src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJ_GeomP1P1_avx_128_fma_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_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_VdwLJ_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_ElecEw_VdwLJ_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     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 145
 146         fix0             = _mm_setzero_pd();
 147         fiy0             = _mm_setzero_pd();
 148         fiz0             = _mm_setzero_pd();
 149
 150         /* Load parameters for i particles */
 151         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 152         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm_setzero_pd();
 156         vvdwsum          = _mm_setzero_pd();
 157
 158         /* Start inner kernel loop */
 159         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 160         {
 161
 162             /* Get j neighbor index, and coordinate index */
 163             jnrA             = jjnr[jidx];
 164             jnrB             = jjnr[jidx+1];
 165             j_coord_offsetA  = DIM*jnrA;
 166             j_coord_offsetB  = DIM*jnrB;
 167
 168             /* load j atom coordinates */
 169             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 170                                               &jx0,&jy0,&jz0);
 171
 172             /* Calculate displacement vector */
 173             dx00             = _mm_sub_pd(ix0,jx0);
 174             dy00             = _mm_sub_pd(iy0,jy0);
 175             dz00             = _mm_sub_pd(iz0,jz0);
 176
 177             /* Calculate squared distance and things based on it */
 178             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 179
 180             rinv00           = avx128fma_invsqrt_d(rsq00);
 181
 182             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 183
 184             /* Load parameters for j particles */
 185             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 186             vdwjidx0A        = 2*vdwtype[jnrA+0];
 187             vdwjidx0B        = 2*vdwtype[jnrB+0];
 188
 189             /**************************
 190              * CALCULATE INTERACTIONS *
 191              **************************/
 192
 193             r00              = _mm_mul_pd(rsq00,rinv00);
 194
 195             /* Compute parameters for interactions between i and j atoms */
 196             qq00             = _mm_mul_pd(iq0,jq0);
 197             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 198                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 199
 200             /* EWALD ELECTROSTATICS */
 201
 202             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 203             ewrt             = _mm_mul_pd(r00,ewtabscale);
 204             ewitab           = _mm_cvttpd_epi32(ewrt);
 205 #ifdef __XOP__
 206             eweps            = _mm_frcz_pd(ewrt);
 207 #else
 208             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 209 #endif
 210             twoeweps         = _mm_add_pd(eweps,eweps);
 211             ewitab           = _mm_slli_epi32(ewitab,2);
 212             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 213             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 214             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 215             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 216             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 217             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 218             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 219             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 220             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 221             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 222
 223             /* LENNARD-JONES DISPERSION/REPULSION */
 224
 225             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 226             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 227             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 228             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 229             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 230
 231             /* Update potential sum for this i atom from the interaction with this j atom. */
 232             velecsum         = _mm_add_pd(velecsum,velec);
 233             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 234
 235             fscal            = _mm_add_pd(felec,fvdw);
 236
 237             /* Update vectorial force */
 238             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 239             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 240             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 241
 242             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 243                                                    _mm_mul_pd(dx00,fscal),
 244                                                    _mm_mul_pd(dy00,fscal),
 245                                                    _mm_mul_pd(dz00,fscal));
 246
 247             /* Inner loop uses 56 flops */
 248         }
 249
 250         if(jidx<j_index_end)
 251         {
 252
 253             jnrA             = jjnr[jidx];
 254             j_coord_offsetA  = DIM*jnrA;
 255
 256             /* load j atom coordinates */
 257             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 258                                               &jx0,&jy0,&jz0);
 259
 260             /* Calculate displacement vector */
 261             dx00             = _mm_sub_pd(ix0,jx0);
 262             dy00             = _mm_sub_pd(iy0,jy0);
 263             dz00             = _mm_sub_pd(iz0,jz0);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 267
 268             rinv00           = avx128fma_invsqrt_d(rsq00);
 269
 270             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 271
 272             /* Load parameters for j particles */
 273             jq0              = _mm_load_sd(charge+jnrA+0);
 274             vdwjidx0A        = 2*vdwtype[jnrA+0];
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             r00              = _mm_mul_pd(rsq00,rinv00);
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             qq00             = _mm_mul_pd(iq0,jq0);
 284             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 285
 286             /* EWALD ELECTROSTATICS */
 287
 288             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 289             ewrt             = _mm_mul_pd(r00,ewtabscale);
 290             ewitab           = _mm_cvttpd_epi32(ewrt);
 291 #ifdef __XOP__
 292             eweps            = _mm_frcz_pd(ewrt);
 293 #else
 294             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 295 #endif
 296             twoeweps         = _mm_add_pd(eweps,eweps);
 297             ewitab           = _mm_slli_epi32(ewitab,2);
 298             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 299             ewtabD           = _mm_setzero_pd();
 300             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 301             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 302             ewtabFn          = _mm_setzero_pd();
