src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwNone_GeomW4P1_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_GeomW4P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwNone_GeomW4P1_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              vdwioffset1;
  67     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  68     int              vdwioffset2;
  69     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  70     int              vdwioffset3;
  71     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  75     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  76     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m128i          ewitab;
  80     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  81     real             *ewtab;
  82     __m128d          dummy_mask,cutoff_mask;
  83     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  84     __m128d          one     = _mm_set1_pd(1.0);
  85     __m128d          two     = _mm_set1_pd(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     facel            = _mm_set1_pd(fr->epsfac);
  98     charge           = mdatoms->chargeA;
  99
 100     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 101     ewtab            = fr->ic->tabq_coul_FDV0;
 102     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 103     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 104
 105     /* Setup water-specific parameters */
 106     inr              = nlist->iinr[0];
 107     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 108     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 109     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 110
 111     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115
 116     outeriter        = 0;
 117     inneriter        = 0;
 118
 119     /* Start outer loop over neighborlists */
 120     for(iidx=0; iidx<nri; iidx++)
 121     {
 122         /* Load shift vector for this list */
 123         i_shift_offset   = DIM*shiftidx[iidx];
 124
 125         /* Load limits for loop over neighbors */
 126         j_index_start    = jindex[iidx];
 127         j_index_end      = jindex[iidx+1];
 128
 129         /* Get outer coordinate index */
 130         inr              = iinr[iidx];
 131         i_coord_offset   = DIM*inr;
 132
 133         /* Load i particle coords and add shift vector */
 134         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 135                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 136
 137         fix1             = _mm_setzero_pd();
 138         fiy1             = _mm_setzero_pd();
 139         fiz1             = _mm_setzero_pd();
 140         fix2             = _mm_setzero_pd();
 141         fiy2             = _mm_setzero_pd();
 142         fiz2             = _mm_setzero_pd();
 143         fix3             = _mm_setzero_pd();
 144         fiy3             = _mm_setzero_pd();
 145         fiz3             = _mm_setzero_pd();
 146
 147         /* Reset potential sums */
 148         velecsum         = _mm_setzero_pd();
 149
 150         /* Start inner kernel loop */
 151         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 152         {
 153
 154             /* Get j neighbor index, and coordinate index */
 155             jnrA             = jjnr[jidx];
 156             jnrB             = jjnr[jidx+1];
 157             j_coord_offsetA  = DIM*jnrA;
 158             j_coord_offsetB  = DIM*jnrB;
 159
 160             /* load j atom coordinates */
 161             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 162                                               &jx0,&jy0,&jz0);
 163
 164             /* Calculate displacement vector */
 165             dx10             = _mm_sub_pd(ix1,jx0);
 166             dy10             = _mm_sub_pd(iy1,jy0);
 167             dz10             = _mm_sub_pd(iz1,jz0);
 168             dx20             = _mm_sub_pd(ix2,jx0);
 169             dy20             = _mm_sub_pd(iy2,jy0);
 170             dz20             = _mm_sub_pd(iz2,jz0);
 171             dx30             = _mm_sub_pd(ix3,jx0);
 172             dy30             = _mm_sub_pd(iy3,jy0);
 173             dz30             = _mm_sub_pd(iz3,jz0);
 174
 175             /* Calculate squared distance and things based on it */
 176             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 177             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 178             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 179
 180             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 181             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 182             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 183
 184             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 185             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 186             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 187
 188             /* Load parameters for j particles */
 189             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 190
 191             fjx0             = _mm_setzero_pd();
 192             fjy0             = _mm_setzero_pd();
 193             fjz0             = _mm_setzero_pd();
 194
 195             /**************************
 196              * CALCULATE INTERACTIONS *
 197              **************************/
 198
 199             r10              = _mm_mul_pd(rsq10,rinv10);
 200
 201             /* Compute parameters for interactions between i and j atoms */
 202             qq10             = _mm_mul_pd(iq1,jq0);
 203
 204             /* EWALD ELECTROSTATICS */
 205
 206             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 207             ewrt             = _mm_mul_pd(r10,ewtabscale);
 208             ewitab           = _mm_cvttpd_epi32(ewrt);
 209 #ifdef __XOP__
 210             eweps            = _mm_frcz_pd(ewrt);
 211 #else
 212             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 213 #endif
 214             twoeweps         = _mm_add_pd(eweps,eweps);
 215             ewitab           = _mm_slli_epi32(ewitab,2);
 216             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 217             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 218             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 219             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 220             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 221             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 222             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 223             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 224             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 225             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             velecsum         = _mm_add_pd(velecsum,velec);
 229
 230             fscal            = felec;
 231
 232             /* Update vectorial force */
 233             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 234             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 235             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 236
 237             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 238             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 239             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             r20              = _mm_mul_pd(rsq20,rinv20);
