cutoff_mask;
 101     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 102     __m128           one     = _mm_set1_ps(1.0);
 103     __m128           two     = _mm_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_ps(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 122     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 123     beta2            = _mm_mul_ps(beta,beta);
 124     beta3            = _mm_mul_ps(beta,beta2);
 125     ewtab            = fr->ic->tabq_coul_FDV0;
 126     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 127     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 128
 129     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 130     rcutoff_scalar   = fr->rcoulomb;
 131     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 132     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 133
 134     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 135     rvdw             = _mm_set1_ps(fr->rvdw);
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 168
 169         fix0             = _mm_setzero_ps();
 170         fiy0             = _mm_setzero_ps();
 171         fiz0             = _mm_setzero_ps();
 172
 173         /* Load parameters for i particles */
 174         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 175         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm_setzero_ps();
 179         vvdwsum          = _mm_setzero_ps();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             jnrC             = jjnr[jidx+2];
 189             jnrD             = jjnr[jidx+3];
 190             j_coord_offsetA  = DIM*jnrA;
 191             j_coord_offsetB  = DIM*jnrB;
 192             j_coord_offsetC  = DIM*jnrC;
 193             j_coord_offsetD  = DIM*jnrD;
 194
 195             /* load j atom coordinates */
 196             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                               x+j_coord_offsetC,x+j_coord_offsetD,
 198                                               &jx0,&jy0,&jz0);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _mm_sub_ps(ix0,jx0);
 202             dy00             = _mm_sub_ps(iy0,jy0);
 203             dz00             = _mm_sub_ps(iz0,jz0);
 204
 205             /* Calculate squared distance and things based on it */
 206             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 207
 208             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 209
 210             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 214                                                               charge+jnrC+0,charge+jnrD+0);
 215             vdwjidx0A        = 2*vdwtype[jnrA+0];
 216             vdwjidx0B        = 2*vdwtype[jnrB+0];
 217             vdwjidx0C        = 2*vdwtype[jnrC+0];
 218             vdwjidx0D        = 2*vdwtype[jnrD+0];
 219
 220             /**************************
 221              * CALCULATE INTERACTIONS *
 222              **************************/
 223
 224             if (gmx_mm_any_lt(rsq00,rcutoff2))
 225             {
 226
 227             r00              = _mm_mul_ps(rsq00,rinv00);
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq00             = _mm_mul_ps(iq0,jq0);
 231             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 232                                          vdwparam+vdwioffset0+vdwjidx0B,
 233                                          vdwparam+vdwioffset0+vdwjidx0C,
 234                                          vdwparam+vdwioffset0+vdwjidx0D,
 235                                          &c6_00,&c12_00);
 236
 237             /* EWALD ELECTROSTATICS */
 238
 239             /* Analytical PME correction */
 240             zeta2            = _mm_mul_ps(beta2,rsq00);
 241             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 242             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 243             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 244             felec            = _mm_mul_ps(qq00,felec);
 245             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 246             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 247             velec            = _mm_mul_ps(qq00,velec);
 248
 249             /* LENNARD-JONES DISPERSION/REPULSION */
 250
 251             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 252             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 253             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 254             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 255                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 256             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 257
 258             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 259
 260             /* Update potential sum for this i atom from the interaction with this j atom. */
 261             velec            = _mm_and_ps(velec,cutoff_mask);
 262             velecsum         = _mm_add_ps(velecsum,velec);
 263             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 264             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 265
 266             fscal            = _mm_add_ps(felec,fvdw);
 267
 268             fscal            = _mm_and_ps(fscal,cutoff_mask);
 269
 270              /* Update vectorial force */
 271             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 272             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 273             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 274
 275             fjptrA             = f+j_coord_offsetA;
 276             fjptrB             = f+j_coord_offsetB;
 277             fjptrC             = f+j_coord_offsetC;
 278             fjptrD             = f+j_coord_offsetD;
 279             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 280                                                    _mm_mul_ps(dx00,fscal),
 281                                                    _mm_mul_ps(dy00,fscal),
 282                                                    _mm_mul_ps(dz00,fscal));
 283
 284             }
 285
 286             /* Inner loop uses 51 flops */
 287         }
 288
 289         if(jidx<j_index_end)
 290         {
 291
 292             /* Get j neighbor index, and coordinate index */
 293             jnrlistA         = jjnr[jidx];
 294             jnrlistB         = jjnr[jidx+1];
 295             jnrlistC         = jjnr[jidx+2];
 296             jnrlistD         = jjnr[jidx+3];
 297             /* Sign of each element will be negative for non-real atoms.