 303             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 304             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 305             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 306             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 307             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 308
 309             /* LENNARD-JONES DISPERSION/REPULSION */
 310
 311             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 312             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 313             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 314             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 315             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 319             velecsum         = _mm_add_pd(velecsum,velec);
 320             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 321             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 322
 323             fscal            = _mm_add_pd(felec,fvdw);
 324
 325             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 326
 327             /* Update vectorial force */
 328             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 329             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 330             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 331
 332             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 333                                                    _mm_mul_pd(dx00,fscal),
 334                                                    _mm_mul_pd(dy00,fscal),
 335                                                    _mm_mul_pd(dz00,fscal));
 336
 337             /* Inner loop uses 56 flops */
 338         }
 339
 340         /* End of innermost loop */
 341
 342         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 343                                               f+i_coord_offset,fshift+i_shift_offset);
 344
 345         ggid                        = gid[iidx];
 346         /* Update potential energies */
 347         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 348         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 349
 350         /* Increment number of inner iterations */
 351         inneriter                  += j_index_end - j_index_start;
 352
 353         /* Outer loop uses 9 flops */
 354     }
 355
 356     /* Increment number of outer iterations */
 357     outeriter        += nri;
 358
 359     /* Update outer/inner flops */
 360
 361     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*56);
 362 }
 363 /*
 364  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_128_fma_double
 365  * Electrostatics interaction: Ewald
 366  * VdW interaction:            LennardJones
 367  * Geometry:                   Particle-Particle
 368  * Calculate force/pot:        Force
 369  */
 370 void
 371 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_128_fma_double
 372                     (t_nblist                    * gmx_restrict       nlist,
 373                      rvec                        * gmx_restrict          xx,
 374                      rvec                        * gmx_restrict          ff,
 375                      struct t_forcerec           * gmx_restrict          fr,
 376                      t_mdatoms                   * gmx_restrict     mdatoms,
 377                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 378                      t_nrnb                      * gmx_restrict        nrnb)
 379 {
 380     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 381      * just 0 for non-waters.
 382      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 383      * jnr indices corresponding to data put in the four positions in the SIMD register.
 384      */
 385     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 386     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 387     int              jnrA,jnrB;
 388     int              j_coord_offsetA,j_coord_offsetB;
 389     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 390     real             rcutoff_scalar;
 391     real             *shiftvec,*fshift,*x,*f;
 392     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 393     int              vdwioffset0;
 394     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 395     int              vdwjidx0A,vdwjidx0B;
 396     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 397     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 398     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 399     real             *charge;
 400     int              nvdwtype;
 401     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 402     int              *vdwtype;
 403     real             *vdwparam;
 404     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 405     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 406     __m128i          ewitab;
 407     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 408     real             *ewtab;
 409     __m128d          dummy_mask,cutoff_mask;
 410     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 411     __m128d          one     = _mm_set1_pd(1.0);
 412     __m128d          two     = _mm_set1_pd(2.0);
 413     x                = xx[0];
 414     f                = ff[0];
 415
 416     nri              = nlist->nri;
 417     iinr             = nlist->iinr;
 418     jindex           = nlist->jindex;
 419     jjnr             = nlist->jjnr;
 420     shiftidx         = nlist->shift;
 421     gid              = nlist->gid;
 422     shiftvec         = fr->shift_vec[0];
 423     fshift           = fr->fshift[0];
 424     facel            = _mm_set1_pd(fr->ic->epsfac);
 425     charge           = mdatoms->chargeA;
 426     nvdwtype         = fr->ntype;
 427     vdwparam         = fr->nbfp;
 428     vdwtype          = mdatoms->typeA;
 429
 430     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 431     ewtab            = fr->ic->tabq_coul_F;
 432     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 433     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 434
 435     /* Avoid stupid compiler warnings */
 436     jnrA = jnrB = 0;
 437     j_coord_offsetA = 0;
 438     j_coord_offsetB = 0;
 439
 440     outeriter        = 0;
 441     inneriter        = 0;
 442
 443     /* Start outer loop over neighborlists */
 444     for(iidx=0; iidx<nri; iidx++)
 445     {
 446         /* Load shift vector for this list */
 447         i_shift_offset   = DIM*shiftidx[iidx];
 448
 449         /* Load limits for loop over neighbors */
 450         j_index_start    = jindex[iidx];
 451         j_index_end      = jindex[iidx+1];
 452
 453         /* Get outer coordinate index */
 454         inr              = iinr[iidx];
 455         i_coord_offset   = DIM*inr;
 456
 457         /* Load i particle coords and add shift vector */
 458         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 459
 460         fix0             = _mm_setzero_pd();
 461         fiy0             = _mm_setzero_pd();
 462         fiz0             = _mm_setzero_pd();
