src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwNone_GeomW4P1_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_avx_128_fma_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_unused * 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              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128i          ewitab;
  96     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     real             *ewtab;
  98     __m128d          dummy_mask,cutoff_mask;
  99     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 100     __m128d          one     = _mm_set1_pd(1.0);
 101     __m128d          two     = _mm_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115
 116     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 117     ewtab            = fr->ic->tabq_coul_FDV0;
 118     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 119     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 124     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 125     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 151                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 152
 153         fix1             = _mm_setzero_pd();
 154         fiy1             = _mm_setzero_pd();
 155         fiz1             = _mm_setzero_pd();
 156         fix2             = _mm_setzero_pd();
 157         fiy2             = _mm_setzero_pd();
 158         fiz2             = _mm_setzero_pd();
 159         fix3             = _mm_setzero_pd();
 160         fiy3             = _mm_setzero_pd();
 161         fiz3             = _mm_setzero_pd();
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx10             = _mm_sub_pd(ix1,jx0);
 182             dy10             = _mm_sub_pd(iy1,jy0);
 183             dz10             = _mm_sub_pd(iz1,jz0);
 184             dx20             = _mm_sub_pd(ix2,jx0);
 185             dy20             = _mm_sub_pd(iy2,jy0);
 186             dz20             = _mm_sub_pd(iz2,jz0);
 187             dx30             = _mm_sub_pd(ix3,jx0);
 188             dy30             = _mm_sub_pd(iy3,jy0);
 189             dz30             = _mm_sub_pd(iz3,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 193             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 194             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 195
 196             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 197             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 198             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 199
 200             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 201             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 202             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 203
 204             /* Load parameters for j particles */
 205             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 206
 207             fjx0             = _mm_setzero_pd();
 208             fjy0             = _mm_setzero_pd();
 209             fjz0             = _mm_setzero_pd();
 210
 211             /**************************
 212              * CALCULATE INTERACTIONS *
 213              **************************/
 214
 215             r10              = _mm_mul_pd(rsq10,rinv10);
 216
 217             /* Compute parameters for interactions between i and j atoms */
 218             qq10             = _mm_mul_pd(iq1,jq0);
 219
 220             /* EWALD ELECTROSTATICS */
 221
 222             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 223             ewrt             = _mm_mul_pd(r10,ewtabscale);
 224             ewitab           = _mm_cvttpd_epi32(ewrt);
 225 #ifdef __XOP__
 226             eweps            = _mm_frcz_pd(ewrt);
 227 #else
 228             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 229 #endif
 230             twoeweps         = _mm_add_pd(eweps,eweps);
 231             ewitab           = _mm_slli_epi32(ewitab,2);
 232             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 233             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 234             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 235             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 236             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 237             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 238             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 239             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 240             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 241             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             velecsum         = _mm_add_pd(velecsum,velec);
 245
 246             fscal            = felec;
 247
 248             /* Update vectorial force */
 249             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 250             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 251             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 252
 253             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 254             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 255             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             r20              = _mm_mul_pd(rsq20,rinv20);
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             qq20             = _mm_mul_pd(iq2,jq0);
 265
 266             /* EWALD ELECTROSTATICS */
 267
 268             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 269             ewrt             = _mm_mul_pd(r20,ewtabscale);
 270             ewitab           = _mm_cvttpd_epi32(ewrt);
 271 #ifdef __XOP__
 272             eweps            = _mm_frcz_pd(ewrt);
 273 #else
 274             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 275 #endif
 276             twoeweps         = _mm_add_pd(eweps,eweps);
 277             ewitab           = _mm_slli_epi32(ewitab,2);
 278             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 279             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 280             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 281             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 282             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 283             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 284             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 285             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 286             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 287             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             velecsum         = _mm_add_pd(velecsum,velec);
 291
 292             fscal            = felec;
 293
