src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJ_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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     int              nvdwtype;
  88     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  92     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  93     __m128i          ewitab;
  94     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  95     real             *ewtab;
  96     __m128d          dummy_mask,cutoff_mask;
  97     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  98     __m128d          one     = _mm_set1_pd(1.0);
  99     __m128d          two     = _mm_set1_pd(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_pd(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 118     ewtab            = fr->ic->tabq_coul_FDV0;
 119     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 120     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _mm_setzero_pd();
 148         fiy0             = _mm_setzero_pd();
 149         fiz0             = _mm_setzero_pd();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 153         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155         /* Reset potential sums */
 156         velecsum         = _mm_setzero_pd();
 157         vvdwsum          = _mm_setzero_pd();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _mm_sub_pd(ix0,jx0);
 175             dy00             = _mm_sub_pd(iy0,jy0);
 176             dz00             = _mm_sub_pd(iz0,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 180
 181             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 182
 183             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 184
 185             /* Load parameters for j particles */
 186             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 187             vdwjidx0A        = 2*vdwtype[jnrA+0];
 188             vdwjidx0B        = 2*vdwtype[jnrB+0];
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             r00              = _mm_mul_pd(rsq00,rinv00);
 195
 196             /* Compute parameters for interactions between i and j atoms */
 197             qq00             = _mm_mul_pd(iq0,jq0);
 198             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 199                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 200
 201             /* EWALD ELECTROSTATICS */
 202
 203             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 204             ewrt             = _mm_mul_pd(r00,ewtabscale);
 205             ewitab           = _mm_cvttpd_epi32(ewrt);
 206 #ifdef __XOP__
 207             eweps            = _mm_frcz_pd(ewrt);
 208 #else
 209             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 210 #endif
 211             twoeweps         = _mm_add_pd(eweps,eweps);
 212             ewitab           = _mm_slli_epi32(ewitab,2);
 213             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 214             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 215             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 216             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 217             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 218             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 219             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 220             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 221             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 222             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 223
 224             /* LENNARD-JONES DISPERSION/REPULSION */
 225
 226             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 227             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 228             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 229             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 230             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 231
 232             /* Update potential sum for this i atom from the interaction with this j atom. */
 233             velecsum         = _mm_add_pd(velecsum,velec);
 234             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 235
 236             fscal            = _mm_add_pd(felec,fvdw);
 237
 238             /* Update vectorial force */
 239             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 240             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 241             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 242
 243             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 244                                                    _mm_mul_pd(dx00,fscal),
 245                                                    _mm_mul_pd(dy00,fscal),
 246                                                    _mm_mul_pd(dz00,fscal));
 247
 248             /* Inner loop uses 56 flops */
 249         }
 250
 251         if(jidx<j_index_end)
 252         {
 253
 254             jnrA             = jjnr[jidx];
 255             j_coord_offsetA  = DIM*jnrA;
 256
 257             /* load j atom coordinates */
 258             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 259                                               &jx0,&jy0,&jz0);
 260
 261             /* Calculate displacement vector */
 262             dx00             = _mm_sub_pd(ix0,jx0);
 263             dy00             = _mm_sub_pd(iy0,jy0);
 264             dz00             = _mm_sub_pd(iz0,jz0);
 265
 266             /* Calculate squared distance and things based on it */
 267             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 268
 269             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 270
 271             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 272
 273             /* Load parameters for j particles */
 274             jq0              = _mm_load_sd(charge+jnrA+0);
 275             vdwjidx0A        = 2*vdwtype[jnrA+0];
 276
 277             /**************************
 278              * CALCULATE INTERACTIONS *
 279              **************************/
 280
 281             r00              = _mm_mul_pd(rsq00,rinv00);
 282
 283             /* Compute parameters for interactions between i and j atoms */
 284             qq00             = _mm_mul_pd(iq0,jq0);
 285             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 286
 287             /* EWALD ELECTROSTATICS */
 288
 289             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 290             ewrt             = _mm_mul_pd(r00,ewtabscale);
 291             ewitab           = _mm_cvttpd_epi32(ewrt);
 292 #ifdef __XOP__
 293             eweps            = _mm_frcz_pd(ewrt);
 294 #else
 295             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 296 #endif
 297             twoeweps         = _mm_add_pd(eweps,eweps);
 298             ewitab           = _mm_slli_epi32(ewitab,2);
 299             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 300             ewtabD           = _mm_setzero_pd();
 301             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 302             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 303             ewtabFn          = _mm_setzero_pd();