 246
 247             /* Compute parameters for interactions between i and j atoms */
 248             qq20             = _mm_mul_pd(iq2,jq0);
 249
 250             /* EWALD ELECTROSTATICS */
 251
 252             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 253             ewrt             = _mm_mul_pd(r20,ewtabscale);
 254             ewitab           = _mm_cvttpd_epi32(ewrt);
 255 #ifdef __XOP__
 256             eweps            = _mm_frcz_pd(ewrt);
 257 #else
 258             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 259 #endif
 260             twoeweps         = _mm_add_pd(eweps,eweps);
 261             ewitab           = _mm_slli_epi32(ewitab,2);
 262             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 263             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 264             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 265             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 266             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 267             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 268             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 269             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 270             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 271             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 272
 273             /* Update potential sum for this i atom from the interaction with this j atom. */
 274             velecsum         = _mm_add_pd(velecsum,velec);
 275
 276             fscal            = felec;
 277
 278             /* Update vectorial force */
 279             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 280             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 281             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 282
 283             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 284             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 285             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             r30              = _mm_mul_pd(rsq30,rinv30);
 292
 293             /* Compute parameters for interactions between i and j atoms */
 294             qq30             = _mm_mul_pd(iq3,jq0);
 295
 296             /* EWALD ELECTROSTATICS */
 297
 298             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 299             ewrt             = _mm_mul_pd(r30,ewtabscale);
 300             ewitab           = _mm_cvttpd_epi32(ewrt);
 301 #ifdef __XOP__
 302             eweps            = _mm_frcz_pd(ewrt);
 303 #else
 304             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 305 #endif
 306             twoeweps         = _mm_add_pd(eweps,eweps);
 307             ewitab           = _mm_slli_epi32(ewitab,2);
 308             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 309             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 310             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 311             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 312             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 313             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 314             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 315             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 316             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 317             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velecsum         = _mm_add_pd(velecsum,velec);
 321
 322             fscal            = felec;
 323
 324             /* Update vectorial force */
 325             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 326             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 327             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 328
 329             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 330             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 331             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 332
 333             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 334
 335             /* Inner loop uses 135 flops */
 336         }
 337
 338         if(jidx<j_index_end)
 339         {
 340
 341             jnrA             = jjnr[jidx];
 342             j_coord_offsetA  = DIM*jnrA;
 343
 344             /* load j atom coordinates */
 345             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 346                                               &jx0,&jy0,&jz0);
 347
 348             /* Calculate displacement vector */
 349             dx10             = _mm_sub_pd(ix1,jx0);
 350             dy10             = _mm_sub_pd(iy1,jy0);
 351             dz10             = _mm_sub_pd(iz1,jz0);
 352             dx20             = _mm_sub_pd(ix2,jx0);
 353             dy20             = _mm_sub_pd(iy2,jy0);
 354             dz20             = _mm_sub_pd(iz2,jz0);
 355             dx30             = _mm_sub_pd(ix3,jx0);
 356             dy30             = _mm_sub_pd(iy3,jy0);
 357             dz30             = _mm_sub_pd(iz3,jz0);
 358
 359             /* Calculate squared distance and things based on it */
 360             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 361             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 362             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 363
 364             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 365             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 366             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 367
 368             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 369             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 370             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 371
 372             /* Load parameters for j particles */
 373             jq0              = _mm_load_sd(charge+jnrA+0);
 374
 375             fjx0             = _mm_setzero_pd();
 376             fjy0             = _mm_setzero_pd();
 377             fjz0             = _mm_setzero_pd();
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             r10              = _mm_mul_pd(rsq10,rinv10);
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq10             = _mm_mul_pd(iq1,jq0);
 387
 388             /* EWALD ELECTROSTATICS */
 389
 390             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 391             ewrt             = _mm_mul_pd(r10,ewtabscale);
 392             ewitab           = _mm_cvttpd_epi32(ewrt);
 393 #ifdef __XOP__
 394             eweps            = _mm_frcz_pd(ewrt);
 395 #else
 396             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 397 #endif
 398             twoeweps         = _mm_add_pd(eweps,eweps);
 399             ewitab           = _mm_slli_epi32(ewitab,2);
 400             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 401             ewtabD           = _mm_setzero_pd();
 402             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 403             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 404             ewtabFn          = _mm_setzero_pd();
