src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJ_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  97     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  98     __m128i          ewitab;
  99     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 100     real             *ewtab;
 101     __m128d          dummy_mask,cutoff_mask;
 102     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 103     __m128d          one     = _mm_set1_pd(1.0);
 104     __m128d          two     = _mm_set1_pd(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_pd(fr->ic->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 123     ewtab            = fr->ic->tabq_coul_FDV0;
 124     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 125     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 130     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 131     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 132     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 158                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 159
 160         fix0             = _mm_setzero_pd();
 161         fiy0             = _mm_setzero_pd();
 162         fiz0             = _mm_setzero_pd();
 163         fix1             = _mm_setzero_pd();
 164         fiy1             = _mm_setzero_pd();
 165         fiz1             = _mm_setzero_pd();
 166         fix2             = _mm_setzero_pd();
 167         fiy2             = _mm_setzero_pd();
 168         fiz2             = _mm_setzero_pd();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_pd();
 172         vvdwsum          = _mm_setzero_pd();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_pd(ix0,jx0);
 190             dy00             = _mm_sub_pd(iy0,jy0);
 191             dz00             = _mm_sub_pd(iz0,jz0);
 192             dx10             = _mm_sub_pd(ix1,jx0);
 193             dy10             = _mm_sub_pd(iy1,jy0);
 194             dz10             = _mm_sub_pd(iz1,jz0);
 195             dx20             = _mm_sub_pd(ix2,jx0);
 196             dy20             = _mm_sub_pd(iy2,jy0);
 197             dz20             = _mm_sub_pd(iz2,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 201             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 202             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 203
 204             rinv00           = avx128fma_invsqrt_d(rsq00);
 205             rinv10           = avx128fma_invsqrt_d(rsq10);
 206             rinv20           = avx128fma_invsqrt_d(rsq20);
 207
 208             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 209             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 210             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216
 217             fjx0             = _mm_setzero_pd();
 218             fjy0             = _mm_setzero_pd();
 219             fjz0             = _mm_setzero_pd();
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             r00              = _mm_mul_pd(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             qq00             = _mm_mul_pd(iq0,jq0);
 229             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 230                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 231
 232             /* EWALD ELECTROSTATICS */
 233
 234             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 235             ewrt             = _mm_mul_pd(r00,ewtabscale);
 236             ewitab           = _mm_cvttpd_epi32(ewrt);
 237 #ifdef __XOP__
 238             eweps            = _mm_frcz_pd(ewrt);
 239 #else
 240             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 241 #endif
 242             twoeweps         = _mm_add_pd(eweps,eweps);
 243             ewitab           = _mm_slli_epi32(ewitab,2);
 244             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 245             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 246             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 247             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 248             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 249             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 250             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 251             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 252             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 253             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 254
 255             /* LENNARD-JONES DISPERSION/REPULSION */
 256
 257             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 258             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 259             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 260             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 261             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             velecsum         = _mm_add_pd(velecsum,velec);
 265             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 266
 267             fscal            = _mm_add_pd(felec,fvdw);
 268
 269             /* Update vectorial force */
 270             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 271             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 272             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 273
 274             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 275             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 276             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             r10              = _mm_mul_pd(rsq10,rinv10);
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             qq10             = _mm_mul_pd(iq1,jq0);
 286
 287             /* EWALD ELECTROSTATICS */
 288
 289             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 290             ewrt             = _mm_mul_pd(r10,ewtabscale);
 291             ewitab           = _mm_cvttpd_epi32(ewrt);
 292 #ifdef __XOP__
 293             eweps            = _mm_frcz_pd(ewrt);
 294 #else
 295             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 296 #endif
 297             twoeweps         = _mm_add_pd(eweps,eweps);
 298             ewitab           = _mm_slli_epi32(ewitab,2);
