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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          ewitab;
 102     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 103     real             *ewtab;
 104     __m128d          dummy_mask,cutoff_mask;
 105     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 106     __m128d          one     = _mm_set1_pd(1.0);
 107     __m128d          two     = _mm_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 126     ewtab            = fr->ic->tabq_coul_FDV0;
 127     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 128     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 133     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 134     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 161                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 162
 163         fix0             = _mm_setzero_pd();
 164         fiy0             = _mm_setzero_pd();
 165         fiz0             = _mm_setzero_pd();
 166         fix1             = _mm_setzero_pd();
 167         fiy1             = _mm_setzero_pd();
 168         fiz1             = _mm_setzero_pd();
 169         fix2             = _mm_setzero_pd();
 170         fiy2             = _mm_setzero_pd();
 171         fiz2             = _mm_setzero_pd();
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm_setzero_pd();
 175         vvdwsum          = _mm_setzero_pd();
 176
 177         /* Start inner kernel loop */
 178         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 179         {
 180
 181             /* Get j neighbor index, and coordinate index */
 182             jnrA             = jjnr[jidx];
 183             jnrB             = jjnr[jidx+1];
 184             j_coord_offsetA  = DIM*jnrA;
 185             j_coord_offsetB  = DIM*jnrB;
 186
 187             /* load j atom coordinates */
 188             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                               &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm_sub_pd(ix0,jx0);
 193             dy00             = _mm_sub_pd(iy0,jy0);
 194             dz00             = _mm_sub_pd(iz0,jz0);
 195             dx10             = _mm_sub_pd(ix1,jx0);
 196             dy10             = _mm_sub_pd(iy1,jy0);
 197             dz10             = _mm_sub_pd(iz1,jz0);
 198             dx20             = _mm_sub_pd(ix2,jx0);
 199             dy20             = _mm_sub_pd(iy2,jy0);
 200             dz20             = _mm_sub_pd(iz2,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 204             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 205             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 206
 207             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 208             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 209             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 210
 211             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 212             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 213             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 214
 215             /* Load parameters for j particles */
 216             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 217             vdwjidx0A        = 2*vdwtype[jnrA+0];
 218             vdwjidx0B        = 2*vdwtype[jnrB+0];
 219
 220             fjx0             = _mm_setzero_pd();
 221             fjy0             = _mm_setzero_pd();
 222             fjz0             = _mm_setzero_pd();
 223
 224             /**************************
 225              * CALCULATE INTERACTIONS *
 226              **************************/
 227
 228             r00              = _mm_mul_pd(rsq00,rinv00);
 229
 230             /* Compute parameters for interactions between i and j atoms */
 231             qq00             = _mm_mul_pd(iq0,jq0);
 232             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 233                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 234
 235             /* EWALD ELECTROSTATICS */
 236
 237             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 238             ewrt             = _mm_mul_pd(r00,ewtabscale);
 239             ewitab           = _mm_cvttpd_epi32(ewrt);
 240 #ifdef __XOP__
 241             eweps            = _mm_frcz_pd(ewrt);
 242 #else
 243             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 244 #endif
 245             twoeweps         = _mm_add_pd(eweps,eweps);
 246             ewitab           = _mm_slli_epi32(ewitab,2);
 247             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 248             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 249             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 250             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 251             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 252             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 253             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 254             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 255             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 256             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 257
 258             /* LENNARD-JONES DISPERSION/REPULSION */
 259
 260             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 261             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 262             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 263             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 264             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 265
 266             /* Update potential sum for this i atom from the interaction with this j atom. */
 267             velecsum         = _mm_add_pd(velecsum,velec);
 268             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 269
 270             fscal            = _mm_add_pd(felec,fvdw);
 271
 272             /* Update vectorial force */
 273             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 274             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 275             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 276
 277             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 278             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 279             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             r10              = _mm_mul_pd(rsq10,rinv10);
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq10             = _mm_mul_pd(iq1,jq0);
 289
 290             /* EWALD ELECTROSTATICS */
 291
 292             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 293             ewrt             = _mm_mul_pd(r10,ewtabscale);
 294             ewitab           = _mm_cvttpd_epi32(ewrt);
 295 #ifdef __XOP__
 296             eweps            = _mm_frcz_pd(ewrt);
 297 #else
 298             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 299 #endif
 300             twoeweps         = _mm_add_pd(eweps,eweps);
 301             ewitab           = _mm_slli_epi32(ewitab,2);
