src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJEw_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double 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_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_avx_128_fma_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  91     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  92     __m128d           c6grid_00;
  93     real             *vdwgridparam;
  94     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  95     __m128d           one_half  = _mm_set1_pd(0.5);
  96     __m128d           minus_one = _mm_set1_pd(-1.0);
  97     __m128i          ewitab;
  98     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  99     real             *ewtab;
 100     __m128d          dummy_mask,cutoff_mask;
 101     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 102     __m128d          one     = _mm_set1_pd(1.0);
 103     __m128d          two     = _mm_set1_pd(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_pd(fr->ic->epsfac);
 116     charge           = mdatoms->chargeA;
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120     vdwgridparam     = fr->ljpme_c6grid;
 121     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 122     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 123     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 124
 125     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 126     ewtab            = fr->ic->tabq_coul_FDV0;
 127     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 128     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _mm_setzero_pd();
 156         fiy0             = _mm_setzero_pd();
 157         fiz0             = _mm_setzero_pd();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 161         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm_setzero_pd();
 165         vvdwsum          = _mm_setzero_pd();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm_sub_pd(ix0,jx0);
 183             dy00             = _mm_sub_pd(iy0,jy0);
 184             dz00             = _mm_sub_pd(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 188
 189             rinv00           = avx128fma_invsqrt_d(rsq00);
 190
 191             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 195             vdwjidx0A        = 2*vdwtype[jnrA+0];
 196             vdwjidx0B        = 2*vdwtype[jnrB+0];
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             r00              = _mm_mul_pd(rsq00,rinv00);
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             qq00             = _mm_mul_pd(iq0,jq0);
 206             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 207                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 208             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 209                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 210
 211             /* EWALD ELECTROSTATICS */
 212
 213             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 214             ewrt             = _mm_mul_pd(r00,ewtabscale);
 215             ewitab           = _mm_cvttpd_epi32(ewrt);
 216 #ifdef __XOP__
 217             eweps            = _mm_frcz_pd(ewrt);
 218 #else
 219             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 220 #endif
 221             twoeweps         = _mm_add_pd(eweps,eweps);
 222             ewitab           = _mm_slli_epi32(ewitab,2);
 223             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 224             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 225             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 226             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 227             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 228             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 229             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 230             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 231             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 232             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 233
 234             /* Analytical LJ-PME */
 235             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 236             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 237             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 238             exponent         = avx128fma_exp_d(ewcljrsq);
 239             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 240             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 241             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 242             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 243             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 244             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 245             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 246             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 247
 248             /* Update potential sum for this i atom from the interaction with this j atom. */
 249             velecsum         = _mm_add_pd(velecsum,velec);
 250             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 251
 252             fscal            = _mm_add_pd(felec,fvdw);
 253
 254             /* Update vectorial force */
 255             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 256             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 257             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 258
 259             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 260                                                    _mm_mul_pd(dx00,fscal),
 261                                                    _mm_mul_pd(dy00,fscal),
 262                                                    _mm_mul_pd(dz00,fscal));
 263
 264             /* Inner loop uses 68 flops */
 265         }
 266
 267         if(jidx<j_index_end)
 268         {
 269
 270             jnrA             = jjnr[jidx];
 271             j_coord_offsetA  = DIM*jnrA;
 272
 273             /* load j atom coordinates */
 274             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 275                                               &jx0,&jy0,&jz0);
 276
 277             /* Calculate displacement vector */
 278             dx00             = _mm_sub_pd(ix0,jx0);
 279             dy00             = _mm_sub_pd(iy0,jy0);
 280             dz00             = _mm_sub_pd(iz0,jz0);
 281
 282             /* Calculate squared distance and things based on it */
 283             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 284
 285             rinv00           = avx128fma_invsqrt_d(rsq00);
 286
 287             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 288
 289             /* Load parameters for j particles */
 290             jq0              = _mm_load_sd(charge+jnrA+0);
 291             vdwjidx0A        = 2*vdwtype[jnrA+0];
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             r00              = _mm_mul_pd(rsq00,rinv00);
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq00             = _mm_mul_pd(iq0,jq0);
 301             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 302             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 303
 304             /* EWALD ELECTROSTATICS */
 305
 306             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 307             ewrt             = _mm_mul_pd(r00,ewtabscale);
 308             ewitab           = _mm_cvttpd_epi32(ewrt);
 309 #ifdef __XOP__
 310             eweps            = _mm_frcz_pd(ewrt);
 311 #else
 312             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 313 #endif
 314             twoeweps         = _mm_add_pd(eweps,eweps);
 315             ewitab           = _mm_slli_epi32(ewitab,2);
 316             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 317             ewtabD           = _mm_setzero_pd();
 318             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 319             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 320             ewtabFn          = _mm_setzero_pd();
