src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecGB_VdwNone_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: GeneralizedBorn
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecGB_VdwNone_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     __m128i          gbitab;
  88     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
  89     __m128d          minushalf = _mm_set1_pd(-0.5);
  90     real             *invsqrta,*dvda,*gbtab;
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  94     real             *vftab;
  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->epsfac);
 111     charge           = mdatoms->chargeA;
 112
 113     invsqrta         = fr->invsqrta;
 114     dvda             = fr->dvda;
 115     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 116     gbtab            = fr->gbtab.data;
 117     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     /* Start outer loop over neighborlists */
 128     for(iidx=0; iidx<nri; iidx++)
 129     {
 130         /* Load shift vector for this list */
 131         i_shift_offset   = DIM*shiftidx[iidx];
 132
 133         /* Load limits for loop over neighbors */
 134         j_index_start    = jindex[iidx];
 135         j_index_end      = jindex[iidx+1];
 136
 137         /* Get outer coordinate index */
 138         inr              = iinr[iidx];
 139         i_coord_offset   = DIM*inr;
 140
 141         /* Load i particle coords and add shift vector */
 142         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 143
 144         fix0             = _mm_setzero_pd();
 145         fiy0             = _mm_setzero_pd();
 146         fiz0             = _mm_setzero_pd();
 147
 148         /* Load parameters for i particles */
 149         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 150         isai0            = _mm_load1_pd(invsqrta+inr+0);
 151
 152         /* Reset potential sums */
 153         velecsum         = _mm_setzero_pd();
 154         vgbsum           = _mm_setzero_pd();
 155         dvdasum          = _mm_setzero_pd();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166
 167             /* load j atom coordinates */
 168             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _mm_sub_pd(ix0,jx0);
 173             dy00             = _mm_sub_pd(iy0,jy0);
 174             dz00             = _mm_sub_pd(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 178
 179             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 180
 181             /* Load parameters for j particles */
 182             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 183             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 184
 185             /**************************
 186              * CALCULATE INTERACTIONS *
 187              **************************/
 188
 189             r00              = _mm_mul_pd(rsq00,rinv00);
 190
 191             /* Compute parameters for interactions between i and j atoms */
 192             qq00             = _mm_mul_pd(iq0,jq0);
 193
 194             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 195             isaprod          = _mm_mul_pd(isai0,isaj0);
 196             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 197             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 198
 199             /* Calculate generalized born table index - this is a separate table from the normal one,
 200              * but we use the same procedure by multiplying r with scale and truncating to integer.
 201              */
 202             rt               = _mm_mul_pd(r00,gbscale);
 203             gbitab           = _mm_cvttpd_epi32(rt);
 204 #ifdef __XOP__
 205             gbeps            = _mm_frcz_pd(rt);
 206 #else
 207             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 208 #endif
 209             gbitab           = _mm_slli_epi32(gbitab,2);
 210
 211             Y                = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,0) );
 212             F                = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,1) );
 213             GMX_MM_TRANSPOSE2_PD(Y,F);
 214             G                = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,0) +2);
 215             H                = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,1) +2);
 216             GMX_MM_TRANSPOSE2_PD(G,H);
 217             Fp               = _mm_macc_pd(gbeps,_mm_macc_pd(gbeps,H,G),F);
 218             VV               = _mm_macc_pd(gbeps,Fp,Y);
 219             vgb              = _mm_mul_pd(gbqqfactor,VV);
 220
 221             twogbeps         = _mm_add_pd(gbeps,gbeps);
 222             FF               = _mm_macc_pd(_mm_macc_pd(twogbeps,H,G),gbeps,Fp);
 223             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 224             dvdatmp          = _mm_mul_pd(minushalf,_mm_macc_pd(fgb,r00,vgb));
 225             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 226             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 227             velec            = _mm_mul_pd(qq00,rinv00);
 228             felec            = _mm_mul_pd(_mm_msub_pd(velec,rinv00,fgb),rinv00);
 229
 230             /* Update potential sum for this i atom from the interaction with this j atom. */
 231             velecsum         = _mm_add_pd(velecsum,velec);
 232             vgbsum           = _mm_add_pd(vgbsum,vgb);
 233
 234             fscal            = felec;
 235
 236             /* Update vectorial force */
 237             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 238             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 239             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 240
 241             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 242                                                    _mm_mul_pd(dx00,fscal),
 243                                                    _mm_mul_pd(dy00,fscal),
 244                                                    _mm_mul_pd(dz00,fscal));
 245
 246             /* Inner loop uses 61 flops */
 247         }
 248
 249         if(jidx<j_index_end)
 250         {
 251
 252             jnrA             = jjnr[jidx];
 253             j_coord_offsetA  = DIM*jnrA;
 254
 255             /* load j atom coordinates */
 256             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 257                                               &jx0,&jy0,&jz0);
 258
 259             /* Calculate displacement vector */
 260             dx00             = _mm_sub_pd(ix0,jx0);
 261             dy00             = _mm_sub_pd(iy0,jy0);
 262             dz00             = _mm_sub_pd(iz0,jz0);
 263
 264             /* Calculate squared distance and things based on it */
 265             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 266
 267             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 268
 269             /* Load parameters for j particles */
 270             jq0              = _mm_load_sd(charge+jnrA+0);
 271             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             r00              = _mm_mul_pd(rsq00,rinv00);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq00             = _mm_mul_pd(iq0,jq0);
 281
 282             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 283             isaprod          = _mm_mul_pd(isai0,isaj0);
 284             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 285             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 286
 287             /* Calculate generalized born table index - this is a separate table from the normal one,
 288              * but we use the same procedure by multiplying r with scale and truncating to integer.