 299             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 300             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 301             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 302             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 303             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 304             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 305             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 306             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 307             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 308             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             velecsum         = _mm_add_pd(velecsum,velec);
 312
 313             fscal            = felec;
 314
 315             /* Update vectorial force */
 316             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 317             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 318             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 319
 320             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 321             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 322             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             r20              = _mm_mul_pd(rsq20,rinv20);
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             qq20             = _mm_mul_pd(iq2,jq0);
 332
 333             /* EWALD ELECTROSTATICS */
 334
 335             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 336             ewrt             = _mm_mul_pd(r20,ewtabscale);
 337             ewitab           = _mm_cvttpd_epi32(ewrt);
 338 #ifdef __XOP__
 339             eweps            = _mm_frcz_pd(ewrt);
 340 #else
 341             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 342 #endif
 343             twoeweps         = _mm_add_pd(eweps,eweps);
 344             ewitab           = _mm_slli_epi32(ewitab,2);
 345             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 346             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 347             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 348             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 349             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 350             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 351             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 352             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 353             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 354             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velecsum         = _mm_add_pd(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361             /* Update vectorial force */
 362             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 363             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 364             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 365
 366             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 367             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 368             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 369
 370             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 371
 372             /* Inner loop uses 147 flops */
 373         }
 374
 375         if(jidx<j_index_end)
 376         {
 377
 378             jnrA             = jjnr[jidx];
 379             j_coord_offsetA  = DIM*jnrA;
 380
 381             /* load j atom coordinates */
 382             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 383                                               &jx0,&jy0,&jz0);
 384
 385             /* Calculate displacement vector */
 386             dx00             = _mm_sub_pd(ix0,jx0);
 387             dy00             = _mm_sub_pd(iy0,jy0);
 388             dz00             = _mm_sub_pd(iz0,jz0);
 389             dx10             = _mm_sub_pd(ix1,jx0);
 390             dy10             = _mm_sub_pd(iy1,jy0);
 391             dz10             = _mm_sub_pd(iz1,jz0);
 392             dx20             = _mm_sub_pd(ix2,jx0);
 393             dy20             = _mm_sub_pd(iy2,jy0);
 394             dz20             = _mm_sub_pd(iz2,jz0);
 395
 396             /* Calculate squared distance and things based on it */
 397             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 398             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 399             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 400
 401             rinv00           = avx128fma_invsqrt_d(rsq00);
 402             rinv10           = avx128fma_invsqrt_d(rsq10);
 403             rinv20           = avx128fma_invsqrt_d(rsq20);
 404
 405             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 406             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 407             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 408
 409             /* Load parameters for j particles */
 410             jq0              = _mm_load_sd(charge+jnrA+0);
 411             vdwjidx0A        = 2*vdwtype[jnrA+0];
 412
 413             fjx0             = _mm_setzero_pd();
 414             fjy0             = _mm_setzero_pd();
 415             fjz0             = _mm_setzero_pd();
 416
 417             /**************************
 418              * CALCULATE INTERACTIONS *
 419              **************************/
 420
 421             r00              = _mm_mul_pd(rsq00,rinv00);
 422
 423             /* Compute parameters for interactions between i and j atoms */
 424             qq00             = _mm_mul_pd(iq0,jq0);
 425             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 426
 427             /* EWALD ELECTROSTATICS */
 428
 429             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 430             ewrt             = _mm_mul_pd(r00,ewtabscale);
 431             ewitab           = _mm_cvttpd_epi32(ewrt);
 432 #ifdef __XOP__
 433             eweps            = _mm_frcz_pd(ewrt);
 434 #else
 435             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 436 #endif
 437             twoeweps         = _mm_add_pd(eweps,eweps);
 438             ewitab           = _mm_slli_epi32(ewitab,2);
 439             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 440             ewtabD           = _mm_setzero_pd();
 441             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 442             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 443             ewtabFn          = _mm_setzero_pd();