 298              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 299              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 300              */
 301             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 302             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 303             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 304             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 305             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 306             j_coord_offsetA  = DIM*jnrA;
 307             j_coord_offsetB  = DIM*jnrB;
 308             j_coord_offsetC  = DIM*jnrC;
 309             j_coord_offsetD  = DIM*jnrD;
 310
 311             /* load j atom coordinates */
 312             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 313                                               x+j_coord_offsetC,x+j_coord_offsetD,
 314                                               &jx0,&jy0,&jz0);
 315
 316             /* Calculate displacement vector */
 317             dx00             = _mm_sub_ps(ix0,jx0);
 318             dy00             = _mm_sub_ps(iy0,jy0);
 319             dz00             = _mm_sub_ps(iz0,jz0);
 320
 321             /* Calculate squared distance and things based on it */
 322             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 323
 324             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 325
 326             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 327
 328             /* Load parameters for j particles */
 329             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 330                                                               charge+jnrC+0,charge+jnrD+0);
 331             vdwjidx0A        = 2*vdwtype[jnrA+0];
 332             vdwjidx0B        = 2*vdwtype[jnrB+0];
 333             vdwjidx0C        = 2*vdwtype[jnrC+0];
 334             vdwjidx0D        = 2*vdwtype[jnrD+0];
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             if (gmx_mm_any_lt(rsq00,rcutoff2))
 341             {
 342
 343             r00              = _mm_mul_ps(rsq00,rinv00);
 344             r00              = _mm_andnot_ps(dummy_mask,r00);
 345
 346             /* Compute parameters for interactions between i and j atoms */
 347             qq00             = _mm_mul_ps(iq0,jq0);
 348             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 349                                          vdwparam+vdwioffset0+vdwjidx0B,
 350                                          vdwparam+vdwioffset0+vdwjidx0C,
 351                                          vdwparam+vdwioffset0+vdwjidx0D,
 352                                          &c6_00,&c12_00);
 353
 354             /* EWALD ELECTROSTATICS */
 355
 356             /* Analytical PME correction */
 357             zeta2            = _mm_mul_ps(beta2,rsq00);
 358             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 359             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 360             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 361             felec            = _mm_mul_ps(qq00,felec);
 362             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 363             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 364             velec            = _mm_mul_ps(qq00,velec);
 365
 366             /* LENNARD-JONES DISPERSION/REPULSION */
 367
 368             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 369             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 370             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 371             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 372                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 373             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 374
 375             cutoff_mask      = _mm_cmplt_ps(rsq00,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             velec            = _mm_andnot_ps(dummy_mask,velec);
 380             velecsum         = _mm_add_ps(velecsum,velec);
 381             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 382             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 383             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 384
 385             fscal            = _mm_add_ps(felec,fvdw);
 386
 387             fscal            = _mm_and_ps(fscal,cutoff_mask);
 388
 389             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 390
 391              /* Update vectorial force */
 392             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 393             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 394             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 395
 396             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 397             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 398             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 399             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 400             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 401                                                    _mm_mul_ps(dx00,fscal),
 402                                                    _mm_mul_ps(dy00,fscal),
 403                                                    _mm_mul_ps(dz00,fscal));
 404
 405             }
 406
 407             /* Inner loop uses 52 flops */
 408         }
 409
 410         /* End of innermost loop */
 411
