src/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecGB_VdwNone_GeomP1P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012, by the GROMACS development team, led by
   5  * David van der Spoel, Berk Hess, Erik Lindahl, and including many
   6  * others, as listed in the AUTHORS file in the top-level source
   7  * 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 sse2_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sse2_double
  54  * Electrostatics interaction: GeneralizedBorn
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sse2_double
  61                     (t_nblist * gmx_restrict                nlist,
  62                      rvec * gmx_restrict                    xx,
  63                      rvec * gmx_restrict                    ff,
  64                      t_forcerec * gmx_restrict              fr,
  65                      t_mdatoms * gmx_restrict               mdatoms,
  66                      nb_kernel_data_t * gmx_restrict        kernel_data,
  67                      t_nrnb * gmx_restrict                  nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128i          gbitab;
  90     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
  91     __m128d          minushalf = _mm_set1_pd(-0.5);
  92     real             *invsqrta,*dvda,*gbtab;
  93     __m128i          vfitab;
  94     __m128i          ifour       = _mm_set1_epi32(4);
  95     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  96     real             *vftab;
  97     __m128d          dummy_mask,cutoff_mask;
  98     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  99     __m128d          one     = _mm_set1_pd(1.0);
 100     __m128d          two     = _mm_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_pd(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114
 115     invsqrta         = fr->invsqrta;
 116     dvda             = fr->dvda;
 117     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 118     gbtab            = fr->gbtab.data;
 119     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 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         isai0            = _mm_load1_pd(invsqrta+inr+0);
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm_setzero_pd();
 156         vgbsum           = _mm_setzero_pd();
 157         dvdasum          = _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             /* Load parameters for j particles */
 184             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 185             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 186
 187             /**************************
 188              * CALCULATE INTERACTIONS *
 189              **************************/
 190
 191             r00              = _mm_mul_pd(rsq00,rinv00);
 192
 193             /* Compute parameters for interactions between i and j atoms */
 194             qq00             = _mm_mul_pd(iq0,jq0);
 195
 196             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 197             isaprod          = _mm_mul_pd(isai0,isaj0);
 198             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 199             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 200
 201             /* Calculate generalized born table index - this is a separate table from the normal one,
 202              * but we use the same procedure by multiplying r with scale and truncating to integer.
 203              */
 204             rt               = _mm_mul_pd(r00,gbscale);
 205             gbitab           = _mm_cvttpd_epi32(rt);
 206             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 207             gbitab           = _mm_slli_epi32(gbitab,2);
 208
 209             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 210             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 211             GMX_MM_TRANSPOSE2_PD(Y,F);
 212             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 213             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 214             GMX_MM_TRANSPOSE2_PD(G,H);
 215             Heps             = _mm_mul_pd(gbeps,H);
 216             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 217             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 218             vgb              = _mm_mul_pd(gbqqfactor,VV);
 219
 220             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 221             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 222             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 223             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 224             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 225             velec            = _mm_mul_pd(qq00,rinv00);
 226             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 227
 228             /* Update potential sum for this i atom from the interaction with this j atom. */
 229             velecsum         = _mm_add_pd(velecsum,velec);
 230             vgbsum           = _mm_add_pd(vgbsum,vgb);
 231
 232             fscal            = felec;
 233
 234             /* Calculate temporary vectorial force */
 235             tx               = _mm_mul_pd(fscal,dx00);
 236             ty               = _mm_mul_pd(fscal,dy00);
 237             tz               = _mm_mul_pd(fscal,dz00);
 238
 239             /* Update vectorial force */
 240             fix0             = _mm_add_pd(fix0,tx);
 241             fiy0             = _mm_add_pd(fiy0,ty);
 242             fiz0             = _mm_add_pd(fiz0,tz);
 243
 244             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 245
 246             /* Inner loop uses 58 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             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 293             gbitab           = _mm_slli_epi32(gbitab,2);
 294
 295             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 296             F                = _mm_setzero_pd();