 289              */
 290             rt               = _mm_mul_pd(r00,gbscale);
 291             gbitab           = _mm_cvttpd_epi32(rt);
 292 #ifdef __XOP__
 293             gbeps            = _mm_frcz_pd(rt);
 294 #else
 295             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 296 #endif
 297             gbitab           = _mm_slli_epi32(gbitab,2);
 298
 299             Y                = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,0) );
 300             F                = _mm_setzero_pd();
 301             GMX_MM_TRANSPOSE2_PD(Y,F);
 302             G                = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,0) +2);
 303             H                = _mm_setzero_pd();
 304             GMX_MM_TRANSPOSE2_PD(G,H);
 305             Fp               = _mm_macc_pd(gbeps,_mm_macc_pd(gbeps,H,G),F);
 306             VV               = _mm_macc_pd(gbeps,Fp,Y);
 307             vgb              = _mm_mul_pd(gbqqfactor,VV);
 308
 309             twogbeps         = _mm_add_pd(gbeps,gbeps);
 310             FF               = _mm_macc_pd(_mm_macc_pd(twogbeps,H,G),gbeps,Fp);
 311             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 312             dvdatmp          = _mm_mul_pd(minushalf,_mm_macc_pd(fgb,r00,vgb));
 313             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 314             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 315             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 316             velec            = _mm_mul_pd(qq00,rinv00);
 317             felec            = _mm_mul_pd(_mm_msub_pd(velec,rinv00,fgb),rinv00);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 321             velecsum         = _mm_add_pd(velecsum,velec);
 322             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 323             vgbsum           = _mm_add_pd(vgbsum,vgb);
 324
 325             fscal            = felec;
 326
 327             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 328
 329             /* Update vectorial force */
 330             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 331             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 332             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 333
 334             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 335                                                    _mm_mul_pd(dx00,fscal),
 336                                                    _mm_mul_pd(dy00,fscal),
 337                                                    _mm_mul_pd(dz00,fscal));
 338
 339             /* Inner loop uses 61 flops */
 340         }
 341
 342         /* End of innermost loop */
 343
 344         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 345                                               f+i_coord_offset,fshift+i_shift_offset);
 346
 347         ggid                        = gid[iidx];
 348         /* Update potential energies */
 349         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 350         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 351         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 352         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 353
 354         /* Increment number of inner iterations */
 355         inneriter                  += j_index_end - j_index_start;
 356
 357         /* Outer loop uses 9 flops */
 358     }
 359
 360     /* Increment number of outer iterations */
 361     outeriter        += nri;
 362
 363     /* Update outer/inner flops */
 364
 365     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*9 + inneriter*61);
 366 }
 367 /*
 368  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_F_avx_128_fma_double
 369  * Electrostatics interaction: GeneralizedBorn
 370  * VdW interaction:            None
 371  * Geometry:                   Particle-Particle
 372  * Calculate force/pot:        Force
 373  */
 374 void
 375 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_avx_128_fma_double
 376                     (t_nblist                    * gmx_restrict       nlist,
 377                      rvec                        * gmx_restrict          xx,
 378                      rvec                        * gmx_restrict          ff,
 379                      t_forcerec                  * gmx_restrict          fr,
 380                      t_mdatoms                   * gmx_restrict     mdatoms,
 381                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 382                      t_nrnb                      * gmx_restrict        nrnb)
 383 {
 384     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 385      * just 0 for non-waters.