 444             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 445             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 446             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 447             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 448             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 449
 450             /* LENNARD-JONES DISPERSION/REPULSION */
 451
 452             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 453             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 454             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 455             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 456             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 460             velecsum         = _mm_add_pd(velecsum,velec);
 461             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 462             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 463
 464             fscal            = _mm_add_pd(felec,fvdw);
 465
 466             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 467
 468             /* Update vectorial force */
 469             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 470             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 471             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 472
 473             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 474             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 475             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             r10              = _mm_mul_pd(rsq10,rinv10);
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq10             = _mm_mul_pd(iq1,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(r10,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(qq10,_mm_sub_pd(rinv10,velec));
 507             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,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             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 519             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 520             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 521
 522             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 523             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 524             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             r20              = _mm_mul_pd(rsq20,rinv20);
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq20             = _mm_mul_pd(iq2,jq0);
 534
 535             /* EWALD ELECTROSTATICS */
 536
 537             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 538             ewrt             = _mm_mul_pd(r20,ewtabscale);
 539             ewitab           = _mm_cvttpd_epi32(ewrt);
 540 #ifdef __XOP__
 541             eweps            = _mm_frcz_pd(ewrt);
 542 #else
 543             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 544 #endif
 545             twoeweps         = _mm_add_pd(eweps,eweps);
 546             ewitab           = _mm_slli_epi32(ewitab,2);
 547             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 548             ewtabD           = _mm_setzero_pd();
 549             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 550             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 551             ewtabFn          = _mm_setzero_pd();
 552             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 553             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 554             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 555             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 556             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 560             velecsum         = _mm_add_pd(velecsum,velec);
 561
 562             fscal            = felec;
 563
 564             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 565
 566             /* Update vectorial force */
 567             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 568             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 569             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 570
 571             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 572             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 573             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 574
 575             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 576
 577             /* Inner loop uses 147 flops */
 578         }
 579
 580         /* End of innermost loop */
 581
 582         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 583                                               f+i_coord_offset,fshift+i_shift_offset);
 584
 585         ggid                        = gid[iidx];
 586         /* Update potential energies */
 587         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 588         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 589
 590         /* Increment number of inner iterations */
 591         inneriter                  += j_index_end - j_index_start;
 592
 593         /* Outer loop uses 20 flops */
 594     }
 595
 596     /* Increment number of outer iterations */
 597     outeriter        += nri;
 598
 599     /* Update outer/inner flops */
 600
 601     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*147);
 602 }
 603 /*
 604  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_double
 605  * Electrostatics interaction: Ewald
 606  * VdW interaction:            LennardJones
 607  * Geometry:                   Water3-Particle
 608  * Calculate force/pot:        Force
 609  */
 610 void
 611 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_double
 612                     (t_nblist                    * gmx_restrict       nlist,
 613                      rvec                        * gmx_restrict          xx,
 614                      rvec                        * gmx_restrict          ff,
 615                      struct t_forcerec           * gmx_restrict          fr,
 616                      t_mdatoms                   * gmx_restrict     mdatoms,
 617                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 618                      t_nrnb                      * gmx_restrict        nrnb)
 619 {
 620     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 621      * just 0 for non-waters.
 622      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 623      * jnr indices corresponding to data put in the four positions in the SIMD register.