 294             /* Update vectorial force */
 295             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 296             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 297             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 298
 299             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 300             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 301             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             r30              = _mm_mul_pd(rsq30,rinv30);
 308
 309             /* Compute parameters for interactions between i and j atoms */
 310             qq30             = _mm_mul_pd(iq3,jq0);
 311
 312             /* EWALD ELECTROSTATICS */
 313
 314             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 315             ewrt             = _mm_mul_pd(r30,ewtabscale);
 316             ewitab           = _mm_cvttpd_epi32(ewrt);
 317 #ifdef __XOP__
 318             eweps            = _mm_frcz_pd(ewrt);
 319 #else
 320             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 321 #endif
 322             twoeweps         = _mm_add_pd(eweps,eweps);
 323             ewitab           = _mm_slli_epi32(ewitab,2);
 324             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 325             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 326             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 327             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 328             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 329             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 330             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 331             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 332             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 333             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velecsum         = _mm_add_pd(velecsum,velec);
 337
 338             fscal            = felec;
 339
 340             /* Update vectorial force */
 341             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 342             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 343             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 344
 345             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 346             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 347             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 348
 349             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 350
 351             /* Inner loop uses 135 flops */
 352         }
 353
 354         if(jidx<j_index_end)
 355         {
 356
 357             jnrA             = jjnr[jidx];
 358             j_coord_offsetA  = DIM*jnrA;
 359
 360             /* load j atom coordinates */
 361             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 362                                               &jx0,&jy0,&jz0);
 363
 364             /* Calculate displacement vector */
 365             dx10             = _mm_sub_pd(ix1,jx0);
 366             dy10             = _mm_sub_pd(iy1,jy0);
 367             dz10             = _mm_sub_pd(iz1,jz0);
 368             dx20             = _mm_sub_pd(ix2,jx0);
 369             dy20             = _mm_sub_pd(iy2,jy0);
 370             dz20             = _mm_sub_pd(iz2,jz0);
 371             dx30             = _mm_sub_pd(ix3,jx0);
 372             dy30             = _mm_sub_pd(iy3,jy0);
 373             dz30             = _mm_sub_pd(iz3,jz0);
 374
 375             /* Calculate squared distance and things based on it */
 376             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 377             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 378             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 379
 380             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 381             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 382             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 383
 384             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 385             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 386             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 387
 388             /* Load parameters for j particles */
 389             jq0              = _mm_load_sd(charge+jnrA+0);
 390
 391             fjx0             = _mm_setzero_pd();
 392             fjy0             = _mm_setzero_pd();
 393             fjz0             = _mm_setzero_pd();
 394
 395             /**************************
 396              * CALCULATE INTERACTIONS *
 397              **************************/
 398
 399             r10              = _mm_mul_pd(rsq10,rinv10);
 400
 401             /* Compute parameters for interactions between i and j atoms */
 402             qq10             = _mm_mul_pd(iq1,jq0);
 403
 404             /* EWALD ELECTROSTATICS */
 405
 406             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 407             ewrt             = _mm_mul_pd(r10,ewtabscale);
 408             ewitab           = _mm_cvttpd_epi32(ewrt);
 409 #ifdef __XOP__
 410             eweps            = _mm_frcz_pd(ewrt);
 411 #else
 412             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 413 #endif
 414             twoeweps         = _mm_add_pd(eweps,eweps);
 415             ewitab           = _mm_slli_epi32(ewitab,2);
 416             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 417             ewtabD           = _mm_setzero_pd();
 418             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 419             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 420             ewtabFn          = _mm_setzero_pd();
 421             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 422             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 423             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 424             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 425             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 426
 427             /* Update potential sum for this i atom from the interaction with this j atom. */
 428             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 429             velecsum         = _mm_add_pd(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 434
 435             /* Update vectorial force */
 436             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 437             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 438             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 439
 440             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 441             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 442             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 443
