src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwNone_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_avx_128_fma_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     __m128i          ewitab;
  74     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  75     real             *ewtab;
  76     __m128d          dummy_mask,cutoff_mask;
  77     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  78     __m128d          one     = _mm_set1_pd(1.0);
  79     __m128d          two     = _mm_set1_pd(2.0);
  80     x                = xx[0];
  81     f                = ff[0];
  82
  83     nri              = nlist->nri;
  84     iinr             = nlist->iinr;
  85     jindex           = nlist->jindex;
  86     jjnr             = nlist->jjnr;
  87     shiftidx         = nlist->shift;
  88     gid              = nlist->gid;
  89     shiftvec         = fr->shift_vec[0];
  90     fshift           = fr->fshift[0];
  91     facel            = _mm_set1_pd(fr->epsfac);
  92     charge           = mdatoms->chargeA;
  93
  94     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
  95     ewtab            = fr->ic->tabq_coul_FDV0;
  96     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
  97     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
  98
  99     /* Avoid stupid compiler warnings */
 100     jnrA = jnrB = 0;
 101     j_coord_offsetA = 0;
 102     j_coord_offsetB = 0;
 103
 104     outeriter        = 0;
 105     inneriter        = 0;
 106
 107     /* Start outer loop over neighborlists */
 108     for(iidx=0; iidx<nri; iidx++)
 109     {
 110         /* Load shift vector for this list */
 111         i_shift_offset   = DIM*shiftidx[iidx];
 112
 113         /* Load limits for loop over neighbors */
 114         j_index_start    = jindex[iidx];
 115         j_index_end      = jindex[iidx+1];
 116
 117         /* Get outer coordinate index */
 118         inr              = iinr[iidx];
 119         i_coord_offset   = DIM*inr;
 120
 121         /* Load i particle coords and add shift vector */
 122         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 123
 124         fix0             = _mm_setzero_pd();
 125         fiy0             = _mm_setzero_pd();
 126         fiz0             = _mm_setzero_pd();
 127
 128         /* Load parameters for i particles */
 129         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 130
 131         /* Reset potential sums */
 132         velecsum         = _mm_setzero_pd();
 133
 134         /* Start inner kernel loop */
 135         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 136         {
 137
 138             /* Get j neighbor index, and coordinate index */
 139             jnrA             = jjnr[jidx];
 140             jnrB             = jjnr[jidx+1];
 141             j_coord_offsetA  = DIM*jnrA;
 142             j_coord_offsetB  = DIM*jnrB;
 143
 144             /* load j atom coordinates */
 145             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 146                                               &jx0,&jy0,&jz0);
 147
 148             /* Calculate displacement vector */
 149             dx00             = _mm_sub_pd(ix0,jx0);
 150             dy00             = _mm_sub_pd(iy0,jy0);
 151             dz00             = _mm_sub_pd(iz0,jz0);
 152
 153             /* Calculate squared distance and things based on it */
 154             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 155
 156             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 157
 158             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 159
 160             /* Load parameters for j particles */
 161             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 162
 163             /**************************
 164              * CALCULATE INTERACTIONS *
 165              **************************/
 166
 167             r00              = _mm_mul_pd(rsq00,rinv00);
 168
 169             /* Compute parameters for interactions between i and j atoms */
 170             qq00             = _mm_mul_pd(iq0,jq0);
 171
 172             /* EWALD ELECTROSTATICS */
 173
 174             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 175             ewrt             = _mm_mul_pd(r00,ewtabscale);
 176             ewitab           = _mm_cvttpd_epi32(ewrt);
 177 #ifdef __XOP__
 178             eweps            = _mm_frcz_pd(ewrt);
 179 #else
 180             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 181 #endif
 182             twoeweps         = _mm_add_pd(eweps,eweps);
 183             ewitab           = _mm_slli_epi32(ewitab,2);
 184             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 185             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 186             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 187             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 188             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 189             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 190             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 191             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 192             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 193             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 194
 195             /* Update potential sum for this i atom from the interaction with this j atom. */
 196             velecsum         = _mm_add_pd(velecsum,velec);
 197
 198             fscal            = felec;
 199
 200             /* Update vectorial force */
 201             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 202             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 203             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 204