 297             GMX_MM_TRANSPOSE2_PD(Y,F);
 298             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 299             H                = _mm_setzero_pd();
 300             GMX_MM_TRANSPOSE2_PD(G,H);
 301             Heps             = _mm_mul_pd(gbeps,H);
 302             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 303             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 304             vgb              = _mm_mul_pd(gbqqfactor,VV);
 305
 306             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 307             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 308             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 309             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 310             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 311             velec            = _mm_mul_pd(qq00,rinv00);
 312             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 316             velecsum         = _mm_add_pd(velecsum,velec);
 317             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 318             vgbsum           = _mm_add_pd(vgbsum,vgb);
 319
 320             fscal            = felec;
 321
 322             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 323
 324             /* Calculate temporary vectorial force */
 325             tx               = _mm_mul_pd(fscal,dx00);
 326             ty               = _mm_mul_pd(fscal,dy00);
 327             tz               = _mm_mul_pd(fscal,dz00);
 328
 329             /* Update vectorial force */
 330             fix0             = _mm_add_pd(fix0,tx);
 331             fiy0             = _mm_add_pd(fiy0,ty);
 332             fiz0             = _mm_add_pd(fiz0,tz);
 333
 334             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 335
 336             /* Inner loop uses 58 flops */
 337         }
 338
 339         /* End of innermost loop */
 340
 341         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 342                                               f+i_coord_offset,fshift+i_shift_offset);
 343
 344         ggid                        = gid[iidx];
 345         /* Update potential energies */
 346         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 347         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 348         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 349         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 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_VF,outeriter*9 + inneriter*58);
 363 }
 364 /*
 365  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse2_double
 366  * Electrostatics interaction: GeneralizedBorn
 367  * VdW interaction:            None
 368  * Geometry:                   Particle-Particle
 369  * Calculate force/pot:        Force
 370  */
 371 void
 372 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse2_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_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     __m128i          gbitab;
 402     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 403     __m128d          minushalf = _mm_set1_pd(-0.5);
 404     real             *invsqrta,*dvda,*gbtab;
 405     __m128i          vfitab;
 406     __m128i          ifour       = _mm_set1_epi32(4);
 407     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 408     real             *vftab;
 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->epsfac);
 425     charge           = mdatoms->chargeA;
 426
 427     invsqrta         = fr->invsqrta;
 428     dvda             = fr->dvda;
 429     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 430     gbtab            = fr->gbtab.data;
 431     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 432
 433     /* Avoid stupid compiler warnings */
 434     jnrA = jnrB = 0;
 435     j_coord_offsetA = 0;
 436     j_coord_offsetB = 0;
 437
 438     outeriter        = 0;
 439     inneriter        = 0;
 440
 441     /* Start outer loop over neighborlists */
 442     for(iidx=0; iidx<nri; iidx++)
 443     {
 444         /* Load shift vector for this list */
 445         i_shift_offset   = DIM*shiftidx[iidx];
 446
 447         /* Load limits for loop over neighbors */
 448         j_index_start    = jindex[iidx];
 449         j_index_end      = jindex[iidx+1];
 450
 451         /* Get outer coordinate index */
 452         inr              = iinr[iidx];
 453         i_coord_offset   = DIM*inr;
 454
 455         /* Load i particle coords and add shift vector */
 456         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 457
 458         fix0             = _mm_setzero_pd();
 459         fiy0             = _mm_setzero_pd();
 460         fiz0             = _mm_setzero_pd();
 461
 462         /* Load parameters for i particles */
 463         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 464         isai0            = _mm_load1_pd(invsqrta+inr+0);
 465
 466         dvdasum          = _mm_setzero_pd();
 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           = gmx_mm_invsqrt_pd(rsq00);
 491
 492             /* Load parameters for j particles */
 493             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 494             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             r00              = _mm_mul_pd(rsq00,rinv00);
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             qq00             = _mm_mul_pd(iq0,jq0);
 504
 505             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 506             isaprod          = _mm_mul_pd(isai0,isaj0);
 507             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 508             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 509
 510             /* Calculate generalized born table index - this is a separate table from the normal one,
 511              * but we use the same procedure by multiplying r with scale and truncating to integer.