 302             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 303             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 304             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 305             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 306             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 307             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 308             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 309             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 310             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 311             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velecsum         = _mm_add_pd(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             /* Update vectorial force */
 319             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 320             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 321             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 322
 323             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 324             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 325             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             r20              = _mm_mul_pd(rsq20,rinv20);
 332
 333             /* Compute parameters for interactions between i and j atoms */
 334             qq20             = _mm_mul_pd(iq2,jq0);
 335
 336             /* EWALD ELECTROSTATICS */
 337
 338             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 339             ewrt             = _mm_mul_pd(r20,ewtabscale);
 340             ewitab           = _mm_cvttpd_epi32(ewrt);
 341 #ifdef __XOP__
 342             eweps            = _mm_frcz_pd(ewrt);
 343 #else
 344             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 345 #endif
 346             twoeweps         = _mm_add_pd(eweps,eweps);
 347             ewitab           = _mm_slli_epi32(ewitab,2);
 348             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 349             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 350             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 351             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 352             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 353             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 354             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 355             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 356             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 357             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velecsum         = _mm_add_pd(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             /* Update vectorial force */
 365             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 366             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 367             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 368
 369             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 370             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 371             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 372
 373             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 374
 375             /* Inner loop uses 147 flops */
 376         }
 377
 378         if(jidx<j_index_end)
 379         {
 380
 381             jnrA             = jjnr[jidx];
 382             j_coord_offsetA  = DIM*jnrA;
 383
 384             /* load j atom coordinates */
 385             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 386                                               &jx0,&jy0,&jz0);
 387
 388             /* Calculate displacement vector */
 389             dx00             = _mm_sub_pd(ix0,jx0);
 390             dy00             = _mm_sub_pd(iy0,jy0);
 391             dz00             = _mm_sub_pd(iz0,jz0);
 392             dx10             = _mm_sub_pd(ix1,jx0);
 393             dy10             = _mm_sub_pd(iy1,jy0);
 394             dz10             = _mm_sub_pd(iz1,jz0);
 395             dx20             = _mm_sub_pd(ix2,jx0);
 396             dy20             = _mm_sub_pd(iy2,jy0);
 397             dz20             = _mm_sub_pd(iz2,jz0);
 398
 399             /* Calculate squared distance and things based on it */
 400             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 401             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 402             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 403
 404             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 405             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 406             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 407
 408             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 409             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 410             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 411
 412             /* Load parameters for j particles */
 413             jq0              = _mm_load_sd(charge+jnrA+0);
 414             vdwjidx0A        = 2*vdwtype[jnrA+0];
 415
 416             fjx0             = _mm_setzero_pd();
 417             fjy0             = _mm_setzero_pd();
 418             fjz0             = _mm_setzero_pd();
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             r00              = _mm_mul_pd(rsq00,rinv00);
 425
 426             /* Compute parameters for interactions between i and j atoms */
 427             qq00             = _mm_mul_pd(iq0,jq0);
 428             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 429
 430             /* EWALD ELECTROSTATICS */
 431
 432             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 433             ewrt             = _mm_mul_pd(r00,ewtabscale);
 434             ewitab           = _mm_cvttpd_epi32(ewrt);
 435 #ifdef __XOP__
 436             eweps            = _mm_frcz_pd(ewrt);
 437 #else
 438             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 439 #endif
 440             twoeweps         = _mm_add_pd(eweps,eweps);
 441             ewitab           = _mm_slli_epi32(ewitab,2);
 442             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 443             ewtabD           = _mm_setzero_pd();
 444             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 445             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 446             ewtabFn          = _mm_setzero_pd();
 447             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 448             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 449             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 450             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 451             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 452
 453             /* LENNARD-JONES DISPERSION/REPULSION */
 454
 455             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 456             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 457             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 458             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 459             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 460
 461             /* Update potential sum for this i atom from the interaction with this j atom. */
 462             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 463             velecsum         = _mm_add_pd(velecsum,velec);