 386      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 387      * jnr indices corresponding to data put in the four positions in the SIMD register.
 388      */
 389     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 390     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 391     int              jnrA,jnrB;
 392     int              j_coord_offsetA,j_coord_offsetB;
 393     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 394     real             rcutoff_scalar;
 395     real             *shiftvec,*fshift,*x,*f;
 396     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 397     int              vdwioffset0;
 398     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 399     int              vdwjidx0A,vdwjidx0B;
 400     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 401     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 402     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 403     real             *charge;
 404     __m128i          gbitab;
 405     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
 406     __m128d          minushalf = _mm_set1_pd(-0.5);
 407     real             *invsqrta,*dvda,*gbtab;
 408     __m128i          vfitab;
 409     __m128i          ifour       = _mm_set1_epi32(4);
 410     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 411     real             *vftab;
 412     __m128d          dummy_mask,cutoff_mask;
 413     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 414     __m128d          one     = _mm_set1_pd(1.0);
 415     __m128d          two     = _mm_set1_pd(2.0);
 416     x                = xx[0];
 417     f                = ff[0];
 418
 419     nri              = nlist->nri;
 420     iinr             = nlist->iinr;
 421     jindex           = nlist->jindex;
 422     jjnr             = nlist->jjnr;
 423     shiftidx         = nlist->shift;
 424     gid              = nlist->gid;
 425     shiftvec         = fr->shift_vec[0];
 426     fshift           = fr->fshift[0];
 427     facel            = _mm_set1_pd(fr->epsfac);
 428     charge           = mdatoms->chargeA;
 429
 430     invsqrta         = fr->invsqrta;
 431     dvda             = fr->dvda;
 432     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 433     gbtab            = fr->gbtab.data;
 434     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 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         isai0            = _mm_load1_pd(invsqrta+inr+0);
 468
 469         dvdasum          = _mm_setzero_pd();
 470
 471         /* Start inner kernel loop */
 472         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 473         {
 474
 475             /* Get j neighbor index, and coordinate index */
 476             jnrA             = jjnr[jidx];
 477             jnrB             = jjnr[jidx+1];
 478             j_coord_offsetA  = DIM*jnrA;
 479             j_coord_offsetB  = DIM*jnrB;
 480
 481             /* load j atom coordinates */
 482             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 483                                               &jx0,&jy0,&jz0);
 484
 485             /* Calculate displacement vector */
 486             dx00             = _mm_sub_pd(ix0,jx0);
 487             dy00             = _mm_sub_pd(iy0,jy0);
 488             dz00             = _mm_sub_pd(iz0,jz0);
 489
 490             /* Calculate squared distance and things based on it */
 491             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 492
 493             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 494
 495             /* Load parameters for j particles */
 496             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 497             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+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
 508             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 509             isaprod          = _mm_mul_pd(isai0,isaj0);
 510             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 511             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 512
 513             /* Calculate generalized born table index - this is a separate table from the normal one,
 514              * but we use the same procedure by multiplying r with scale and truncating to integer.