 412         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 413                                               f+i_coord_offset,fshift+i_shift_offset);
 414
 415         ggid                        = gid[iidx];
 416         /* Update potential energies */
 417         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 418         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 419
 420         /* Increment number of inner iterations */
 421         inneriter                  += j_index_end - j_index_start;
 422
 423         /* Outer loop uses 9 flops */
 424     }
 425
 426     /* Increment number of outer iterations */
 427     outeriter        += nri;
 428
 429     /* Update outer/inner flops */
 430
 431     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*52);
 432 }
 433 /*
 434  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_128_fma_single
 435  * Electrostatics interaction: Ewald
 436  * VdW interaction:            LennardJones
 437  * Geometry:                   Particle-Particle
 438  * Calculate force/pot:        Force
 439  */
 440 void
 441 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_128_fma_single
 442                     (t_nblist                    * gmx_restrict       nlist,
 443                      rvec                        * gmx_restrict          xx,
 444                      rvec                        * gmx_restrict          ff,
 445                      t_forcerec                  * gmx_restrict          fr,
 446                      t_mdatoms                   * gmx_restrict     mdatoms,
 447                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 448                      t_nrnb                      * gmx_restrict        nrnb)
 449 {
 450     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 451      * just 0 for non-waters.
 452      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 453      * jnr indices corresponding to data put in the four positions in the SIMD register.
 454      */
 455     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 456     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 457     int              jnrA,jnrB,jnrC,jnrD;
 458     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 459     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 460     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 461     real             rcutoff_scalar;
 462     real             *shiftvec,*fshift,*x,*f;
 463     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 464     real             scratch[4*DIM];
 465     __m128           fscal,rcutoff,rcutoff2,jidxall;
 466     int              vdwioffset0;
 467     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 468     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 469     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 470     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 471     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 472     real             *charge;
 473     int              nvdwtype;
 474     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 475     int              *vdwtype;
 476     real             *vdwparam;
 477     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 478     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 479     __m128i          ewitab;
 480     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 481     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 482     real             *ewtab;
 483     __m128           dummy_mask,cutoff_mask;
 484     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 485     __m128           one     = _mm_set1_ps(1.0);
 486     __m128           two     = _mm_set1_ps(2.0);
 487     x                = xx[0];
 488     f                = ff[0];
 489
 490     nri              = nlist->nri;
 491     iinr             = nlist->iinr;
 492     jindex           = nlist->jindex;
 493     jjnr             = nlist->jjnr;
 494     shiftidx         = nlist->shift;
 495     gid              = nlist->gid;
 496     shiftvec         = fr->shift_vec[0];
 497     fshift           = fr->fshift[0];
 498     facel            = _mm_set1_ps(fr->epsfac);
 499     charge           = mdatoms->chargeA;
 500     nvdwtype         = fr->ntype;
 501     vdwparam         = fr->nbfp;
 502     vdwtype          = mdatoms->typeA;
 503
 504     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 505     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 506     beta2            = _mm_mul_ps(beta,beta);
 507     beta3            = _mm_mul_ps(beta,beta2);
 508     ewtab            = fr->ic->tabq_coul_F;
 509     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 510     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 511
 512     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 513     rcutoff_scalar   = fr->rcoulomb;
 514     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 515     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 516
 517     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 518     rvdw             = _mm_set1_ps(fr->rvdw);
 519
 520     /* Avoid stupid compiler warnings */
 521     jnrA = jnrB = jnrC = jnrD = 0;
 522     j_coord_offsetA = 0;
 523     j_coord_offsetB = 0;
 524     j_coord_offsetC = 0;
 525     j_coord_offsetD = 0;
 526
 527     outeriter        = 0;
 528     inneriter        = 0;
 529
 530     for(iidx=0;iidx<4*DIM;iidx++)
 531     {
 532         scratch[iidx] = 0.0;
 533     }
 534
 535     /* Start outer loop over neighborlists */
 536     for(iidx=0; iidx<nri; iidx++)
 537     {
 538         /* Load shift vector for this list */