 464             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 465             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 466
 467             fscal            = _mm_add_pd(felec,fvdw);
 468
 469             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 470
 471             /* Update vectorial force */
 472             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 473             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 474             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 475
 476             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 477             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 478             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r10              = _mm_mul_pd(rsq10,rinv10);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             qq10             = _mm_mul_pd(iq1,jq0);
 488
 489             /* EWALD ELECTROSTATICS */
 490
 491             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 492             ewrt             = _mm_mul_pd(r10,ewtabscale);
 493             ewitab           = _mm_cvttpd_epi32(ewrt);
 494 #ifdef __XOP__
 495             eweps            = _mm_frcz_pd(ewrt);
 496 #else
 497             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 498 #endif
 499             twoeweps         = _mm_add_pd(eweps,eweps);
 500             ewitab           = _mm_slli_epi32(ewitab,2);
 501             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 502             ewtabD           = _mm_setzero_pd();
 503             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 504             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 505             ewtabFn          = _mm_setzero_pd();
 506             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 507             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 508             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 509             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 510             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 511
 512             /* Update potential sum for this i atom from the interaction with this j atom. */
 513             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 514             velecsum         = _mm_add_pd(velecsum,velec);
 515
 516             fscal            = felec;
 517
 518             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 519
 520             /* Update vectorial force */
 521             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 522             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 523             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 524
 525             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 526             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 527             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             r20              = _mm_mul_pd(rsq20,rinv20);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq20             = _mm_mul_pd(iq2,jq0);
 537
 538             /* EWALD ELECTROSTATICS */
 539
 540             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 541             ewrt             = _mm_mul_pd(r20,ewtabscale);
 542             ewitab           = _mm_cvttpd_epi32(ewrt);
 543 #ifdef __XOP__
 544             eweps            = _mm_frcz_pd(ewrt);
 545 #else
 546             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 547 #endif
 548             twoeweps         = _mm_add_pd(eweps,eweps);
 549             ewitab           = _mm_slli_epi32(ewitab,2);
 550             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 551             ewtabD           = _mm_setzero_pd();
 552             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 553             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 554             ewtabFn          = _mm_setzero_pd();
 555             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 556             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 557             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 558             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 559             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 560
 561             /* Update potential sum for this i atom from the interaction with this j atom. */
 562             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 563             velecsum         = _mm_add_pd(velecsum,velec);
 564
 565             fscal            = felec;
 566
 567             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 568
 569             /* Update vectorial force */
 570             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 571             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 572             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 573
 574             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 575             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 576             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 577
 578             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 579
 580             /* Inner loop uses 147 flops */
 581         }
 582
 583         /* End of innermost loop */
 584
 585         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 586                                               f+i_coord_offset,fshift+i_shift_offset);
 587
 588         ggid                        = gid[iidx];
 589         /* Update potential energies */
 590         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 591         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 592
 593         /* Increment number of inner iterations */
 594         inneriter                  += j_index_end - j_index_start;
 595
 596         /* Outer loop uses 20 flops */
 597     }
 598
 599     /* Increment number of outer iterations */
 600     outeriter        += nri;
 601
 602     /* Update outer/inner flops */
 603
 604     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*147);
 605 }
 606 /*
 607  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_double
 608  * Electrostatics interaction: Ewald
 609  * VdW interaction:            LennardJones
 610  * Geometry:                   Water3-Particle
 611  * Calculate force/pot:        Force
 612  */
 613 void
 614 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_double
 615                     (t_nblist                    * gmx_restrict       nlist,
 616                      rvec                        * gmx_restrict          xx,
 617                      rvec                        * gmx_restrict          ff,
 618                      t_forcerec                  * gmx_restrict          fr,
 619                      t_mdatoms                   * gmx_restrict     mdatoms,
 620                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 621                      t_nrnb                      * gmx_restrict        nrnb)
 622 {
 623     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 624      * just 0 for non-waters.
 625      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 626      * jnr indices corresponding to data put in the four positions in the SIMD register.