 304             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 305             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 306             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 307             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 308             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 309
 310             /* LENNARD-JONES DISPERSION/REPULSION */
 311
 312             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 313             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 314             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 315             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 316             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 320             velecsum         = _mm_add_pd(velecsum,velec);
 321             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 322             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 323
 324             fscal            = _mm_add_pd(felec,fvdw);
 325
 326             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 327
 328             /* Update vectorial force */
 329             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 330             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 331             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 332
 333             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 334                                                    _mm_mul_pd(dx00,fscal),
 335                                                    _mm_mul_pd(dy00,fscal),
 336                                                    _mm_mul_pd(dz00,fscal));
 337
 338             /* Inner loop uses 56 flops */
 339         }
 340
 341         /* End of innermost loop */
 342
 343         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 344                                               f+i_coord_offset,fshift+i_shift_offset);
 345
 346         ggid                        = gid[iidx];
 347         /* Update potential energies */
 348         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 349         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 350
 351         /* Increment number of inner iterations */
 352         inneriter                  += j_index_end - j_index_start;
 353
 354         /* Outer loop uses 9 flops */
 355     }
 356
 357     /* Increment number of outer iterations */
 358     outeriter        += nri;
 359
 360     /* Update outer/inner flops */
 361
 362     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*56);
 363 }
 364 /*
 365  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_128_fma_double
 366  * Electrostatics interaction: Ewald
 367  * VdW interaction:            LennardJones
 368  * Geometry:                   Particle-Particle
 369  * Calculate force/pot:        Force
 370  */
 371 void
 372 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_128_fma_double
 373                     (t_nblist                    * gmx_restrict       nlist,
 374                      rvec                        * gmx_restrict          xx,
 375                      rvec                        * gmx_restrict          ff,
 376                      t_forcerec                  * gmx_restrict          fr,
 377                      t_mdatoms                   * gmx_restrict     mdatoms,
 378                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 379                      t_nrnb                      * gmx_restrict        nrnb)
 380 {
 381     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 382      * just 0 for non-waters.
 383      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 384      * jnr indices corresponding to data put in the four positions in the SIMD register.
 385      */
 386     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 387     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 388     int              jnrA,jnrB;
 389     int              j_coord_offsetA,j_coord_offsetB;
 390     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 391     real             rcutoff_scalar;
 392     real             *shiftvec,*fshift,*x,*f;
 393     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 394     int              vdwioffset0;
 395     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 396     int              vdwjidx0A,vdwjidx0B;
 397     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 398     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 399     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 400     real             *charge;
 401     int              nvdwtype;
 402     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 403     int              *vdwtype;
 404     real             *vdwparam;
 405     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 406     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 407     __m128i          ewitab;
 408     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 409     real             *ewtab;
 410     __m128d          dummy_mask,cutoff_mask;
 411     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 412     __m128d          one     = _mm_set1_pd(1.0);
 413     __m128d          two     = _mm_set1_pd(2.0);
 414     x                = xx[0];
 415     f                = ff[0];
 416
 417     nri              = nlist->nri;
 418     iinr             = nlist->iinr;
 419     jindex           = nlist->jindex;
 420     jjnr             = nlist->jjnr;
 421     shiftidx         = nlist->shift;
 422     gid              = nlist->gid;
 423     shiftvec         = fr->shift_vec[0];
 424     fshift           = fr->fshift[0];
 425     facel            = _mm_set1_pd(fr->epsfac);
 426     charge           = mdatoms->chargeA;
 427     nvdwtype         = fr->ntype;
 428     vdwparam         = fr->nbfp;
 429     vdwtype          = mdatoms->typeA;
 430
 431     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 432     ewtab            = fr->ic->tabq_coul_F;
 433     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 434     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 435
 436     /* Avoid stupid compiler warnings */
 437     jnrA = jnrB = 0;
 438     j_coord_offsetA = 0;
 439     j_coord_offsetB = 0;
 440
 441     outeriter        = 0;
 442     inneriter        = 0;
 443
 444     /* Start outer loop over neighborlists */
 445     for(iidx=0; iidx<nri; iidx++)
 446     {
 447         /* Load shift vector for this list */
 448         i_shift_offset   = DIM*shiftidx[iidx];
 449
 450         /* Load limits for loop over neighbors */
 451         j_index_start    = jindex[iidx];
 452         j_index_end      = jindex[iidx+1];
 453
 454         /* Get outer coordinate index */
 455         inr              = iinr[iidx];
 456         i_coord_offset   = DIM*inr;
 457
 458         /* Load i particle coords and add shift vector */
 459         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 460
 461         fix0             = _mm_setzero_pd();
 462         fiy0             = _mm_setzero_pd();
 463         fiz0             = _mm_setzero_pd();
 464
 465         /* Load parameters for i particles */
