src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_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_ElecCSTab_VdwNone_GeomP1P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_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          vfitab;
  88     __m128i          ifour       = _mm_set1_epi32(4);
  89     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  90     real             *vftab;
  91     __m128d          dummy_mask,cutoff_mask;
  92     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  93     __m128d          one     = _mm_set1_pd(1.0);
  94     __m128d          two     = _mm_set1_pd(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_pd(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108
 109     vftab            = kernel_data->table_elec->data;
 110     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 111
 112     /* Avoid stupid compiler warnings */
 113     jnrA = jnrB = 0;
 114     j_coord_offsetA = 0;
 115     j_coord_offsetB = 0;
 116
 117     outeriter        = 0;
 118     inneriter        = 0;
 119
 120     /* Start outer loop over neighborlists */
 121     for(iidx=0; iidx<nri; iidx++)
 122     {
 123         /* Load shift vector for this list */
 124         i_shift_offset   = DIM*shiftidx[iidx];
 125
 126         /* Load limits for loop over neighbors */
 127         j_index_start    = jindex[iidx];
 128         j_index_end      = jindex[iidx+1];
 129
 130         /* Get outer coordinate index */
 131         inr              = iinr[iidx];
 132         i_coord_offset   = DIM*inr;
 133
 134         /* Load i particle coords and add shift vector */
 135         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 136
 137         fix0             = _mm_setzero_pd();
 138         fiy0             = _mm_setzero_pd();
 139         fiz0             = _mm_setzero_pd();
 140
 141         /* Load parameters for i particles */
 142         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 143
 144         /* Reset potential sums */
 145         velecsum         = _mm_setzero_pd();
 146
 147         /* Start inner kernel loop */
 148         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 149         {
 150
 151             /* Get j neighbor index, and coordinate index */
 152             jnrA             = jjnr[jidx];
 153             jnrB             = jjnr[jidx+1];
 154             j_coord_offsetA  = DIM*jnrA;
 155             j_coord_offsetB  = DIM*jnrB;
 156
 157             /* load j atom coordinates */
 158             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 159                                               &jx0,&jy0,&jz0);
 160
 161             /* Calculate displacement vector */
 162             dx00             = _mm_sub_pd(ix0,jx0);
 163             dy00             = _mm_sub_pd(iy0,jy0);
 164             dz00             = _mm_sub_pd(iz0,jz0);
 165
 166             /* Calculate squared distance and things based on it */
 167             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 168
 169             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 170
 171             /* Load parameters for j particles */
 172             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 173
 174             /**************************
 175              * CALCULATE INTERACTIONS *
 176              **************************/
 177
 178             r00              = _mm_mul_pd(rsq00,rinv00);
 179
 180             /* Compute parameters for interactions between i and j atoms */
 181             qq00             = _mm_mul_pd(iq0,jq0);
 182
 183             /* Calculate table index by multiplying r with table scale and truncate to integer */
 184             rt               = _mm_mul_pd(r00,vftabscale);
 185             vfitab           = _mm_cvttpd_epi32(rt);
 186 #ifdef __XOP__
 187             vfeps            = _mm_frcz_pd(rt);
 188 #else
 189             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 190 #endif
 191             twovfeps         = _mm_add_pd(vfeps,vfeps);
 192             vfitab           = _mm_slli_epi32(vfitab,2);
 193
 194             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 195             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 196             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 197             GMX_MM_TRANSPOSE2_PD(Y,F);
 198             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 199             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 200             GMX_MM_TRANSPOSE2_PD(G,H);
 201             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 202             VV               = _mm_macc_pd(vfeps,Fp,Y);
 203             velec            = _mm_mul_pd(qq00,VV);
 204             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 205             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 206
 207             /* Update potential sum for this i atom from the interaction with this j atom. */
 208             velecsum         = _mm_add_pd(velecsum,velec);
 209
 210             fscal            = felec;
 211
 212             /* Update vectorial force */
 213             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 214             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 215             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 216
 217             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 218                                                    _mm_mul_pd(dx00,fscal),
 219                                                    _mm_mul_pd(dy00,fscal),
 220                                                    _mm_mul_pd(dz00,fscal));
 221
 222             /* Inner loop uses 46 flops */
 223         }
 224
 225         if(jidx<j_index_end)
 226         {
 227
 228             jnrA             = jjnr[jidx];
 229             j_coord_offsetA  = DIM*jnrA;
 230
 231             /* load j atom coordinates */
 232             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 233                                               &jx0,&jy0,&jz0);
 234
 235             /* Calculate displacement vector */
 236             dx00             = _mm_sub_pd(ix0,jx0);
 237             dy00             = _mm_sub_pd(iy0,jy0);
 238             dz00             = _mm_sub_pd(iz0,jz0);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 242
 243             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 244
 245             /* Load parameters for j particles */
 246             jq0              = _mm_load_sd(charge+jnrA+0);
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             r00              = _mm_mul_pd(rsq00,rinv00);
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             qq00             = _mm_mul_pd(iq0,jq0);
 256
 257             /* Calculate table index by multiplying r with table scale and truncate to integer */