 627      */
 628     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 629     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 630     int              jnrA,jnrB;
 631     int              j_coord_offsetA,j_coord_offsetB;
 632     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 633     real             rcutoff_scalar;
 634     real             *shiftvec,*fshift,*x,*f;
 635     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 636     int              vdwioffset0;
 637     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 638     int              vdwioffset1;
 639     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 640     int              vdwioffset2;
 641     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 642     int              vdwjidx0A,vdwjidx0B;
 643     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 644     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 645     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 646     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 647     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 648     real             *charge;
 649     int              nvdwtype;
 650     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 651     int              *vdwtype;
 652     real             *vdwparam;
 653     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 654     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 655     __m128i          ewitab;
 656     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 657     real             *ewtab;
 658     __m128d          dummy_mask,cutoff_mask;
 659     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 660     __m128d          one     = _mm_set1_pd(1.0);
 661     __m128d          two     = _mm_set1_pd(2.0);
 662     x                = xx[0];
 663     f                = ff[0];
 664
 665     nri              = nlist->nri;
 666     iinr             = nlist->iinr;
 667     jindex           = nlist->jindex;
 668     jjnr             = nlist->jjnr;
 669     shiftidx         = nlist->shift;
 670     gid              = nlist->gid;
 671     shiftvec         = fr->shift_vec[0];
 672     fshift           = fr->fshift[0];
 673     facel            = _mm_set1_pd(fr->epsfac);
 674     charge           = mdatoms->chargeA;
 675     nvdwtype         = fr->ntype;
 676     vdwparam         = fr->nbfp;
 677     vdwtype          = mdatoms->typeA;
 678
 679     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 680     ewtab            = fr->ic->tabq_coul_F;
 681     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 682     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 683
 684     /* Setup water-specific parameters */
 685     inr              = nlist->iinr[0];
 686     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 687     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 688     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 689     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 690
 691     /* Avoid stupid compiler warnings */
 692     jnrA = jnrB = 0;
 693     j_coord_offsetA = 0;
 694     j_coord_offsetB = 0;
 695
 696     outeriter        = 0;
 697     inneriter        = 0;
 698
 699     /* Start outer loop over neighborlists */
 700     for(iidx=0; iidx<nri; iidx++)
 701     {
 702         /* Load shift vector for this list */
 703         i_shift_offset   = DIM*shiftidx[iidx];
 704
 705         /* Load limits for loop over neighbors */
 706         j_index_start    = jindex[iidx];
 707         j_index_end      = jindex[iidx+1];
 708
 709         /* Get outer coordinate index */
 710         inr              = iinr[iidx];
 711         i_coord_offset   = DIM*inr;
 712
 713         /* Load i particle coords and add shift vector */
 714         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 715                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 716
 717         fix0             = _mm_setzero_pd();
 718         fiy0             = _mm_setzero_pd();
 719         fiz0             = _mm_setzero_pd();
 720         fix1             = _mm_setzero_pd();
 721         fiy1             = _mm_setzero_pd();
 722         fiz1             = _mm_setzero_pd();
 723         fix2             = _mm_setzero_pd();
 724         fiy2             = _mm_setzero_pd();
 725         fiz2             = _mm_setzero_pd();
 726
 727         /* Start inner kernel loop */
 728         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 729         {
 730
 731             /* Get j neighbor index, and coordinate index */
 732             jnrA             = jjnr[jidx];
 733             jnrB             = jjnr[jidx+1];
 734             j_coord_offsetA  = DIM*jnrA;
 735             j_coord_offsetB  = DIM*jnrB;
 736
 737             /* load j atom coordinates */
 738             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 739                                               &jx0,&jy0,&jz0);
 740
 741             /* Calculate displacement vector */
 742             dx00             = _mm_sub_pd(ix0,jx0);
 743             dy00             = _mm_sub_pd(iy0,jy0);
 744             dz00             = _mm_sub_pd(iz0,jz0);
 745             dx10             = _mm_sub_pd(ix1,jx0);
 746             dy10             = _mm_sub_pd(iy1,jy0);
 747             dz10             = _mm_sub_pd(iz1,jz0);
 748             dx20             = _mm_sub_pd(ix2,jx0);
 749             dy20             = _mm_sub_pd(iy2,jy0);
 750             dz20             = _mm_sub_pd(iz2,jz0);
 751
 752             /* Calculate squared distance and things based on it */
 753             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 754             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 755             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 756
 757             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 758             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 759             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 760
 761             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 762             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 763             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 764
 765             /* Load parameters for j particles */
 766             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 767             vdwjidx0A        = 2*vdwtype[jnrA+0];
 768             vdwjidx0B        = 2*vdwtype[jnrB+0];
 769