 624      */
 625     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 626     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 627     int              jnrA,jnrB;
 628     int              j_coord_offsetA,j_coord_offsetB;
 629     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 630     real             rcutoff_scalar;
 631     real             *shiftvec,*fshift,*x,*f;
 632     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 633     int              vdwioffset0;
 634     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 635     int              vdwioffset1;
 636     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 637     int              vdwioffset2;
 638     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 639     int              vdwjidx0A,vdwjidx0B;
 640     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 641     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 642     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 643     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 644     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 645     real             *charge;
 646     int              nvdwtype;
 647     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 648     int              *vdwtype;
 649     real             *vdwparam;
 650     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 651     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 652     __m128i          ewitab;
 653     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 654     real             *ewtab;
 655     __m128d          dummy_mask,cutoff_mask;
 656     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 657     __m128d          one     = _mm_set1_pd(1.0);
 658     __m128d          two     = _mm_set1_pd(2.0);
 659     x                = xx[0];
 660     f                = ff[0];
 661
 662     nri              = nlist->nri;
 663     iinr             = nlist->iinr;
 664     jindex           = nlist->jindex;
 665     jjnr             = nlist->jjnr;
 666     shiftidx         = nlist->shift;
 667     gid              = nlist->gid;
 668     shiftvec         = fr->shift_vec[0];
 669     fshift           = fr->fshift[0];
 670     facel            = _mm_set1_pd(fr->ic->epsfac);
 671     charge           = mdatoms->chargeA;
 672     nvdwtype         = fr->ntype;
 673     vdwparam         = fr->nbfp;
 674     vdwtype          = mdatoms->typeA;
 675
 676     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 677     ewtab            = fr->ic->tabq_coul_F;
 678     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 679     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 680
 681     /* Setup water-specific parameters */
 682     inr              = nlist->iinr[0];
 683     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 684     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 685     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 686     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 687
 688     /* Avoid stupid compiler warnings */
 689     jnrA = jnrB = 0;
 690     j_coord_offsetA = 0;
 691     j_coord_offsetB = 0;
 692
 693     outeriter        = 0;
 694     inneriter        = 0;
 695
 696     /* Start outer loop over neighborlists */
 697     for(iidx=0; iidx<nri; iidx++)
 698     {
 699         /* Load shift vector for this list */
 700         i_shift_offset   = DIM*shiftidx[iidx];
 701
 702         /* Load limits for loop over neighbors */
 703         j_index_start    = jindex[iidx];
 704         j_index_end      = jindex[iidx+1];
 705
 706         /* Get outer coordinate index */
 707         inr              = iinr[iidx];
 708         i_coord_offset   = DIM*inr;
 709
 710         /* Load i particle coords and add shift vector */
 711         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 712                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 713
 714         fix0             = _mm_setzero_pd();
 715         fiy0             = _mm_setzero_pd();
 716         fiz0             = _mm_setzero_pd();
 717         fix1             = _mm_setzero_pd();
 718         fiy1             = _mm_setzero_pd();
 719         fiz1             = _mm_setzero_pd();
 720         fix2             = _mm_setzero_pd();
 721         fiy2             = _mm_setzero_pd();
 722         fiz2             = _mm_setzero_pd();
 723
 724         /* Start inner kernel loop */
 725         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 726         {
 727
 728             /* Get j neighbor index, and coordinate index */
 729             jnrA             = jjnr[jidx];
 730             jnrB             = jjnr[jidx+1];
 731             j_coord_offsetA  = DIM*jnrA;
 732             j_coord_offsetB  = DIM*jnrB;
 733
 734             /* load j atom coordinates */
 735             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 736                                               &jx0,&jy0,&jz0);
 737
 738             /* Calculate displacement vector */
 739             dx00             = _mm_sub_pd(ix0,jx0);
 740             dy00             = _mm_sub_pd(iy0,jy0);
 741             dz00             = _mm_sub_pd(iz0,jz0);
 742             dx10             = _mm_sub_pd(ix1,jx0);
 743             dy10             = _mm_sub_pd(iy1,jy0);
 744             dz10             = _mm_sub_pd(iz1,jz0);
 745             dx20             = _mm_sub_pd(ix2,jx0);
 746             dy20             = _mm_sub_pd(iy2,jy0);
 747             dz20             = _mm_sub_pd(iz2,jz0);
 748
 749             /* Calculate squared distance and things based on it */
 750             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 751             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 752             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 753
 754             rinv00           = avx128fma_invsqrt_d(rsq00);
 755             rinv10           = avx128fma_invsqrt_d(rsq10);
 756             rinv20           = avx128fma_invsqrt_d(rsq20);
 757
 758             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 759             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 760             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 761
 762             /* Load parameters for j particles */
 763             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 764             vdwjidx0A        = 2*vdwtype[jnrA+0];
 765             vdwjidx0B        = 2*vdwtype[jnrB+0];
 766
 767             fjx0             = _mm_setzero_pd();