 512              */
 513             rt               = _mm_mul_pd(r00,gbscale);
 514             gbitab           = _mm_cvttpd_epi32(rt);
 515             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 516             gbitab           = _mm_slli_epi32(gbitab,2);
 517
 518             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 519             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 520             GMX_MM_TRANSPOSE2_PD(Y,F);
 521             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 522             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 523             GMX_MM_TRANSPOSE2_PD(G,H);
 524             Heps             = _mm_mul_pd(gbeps,H);
 525             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 526             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 527             vgb              = _mm_mul_pd(gbqqfactor,VV);
 528
 529             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 530             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 531             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 532             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 533             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 534             velec            = _mm_mul_pd(qq00,rinv00);
 535             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 536
 537             fscal            = felec;
 538
 539             /* Calculate temporary vectorial force */
 540             tx               = _mm_mul_pd(fscal,dx00);
 541             ty               = _mm_mul_pd(fscal,dy00);
 542             tz               = _mm_mul_pd(fscal,dz00);
 543
 544             /* Update vectorial force */
 545             fix0             = _mm_add_pd(fix0,tx);
 546             fiy0             = _mm_add_pd(fiy0,ty);
 547             fiz0             = _mm_add_pd(fiz0,tz);
 548
 549             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 550
 551             /* Inner loop uses 56 flops */
 552         }
 553
 554         if(jidx<j_index_end)
 555         {
 556
 557             jnrA             = jjnr[jidx];
 558             j_coord_offsetA  = DIM*jnrA;
 559
 560             /* load j atom coordinates */
 561             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 562                                               &jx0,&jy0,&jz0);
 563
 564             /* Calculate displacement vector */
 565             dx00             = _mm_sub_pd(ix0,jx0);
 566             dy00             = _mm_sub_pd(iy0,jy0);
 567             dz00             = _mm_sub_pd(iz0,jz0);
 568
 569             /* Calculate squared distance and things based on it */
 570             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 571
 572             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 573
 574             /* Load parameters for j particles */
 575             jq0              = _mm_load_sd(charge+jnrA+0);
 576             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             r00              = _mm_mul_pd(rsq00,rinv00);
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq00             = _mm_mul_pd(iq0,jq0);
 586
 587             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 588             isaprod          = _mm_mul_pd(isai0,isaj0);
 589             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 590             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 591
 592             /* Calculate generalized born table index - this is a separate table from the normal one,
 593              * but we use the same procedure by multiplying r with scale and truncating to integer.
 594              */
 595             rt               = _mm_mul_pd(r00,gbscale);
 596             gbitab           = _mm_cvttpd_epi32(rt);
 597             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 598             gbitab           = _mm_slli_epi32(gbitab,2);
 599
 600             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 601             F                = _mm_setzero_pd();
 602             GMX_MM_TRANSPOSE2_PD(Y,F);
 603             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 604             H                = _mm_setzero_pd();
 605             GMX_MM_TRANSPOSE2_PD(G,H);
 606             Heps             = _mm_mul_pd(gbeps,H);
 607             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 608             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 609             vgb              = _mm_mul_pd(gbqqfactor,VV);
 610
 611             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 612             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 613             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 614             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 615             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 616             velec            = _mm_mul_pd(qq00,rinv00);
 617             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 618
 619             fscal            = felec;
 620
 621             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm_mul_pd(fscal,dx00);
 625             ty               = _mm_mul_pd(fscal,dy00);
 626             tz               = _mm_mul_pd(fscal,dz00);
 627
 628             /* Update vectorial force */
 629             fix0             = _mm_add_pd(fix0,tx);
 630             fiy0             = _mm_add_pd(fiy0,ty);
 631             fiz0             = _mm_add_pd(fiz0,tz);
 632
 633             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 634
 635             /* Inner loop uses 56 flops */
 636         }
 637
 638         /* End of innermost loop */
 639
 640         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 641                                               f+i_coord_offset,fshift+i_shift_offset);
 642
 643         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 644         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 645
 646         /* Increment number of inner iterations */
 647         inneriter                  += j_index_end - j_index_start;
 648
 649         /* Outer loop uses 7 flops */
 650     }
 651
 652     /* Increment number of outer iterations */
 653     outeriter        += nri;
 654
 655     /* Update outer/inner flops */
 656
 657     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*56);
 658 }