 770             fjx0             = _mm_setzero_pd();
 771             fjy0             = _mm_setzero_pd();
 772             fjz0             = _mm_setzero_pd();
 773
 774             /**************************
 775              * CALCULATE INTERACTIONS *
 776              **************************/
 777
 778             r00              = _mm_mul_pd(rsq00,rinv00);
 779
 780             /* Compute parameters for interactions between i and j atoms */
 781             qq00             = _mm_mul_pd(iq0,jq0);
 782             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 783                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 784
 785             /* EWALD ELECTROSTATICS */
 786
 787             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 788             ewrt             = _mm_mul_pd(r00,ewtabscale);
 789             ewitab           = _mm_cvttpd_epi32(ewrt);
 790 #ifdef __XOP__
 791             eweps            = _mm_frcz_pd(ewrt);
 792 #else
 793             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 794 #endif
 795             twoeweps         = _mm_add_pd(eweps,eweps);
 796             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 797                                          &ewtabF,&ewtabFn);
 798             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 799             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 800
 801             /* LENNARD-JONES DISPERSION/REPULSION */
 802
 803             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 804             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 805
 806             fscal            = _mm_add_pd(felec,fvdw);
 807
 808             /* Update vectorial force */
 809             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 810             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 811             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 812
 813             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 814             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 815             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 816
 817             /**************************
 818              * CALCULATE INTERACTIONS *
 819              **************************/
 820
 821             r10              = _mm_mul_pd(rsq10,rinv10);
 822
 823             /* Compute parameters for interactions between i and j atoms */
 824             qq10             = _mm_mul_pd(iq1,jq0);
 825
 826             /* EWALD ELECTROSTATICS */
 827
 828             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 829             ewrt             = _mm_mul_pd(r10,ewtabscale);
 830             ewitab           = _mm_cvttpd_epi32(ewrt);
 831 #ifdef __XOP__
 832             eweps            = _mm_frcz_pd(ewrt);
 833 #else
 834             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 835 #endif
 836             twoeweps         = _mm_add_pd(eweps,eweps);
 837             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 838                                          &ewtabF,&ewtabFn);
 839             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 840             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 841
 842             fscal            = felec;
 843
 844             /* Update vectorial force */
 845             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 846             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 847             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 848
 849             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 850             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 851             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 852
 853             /**************************
 854              * CALCULATE INTERACTIONS *
 855              **************************/
 856
 857             r20              = _mm_mul_pd(rsq20,rinv20);
 858
 859             /* Compute parameters for interactions between i and j atoms */
 860             qq20             = _mm_mul_pd(iq2,jq0);
 861
 862             /* EWALD ELECTROSTATICS */
 863
 864             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 865             ewrt             = _mm_mul_pd(r20,ewtabscale);
 866             ewitab           = _mm_cvttpd_epi32(ewrt);
 867 #ifdef __XOP__
 868             eweps            = _mm_frcz_pd(ewrt);
 869 #else
 870             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 871 #endif
 872             twoeweps         = _mm_add_pd(eweps,eweps);
 873             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 874                                          &ewtabF,&ewtabFn);
 875             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 876             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 877
 878             fscal            = felec;
 879
 880             /* Update vectorial force */
 881             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 882             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 883             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 884
 885             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 886             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 887             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 888
 889             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 890
 891             /* Inner loop uses 127 flops */
 892         }
 893
 894         if(jidx<j_index_end)
 895         {
 896
 897             jnrA             = jjnr[jidx];
 898             j_coord_offsetA  = DIM*jnrA;
 899
 900             /* load j atom coordinates */
 901             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 902                                               &jx0,&jy0,&jz0);
 903
 904             /* Calculate displacement vector */
 905             dx00             = _mm_sub_pd(ix0,jx0);
 906             dy00             = _mm_sub_pd(iy0,jy0);
 907             dz00             = _mm_sub_pd(iz0,jz0);
 908             dx10             = _mm_sub_pd(ix1,jx0);
 909             dy10             = _mm_sub_pd(iy1,jy0);
 910             dz10             = _mm_sub_pd(iz1,jz0);
 911             dx20             = _mm_sub_pd(ix2,jx0);
 912             dy20             = _mm_sub_pd(iy2,jy0);
 913             dz20             = _mm_sub_pd(iz2,jz0);
 914