 768             fjy0             = _mm_setzero_pd();
 769             fjz0             = _mm_setzero_pd();
 770
 771             /**************************
 772              * CALCULATE INTERACTIONS *
 773              **************************/
 774
 775             r00              = _mm_mul_pd(rsq00,rinv00);
 776
 777             /* Compute parameters for interactions between i and j atoms */
 778             qq00             = _mm_mul_pd(iq0,jq0);
 779             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 780                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 781
 782             /* EWALD ELECTROSTATICS */
 783
 784             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 785             ewrt             = _mm_mul_pd(r00,ewtabscale);
 786             ewitab           = _mm_cvttpd_epi32(ewrt);
 787 #ifdef __XOP__
 788             eweps            = _mm_frcz_pd(ewrt);
 789 #else
 790             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 791 #endif
 792             twoeweps         = _mm_add_pd(eweps,eweps);
 793             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 794                                          &ewtabF,&ewtabFn);
 795             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 796             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 797
 798             /* LENNARD-JONES DISPERSION/REPULSION */
 799
 800             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 801             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 802
 803             fscal            = _mm_add_pd(felec,fvdw);
 804
 805             /* Update vectorial force */
 806             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 807             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 808             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 809
 810             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 811             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 812             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 813
 814             /**************************
 815              * CALCULATE INTERACTIONS *
 816              **************************/
 817
 818             r10              = _mm_mul_pd(rsq10,rinv10);
 819
 820             /* Compute parameters for interactions between i and j atoms */
 821             qq10             = _mm_mul_pd(iq1,jq0);
 822
 823             /* EWALD ELECTROSTATICS */
 824
 825             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 826             ewrt             = _mm_mul_pd(r10,ewtabscale);
 827             ewitab           = _mm_cvttpd_epi32(ewrt);
 828 #ifdef __XOP__
 829             eweps            = _mm_frcz_pd(ewrt);
 830 #else
 831             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 832 #endif
 833             twoeweps         = _mm_add_pd(eweps,eweps);
 834             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 835                                          &ewtabF,&ewtabFn);
 836             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 837             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 838
 839             fscal            = felec;
 840
 841             /* Update vectorial force */
 842             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 843             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 844             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 845
 846             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 847             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 848             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             r20              = _mm_mul_pd(rsq20,rinv20);
 855
 856             /* Compute parameters for interactions between i and j atoms */
 857             qq20             = _mm_mul_pd(iq2,jq0);
 858
 859             /* EWALD ELECTROSTATICS */
 860
 861             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 862             ewrt             = _mm_mul_pd(r20,ewtabscale);
 863             ewitab           = _mm_cvttpd_epi32(ewrt);
 864 #ifdef __XOP__
 865             eweps            = _mm_frcz_pd(ewrt);
 866 #else
 867             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 868 #endif
 869             twoeweps         = _mm_add_pd(eweps,eweps);
 870             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 871                                          &ewtabF,&ewtabFn);
 872             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 873             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 874
 875             fscal            = felec;
 876
 877             /* Update vectorial force */
 878             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 879             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 880             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 881
 882             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 883             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 884             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 885
 886             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 887
 888             /* Inner loop uses 127 flops */
 889         }
 890
 891         if(jidx<j_index_end)
 892         {
 893
 894             jnrA             = jjnr[jidx];
 895             j_coord_offsetA  = DIM*jnrA;
 896
 897             /* load j atom coordinates */
 898             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 899                                               &jx0,&jy0,&jz0);
 900
 901             /* Calculate displacement vector */
 902             dx00             = _mm_sub_pd(ix0,jx0);
 903             dy00             = _mm_sub_pd(iy0,jy0);
 904             dz00             = _mm_sub_pd(iz0,jz0);
 905             dx10             = _mm_sub_pd(ix1,jx0);
 906             dy10             = _mm_sub_pd(iy1,jy0);
 907             dz10             = _mm_sub_pd(iz1,jz0);
 908             dx20             = _mm_sub_pd(ix2,jx0);
 909             dy20             = _mm_sub_pd(iy2,jy0);
 910             dz20             = _mm_sub_pd(iz2,jz0);
 911
 912             /* Calculate squared distance and things based on it */
 913             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 914             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 915             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 916
