src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwLJSh_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_ElecEwSh_VdwLJSh_GeomP1P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJSh_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     __m128i          ewitab;
  93     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     real             *ewtab;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_pd(fr->ic->epsfac);
 111     charge           = mdatoms->chargeA;
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 117     ewtab            = fr->ic->tabq_coul_FDV0;
 118     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 119     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 120
 121     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 122     rcutoff_scalar   = fr->ic->rcoulomb;
 123     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 124     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 125
 126     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 127     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _mm_setzero_pd();
 155         fiy0             = _mm_setzero_pd();
 156         fiz0             = _mm_setzero_pd();
 157
 158         /* Load parameters for i particles */
 159         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 160         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164         vvdwsum          = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 187
 188             rinv00           = avx128fma_invsqrt_d(rsq00);
 189
 190             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 191
 192             /* Load parameters for j particles */
 193             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 194             vdwjidx0A        = 2*vdwtype[jnrA+0];
 195             vdwjidx0B        = 2*vdwtype[jnrB+0];
 196
 197             /**************************
 198              * CALCULATE INTERACTIONS *
 199              **************************/
 200
 201             if (gmx_mm_any_lt(rsq00,rcutoff2))
 202             {
 203
 204             r00              = _mm_mul_pd(rsq00,rinv00);
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq00             = _mm_mul_pd(iq0,jq0);
 208             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 209                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 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(_mm_sub_pd(rinv00,sh_ewald),velec));
 232             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 233
 234             /* LENNARD-JONES DISPERSION/REPULSION */
 235
 236             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 237             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 238             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 239             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 240                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 241             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 242
 243             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 244
 245             /* Update potential sum for this i atom from the interaction with this j atom. */
 246             velec            = _mm_and_pd(velec,cutoff_mask);
 247             velecsum         = _mm_add_pd(velecsum,velec);
 248             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 249             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 250
 251             fscal            = _mm_add_pd(felec,fvdw);
 252
 253             fscal            = _mm_and_pd(fscal,cutoff_mask);
 254
 255             /* Update vectorial force */
 256             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 257             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 258             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 259
 260             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 261                                                    _mm_mul_pd(dx00,fscal),
 262                                                    _mm_mul_pd(dy00,fscal),
 263                                                    _mm_mul_pd(dz00,fscal));
 264
 265             }
 266
 267             /* Inner loop uses 67 flops */
 268         }
 269
 270         if(jidx<j_index_end)
 271         {
 272
 273             jnrA             = jjnr[jidx];
 274             j_coord_offsetA  = DIM*jnrA;
 275
 276             /* load j atom coordinates */
 277             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 278                                               &jx0,&jy0,&jz0);
 279
 280             /* Calculate displacement vector */
 281             dx00             = _mm_sub_pd(ix0,jx0);
 282             dy00             = _mm_sub_pd(iy0,jy0);
 283             dz00             = _mm_sub_pd(iz0,jz0);
 284
 285             /* Calculate squared distance and things based on it */
 286             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 287
 288             rinv00           = avx128fma_invsqrt_d(rsq00);
 289
 290             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 291
 292             /* Load parameters for j particles */
 293             jq0              = _mm_load_sd(charge+jnrA+0);
 294             vdwjidx0A        = 2*vdwtype[jnrA+0];
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             if (gmx_mm_any_lt(rsq00,rcutoff2))
 301             {
 302
 303             r00              = _mm_mul_pd(rsq00,rinv00);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq00             = _mm_mul_pd(iq0,jq0);
 307             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 308
 309             /* EWALD ELECTROSTATICS */
 310
 311             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 312             ewrt             = _mm_mul_pd(r00,ewtabscale);
 313             ewitab           = _mm_cvttpd_epi32(ewrt);
 314 #ifdef __XOP__
 315             eweps            = _mm_frcz_pd(ewrt);
 316 #else
 317             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 318 #endif
 319             twoeweps         = _mm_add_pd(eweps,eweps);
 320             ewitab           = _mm_slli_epi32(ewitab,2);
 321             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 322             ewtabD           = _mm_setzero_pd();
 323             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 324             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 325             ewtabFn          = _mm_setzero_pd();
 326             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 327             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 328             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 329             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 330             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 331
 332             /* LENNARD-JONES DISPERSION/REPULSION */
 333
 334             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 335             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 336             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 337             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 338                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 339             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 340
 341             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velec            = _mm_and_pd(velec,cutoff_mask);
 345             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 346             velecsum         = _mm_add_pd(velecsum,velec);
 347             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 348             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 349             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 350
 351             fscal            = _mm_add_pd(felec,fvdw);
 352
 353             fscal            = _mm_and_pd(fscal,cutoff_mask);
 354
 355             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 356
 357             /* Update vectorial force */
 358             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 359             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 360             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 361
 362             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 363                                                    _mm_mul_pd(dx00,fscal),
 364                                                    _mm_mul_pd(dy00,fscal),
 365                                                    _mm_mul_pd(dz00,fscal));
 366
 367             }
 368
 369             /* Inner loop uses 67 flops */
 370         }
 371
 372         /* End of innermost loop */
 373
 374         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 375                                               f+i_coord_offset,fshift+i_shift_offset);
 376
 377         ggid                        = gid[iidx];
 378         /* Update potential energies */
 379         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 380         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 381
 382         /* Increment number of inner iterations */
 383         inneriter                  += j_index_end - j_index_start;
 384
 385         /* Outer loop uses 9 flops */
 386     }
 387
 388     /* Increment number of outer iterations */
 389     outeriter        += nri;
 390
 391     /* Update outer/inner flops */
 392
 393     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*67);
 394 }
 395 /*
 396  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 397  * Electrostatics interaction: Ewald
 398  * VdW interaction:            LennardJones
 399  * Geometry:                   Particle-Particle
 400  * Calculate force/pot:        Force
 401  */
 402 void
 403 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 404                     (t_nblist                    * gmx_restrict       nlist,
 405                      rvec                        * gmx_restrict          xx,
 406                      rvec                        * gmx_restrict          ff,
 407                      struct t_forcerec           * gmx_restrict          fr,
 408                      t_mdatoms                   * gmx_restrict     mdatoms,
 409                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 410                      t_nrnb                      * gmx_restrict        nrnb)
 411 {
 412     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 413      * just 0 for non-waters.