 258             rt               = _mm_mul_pd(r00,vftabscale);
 259             vfitab           = _mm_cvttpd_epi32(rt);
 260 #ifdef __XOP__
 261             vfeps            = _mm_frcz_pd(rt);
 262 #else
 263             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 264 #endif
 265             twovfeps         = _mm_add_pd(vfeps,vfeps);
 266             vfitab           = _mm_slli_epi32(vfitab,2);
 267
 268             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 269             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 270             F                = _mm_setzero_pd();
 271             GMX_MM_TRANSPOSE2_PD(Y,F);
 272             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 273             H                = _mm_setzero_pd();
 274             GMX_MM_TRANSPOSE2_PD(G,H);
 275             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 276             VV               = _mm_macc_pd(vfeps,Fp,Y);
 277             velec            = _mm_mul_pd(qq00,VV);
 278             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 279             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 283             velecsum         = _mm_add_pd(velecsum,velec);
 284
 285             fscal            = felec;
 286
 287             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 288
 289             /* Update vectorial force */
 290             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 291             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 292             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 293
 294             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 295                                                    _mm_mul_pd(dx00,fscal),
 296                                                    _mm_mul_pd(dy00,fscal),
 297                                                    _mm_mul_pd(dz00,fscal));
 298
 299             /* Inner loop uses 46 flops */
 300         }
 301
 302         /* End of innermost loop */
 303
 304         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 305                                               f+i_coord_offset,fshift+i_shift_offset);
 306
 307         ggid                        = gid[iidx];
 308         /* Update potential energies */
 309         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 310
 311         /* Increment number of inner iterations */
 312         inneriter                  += j_index_end - j_index_start;
 313
 314         /* Outer loop uses 8 flops */
 315     }
 316
 317     /* Increment number of outer iterations */
 318     outeriter        += nri;
 319
 320     /* Update outer/inner flops */
 321
 322     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*46);
 323 }
 324 /*
 325  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_avx_128_fma_double
 326  * Electrostatics interaction: CubicSplineTable
 327  * VdW interaction:            None
 328  * Geometry:                   Particle-Particle
 329  * Calculate force/pot:        Force
 330  */
 331 void
 332 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_avx_128_fma_double
 333                     (t_nblist                    * gmx_restrict       nlist,
 334                      rvec                        * gmx_restrict          xx,
 335                      rvec                        * gmx_restrict          ff,
 336                      t_forcerec                  * gmx_restrict          fr,
 337                      t_mdatoms                   * gmx_restrict     mdatoms,
 338                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 339                      t_nrnb                      * gmx_restrict        nrnb)
 340 {
 341     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 342      * just 0 for non-waters.
 343      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 344      * jnr indices corresponding to data put in the four positions in the SIMD register.
 345      */
 346     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 347     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 348     int              jnrA,jnrB;
 349     int              j_coord_offsetA,j_coord_offsetB;
 350     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 351     real             rcutoff_scalar;
 352     real             *shiftvec,*fshift,*x,*f;
 353     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 354     int              vdwioffset0;
 355     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 356     int              vdwjidx0A,vdwjidx0B;
 357     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 358     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 359     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 360     real             *charge;
 361     __m128i          vfitab;
 362     __m128i          ifour       = _mm_set1_epi32(4);
 363     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 364     real             *vftab;
 365     __m128d          dummy_mask,cutoff_mask;
 366     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 367     __m128d          one     = _mm_set1_pd(1.0);
 368     __m128d          two     = _mm_set1_pd(2.0);
 369     x                = xx[0];
 370     f                = ff[0];
 371
 372     nri              = nlist->nri;
 373     iinr             = nlist->iinr;
 374     jindex           = nlist->jindex;
 375     jjnr             = nlist->jjnr;
 376     shiftidx         = nlist->shift;
 377     gid              = nlist->gid;
 378     shiftvec         = fr->shift_vec[0];
 379     fshift           = fr->fshift[0];
 380     facel            = _mm_set1_pd(fr->epsfac);
 381     charge           = mdatoms->chargeA;
 382
 383     vftab            = kernel_data->table_elec->data;
 384     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 385
 386     /* Avoid stupid compiler warnings */
 387     jnrA = jnrB = 0;
 388     j_coord_offsetA = 0;
 389     j_coord_offsetB = 0;
 390
 391     outeriter        = 0;
 392     inneriter        = 0;
 393
 394     /* Start outer loop over neighborlists */
 395     for(iidx=0; iidx<nri; iidx++)
 396     {
 397         /* Load shift vector for this list */
 398         i_shift_offset   = DIM*shiftidx[iidx];
 399
 400         /* Load limits for loop over neighbors */
 401         j_index_start    = jindex[iidx];
 402         j_index_end      = jindex[iidx+1];
 403
 404         /* Get outer coordinate index */
 405         inr              = iinr[iidx];
 406         i_coord_offset   = DIM*inr;
 407
 408         /* Load i particle coords and add shift vector */
 409         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 410
 411         fix0             = _mm_setzero_pd();
 412         fiy0             = _mm_setzero_pd();
 413         fiz0             = _mm_setzero_pd();
 414
 415         /* Load parameters for i particles */
 416         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 417