 917             rinv00           = avx128fma_invsqrt_d(rsq00);
 918             rinv10           = avx128fma_invsqrt_d(rsq10);
 919             rinv20           = avx128fma_invsqrt_d(rsq20);
 920
 921             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 922             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 923             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 924
 925             /* Load parameters for j particles */
 926             jq0              = _mm_load_sd(charge+jnrA+0);
 927             vdwjidx0A        = 2*vdwtype[jnrA+0];
 928
 929             fjx0             = _mm_setzero_pd();
 930             fjy0             = _mm_setzero_pd();
 931             fjz0             = _mm_setzero_pd();
 932
 933             /**************************
 934              * CALCULATE INTERACTIONS *
 935              **************************/
 936
 937             r00              = _mm_mul_pd(rsq00,rinv00);
 938
 939             /* Compute parameters for interactions between i and j atoms */
 940             qq00             = _mm_mul_pd(iq0,jq0);
 941             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 942
 943             /* EWALD ELECTROSTATICS */
 944
 945             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 946             ewrt             = _mm_mul_pd(r00,ewtabscale);
 947             ewitab           = _mm_cvttpd_epi32(ewrt);
 948 #ifdef __XOP__
 949             eweps            = _mm_frcz_pd(ewrt);
 950 #else
 951             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 952 #endif
 953             twoeweps         = _mm_add_pd(eweps,eweps);
 954             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 955             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 956             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 957
 958             /* LENNARD-JONES DISPERSION/REPULSION */
 959
 960             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 961             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 962
 963             fscal            = _mm_add_pd(felec,fvdw);
 964
 965             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 966
 967             /* Update vectorial force */
 968             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 969             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 970             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 971
 972             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 973             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 974             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 975
 976             /**************************
 977              * CALCULATE INTERACTIONS *
 978              **************************/
 979
 980             r10              = _mm_mul_pd(rsq10,rinv10);
 981
 982             /* Compute parameters for interactions between i and j atoms */
 983             qq10             = _mm_mul_pd(iq1,jq0);
 984
 985             /* EWALD ELECTROSTATICS */
 986
 987             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 988             ewrt             = _mm_mul_pd(r10,ewtabscale);
 989             ewitab           = _mm_cvttpd_epi32(ewrt);
 990 #ifdef __XOP__
 991             eweps            = _mm_frcz_pd(ewrt);
 992 #else
 993             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 994 #endif
 995             twoeweps         = _mm_add_pd(eweps,eweps);
 996             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 997             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 998             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 999
1000             fscal            = felec;
1001
1002             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1003
1004             /* Update vectorial force */
1005             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1006             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1007             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1008
1009             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1010             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1011             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             r20              = _mm_mul_pd(rsq20,rinv20);
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             qq20             = _mm_mul_pd(iq2,jq0);
1021
1022             /* EWALD ELECTROSTATICS */
1023
1024             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1025             ewrt             = _mm_mul_pd(r20,ewtabscale);
1026             ewitab           = _mm_cvttpd_epi32(ewrt);
1027 #ifdef __XOP__
1028             eweps            = _mm_frcz_pd(ewrt);
1029 #else
1030             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1031 #endif
1032             twoeweps         = _mm_add_pd(eweps,eweps);
1033             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1034             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1035             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1036
1037             fscal            = felec;
1038
1039             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1040
1041             /* Update vectorial force */
1042             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1043             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1044             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1045
1046             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1047             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1048             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1049
1050             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1051
1052             /* Inner loop uses 127 flops */
1053         }
1054
1055         /* End of innermost loop */
1056
1057         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1058                                               f+i_coord_offset,fshift+i_shift_offset);
1059
1060         /* Increment number of inner iterations */
1061         inneriter                  += j_index_end - j_index_start;
1062
1063         /* Outer loop uses 18 flops */
1064     }
1065
1066     /* Increment number of outer iterations */
1067     outeriter        += nri;
1068
1069     /* Update outer/inner flops */
1070
1071     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);
1072 }