 321             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 322             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 323             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 324             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 325             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 326
 327             /* Analytical LJ-PME */
 328             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 329             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 330             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 331             exponent         = avx128fma_exp_d(ewcljrsq);
 332             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 333             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 334             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 335             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 336             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 337             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 338             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 339             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 343             velecsum         = _mm_add_pd(velecsum,velec);
 344             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 345             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 346
 347             fscal            = _mm_add_pd(felec,fvdw);
 348
 349             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 350
 351             /* Update vectorial force */
 352             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 353             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 354             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 355
 356             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 357                                                    _mm_mul_pd(dx00,fscal),
 358                                                    _mm_mul_pd(dy00,fscal),
 359                                                    _mm_mul_pd(dz00,fscal));
 360
 361             /* Inner loop uses 68 flops */
 362         }
 363
 364         /* End of innermost loop */
 365
 366         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 367                                               f+i_coord_offset,fshift+i_shift_offset);
 368
 369         ggid                        = gid[iidx];
 370         /* Update potential energies */
 371         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 372         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 373
 374         /* Increment number of inner iterations */
 375         inneriter                  += j_index_end - j_index_start;
 376
 377         /* Outer loop uses 9 flops */
 378     }
 379
 380     /* Increment number of outer iterations */
 381     outeriter        += nri;
 382
 383     /* Update outer/inner flops */
 384
 385     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
 386 }
 387 /*
 388  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_avx_128_fma_double
 389  * Electrostatics interaction: Ewald
 390  * VdW interaction:            LJEwald
 391  * Geometry:                   Particle-Particle
 392  * Calculate force/pot:        Force
 393  */
 394 void
 395 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_avx_128_fma_double
 396                     (t_nblist                    * gmx_restrict       nlist,
 397                      rvec                        * gmx_restrict          xx,
 398                      rvec                        * gmx_restrict          ff,
 399                      struct t_forcerec           * gmx_restrict          fr,
 400                      t_mdatoms                   * gmx_restrict     mdatoms,
 401                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 402                      t_nrnb                      * gmx_restrict        nrnb)
 403 {
 404     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 405      * just 0 for non-waters.
 406      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 407      * jnr indices corresponding to data put in the four positions in the SIMD register.
 408      */
 409     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 410     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 411     int              jnrA,jnrB;
 412     int              j_coord_offsetA,j_coord_offsetB;
 413     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 414     real             rcutoff_scalar;
 415     real             *shiftvec,*fshift,*x,*f;
 416     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 417     int              vdwioffset0;
 418     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 419     int              vdwjidx0A,vdwjidx0B;
 420     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 421     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 422     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 423     real             *charge;
 424     int              nvdwtype;
 425     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 426     int              *vdwtype;
 427     real             *vdwparam;
 428     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 429     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 430     __m128d           c6grid_00;
 431     real             *vdwgridparam;
 432     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 433     __m128d           one_half  = _mm_set1_pd(0.5);
 434     __m128d           minus_one = _mm_set1_pd(-1.0);
 435     __m128i          ewitab;
 436     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 437     real             *ewtab;
 438     __m128d          dummy_mask,cutoff_mask;
 439     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 440     __m128d          one     = _mm_set1_pd(1.0);
 441     __m128d          two     = _mm_set1_pd(2.0);
 442     x                = xx[0];
 443     f                = ff[0];
 444
 445     nri              = nlist->nri;
 446     iinr             = nlist->iinr;
 447     jindex           = nlist->jindex;
 448     jjnr             = nlist->jjnr;
 449     shiftidx         = nlist->shift;
 450     gid              = nlist->gid;
 451     shiftvec         = fr->shift_vec[0];
 452     fshift           = fr->fshift[0];
 453     facel            = _mm_set1_pd(fr->ic->epsfac);
 454     charge           = mdatoms->chargeA;
 455     nvdwtype         = fr->ntype;
 456     vdwparam         = fr->nbfp;
 457     vdwtype          = mdatoms->typeA;
 458     vdwgridparam     = fr->ljpme_c6grid;
 459     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 460     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 461     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 462
 463     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 464     ewtab            = fr->ic->tabq_coul_F;
 465     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 466     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 467
 468     /* Avoid stupid compiler warnings */
 469     jnrA = jnrB = 0;
 470     j_coord_offsetA = 0;
 471     j_coord_offsetB = 0;
 472
 473     outeriter        = 0;
 474     inneriter        = 0;
 475
 476     /* Start outer loop over neighborlists */
 477     for(iidx=0; iidx<nri; iidx++)
 478     {
 479         /* Load shift vector for this list */
 480         i_shift_offset   = DIM*shiftidx[iidx];
 481
 482         /* Load limits for loop over neighbors */
 483         j_index_start    = jindex[iidx];
 484         j_index_end      = jindex[iidx+1];
 485
 486         /* Get outer coordinate index */
 487         inr              = iinr[iidx];
 488         i_coord_offset   = DIM*inr;
 489
 490         /* Load i particle coords and add shift vector */
 491         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 492
 493         fix0             = _mm_setzero_pd();
 494         fiy0             = _mm_setzero_pd();
 495         fiz0             = _mm_setzero_pd();
 496
 497         /* Load parameters for i particles */
 498         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 499         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 500
 501         /* Start inner kernel loop */
 502         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 503         {
 504