 466         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 467         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 468
 469         /* Start inner kernel loop */
 470         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 471         {
 472
 473             /* Get j neighbor index, and coordinate index */
 474             jnrA             = jjnr[jidx];
 475             jnrB             = jjnr[jidx+1];
 476             j_coord_offsetA  = DIM*jnrA;
 477             j_coord_offsetB  = DIM*jnrB;
 478
 479             /* load j atom coordinates */
 480             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 481                                               &jx0,&jy0,&jz0);
 482
 483             /* Calculate displacement vector */
 484             dx00             = _mm_sub_pd(ix0,jx0);
 485             dy00             = _mm_sub_pd(iy0,jy0);
 486             dz00             = _mm_sub_pd(iz0,jz0);
 487
 488             /* Calculate squared distance and things based on it */
 489             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 490
 491             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 492
 493             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 494
 495             /* Load parameters for j particles */
 496             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 497             vdwjidx0A        = 2*vdwtype[jnrA+0];
 498             vdwjidx0B        = 2*vdwtype[jnrB+0];
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             r00              = _mm_mul_pd(rsq00,rinv00);
 505
 506             /* Compute parameters for interactions between i and j atoms */
 507             qq00             = _mm_mul_pd(iq0,jq0);
 508             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 509                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 510
 511             /* EWALD ELECTROSTATICS */
 512
 513             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 514             ewrt             = _mm_mul_pd(r00,ewtabscale);
 515             ewitab           = _mm_cvttpd_epi32(ewrt);
 516 #ifdef __XOP__
 517             eweps            = _mm_frcz_pd(ewrt);
 518 #else
 519             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 520 #endif
 521             twoeweps         = _mm_add_pd(eweps,eweps);
 522             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 523                                          &ewtabF,&ewtabFn);
 524             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 525             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 526
 527             /* LENNARD-JONES DISPERSION/REPULSION */
 528
 529             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 530             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 531
 532             fscal            = _mm_add_pd(felec,fvdw);
 533
 534             /* Update vectorial force */
 535             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 536             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 537             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 538
 539             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 540                                                    _mm_mul_pd(dx00,fscal),
 541                                                    _mm_mul_pd(dy00,fscal),
 542                                                    _mm_mul_pd(dz00,fscal));
 543
 544             /* Inner loop uses 46 flops */
 545         }
 546
 547         if(jidx<j_index_end)
 548         {
 549
 550             jnrA             = jjnr[jidx];
 551             j_coord_offsetA  = DIM*jnrA;
 552
 553             /* load j atom coordinates */
 554             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 555                                               &jx0,&jy0,&jz0);
 556
 557             /* Calculate displacement vector */
 558             dx00             = _mm_sub_pd(ix0,jx0);
 559             dy00             = _mm_sub_pd(iy0,jy0);
 560             dz00             = _mm_sub_pd(iz0,jz0);
 561
 562             /* Calculate squared distance and things based on it */
 563             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 564
 565             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 566
 567             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 568
 569             /* Load parameters for j particles */
 570             jq0              = _mm_load_sd(charge+jnrA+0);
 571             vdwjidx0A        = 2*vdwtype[jnrA+0];
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             r00              = _mm_mul_pd(rsq00,rinv00);
 578
 579             /* Compute parameters for interactions between i and j atoms */
 580             qq00             = _mm_mul_pd(iq0,jq0);
 581             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 582
 583             /* EWALD ELECTROSTATICS */
 584
 585             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 586             ewrt             = _mm_mul_pd(r00,ewtabscale);
 587             ewitab           = _mm_cvttpd_epi32(ewrt);
 588 #ifdef __XOP__
 589             eweps            = _mm_frcz_pd(ewrt);
 590 #else
 591             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 592 #endif
 593             twoeweps         = _mm_add_pd(eweps,eweps);
 594             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 595             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 596             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 597
 598             /* LENNARD-JONES DISPERSION/REPULSION */
 599
 600             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 601             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 602
 603             fscal            = _mm_add_pd(felec,fvdw);
 604
 605             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 606
 607             /* Update vectorial force */
 608             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 609             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 610             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 611
 612             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 613                                                    _mm_mul_pd(dx00,fscal),
 614                                                    _mm_mul_pd(dy00,fscal),
 615                                                    _mm_mul_pd(dz00,fscal));
 616
 617             /* Inner loop uses 46 flops */
 618         }
 619
 620         /* End of innermost loop */
 621
 622         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 623                                               f+i_coord_offset,fshift+i_shift_offset);
 624
 625         /* Increment number of inner iterations */
 626         inneriter                  += j_index_end - j_index_start;
 627
 628         /* Outer loop uses 7 flops */
 629     }
 630
 631     /* Increment number of outer iterations */
 632     outeriter        += nri;
 633
 634     /* Update outer/inner flops */
 635
 636     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*46);
 637 }