 205             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 206                                                    _mm_mul_pd(dx00,fscal),
 207                                                    _mm_mul_pd(dy00,fscal),
 208                                                    _mm_mul_pd(dz00,fscal));
 209
 210             /* Inner loop uses 44 flops */
 211         }
 212
 213         if(jidx<j_index_end)
 214         {
 215
 216             jnrA             = jjnr[jidx];
 217             j_coord_offsetA  = DIM*jnrA;
 218
 219             /* load j atom coordinates */
 220             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 221                                               &jx0,&jy0,&jz0);
 222
 223             /* Calculate displacement vector */
 224             dx00             = _mm_sub_pd(ix0,jx0);
 225             dy00             = _mm_sub_pd(iy0,jy0);
 226             dz00             = _mm_sub_pd(iz0,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 230
 231             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 232
 233             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 234
 235             /* Load parameters for j particles */
 236             jq0              = _mm_load_sd(charge+jnrA+0);
 237
 238             /**************************
 239              * CALCULATE INTERACTIONS *
 240              **************************/
 241
 242             r00              = _mm_mul_pd(rsq00,rinv00);
 243
 244             /* Compute parameters for interactions between i and j atoms */
 245             qq00             = _mm_mul_pd(iq0,jq0);
 246
 247             /* EWALD ELECTROSTATICS */
 248
 249             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 250             ewrt             = _mm_mul_pd(r00,ewtabscale);
 251             ewitab           = _mm_cvttpd_epi32(ewrt);
 252 #ifdef __XOP__
 253             eweps            = _mm_frcz_pd(ewrt);
 254 #else
 255             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 256 #endif
 257             twoeweps         = _mm_add_pd(eweps,eweps);
 258             ewitab           = _mm_slli_epi32(ewitab,2);
 259             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 260             ewtabD           = _mm_setzero_pd();
 261             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 262             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 263             ewtabFn          = _mm_setzero_pd();
 264             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 265             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 266             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 267             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 268             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 269
 270             /* Update potential sum for this i atom from the interaction with this j atom. */
 271             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 272             velecsum         = _mm_add_pd(velecsum,velec);
 273
 274             fscal            = felec;
 275
 276             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 277
 278             /* Update vectorial force */
 279             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 280             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 281             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 282
 283             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 284                                                    _mm_mul_pd(dx00,fscal),
 285                                                    _mm_mul_pd(dy00,fscal),
 286                                                    _mm_mul_pd(dz00,fscal));
 287
 288             /* Inner loop uses 44 flops */
 289         }
 290
 291         /* End of innermost loop */
 292
 293         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 294                                               f+i_coord_offset,fshift+i_shift_offset);
 295
 296         ggid                        = gid[iidx];
 297         /* Update potential energies */
 298         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 299
 300         /* Increment number of inner iterations */
 301         inneriter                  += j_index_end - j_index_start;
 302
 303         /* Outer loop uses 8 flops */
 304     }
 305
 306     /* Increment number of outer iterations */
 307     outeriter        += nri;
 308
 309     /* Update outer/inner flops */
 310
 311     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*44);
 312 }
 313 /*
 314  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_128_fma_double
 315  * Electrostatics interaction: Ewald
 316  * VdW interaction:            None
 317  * Geometry:                   Particle-Particle
 318  * Calculate force/pot:        Force
 319  */
 320 void
 321 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_avx_128_fma_double
 322                     (t_nblist * gmx_restrict                nlist,
 323                      rvec * gmx_restrict                    xx,
 324                      rvec * gmx_restrict                    ff,
 325                      t_forcerec * gmx_restrict              fr,
 326                      t_mdatoms * gmx_restrict               mdatoms,
 327                      nb_kernel_data_t * gmx_restrict        kernel_data,
 328                      t_nrnb * gmx_restrict                  nrnb)
 329 {
 330     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 331      * just 0 for non-waters.
 332      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 333      * jnr indices corresponding to data put in the four positions in the SIMD register.