 444             /**************************
 445              * CALCULATE INTERACTIONS *
 446              **************************/
 447
 448             r20              = _mm_mul_pd(rsq20,rinv20);
 449
 450             /* Compute parameters for interactions between i and j atoms */
 451             qq20             = _mm_mul_pd(iq2,jq0);
 452
 453             /* EWALD ELECTROSTATICS */
 454
 455             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 456             ewrt             = _mm_mul_pd(r20,ewtabscale);
 457             ewitab           = _mm_cvttpd_epi32(ewrt);
 458 #ifdef __XOP__
 459             eweps            = _mm_frcz_pd(ewrt);
 460 #else
 461             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 462 #endif
 463             twoeweps         = _mm_add_pd(eweps,eweps);
 464             ewitab           = _mm_slli_epi32(ewitab,2);
 465             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 466             ewtabD           = _mm_setzero_pd();
 467             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 468             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 469             ewtabFn          = _mm_setzero_pd();
 470             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 471             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 472             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 473             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 474             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 475
 476             /* Update potential sum for this i atom from the interaction with this j atom. */
 477             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 478             velecsum         = _mm_add_pd(velecsum,velec);
 479
 480             fscal            = felec;
 481
 482             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 483
 484             /* Update vectorial force */
 485             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 486             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 487             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 488
 489             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 490             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 491             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 492
 493             /**************************
 494              * CALCULATE INTERACTIONS *
 495              **************************/
 496
 497             r30              = _mm_mul_pd(rsq30,rinv30);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq30             = _mm_mul_pd(iq3,jq0);
 501
 502             /* EWALD ELECTROSTATICS */
 503
 504             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 505             ewrt             = _mm_mul_pd(r30,ewtabscale);
 506             ewitab           = _mm_cvttpd_epi32(ewrt);
 507 #ifdef __XOP__
 508             eweps            = _mm_frcz_pd(ewrt);
 509 #else
 510             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 511 #endif
 512             twoeweps         = _mm_add_pd(eweps,eweps);
 513             ewitab           = _mm_slli_epi32(ewitab,2);
 514             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 515             ewtabD           = _mm_setzero_pd();
 516             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 517             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 518             ewtabFn          = _mm_setzero_pd();
 519             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 520             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 521             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 522             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 523             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 527             velecsum         = _mm_add_pd(velecsum,velec);
 528
 529             fscal            = felec;
 530
 531             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 532
 533             /* Update vectorial force */
 534             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 535             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 536             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 537
 538             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 539             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 540             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 541
 542             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 543
 544             /* Inner loop uses 135 flops */
 545         }
 546
 547         /* End of innermost loop */
 548
 549         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 550                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 551
 552         ggid                        = gid[iidx];
 553         /* Update potential energies */
 554         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 555
 556         /* Increment number of inner iterations */
 557         inneriter                  += j_index_end - j_index_start;
 558
 559         /* Outer loop uses 19 flops */
 560     }
 561
 562     /* Increment number of outer iterations */
 563     outeriter        += nri;
 564
 565     /* Update outer/inner flops */
 566
 567     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*135);
 568 }
 569 /*
 570  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_128_fma_double
 571  * Electrostatics interaction: Ewald
 572  * VdW interaction:            None
 573  * Geometry:                   Water4-Particle
 574  * Calculate force/pot:        Force
 575  */
 576 void
 577 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_128_fma_double
 578                     (t_nblist                    * gmx_restrict       nlist,
 579                      rvec                        * gmx_restrict          xx,
 580                      rvec                        * gmx_restrict          ff,
 581                      t_forcerec                  * gmx_restrict          fr,
 582                      t_mdatoms                   * gmx_restrict     mdatoms,
 583                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 584                      t_nrnb                      * gmx_restrict        nrnb)
 585 {
 586     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 587      * just 0 for non-waters.
 588      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 589      * jnr indices corresponding to data put in the four positions in the SIMD register.