 505             /* Get j neighbor index, and coordinate index */
 506             jnrA             = jjnr[jidx];
 507             jnrB             = jjnr[jidx+1];
 508             j_coord_offsetA  = DIM*jnrA;
 509             j_coord_offsetB  = DIM*jnrB;
 510
 511             /* load j atom coordinates */
 512             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 513                                               &jx0,&jy0,&jz0);
 514
 515             /* Calculate displacement vector */
 516             dx00             = _mm_sub_pd(ix0,jx0);
 517             dy00             = _mm_sub_pd(iy0,jy0);
 518             dz00             = _mm_sub_pd(iz0,jz0);
 519
 520             /* Calculate squared distance and things based on it */
 521             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 522
 523             rinv00           = avx128fma_invsqrt_d(rsq00);
 524
 525             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 526
 527             /* Load parameters for j particles */
 528             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 529             vdwjidx0A        = 2*vdwtype[jnrA+0];
 530             vdwjidx0B        = 2*vdwtype[jnrB+0];
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             r00              = _mm_mul_pd(rsq00,rinv00);
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             qq00             = _mm_mul_pd(iq0,jq0);
 540             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 541                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 542             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 543                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 544
 545             /* EWALD ELECTROSTATICS */
 546
 547             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 548             ewrt             = _mm_mul_pd(r00,ewtabscale);
 549             ewitab           = _mm_cvttpd_epi32(ewrt);
 550 #ifdef __XOP__
 551             eweps            = _mm_frcz_pd(ewrt);
 552 #else
 553             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 554 #endif
 555             twoeweps         = _mm_add_pd(eweps,eweps);
 556             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 557                                          &ewtabF,&ewtabFn);
 558             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 559             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 560
 561             /* Analytical LJ-PME */
 562             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 563             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 564             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 565             exponent         = avx128fma_exp_d(ewcljrsq);
 566             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 567             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 568             /* f6A = 6 * C6grid * (1 - poly) */
 569             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 570             /* f6B = C6grid * exponent * beta^6 */
 571             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 572             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 573             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 574
 575             fscal            = _mm_add_pd(felec,fvdw);
 576
 577             /* Update vectorial force */
 578             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 579             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 580             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 581
 582             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 583                                                    _mm_mul_pd(dx00,fscal),
 584                                                    _mm_mul_pd(dy00,fscal),
 585                                                    _mm_mul_pd(dz00,fscal));
 586
 587             /* Inner loop uses 60 flops */
 588         }
 589
 590         if(jidx<j_index_end)
 591         {
 592
 593             jnrA             = jjnr[jidx];
 594             j_coord_offsetA  = DIM*jnrA;
 595
 596             /* load j atom coordinates */
 597             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 598                                               &jx0,&jy0,&jz0);
 599
 600             /* Calculate displacement vector */
 601             dx00             = _mm_sub_pd(ix0,jx0);
 602             dy00             = _mm_sub_pd(iy0,jy0);
 603             dz00             = _mm_sub_pd(iz0,jz0);
 604
 605             /* Calculate squared distance and things based on it */
 606             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 607
 608             rinv00           = avx128fma_invsqrt_d(rsq00);
 609
 610             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 611
 612             /* Load parameters for j particles */
 613             jq0              = _mm_load_sd(charge+jnrA+0);
 614             vdwjidx0A        = 2*vdwtype[jnrA+0];
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             r00              = _mm_mul_pd(rsq00,rinv00);
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq00             = _mm_mul_pd(iq0,jq0);
 624             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 625             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 626
 627             /* EWALD ELECTROSTATICS */
 628
 629             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 630             ewrt             = _mm_mul_pd(r00,ewtabscale);
 631             ewitab           = _mm_cvttpd_epi32(ewrt);
 632 #ifdef __XOP__
 633             eweps            = _mm_frcz_pd(ewrt);
 634 #else
 635             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 636 #endif
 637             twoeweps         = _mm_add_pd(eweps,eweps);
 638             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 639             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 640             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 641
 642             /* Analytical LJ-PME */
 643             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 644             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 645             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 646             exponent         = avx128fma_exp_d(ewcljrsq);
 647             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 648             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 649             /* f6A = 6 * C6grid * (1 - poly) */
 650             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 651             /* f6B = C6grid * exponent * beta^6 */
 652             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 653             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 654             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 655
 656             fscal            = _mm_add_pd(felec,fvdw);
 657
 658             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 659
 660             /* Update vectorial force */
 661             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 662             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 663             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 664
 665             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 666                                                    _mm_mul_pd(dx00,fscal),
 667                                                    _mm_mul_pd(dy00,fscal),
 668                                                    _mm_mul_pd(dz00,fscal));
 669
 670             /* Inner loop uses 60 flops */
 671         }
 672
 673         /* End of innermost loop */
 674
 675         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 676                                               f+i_coord_offset,fshift+i_shift_offset);
 677
 678         /* Increment number of inner iterations */
 679         inneriter                  += j_index_end - j_index_start;
 680
 681         /* Outer loop uses 7 flops */
 682     }
 683
 684     /* Increment number of outer iterations */
 685     outeriter        += nri;
 686
 687     /* Update outer/inner flops */
 688
 689     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 690 }