 418         /* Start inner kernel loop */
 419         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 420         {
 421
 422             /* Get j neighbor index, and coordinate index */
 423             jnrA             = jjnr[jidx];
 424             jnrB             = jjnr[jidx+1];
 425             j_coord_offsetA  = DIM*jnrA;
 426             j_coord_offsetB  = DIM*jnrB;
 427
 428             /* load j atom coordinates */
 429             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 430                                               &jx0,&jy0,&jz0);
 431
 432             /* Calculate displacement vector */
 433             dx00             = _mm_sub_pd(ix0,jx0);
 434             dy00             = _mm_sub_pd(iy0,jy0);
 435             dz00             = _mm_sub_pd(iz0,jz0);
 436
 437             /* Calculate squared distance and things based on it */
 438             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 439
 440             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 441
 442             /* Load parameters for j particles */
 443             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             r00              = _mm_mul_pd(rsq00,rinv00);
 450
 451             /* Compute parameters for interactions between i and j atoms */
 452             qq00             = _mm_mul_pd(iq0,jq0);
 453
 454             /* Calculate table index by multiplying r with table scale and truncate to integer */
 455             rt               = _mm_mul_pd(r00,vftabscale);
 456             vfitab           = _mm_cvttpd_epi32(rt);
 457 #ifdef __XOP__
 458             vfeps            = _mm_frcz_pd(rt);
 459 #else
 460             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 461 #endif
 462             twovfeps         = _mm_add_pd(vfeps,vfeps);
 463             vfitab           = _mm_slli_epi32(vfitab,2);
 464
 465             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 466             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 467             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 468             GMX_MM_TRANSPOSE2_PD(Y,F);
 469             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 470             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 471             GMX_MM_TRANSPOSE2_PD(G,H);
 472             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 473             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 474             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 475
 476             fscal            = felec;
 477
 478             /* Update vectorial force */
 479             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 480             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 481             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 482
 483             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 484                                                    _mm_mul_pd(dx00,fscal),
 485                                                    _mm_mul_pd(dy00,fscal),
 486                                                    _mm_mul_pd(dz00,fscal));
 487
 488             /* Inner loop uses 42 flops */
 489         }
 490
 491         if(jidx<j_index_end)
 492         {
 493
 494             jnrA             = jjnr[jidx];
 495             j_coord_offsetA  = DIM*jnrA;
 496
 497             /* load j atom coordinates */
 498             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 499                                               &jx0,&jy0,&jz0);
 500
 501             /* Calculate displacement vector */
 502             dx00             = _mm_sub_pd(ix0,jx0);
 503             dy00             = _mm_sub_pd(iy0,jy0);
 504             dz00             = _mm_sub_pd(iz0,jz0);
 505
 506             /* Calculate squared distance and things based on it */
 507             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 508
 509             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 510
 511             /* Load parameters for j particles */
 512             jq0              = _mm_load_sd(charge+jnrA+0);
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             r00              = _mm_mul_pd(rsq00,rinv00);
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             qq00             = _mm_mul_pd(iq0,jq0);
 522
 523             /* Calculate table index by multiplying r with table scale and truncate to integer */
 524             rt               = _mm_mul_pd(r00,vftabscale);
 525             vfitab           = _mm_cvttpd_epi32(rt);
 526 #ifdef __XOP__
 527             vfeps            = _mm_frcz_pd(rt);
 528 #else
 529             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 530 #endif
 531             twovfeps         = _mm_add_pd(vfeps,vfeps);
 532             vfitab           = _mm_slli_epi32(vfitab,2);
 533
 534             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 535             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 536             F                = _mm_setzero_pd();
 537             GMX_MM_TRANSPOSE2_PD(Y,F);
 538             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 539             H                = _mm_setzero_pd();
 540             GMX_MM_TRANSPOSE2_PD(G,H);
 541             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 542             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 543             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 544
 545             fscal            = felec;
 546
 547             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 548
 549             /* Update vectorial force */
 550             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 551             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 552             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 553
 554             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 555                                                    _mm_mul_pd(dx00,fscal),
 556                                                    _mm_mul_pd(dy00,fscal),
 557                                                    _mm_mul_pd(dz00,fscal));
 558
 559             /* Inner loop uses 42 flops */
 560         }
 561
 562         /* End of innermost loop */
 563
 564         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 565                                               f+i_coord_offset,fshift+i_shift_offset);
 566
 567         /* Increment number of inner iterations */
 568         inneriter                  += j_index_end - j_index_start;
 569
 570         /* Outer loop uses 7 flops */
 571     }
 572
 573     /* Increment number of outer iterations */
 574     outeriter        += nri;
 575
 576     /* Update outer/inner flops */
 577
 578     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*42);
 579 }