 539         i_shift_offset   = DIM*shiftidx[iidx];
 540
 541         /* Load limits for loop over neighbors */
 542         j_index_start    = jindex[iidx];
 543         j_index_end      = jindex[iidx+1];
 544
 545         /* Get outer coordinate index */
 546         inr              = iinr[iidx];
 547         i_coord_offset   = DIM*inr;
 548
 549         /* Load i particle coords and add shift vector */
 550         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 551
 552         fix0             = _mm_setzero_ps();
 553         fiy0             = _mm_setzero_ps();
 554         fiz0             = _mm_setzero_ps();
 555
 556         /* Load parameters for i particles */
 557         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 558         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 559
 560         /* Start inner kernel loop */
 561         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 562         {
 563
 564             /* Get j neighbor index, and coordinate index */
 565             jnrA             = jjnr[jidx];
 566             jnrB             = jjnr[jidx+1];
 567             jnrC             = jjnr[jidx+2];
 568             jnrD             = jjnr[jidx+3];
 569             j_coord_offsetA  = DIM*jnrA;
 570             j_coord_offsetB  = DIM*jnrB;
 571             j_coord_offsetC  = DIM*jnrC;
 572             j_coord_offsetD  = DIM*jnrD;
 573
 574             /* load j atom coordinates */
 575             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 576                                               x+j_coord_offsetC,x+j_coord_offsetD,
 577                                               &jx0,&jy0,&jz0);
 578
 579             /* Calculate displacement vector */
 580             dx00             = _mm_sub_ps(ix0,jx0);
 581             dy00             = _mm_sub_ps(iy0,jy0);
 582             dz00             = _mm_sub_ps(iz0,jz0);
 583
 584             /* Calculate squared distance and things based on it */
 585             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 586
 587             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 588
 589             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 590
 591             /* Load parameters for j particles */
 592             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 593                                                               charge+jnrC+0,charge+jnrD+0);
 594             vdwjidx0A        = 2*vdwtype[jnrA+0];
 595             vdwjidx0B        = 2*vdwtype[jnrB+0];
 596             vdwjidx0C        = 2*vdwtype[jnrC+0];
 597             vdwjidx0D        = 2*vdwtype[jnrD+0];
 598
 599             /**************************
 600              * CALCULATE INTERACTIONS *
 601              **************************/
 602
 603             if (gmx_mm_any_lt(rsq00,rcutoff2))
 604             {
 605
 606             r00              = _mm_mul_ps(rsq00,rinv00);
 607
 608             /* Compute parameters for interactions between i and j atoms */
 609             qq00             = _mm_mul_ps(iq0,jq0);
 610             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 611                                          vdwparam+vdwioffset0+vdwjidx0B,
 612                                          vdwparam+vdwioffset0+vdwjidx0C,
 613                                          vdwparam+vdwioffset0+vdwjidx0D,
 614                                          &c6_00,&c12_00);
 615
 616             /* EWALD ELECTROSTATICS */
 617
 618             /* Analytical PME correction */
 619             zeta2            = _mm_mul_ps(beta2,rsq00);
 620             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 621             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 622             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 623             felec            = _mm_mul_ps(qq00,felec);
 624
 625             /* LENNARD-JONES DISPERSION/REPULSION */
 626
 627             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 628             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 629
 630             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 631
 632             fscal            = _mm_add_ps(felec,fvdw);
 633
 634             fscal            = _mm_and_ps(fscal,cutoff_mask);
 635
 636              /* Update vectorial force */
 637             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 638             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 639             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 640
 641             fjptrA             = f+j_coord_offsetA;
 642             fjptrB             = f+j_coord_offsetB;
 643             fjptrC             = f+j_coord_offsetC;
 644             fjptrD             = f+j_coord_offsetD;
 645             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 646                                                    _mm_mul_ps(dx00,fscal),
 647                                                    _mm_mul_ps(dy00,fscal),
 648                                                    _mm_mul_ps(dz00,fscal));
 649
 650             }
 651
 652             /* Inner loop uses 38 flops */
 653         }
 654
 655         if(jidx<j_index_end)
 656         {
 657
 658             /* Get j neighbor index, and coordinate index */
 659             jnrlistA         = jjnr[jidx];
 660             jnrlistB         = jjnr[jidx+1];
 661             jnrlistC         = jjnr[jidx+2];
 662             jnrlistD         = jjnr[jidx+3];
 663             /* Sign of each element will be negative for non-real atoms.