 590      */
 591     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 592     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 593     int              jnrA,jnrB;
 594     int              j_coord_offsetA,j_coord_offsetB;
 595     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 596     real             rcutoff_scalar;
 597     real             *shiftvec,*fshift,*x,*f;
 598     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 599     int              vdwioffset1;
 600     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 601     int              vdwioffset2;
 602     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 603     int              vdwioffset3;
 604     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 605     int              vdwjidx0A,vdwjidx0B;
 606     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 607     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 608     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 609     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 610     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 611     real             *charge;
 612     __m128i          ewitab;
 613     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 614     real             *ewtab;
 615     __m128d          dummy_mask,cutoff_mask;
 616     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 617     __m128d          one     = _mm_set1_pd(1.0);
 618     __m128d          two     = _mm_set1_pd(2.0);
 619     x                = xx[0];
 620     f                = ff[0];
 621
 622     nri              = nlist->nri;
 623     iinr             = nlist->iinr;
 624     jindex           = nlist->jindex;
 625     jjnr             = nlist->jjnr;
 626     shiftidx         = nlist->shift;
 627     gid              = nlist->gid;
 628     shiftvec         = fr->shift_vec[0];
 629     fshift           = fr->fshift[0];
 630     facel            = _mm_set1_pd(fr->epsfac);
 631     charge           = mdatoms->chargeA;
 632
 633     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 634     ewtab            = fr->ic->tabq_coul_F;
 635     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 636     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 637
 638     /* Setup water-specific parameters */
 639     inr              = nlist->iinr[0];
 640     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 641     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 642     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 643
 644     /* Avoid stupid compiler warnings */
 645     jnrA = jnrB = 0;
 646     j_coord_offsetA = 0;
 647     j_coord_offsetB = 0;
 648
 649     outeriter        = 0;
 650     inneriter        = 0;
 651
 652     /* Start outer loop over neighborlists */
 653     for(iidx=0; iidx<nri; iidx++)
 654     {
 655         /* Load shift vector for this list */
 656         i_shift_offset   = DIM*shiftidx[iidx];
 657
 658         /* Load limits for loop over neighbors */
 659         j_index_start    = jindex[iidx];
 660         j_index_end      = jindex[iidx+1];
 661
 662         /* Get outer coordinate index */
 663         inr              = iinr[iidx];
 664         i_coord_offset   = DIM*inr;
 665
 666         /* Load i particle coords and add shift vector */
 667         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 668                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 669
 670         fix1             = _mm_setzero_pd();
 671         fiy1             = _mm_setzero_pd();
 672         fiz1             = _mm_setzero_pd();
 673         fix2             = _mm_setzero_pd();
 674         fiy2             = _mm_setzero_pd();
 675         fiz2             = _mm_setzero_pd();
 676         fix3             = _mm_setzero_pd();
 677         fiy3             = _mm_setzero_pd();
 678         fiz3             = _mm_setzero_pd();
 679
 680         /* Start inner kernel loop */
 681         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 682         {
 683
 684             /* Get j neighbor index, and coordinate index */
 685             jnrA             = jjnr[jidx];
 686             jnrB             = jjnr[jidx+1];
 687             j_coord_offsetA  = DIM*jnrA;
 688             j_coord_offsetB  = DIM*jnrB;
 689
 690             /* load j atom coordinates */
 691             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 692                                               &jx0,&jy0,&jz0);
 693
 694             /* Calculate displacement vector */
 695             dx10             = _mm_sub_pd(ix1,jx0);
 696             dy10             = _mm_sub_pd(iy1,jy0);
 697             dz10             = _mm_sub_pd(iz1,jz0);
 698             dx20             = _mm_sub_pd(ix2,jx0);
 699             dy20             = _mm_sub_pd(iy2,jy0);
 700             dz20             = _mm_sub_pd(iz2,jz0);
 701             dx30             = _mm_sub_pd(ix3,jx0);
 702             dy30             = _mm_sub_pd(iy3,jy0);
 703             dz30             = _mm_sub_pd(iz3,jz0);
 704
 705             /* Calculate squared distance and things based on it */
 706             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 707             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 708             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 709
 710             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 711             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 712             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 713
 714             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 715             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 716             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 717
 718             /* Load parameters for j particles */
 719             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 720
 721             fjx0             = _mm_setzero_pd();
 722             fjy0             = _mm_setzero_pd();
 723             fjz0             = _mm_setzero_pd();
 724
 725             /**************************
 726              * CALCULATE INTERACTIONS *