 405             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 406             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 407             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 408             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 409             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 413             velecsum         = _mm_add_pd(velecsum,velec);
 414
 415             fscal            = felec;
 416
 417             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 418
 419             /* Update vectorial force */
 420             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 421             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 422             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 423
 424             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 425             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 426             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             r20              = _mm_mul_pd(rsq20,rinv20);
 433
 434             /* Compute parameters for interactions between i and j atoms */
 435             qq20             = _mm_mul_pd(iq2,jq0);
 436
 437             /* EWALD ELECTROSTATICS */
 438
 439             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 440             ewrt             = _mm_mul_pd(r20,ewtabscale);
 441             ewitab           = _mm_cvttpd_epi32(ewrt);
 442 #ifdef __XOP__
 443             eweps            = _mm_frcz_pd(ewrt);
 444 #else
 445             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 446 #endif
 447             twoeweps         = _mm_add_pd(eweps,eweps);
 448             ewitab           = _mm_slli_epi32(ewitab,2);
 449             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 450             ewtabD           = _mm_setzero_pd();
 451             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 452             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 453             ewtabFn          = _mm_setzero_pd();
 454             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 455             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 456             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 457             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 458             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 459
 460             /* Update potential sum for this i atom from the interaction with this j atom. */
 461             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 462             velecsum         = _mm_add_pd(velecsum,velec);
 463
 464             fscal            = felec;
 465
 466             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 467
 468             /* Update vectorial force */
 469             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 470             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 471             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 472
 473             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 474             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 475             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             r30              = _mm_mul_pd(rsq30,rinv30);
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq30             = _mm_mul_pd(iq3,jq0);
 485
 486             /* EWALD ELECTROSTATICS */
 487
 488             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 489             ewrt             = _mm_mul_pd(r30,ewtabscale);
 490             ewitab           = _mm_cvttpd_epi32(ewrt);
 491 #ifdef __XOP__
 492             eweps            = _mm_frcz_pd(ewrt);
 493 #else
 494             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 495 #endif
 496             twoeweps         = _mm_add_pd(eweps,eweps);
 497             ewitab           = _mm_slli_epi32(ewitab,2);
 498             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 499             ewtabD           = _mm_setzero_pd();
 500             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 501             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 502             ewtabFn          = _mm_setzero_pd();
 503             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 504             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 505             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 506             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 507             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 508
 509             /* Update potential sum for this i atom from the interaction with this j atom. */
 510             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 511             velecsum         = _mm_add_pd(velecsum,velec);
 512
 513             fscal            = felec;
 514
 515             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 516
 517             /* Update vectorial force */
 518             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 519             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 520             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 521
 522             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 523             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 524             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 525
 526             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 527
 528             /* Inner loop uses 135 flops */
 529         }
 530
 531         /* End of innermost loop */
 532
 533         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 534                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 535
 536         ggid                        = gid[iidx];
 537         /* Update potential energies */
 538         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 539
 540         /* Increment number of inner iterations */
 541         inneriter                  += j_index_end - j_index_start;
 542
 543         /* Outer loop uses 19 flops */
 544     }
 545
 546     /* Increment number of outer iterations */
 547     outeriter        += nri;
 548
 549     /* Update outer/inner flops */
 550
 551     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*135);
 552 }
 553 /*
 554  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_128_fma_double
 555  * Electrostatics interaction: Ewald
 556  * VdW interaction:            None
 557  * Geometry:                   Water4-Particle
 558  * Calculate force/pot:        Force
 559  */
 560 void
 561 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_avx_128_fma_double
 562                     (t_nblist * gmx_restrict                nlist,
 563                      rvec * gmx_restrict                    xx,
 564                      rvec * gmx_restrict                    ff,
 565                      t_forcerec * gmx_restrict              fr,
 566                      t_mdatoms * gmx_restrict               mdatoms,
 567                      nb_kernel_data_t * gmx_restrict        kernel_data,
 568                      t_nrnb * gmx_restrict                  nrnb)
 569 {
 570     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 571      * just 0 for non-waters.