 414      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 415      * jnr indices corresponding to data put in the four positions in the SIMD register.
 416      */
 417     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 418     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 419     int              jnrA,jnrB;
 420     int              j_coord_offsetA,j_coord_offsetB;
 421     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 422     real             rcutoff_scalar;
 423     real             *shiftvec,*fshift,*x,*f;
 424     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 425     int              vdwioffset0;
 426     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 427     int              vdwjidx0A,vdwjidx0B;
 428     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 429     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 430     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 431     real             *charge;
 432     int              nvdwtype;
 433     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 434     int              *vdwtype;
 435     real             *vdwparam;
 436     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 437     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 438     __m128i          ewitab;
 439     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 440     real             *ewtab;
 441     __m128d          dummy_mask,cutoff_mask;
 442     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 443     __m128d          one     = _mm_set1_pd(1.0);
 444     __m128d          two     = _mm_set1_pd(2.0);
 445     x                = xx[0];
 446     f                = ff[0];
 447
 448     nri              = nlist->nri;
 449     iinr             = nlist->iinr;
 450     jindex           = nlist->jindex;
 451     jjnr             = nlist->jjnr;
 452     shiftidx         = nlist->shift;
 453     gid              = nlist->gid;
 454     shiftvec         = fr->shift_vec[0];
 455     fshift           = fr->fshift[0];
 456     facel            = _mm_set1_pd(fr->ic->epsfac);
 457     charge           = mdatoms->chargeA;
 458     nvdwtype         = fr->ntype;
 459     vdwparam         = fr->nbfp;
 460     vdwtype          = mdatoms->typeA;
 461
 462     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 463     ewtab            = fr->ic->tabq_coul_F;
 464     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 465     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 466
 467     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 468     rcutoff_scalar   = fr->ic->rcoulomb;
 469     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 470     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 471
 472     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 473     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 474
 475     /* Avoid stupid compiler warnings */
 476     jnrA = jnrB = 0;
 477     j_coord_offsetA = 0;
 478     j_coord_offsetB = 0;
 479
 480     outeriter        = 0;
 481     inneriter        = 0;
 482
 483     /* Start outer loop over neighborlists */
 484     for(iidx=0; iidx<nri; iidx++)
 485     {
 486         /* Load shift vector for this list */
 487         i_shift_offset   = DIM*shiftidx[iidx];
 488
 489         /* Load limits for loop over neighbors */
 490         j_index_start    = jindex[iidx];
 491         j_index_end      = jindex[iidx+1];
 492
 493         /* Get outer coordinate index */
 494         inr              = iinr[iidx];
 495         i_coord_offset   = DIM*inr;
 496
 497         /* Load i particle coords and add shift vector */
 498         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 499
 500         fix0             = _mm_setzero_pd();
 501         fiy0             = _mm_setzero_pd();
 502         fiz0             = _mm_setzero_pd();
 503
 504         /* Load parameters for i particles */
 505         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 506         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 507
 508         /* Start inner kernel loop */
 509         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 510         {
 511
 512             /* Get j neighbor index, and coordinate index */
 513             jnrA             = jjnr[jidx];
 514             jnrB             = jjnr[jidx+1];
 515             j_coord_offsetA  = DIM*jnrA;
 516             j_coord_offsetB  = DIM*jnrB;
 517
 518             /* load j atom coordinates */
 519             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 520                                               &jx0,&jy0,&jz0);
 521
 522             /* Calculate displacement vector */
 523             dx00             = _mm_sub_pd(ix0,jx0);
 524             dy00             = _mm_sub_pd(iy0,jy0);
 525             dz00             = _mm_sub_pd(iz0,jz0);
 526
 527             /* Calculate squared distance and things based on it */
 528             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 529
 530             rinv00           = avx128fma_invsqrt_d(rsq00);
 531
 532             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 533
 534             /* Load parameters for j particles */
 535             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 536             vdwjidx0A        = 2*vdwtype[jnrA+0];
 537             vdwjidx0B        = 2*vdwtype[jnrB+0];
 538
 539             /**************************
 540              * CALCULATE INTERACTIONS *
 541              **************************/
 542
 543             if (gmx_mm_any_lt(rsq00,rcutoff2))
 544             {
 545
 546             r00              = _mm_mul_pd(rsq00,rinv00);
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq00             = _mm_mul_pd(iq0,jq0);
 550             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 551                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 552