 915             /* Calculate squared distance and things based on it */
 916             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 917             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 918             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 919
 920             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 921             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 922             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 923
 924             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 925             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 926             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 927
 928             /* Load parameters for j particles */
 929             jq0              = _mm_load_sd(charge+jnrA+0);
 930             vdwjidx0A        = 2*vdwtype[jnrA+0];
 931
 932             fjx0             = _mm_setzero_pd();
 933             fjy0             = _mm_setzero_pd();
 934             fjz0             = _mm_setzero_pd();
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             r00              = _mm_mul_pd(rsq00,rinv00);
 941
 942             /* Compute parameters for interactions between i and j atoms */
 943             qq00             = _mm_mul_pd(iq0,jq0);
 944             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 945
 946             /* EWALD ELECTROSTATICS */
 947
 948             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 949             ewrt             = _mm_mul_pd(r00,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(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 960
 961             /* LENNARD-JONES DISPERSION/REPULSION */
 962
 963             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 964             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 965
 966             fscal            = _mm_add_pd(felec,fvdw);
 967
 968             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 969
 970             /* Update vectorial force */
 971             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 972             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 973             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 974
 975             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 976             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 977             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 978
 979             /**************************
 980              * CALCULATE INTERACTIONS *
 981              **************************/
 982
 983             r10              = _mm_mul_pd(rsq10,rinv10);
 984
 985             /* Compute parameters for interactions between i and j atoms */
 986             qq10             = _mm_mul_pd(iq1,jq0);
 987
 988             /* EWALD ELECTROSTATICS */
 989
 990             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 991             ewrt             = _mm_mul_pd(r10,ewtabscale);
 992             ewitab           = _mm_cvttpd_epi32(ewrt);
 993 #ifdef __XOP__
 994             eweps            = _mm_frcz_pd(ewrt);
 995 #else
 996             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 997 #endif
 998             twoeweps         = _mm_add_pd(eweps,eweps);
 999             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1000             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1001             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1002
1003             fscal            = felec;
1004
1005             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1006
1007             /* Update vectorial force */
1008             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1009             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1010             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1011
1012             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1013             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1014             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1015
1016             /**************************
1017              * CALCULATE INTERACTIONS *
1018              **************************/
1019
1020             r20              = _mm_mul_pd(rsq20,rinv20);
1021
1022             /* Compute parameters for interactions between i and j atoms */
1023             qq20             = _mm_mul_pd(iq2,jq0);
1024
1025             /* EWALD ELECTROSTATICS */
1026
1027             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1028             ewrt             = _mm_mul_pd(r20,ewtabscale);
1029             ewitab           = _mm_cvttpd_epi32(ewrt);
1030 #ifdef __XOP__
1031             eweps            = _mm_frcz_pd(ewrt);
1032 #else
1033             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1034 #endif
1035             twoeweps         = _mm_add_pd(eweps,eweps);
1036             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1037             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1038             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1039
1040             fscal            = felec;
1041
1042             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1043
1044             /* Update vectorial force */
1045             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1046             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1047             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1048
1049             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1050             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1051             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1052
1053             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1054
1055             /* Inner loop uses 127 flops */
1056         }
1057
1058         /* End of innermost loop */
1059
1060         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1061                                               f+i_coord_offset,fshift+i_shift_offset);
1062
1063         /* Increment number of inner iterations */
1064         inneriter                  += j_index_end - j_index_start;
1065
1066         /* Outer loop uses 18 flops */
1067     }
1068
1069     /* Increment number of outer iterations */
1070     outeriter        += nri;
1071
1072     /* Update outer/inner flops */
1073
1074     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);
1075 }