 515              */
 516             rt               = _mm_mul_pd(r00,gbscale);
 517             gbitab           = _mm_cvttpd_epi32(rt);
 518 #ifdef __XOP__
 519             gbeps            = _mm_frcz_pd(rt);
 520 #else
 521             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 522 #endif
 523             gbitab           = _mm_slli_epi32(gbitab,2);
 524
 525             Y                = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,0) );
 526             F                = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,1) );
 527             GMX_MM_TRANSPOSE2_PD(Y,F);
 528             G                = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,0) +2);
 529             H                = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,1) +2);
 530             GMX_MM_TRANSPOSE2_PD(G,H);
 531             Fp               = _mm_macc_pd(gbeps,_mm_macc_pd(gbeps,H,G),F);
 532             VV               = _mm_macc_pd(gbeps,Fp,Y);
 533             vgb              = _mm_mul_pd(gbqqfactor,VV);
 534
 535             twogbeps         = _mm_add_pd(gbeps,gbeps);
 536             FF               = _mm_macc_pd(_mm_macc_pd(twogbeps,H,G),gbeps,Fp);
 537             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 538             dvdatmp          = _mm_mul_pd(minushalf,_mm_macc_pd(fgb,r00,vgb));
 539             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 540             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 541             velec            = _mm_mul_pd(qq00,rinv00);
 542             felec            = _mm_mul_pd(_mm_msub_pd(velec,rinv00,fgb),rinv00);
 543
 544             fscal            = felec;
 545
 546             /* Update vectorial force */
 547             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 548             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 549             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 550
 551             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 552                                                    _mm_mul_pd(dx00,fscal),
 553                                                    _mm_mul_pd(dy00,fscal),
 554                                                    _mm_mul_pd(dz00,fscal));
 555
 556             /* Inner loop uses 59 flops */
 557         }
 558
 559         if(jidx<j_index_end)
 560         {
 561
 562             jnrA             = jjnr[jidx];
 563             j_coord_offsetA  = DIM*jnrA;
 564
 565             /* load j atom coordinates */
 566             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 567                                               &jx0,&jy0,&jz0);
 568
 569             /* Calculate displacement vector */
 570             dx00             = _mm_sub_pd(ix0,jx0);
 571             dy00             = _mm_sub_pd(iy0,jy0);
 572             dz00             = _mm_sub_pd(iz0,jz0);
 573
 574             /* Calculate squared distance and things based on it */
 575             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 576
 577             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 578
 579             /* Load parameters for j particles */
 580             jq0              = _mm_load_sd(charge+jnrA+0);
 581             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 582
 583             /**************************
 584              * CALCULATE INTERACTIONS *
 585              **************************/
 586
 587             r00              = _mm_mul_pd(rsq00,rinv00);
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq00             = _mm_mul_pd(iq0,jq0);
 591
 592             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 593             isaprod          = _mm_mul_pd(isai0,isaj0);
 594             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 595             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 596
 597             /* Calculate generalized born table index - this is a separate table from the normal one,
 598              * but we use the same procedure by multiplying r with scale and truncating to integer.
 599              */
 600             rt               = _mm_mul_pd(r00,gbscale);
 601             gbitab           = _mm_cvttpd_epi32(rt);
 602 #ifdef __XOP__
 603             gbeps            = _mm_frcz_pd(rt);
 604 #else
 605             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 606 #endif
 607             gbitab           = _mm_slli_epi32(gbitab,2);
 608
 609             Y                = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,0) );
 610             F                = _mm_setzero_pd();
 611             GMX_MM_TRANSPOSE2_PD(Y,F);
 612             G                = _mm_load_pd( gbtab + _mm_extract_epi32(gbitab,0) +2);
 613             H                = _mm_setzero_pd();
 614             GMX_MM_TRANSPOSE2_PD(G,H);
 615             Fp               = _mm_macc_pd(gbeps,_mm_macc_pd(gbeps,H,G),F);
 616             VV               = _mm_macc_pd(gbeps,Fp,Y);
 617             vgb              = _mm_mul_pd(gbqqfactor,VV);
 618
 619             twogbeps         = _mm_add_pd(gbeps,gbeps);
 620             FF               = _mm_macc_pd(_mm_macc_pd(twogbeps,H,G),gbeps,Fp);
 621             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 622             dvdatmp          = _mm_mul_pd(minushalf,_mm_macc_pd(fgb,r00,vgb));
 623             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 624             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 625             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 626             velec            = _mm_mul_pd(qq00,rinv00);
 627             felec            = _mm_mul_pd(_mm_msub_pd(velec,rinv00,fgb),rinv00);
 628
 629             fscal            = felec;
 630
 631             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 632
 633             /* Update vectorial force */
 634             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 635             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 636             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 637
 638             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 639                                                    _mm_mul_pd(dx00,fscal),
 640                                                    _mm_mul_pd(dy00,fscal),
 641                                                    _mm_mul_pd(dz00,fscal));
 642
 643             /* Inner loop uses 59 flops */
 644         }
 645
 646         /* End of innermost loop */
 647
 648         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 649                                               f+i_coord_offset,fshift+i_shift_offset);
 650
 651         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 652         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 653
 654         /* Increment number of inner iterations */
 655         inneriter                  += j_index_end - j_index_start;
 656
 657         /* Outer loop uses 7 flops */
 658     }
 659
 660     /* Increment number of outer iterations */
 661     outeriter        += nri;
 662
 663     /* Update outer/inner flops */
 664
 665     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*59);
 666 }