 664              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 665              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 666              */
 667             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 668             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 669             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 670             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 671             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 672             j_coord_offsetA  = DIM*jnrA;
 673             j_coord_offsetB  = DIM*jnrB;
 674             j_coord_offsetC  = DIM*jnrC;
 675             j_coord_offsetD  = DIM*jnrD;
 676
 677             /* load j atom coordinates */
 678             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 679                                               x+j_coord_offsetC,x+j_coord_offsetD,
 680                                               &jx0,&jy0,&jz0);
 681
 682             /* Calculate displacement vector */
 683             dx00             = _mm_sub_ps(ix0,jx0);
 684             dy00             = _mm_sub_ps(iy0,jy0);
 685             dz00             = _mm_sub_ps(iz0,jz0);
 686
 687             /* Calculate squared distance and things based on it */
 688             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 689
 690             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 691
 692             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 693
 694             /* Load parameters for j particles */
 695             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 696                                                               charge+jnrC+0,charge+jnrD+0);
 697             vdwjidx0A        = 2*vdwtype[jnrA+0];
 698             vdwjidx0B        = 2*vdwtype[jnrB+0];
 699             vdwjidx0C        = 2*vdwtype[jnrC+0];
 700             vdwjidx0D        = 2*vdwtype[jnrD+0];
 701
 702             /**************************
 703              * CALCULATE INTERACTIONS *
 704              **************************/
 705
 706             if (gmx_mm_any_lt(rsq00,rcutoff2))
 707             {
 708
 709             r00              = _mm_mul_ps(rsq00,rinv00);
 710             r00              = _mm_andnot_ps(dummy_mask,r00);
 711
 712             /* Compute parameters for interactions between i and j atoms */
 713             qq00             = _mm_mul_ps(iq0,jq0);
 714             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 715                                          vdwparam+vdwioffset0+vdwjidx0B,
 716                                          vdwparam+vdwioffset0+vdwjidx0C,
 717                                          vdwparam+vdwioffset0+vdwjidx0D,
 718                                          &c6_00,&c12_00);
 719
 720             /* EWALD ELECTROSTATICS */
 721
 722             /* Analytical PME correction */
 723             zeta2            = _mm_mul_ps(beta2,rsq00);
 724             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 725             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 726             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 727             felec            = _mm_mul_ps(qq00,felec);
 728
 729             /* LENNARD-JONES DISPERSION/REPULSION */
 730
 731             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 732             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 733
 734             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 735
 736             fscal            = _mm_add_ps(felec,fvdw);
 737
 738             fscal            = _mm_and_ps(fscal,cutoff_mask);
 739
 740             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 741
 742              /* Update vectorial force */
 743             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 744             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 745             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 746
 747             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 748             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 749             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 750             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 751             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 752                                                    _mm_mul_ps(dx00,fscal),
 753                                                    _mm_mul_ps(dy00,fscal),
 754                                                    _mm_mul_ps(dz00,fscal));
 755
 756             }
 757
 758             /* Inner loop uses 39 flops */
 759         }
 760
 761         /* End of innermost loop */
 762
 763         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 764                                               f+i_coord_offset,fshift+i_shift_offset);
 765
 766         /* Increment number of inner iterations */
 767         inneriter                  += j_index_end - j_index_start;
 768
 769         /* Outer loop uses 7 flops */
 770     }
 771
 772     /* Increment number of outer iterations */
 773     outeriter        += nri;
 774
 775     /* Update outer/inner flops */
 776
 777     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*39);
 778 }