 334      */
 335     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 336     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 337     int              jnrA,jnrB;
 338     int              j_coord_offsetA,j_coord_offsetB;
 339     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 340     real             rcutoff_scalar;
 341     real             *shiftvec,*fshift,*x,*f;
 342     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 343     int              vdwioffset0;
 344     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 345     int              vdwjidx0A,vdwjidx0B;
 346     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 347     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 348     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 349     real             *charge;
 350     __m128i          ewitab;
 351     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 352     real             *ewtab;
 353     __m128d          dummy_mask,cutoff_mask;
 354     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 355     __m128d          one     = _mm_set1_pd(1.0);
 356     __m128d          two     = _mm_set1_pd(2.0);
 357     x                = xx[0];
 358     f                = ff[0];
 359
 360     nri              = nlist->nri;
 361     iinr             = nlist->iinr;
 362     jindex           = nlist->jindex;
 363     jjnr             = nlist->jjnr;
 364     shiftidx         = nlist->shift;
 365     gid              = nlist->gid;
 366     shiftvec         = fr->shift_vec[0];
 367     fshift           = fr->fshift[0];
 368     facel            = _mm_set1_pd(fr->epsfac);
 369     charge           = mdatoms->chargeA;
 370
 371     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 372     ewtab            = fr->ic->tabq_coul_F;
 373     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 374     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 375
 376     /* Avoid stupid compiler warnings */
 377     jnrA = jnrB = 0;
 378     j_coord_offsetA = 0;
 379     j_coord_offsetB = 0;
 380
 381     outeriter        = 0;
 382     inneriter        = 0;
 383
 384     /* Start outer loop over neighborlists */
 385     for(iidx=0; iidx<nri; iidx++)
 386     {
 387         /* Load shift vector for this list */
 388         i_shift_offset   = DIM*shiftidx[iidx];
 389
 390         /* Load limits for loop over neighbors */
 391         j_index_start    = jindex[iidx];
 392         j_index_end      = jindex[iidx+1];
 393
 394         /* Get outer coordinate index */
 395         inr              = iinr[iidx];
 396         i_coord_offset   = DIM*inr;
 397
 398         /* Load i particle coords and add shift vector */
 399         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 400
 401         fix0             = _mm_setzero_pd();
 402         fiy0             = _mm_setzero_pd();
 403         fiz0             = _mm_setzero_pd();
 404
 405         /* Load parameters for i particles */
 406         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 407
 408         /* Start inner kernel loop */
 409         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 410         {
 411
 412             /* Get j neighbor index, and coordinate index */
 413             jnrA             = jjnr[jidx];
 414             jnrB             = jjnr[jidx+1];
 415             j_coord_offsetA  = DIM*jnrA;
 416             j_coord_offsetB  = DIM*jnrB;
 417
 418             /* load j atom coordinates */
 419             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 420                                               &jx0,&jy0,&jz0);
 421
 422             /* Calculate displacement vector */
 423             dx00             = _mm_sub_pd(ix0,jx0);
 424             dy00             = _mm_sub_pd(iy0,jy0);
 425             dz00             = _mm_sub_pd(iz0,jz0);
 426
 427             /* Calculate squared distance and things based on it */
 428             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 429
 430             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 431
 432             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 433
 434             /* Load parameters for j particles */
 435             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 436
 437             /**************************
 438              * CALCULATE INTERACTIONS *
 439              **************************/
 440
 441             r00              = _mm_mul_pd(rsq00,rinv00);
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             qq00             = _mm_mul_pd(iq0,jq0);
 445
 446             /* EWALD ELECTROSTATICS */
 447
 448             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 449             ewrt             = _mm_mul_pd(r00,ewtabscale);
 450             ewitab           = _mm_cvttpd_epi32(ewrt);
 451 #ifdef __XOP__
 452             eweps            = _mm_frcz_pd(ewrt);
 453 #else
 454             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 455 #endif
 456             twoeweps         = _mm_add_pd(eweps,eweps);
 457             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 458                                          &ewtabF,&ewtabFn);
 459             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 460             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 461
 462             fscal            = felec;
 463
 464             /* Update vectorial force */
 465             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 466             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 467             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 468
 469             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 470                                                    _mm_mul_pd(dx00,fscal),
 471                                                    _mm_mul_pd(dy00,fscal),
 472                                                    _mm_mul_pd(dz00,fscal));
 473
 474             /* Inner loop uses 39 flops */
 475         }
 476
 477         if(jidx<j_index_end)
 478         {
 479
 480             jnrA             = jjnr[jidx];
 481             j_coord_offsetA  = DIM*jnrA;
 482
 483             /* load j atom coordinates */
 484             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 485                                               &jx0,&jy0,&jz0);
 486
 487             /* Calculate displacement vector */
 488             dx00             = _mm_sub_pd(ix0,jx0);
 489             dy00             = _mm_sub_pd(iy0,jy0);
 490             dz00             = _mm_sub_pd(iz0,jz0);
 491
 492             /* Calculate squared distance and things based on it */
 493             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 494
 495             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 496
 497             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 498
 499             /* Load parameters for j particles */
 500             jq0              = _mm_load_sd(charge+jnrA+0);
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             r00              = _mm_mul_pd(rsq00,rinv00);
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             qq00             = _mm_mul_pd(iq0,jq0);
 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_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 523             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 524             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 525
 526             fscal            = felec;
 527
 528             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 529
 530             /* Update vectorial force */
 531             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 532             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 533             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 534
 535             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 536                                                    _mm_mul_pd(dx00,fscal),
 537                                                    _mm_mul_pd(dy00,fscal),
 538                                                    _mm_mul_pd(dz00,fscal));
 539
 540             /* Inner loop uses 39 flops */
 541         }
 542
 543         /* End of innermost loop */
 544
 545         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 546                                               f+i_coord_offset,fshift+i_shift_offset);
 547
 548         /* Increment number of inner iterations */
 549         inneriter                  += j_index_end - j_index_start;
 550
 551         /* Outer loop uses 7 flops */
 552     }
 553
 554     /* Increment number of outer iterations */
 555     outeriter        += nri;
 556
 557     /* Update outer/inner flops */
 558
 559     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*39);
 560 }