 572      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 573      * jnr indices corresponding to data put in the four positions in the SIMD register.
 574      */
 575     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 576     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 577     int              jnrA,jnrB;
 578     int              j_coord_offsetA,j_coord_offsetB;
 579     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 580     real             rcutoff_scalar;
 581     real             *shiftvec,*fshift,*x,*f;
 582     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 583     int              vdwioffset1;
 584     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 585     int              vdwioffset2;
 586     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 587     int              vdwioffset3;
 588     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 589     int              vdwjidx0A,vdwjidx0B;
 590     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 591     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 592     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 593     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 594     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 595     real             *charge;
 596     __m128i          ewitab;
 597     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 598     real             *ewtab;
 599     __m128d          dummy_mask,cutoff_mask;
 600     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 601     __m128d          one     = _mm_set1_pd(1.0);
 602     __m128d          two     = _mm_set1_pd(2.0);
 603     x                = xx[0];
 604     f                = ff[0];
 605
 606     nri              = nlist->nri;
 607     iinr             = nlist->iinr;
 608     jindex           = nlist->jindex;
 609     jjnr             = nlist->jjnr;
 610     shiftidx         = nlist->shift;
 611     gid              = nlist->gid;
 612     shiftvec         = fr->shift_vec[0];
 613     fshift           = fr->fshift[0];
 614     facel            = _mm_set1_pd(fr->epsfac);
 615     charge           = mdatoms->chargeA;
 616
 617     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 618     ewtab            = fr->ic->tabq_coul_F;
 619     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 620     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 621
 622     /* Setup water-specific parameters */
 623     inr              = nlist->iinr[0];
 624     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 625     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 626     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 627
 628     /* Avoid stupid compiler warnings */
 629     jnrA = jnrB = 0;
 630     j_coord_offsetA = 0;
 631     j_coord_offsetB = 0;
 632
 633     outeriter        = 0;
 634     inneriter        = 0;
 635
 636     /* Start outer loop over neighborlists */
 637     for(iidx=0; iidx<nri; iidx++)
 638     {
 639         /* Load shift vector for this list */
 640         i_shift_offset   = DIM*shiftidx[iidx];
 641
 642         /* Load limits for loop over neighbors */
 643         j_index_start    = jindex[iidx];
 644         j_index_end      = jindex[iidx+1];
 645
 646         /* Get outer coordinate index */
 647         inr              = iinr[iidx];
 648         i_coord_offset   = DIM*inr;
 649
 650         /* Load i particle coords and add shift vector */
 651         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 652                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 653
 654         fix1             = _mm_setzero_pd();
 655         fiy1             = _mm_setzero_pd();
 656         fiz1             = _mm_setzero_pd();
 657         fix2             = _mm_setzero_pd();
 658         fiy2             = _mm_setzero_pd();
 659         fiz2             = _mm_setzero_pd();
 660         fix3             = _mm_setzero_pd();
 661         fiy3             = _mm_setzero_pd();
 662         fiz3             = _mm_setzero_pd();
 663
 664         /* Start inner kernel loop */
 665         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 666         {
 667
 668             /* Get j neighbor index, and coordinate index */
 669             jnrA             = jjnr[jidx];
 670             jnrB             = jjnr[jidx+1];
 671             j_coord_offsetA  = DIM*jnrA;
 672             j_coord_offsetB  = DIM*jnrB;
 673
 674             /* load j atom coordinates */
 675             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 676                                               &jx0,&jy0,&jz0);
 677
 678             /* Calculate displacement vector */
 679             dx10             = _mm_sub_pd(ix1,jx0);
 680             dy10             = _mm_sub_pd(iy1,jy0);
 681             dz10             = _mm_sub_pd(iz1,jz0);
 682             dx20             = _mm_sub_pd(ix2,jx0);
 683             dy20             = _mm_sub_pd(iy2,jy0);
 684             dz20             = _mm_sub_pd(iz2,jz0);
 685             dx30             = _mm_sub_pd(ix3,jx0);
 686             dy30             = _mm_sub_pd(iy3,jy0);
 687             dz30             = _mm_sub_pd(iz3,jz0);
 688
 689             /* Calculate squared distance and things based on it */
 690             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 691             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 692             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 693
 694             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 695             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 696             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 697
 698             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 699             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 700             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 701
 702             /* Load parameters for j particles */
 703             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 704
 705             fjx0             = _mm_setzero_pd();
 706             fjy0             = _mm_setzero_pd();
 707             fjz0             = _mm_setzero_pd();