 553             /* EWALD ELECTROSTATICS */
 554
 555             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 556             ewrt             = _mm_mul_pd(r00,ewtabscale);
 557             ewitab           = _mm_cvttpd_epi32(ewrt);
 558 #ifdef __XOP__
 559             eweps            = _mm_frcz_pd(ewrt);
 560 #else
 561             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 562 #endif
 563             twoeweps         = _mm_add_pd(eweps,eweps);
 564             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 565                                          &ewtabF,&ewtabFn);
 566             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 567             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 568
 569             /* LENNARD-JONES DISPERSION/REPULSION */
 570
 571             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 572             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 573
 574             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 575
 576             fscal            = _mm_add_pd(felec,fvdw);
 577
 578             fscal            = _mm_and_pd(fscal,cutoff_mask);
 579
 580             /* Update vectorial force */
 581             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 582             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 583             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 584
 585             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 586                                                    _mm_mul_pd(dx00,fscal),
 587                                                    _mm_mul_pd(dy00,fscal),
 588                                                    _mm_mul_pd(dz00,fscal));
 589
 590             }
 591
 592             /* Inner loop uses 49 flops */
 593         }
 594
 595         if(jidx<j_index_end)
 596         {
 597
 598             jnrA             = jjnr[jidx];
 599             j_coord_offsetA  = DIM*jnrA;
 600
 601             /* load j atom coordinates */
 602             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 603                                               &jx0,&jy0,&jz0);
 604
 605             /* Calculate displacement vector */
 606             dx00             = _mm_sub_pd(ix0,jx0);
 607             dy00             = _mm_sub_pd(iy0,jy0);
 608             dz00             = _mm_sub_pd(iz0,jz0);
 609
 610             /* Calculate squared distance and things based on it */
 611             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 612
 613             rinv00           = avx128fma_invsqrt_d(rsq00);
 614
 615             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 616
 617             /* Load parameters for j particles */
 618             jq0              = _mm_load_sd(charge+jnrA+0);
 619             vdwjidx0A        = 2*vdwtype[jnrA+0];
 620
 621             /**************************
 622              * CALCULATE INTERACTIONS *
 623              **************************/
 624
 625             if (gmx_mm_any_lt(rsq00,rcutoff2))
 626             {
 627
 628             r00              = _mm_mul_pd(rsq00,rinv00);
 629
 630             /* Compute parameters for interactions between i and j atoms */
 631             qq00             = _mm_mul_pd(iq0,jq0);
 632             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 633
 634             /* EWALD ELECTROSTATICS */
 635
 636             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 637             ewrt             = _mm_mul_pd(r00,ewtabscale);
 638             ewitab           = _mm_cvttpd_epi32(ewrt);
 639 #ifdef __XOP__
 640             eweps            = _mm_frcz_pd(ewrt);
 641 #else
 642             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 643 #endif
 644             twoeweps         = _mm_add_pd(eweps,eweps);
 645             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 646             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 647             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 648
 649             /* LENNARD-JONES DISPERSION/REPULSION */
 650
 651             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 652             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 653
 654             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 655
 656             fscal            = _mm_add_pd(felec,fvdw);
 657
 658             fscal            = _mm_and_pd(fscal,cutoff_mask);
 659
 660             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 661
 662             /* Update vectorial force */
 663             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 664             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 665             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 666
 667             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 668                                                    _mm_mul_pd(dx00,fscal),
 669                                                    _mm_mul_pd(dy00,fscal),
 670                                                    _mm_mul_pd(dz00,fscal));
 671
 672             }
 673
 674             /* Inner loop uses 49 flops */
 675         }
 676
 677         /* End of innermost loop */
 678
 679         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 680                                               f+i_coord_offset,fshift+i_shift_offset);
 681
 682         /* Increment number of inner iterations */
 683         inneriter                  += j_index_end - j_index_start;
 684
 685         /* Outer loop uses 7 flops */
 686     }
 687
 688     /* Increment number of outer iterations */
 689     outeriter        += nri;
 690
 691     /* Update outer/inner flops */
 692
 693     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*49);
 694 }