 727              **************************/
 728
 729             r10              = _mm_mul_pd(rsq10,rinv10);
 730
 731             /* Compute parameters for interactions between i and j atoms */
 732             qq10             = _mm_mul_pd(iq1,jq0);
 733
 734             /* EWALD ELECTROSTATICS */
 735
 736             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 737             ewrt             = _mm_mul_pd(r10,ewtabscale);
 738             ewitab           = _mm_cvttpd_epi32(ewrt);
 739 #ifdef __XOP__
 740             eweps            = _mm_frcz_pd(ewrt);
 741 #else
 742             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 743 #endif
 744             twoeweps         = _mm_add_pd(eweps,eweps);
 745             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 746                                          &ewtabF,&ewtabFn);
 747             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 748             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 749
 750             fscal            = felec;
 751
 752             /* Update vectorial force */
 753             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 754             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 755             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 756
 757             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 758             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 759             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 760
 761             /**************************
 762              * CALCULATE INTERACTIONS *
 763              **************************/
 764
 765             r20              = _mm_mul_pd(rsq20,rinv20);
 766
 767             /* Compute parameters for interactions between i and j atoms */
 768             qq20             = _mm_mul_pd(iq2,jq0);
 769
 770             /* EWALD ELECTROSTATICS */
 771
 772             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 773             ewrt             = _mm_mul_pd(r20,ewtabscale);
 774             ewitab           = _mm_cvttpd_epi32(ewrt);
 775 #ifdef __XOP__
 776             eweps            = _mm_frcz_pd(ewrt);
 777 #else
 778             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 779 #endif
 780             twoeweps         = _mm_add_pd(eweps,eweps);
 781             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 782                                          &ewtabF,&ewtabFn);
 783             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 784             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 785
 786             fscal            = felec;
 787
 788             /* Update vectorial force */
 789             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 790             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 791             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 792
 793             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 794             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 795             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 796
 797             /**************************
 798              * CALCULATE INTERACTIONS *
 799              **************************/
 800
 801             r30              = _mm_mul_pd(rsq30,rinv30);
 802
 803             /* Compute parameters for interactions between i and j atoms */
 804             qq30             = _mm_mul_pd(iq3,jq0);
 805
 806             /* EWALD ELECTROSTATICS */
 807
 808             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 809             ewrt             = _mm_mul_pd(r30,ewtabscale);
 810             ewitab           = _mm_cvttpd_epi32(ewrt);
 811 #ifdef __XOP__
 812             eweps            = _mm_frcz_pd(ewrt);
 813 #else
 814             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 815 #endif
 816             twoeweps         = _mm_add_pd(eweps,eweps);
 817             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 818                                          &ewtabF,&ewtabFn);
 819             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 820             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 821
 822             fscal            = felec;
 823
 824             /* Update vectorial force */
 825             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 826             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 827             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 828
 829             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 830             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 831             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 832
 833             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 834
 835             /* Inner loop uses 120 flops */
 836         }
 837
 838         if(jidx<j_index_end)
 839         {
 840
 841             jnrA             = jjnr[jidx];
 842             j_coord_offsetA  = DIM*jnrA;
 843
 844             /* load j atom coordinates */
 845             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 846                                               &jx0,&jy0,&jz0);
 847
 848             /* Calculate displacement vector */
 849             dx10             = _mm_sub_pd(ix1,jx0);
 850             dy10             = _mm_sub_pd(iy1,jy0);
 851             dz10             = _mm_sub_pd(iz1,jz0);
 852             dx20             = _mm_sub_pd(ix2,jx0);
 853             dy20             = _mm_sub_pd(iy2,jy0);
 854             dz20             = _mm_sub_pd(iz2,jz0);
 855             dx30             = _mm_sub_pd(ix3,jx0);
 856             dy30             = _mm_sub_pd(iy3,jy0);
 857             dz30             = _mm_sub_pd(iz3,jz0);
 858
 859             /* Calculate squared distance and things based on it */
 860             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 861             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 862             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 863