 708
 709             /**************************
 710              * CALCULATE INTERACTIONS *
 711              **************************/
 712
 713             r10              = _mm_mul_pd(rsq10,rinv10);
 714
 715             /* Compute parameters for interactions between i and j atoms */
 716             qq10             = _mm_mul_pd(iq1,jq0);
 717
 718             /* EWALD ELECTROSTATICS */
 719
 720             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 721             ewrt             = _mm_mul_pd(r10,ewtabscale);
 722             ewitab           = _mm_cvttpd_epi32(ewrt);
 723 #ifdef __XOP__
 724             eweps            = _mm_frcz_pd(ewrt);
 725 #else
 726             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 727 #endif
 728             twoeweps         = _mm_add_pd(eweps,eweps);
 729             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 730                                          &ewtabF,&ewtabFn);
 731             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 732             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 733
 734             fscal            = felec;
 735
 736             /* Update vectorial force */
 737             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 738             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 739             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 740
 741             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 742             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 743             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 744
 745             /**************************
 746              * CALCULATE INTERACTIONS *
 747              **************************/
 748
 749             r20              = _mm_mul_pd(rsq20,rinv20);
 750
 751             /* Compute parameters for interactions between i and j atoms */
 752             qq20             = _mm_mul_pd(iq2,jq0);
 753
 754             /* EWALD ELECTROSTATICS */
 755
 756             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 757             ewrt             = _mm_mul_pd(r20,ewtabscale);
 758             ewitab           = _mm_cvttpd_epi32(ewrt);
 759 #ifdef __XOP__
 760             eweps            = _mm_frcz_pd(ewrt);
 761 #else
 762             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 763 #endif
 764             twoeweps         = _mm_add_pd(eweps,eweps);
 765             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 766                                          &ewtabF,&ewtabFn);
 767             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 768             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 769
 770             fscal            = felec;
 771
 772             /* Update vectorial force */
 773             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 774             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 775             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 776
 777             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 778             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 779             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 780
 781             /**************************
 782              * CALCULATE INTERACTIONS *
 783              **************************/
 784
 785             r30              = _mm_mul_pd(rsq30,rinv30);
 786
 787             /* Compute parameters for interactions between i and j atoms */
 788             qq30             = _mm_mul_pd(iq3,jq0);
 789
 790             /* EWALD ELECTROSTATICS */
 791
 792             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 793             ewrt             = _mm_mul_pd(r30,ewtabscale);
 794             ewitab           = _mm_cvttpd_epi32(ewrt);
 795 #ifdef __XOP__
 796             eweps            = _mm_frcz_pd(ewrt);
 797 #else
 798             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 799 #endif
 800             twoeweps         = _mm_add_pd(eweps,eweps);
 801             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 802                                          &ewtabF,&ewtabFn);
 803             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 804             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 805
 806             fscal            = felec;
 807
 808             /* Update vectorial force */
 809             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 810             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 811             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 812
 813             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 814             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 815             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 816
 817             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 818
 819             /* Inner loop uses 120 flops */
 820         }
 821
 822         if(jidx<j_index_end)
 823         {
 824
 825             jnrA             = jjnr[jidx];
 826             j_coord_offsetA  = DIM*jnrA;
 827
 828             /* load j atom coordinates */
 829             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 830                                               &jx0,&jy0,&jz0);
 831
 832             /* Calculate displacement vector */
 833             dx10             = _mm_sub_pd(ix1,jx0);
 834             dy10             = _mm_sub_pd(iy1,jy0);
 835             dz10             = _mm_sub_pd(iz1,jz0);
 836             dx20             = _mm_sub_pd(ix2,jx0);
 837             dy20             = _mm_sub_pd(iy2,jy0);
 838             dz20             = _mm_sub_pd(iz2,jz0);
 839             dx30             = _mm_sub_pd(ix3,jx0);
 840             dy30             = _mm_sub_pd(iy3,jy0);
 841             dz30             = _mm_sub_pd(iz3,jz0);
 842
 843             /* Calculate squared distance and things based on it */
 844             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 845             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 846             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 847