 463
 464         /* Load parameters for i particles */
 465         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 466         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 467
 468         /* Start inner kernel loop */
 469         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 470         {
 471
 472             /* Get j neighbor index, and coordinate index */
 473             jnrA             = jjnr[jidx];
 474             jnrB             = jjnr[jidx+1];
 475             j_coord_offsetA  = DIM*jnrA;
 476             j_coord_offsetB  = DIM*jnrB;
 477
 478             /* load j atom coordinates */
 479             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 480                                               &jx0,&jy0,&jz0);
 481
 482             /* Calculate displacement vector */
 483             dx00             = _mm_sub_pd(ix0,jx0);
 484             dy00             = _mm_sub_pd(iy0,jy0);
 485             dz00             = _mm_sub_pd(iz0,jz0);
 486
 487             /* Calculate squared distance and things based on it */
 488             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 489
 490             rinv00           = avx128fma_invsqrt_d(rsq00);
 491
 492             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 493
 494             /* Load parameters for j particles */
 495             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 496             vdwjidx0A        = 2*vdwtype[jnrA+0];
 497             vdwjidx0B        = 2*vdwtype[jnrB+0];
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             r00              = _mm_mul_pd(rsq00,rinv00);
 504
 505             /* Compute parameters for interactions between i and j atoms */
 506             qq00             = _mm_mul_pd(iq0,jq0);
 507             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 508                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 509
 510             /* EWALD ELECTROSTATICS */
 511
 512             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 513             ewrt             = _mm_mul_pd(r00,ewtabscale);
 514             ewitab           = _mm_cvttpd_epi32(ewrt);
 515 #ifdef __XOP__
 516             eweps            = _mm_frcz_pd(ewrt);
 517 #else
 518             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 519 #endif
 520             twoeweps         = _mm_add_pd(eweps,eweps);
 521             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 522                                          &ewtabF,&ewtabFn);
 523             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 524             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 525
 526             /* LENNARD-JONES DISPERSION/REPULSION */
 527
 528             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 529             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 530
 531             fscal            = _mm_add_pd(felec,fvdw);
 532
 533             /* Update vectorial force */
 534             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 535             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 536             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 537
 538             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 539                                                    _mm_mul_pd(dx00,fscal),
 540                                                    _mm_mul_pd(dy00,fscal),
 541                                                    _mm_mul_pd(dz00,fscal));
 542
 543             /* Inner loop uses 46 flops */
 544         }
 545
 546         if(jidx<j_index_end)
 547         {
 548
 549             jnrA             = jjnr[jidx];
 550             j_coord_offsetA  = DIM*jnrA;
 551
 552             /* load j atom coordinates */
 553             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 554                                               &jx0,&jy0,&jz0);
 555
 556             /* Calculate displacement vector */
 557             dx00             = _mm_sub_pd(ix0,jx0);
 558             dy00             = _mm_sub_pd(iy0,jy0);
 559             dz00             = _mm_sub_pd(iz0,jz0);
 560
 561             /* Calculate squared distance and things based on it */
 562             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 563
 564             rinv00           = avx128fma_invsqrt_d(rsq00);
 565
 566             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 567
 568             /* Load parameters for j particles */
 569             jq0              = _mm_load_sd(charge+jnrA+0);
 570             vdwjidx0A        = 2*vdwtype[jnrA+0];
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             r00              = _mm_mul_pd(rsq00,rinv00);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             qq00             = _mm_mul_pd(iq0,jq0);
 580             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 581
 582             /* EWALD ELECTROSTATICS */
 583
 584             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 585             ewrt             = _mm_mul_pd(r00,ewtabscale);
 586             ewitab           = _mm_cvttpd_epi32(ewrt);
 587 #ifdef __XOP__
 588             eweps            = _mm_frcz_pd(ewrt);
 589 #else
 590             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 591 #endif
 592             twoeweps         = _mm_add_pd(eweps,eweps);
 593             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 594             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 595             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 596
 597             /* LENNARD-JONES DISPERSION/REPULSION */
 598
 599             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 600             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 601
 602             fscal            = _mm_add_pd(felec,fvdw);
 603
 604             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 605
 606             /* Update vectorial force */
 607             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 608             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 609             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 610
 611             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 612                                                    _mm_mul_pd(dx00,fscal),
 613                                                    _mm_mul_pd(dy00,fscal),
 614                                                    _mm_mul_pd(dz00,fscal));
 615
 616             /* Inner loop uses 46 flops */
 617         }
 618
 619         /* End of innermost loop */
 620
 621         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 622                                               f+i_coord_offset,fshift+i_shift_offset);
 623
 624         /* Increment number of inner iterations */
 625         inneriter                  += j_index_end - j_index_start;
 626
 627         /* Outer loop uses 7 flops */
 628     }
 629
 630     /* Increment number of outer iterations */
 631     outeriter        += nri;
 632
 633     /* Update outer/inner flops */
 634
 635     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*46);
 636 }