 864             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 865             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 866             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 867
 868             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 869             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 870             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 871
 872             /* Load parameters for j particles */
 873             jq0              = _mm_load_sd(charge+jnrA+0);
 874
 875             fjx0             = _mm_setzero_pd();
 876             fjy0             = _mm_setzero_pd();
 877             fjz0             = _mm_setzero_pd();
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             r10              = _mm_mul_pd(rsq10,rinv10);
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             qq10             = _mm_mul_pd(iq1,jq0);
 887
 888             /* EWALD ELECTROSTATICS */
 889
 890             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 891             ewrt             = _mm_mul_pd(r10,ewtabscale);
 892             ewitab           = _mm_cvttpd_epi32(ewrt);
 893 #ifdef __XOP__
 894             eweps            = _mm_frcz_pd(ewrt);
 895 #else
 896             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 897 #endif
 898             twoeweps         = _mm_add_pd(eweps,eweps);
 899             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 900             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 901             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 902
 903             fscal            = felec;
 904
 905             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 906
 907             /* Update vectorial force */
 908             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 909             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 910             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 911
 912             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 913             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 914             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 915
 916             /**************************
 917              * CALCULATE INTERACTIONS *
 918              **************************/
 919
 920             r20              = _mm_mul_pd(rsq20,rinv20);
 921
 922             /* Compute parameters for interactions between i and j atoms */
 923             qq20             = _mm_mul_pd(iq2,jq0);
 924
 925             /* EWALD ELECTROSTATICS */
 926
 927             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 928             ewrt             = _mm_mul_pd(r20,ewtabscale);
 929             ewitab           = _mm_cvttpd_epi32(ewrt);
 930 #ifdef __XOP__
 931             eweps            = _mm_frcz_pd(ewrt);
 932 #else
 933             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 934 #endif
 935             twoeweps         = _mm_add_pd(eweps,eweps);
 936             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 937             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 938             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 939
 940             fscal            = felec;
 941
 942             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 943
 944             /* Update vectorial force */
 945             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 946             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 947             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 948
 949             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 950             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 951             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 952
 953             /**************************
 954              * CALCULATE INTERACTIONS *
 955              **************************/
 956
 957             r30              = _mm_mul_pd(rsq30,rinv30);
 958
 959             /* Compute parameters for interactions between i and j atoms */
 960             qq30             = _mm_mul_pd(iq3,jq0);
 961
 962             /* EWALD ELECTROSTATICS */
 963
 964             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 965             ewrt             = _mm_mul_pd(r30,ewtabscale);
 966             ewitab           = _mm_cvttpd_epi32(ewrt);
 967 #ifdef __XOP__
 968             eweps            = _mm_frcz_pd(ewrt);
 969 #else
 970             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 971 #endif
 972             twoeweps         = _mm_add_pd(eweps,eweps);
 973             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 974             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 975             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 976
 977             fscal            = felec;
 978
 979             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 980
 981             /* Update vectorial force */
 982             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 983             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 984             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 985
 986             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 987             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 988             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 989
 990             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 991
 992             /* Inner loop uses 120 flops */
 993         }
 994
 995         /* End of innermost loop */
 996
 997         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 998                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 999
1000         /* Increment number of inner iterations */
1001         inneriter                  += j_index_end - j_index_start;
1002
1003         /* Outer loop uses 18 flops */
1004     }
1005
1006     /* Increment number of outer iterations */
1007     outeriter        += nri;
1008
1009     /* Update outer/inner flops */
1010
1011     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);
1012 }