 848             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 849             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 850             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 851
 852             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 853             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 854             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 855
 856             /* Load parameters for j particles */
 857             jq0              = _mm_load_sd(charge+jnrA+0);
 858
 859             fjx0             = _mm_setzero_pd();
 860             fjy0             = _mm_setzero_pd();
 861             fjz0             = _mm_setzero_pd();
 862
 863             /**************************
 864              * CALCULATE INTERACTIONS *
 865              **************************/
 866
 867             r10              = _mm_mul_pd(rsq10,rinv10);
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq10             = _mm_mul_pd(iq1,jq0);
 871
 872             /* EWALD ELECTROSTATICS */
 873
 874             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 875             ewrt             = _mm_mul_pd(r10,ewtabscale);
 876             ewitab           = _mm_cvttpd_epi32(ewrt);
 877 #ifdef __XOP__
 878             eweps            = _mm_frcz_pd(ewrt);
 879 #else
 880             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 881 #endif
 882             twoeweps         = _mm_add_pd(eweps,eweps);
 883             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 884             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 885             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 886
 887             fscal            = felec;
 888
 889             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 890
 891             /* Update vectorial force */
 892             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 893             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 894             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 895
 896             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 897             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 898             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             r20              = _mm_mul_pd(rsq20,rinv20);
 905
 906             /* Compute parameters for interactions between i and j atoms */
 907             qq20             = _mm_mul_pd(iq2,jq0);
 908
 909             /* EWALD ELECTROSTATICS */
 910
 911             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 912             ewrt             = _mm_mul_pd(r20,ewtabscale);
 913             ewitab           = _mm_cvttpd_epi32(ewrt);
 914 #ifdef __XOP__
 915             eweps            = _mm_frcz_pd(ewrt);
 916 #else
 917             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 918 #endif
 919             twoeweps         = _mm_add_pd(eweps,eweps);
 920             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 921             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 922             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 923
 924             fscal            = felec;
 925
 926             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 927
 928             /* Update vectorial force */
 929             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 930             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 931             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 932
 933             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 934             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 935             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 936
 937             /**************************
 938              * CALCULATE INTERACTIONS *
 939              **************************/
 940
 941             r30              = _mm_mul_pd(rsq30,rinv30);
 942
 943             /* Compute parameters for interactions between i and j atoms */
 944             qq30             = _mm_mul_pd(iq3,jq0);
 945
 946             /* EWALD ELECTROSTATICS */
 947
 948             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 949             ewrt             = _mm_mul_pd(r30,ewtabscale);
 950             ewitab           = _mm_cvttpd_epi32(ewrt);
 951 #ifdef __XOP__
 952             eweps            = _mm_frcz_pd(ewrt);
 953 #else
 954             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 955 #endif
 956             twoeweps         = _mm_add_pd(eweps,eweps);
 957             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 958             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 959             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 960
 961             fscal            = felec;
 962
 963             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 964
 965             /* Update vectorial force */
 966             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 967             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 968             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 969
 970             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 971             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 972             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 973
 974             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 975
 976             /* Inner loop uses 120 flops */
 977         }
 978
 979         /* End of innermost loop */
 980
 981         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 982                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 983
 984         /* Increment number of inner iterations */
 985         inneriter                  += j_index_end - j_index_start;
 986
 987         /* Outer loop uses 18 flops */
 988     }
 989
 990     /* Increment number of outer iterations */
 991     outeriter        += nri;
 992
 993     /* Update outer/inner flops */
 994
 995     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);
 996 }