src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              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              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128i          ewitab;
 105     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     real             *ewtab;
 107     __m128d          dummy_mask,cutoff_mask;
 108     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 109     __m128d          one     = _mm_set1_pd(1.0);
 110     __m128d          two     = _mm_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_pd(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 136     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 137     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->rcoulomb;
 142     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 143     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 144
 145     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 146     rvdw             = _mm_set1_pd(fr->rvdw);
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 173
 174         fix0             = _mm_setzero_pd();
 175         fiy0             = _mm_setzero_pd();
 176         fiz0             = _mm_setzero_pd();
 177         fix1             = _mm_setzero_pd();
 178         fiy1             = _mm_setzero_pd();
 179         fiz1             = _mm_setzero_pd();
 180         fix2             = _mm_setzero_pd();
 181         fiy2             = _mm_setzero_pd();
 182         fiz2             = _mm_setzero_pd();
 183         fix3             = _mm_setzero_pd();
 184         fiy3             = _mm_setzero_pd();
 185         fiz3             = _mm_setzero_pd();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm_setzero_pd();
 189         vvdwsum          = _mm_setzero_pd();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200
 201             /* load j atom coordinates */
 202             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 203                                               &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm_sub_pd(ix0,jx0);
 207             dy00             = _mm_sub_pd(iy0,jy0);
 208             dz00             = _mm_sub_pd(iz0,jz0);
 209             dx10             = _mm_sub_pd(ix1,jx0);
 210             dy10             = _mm_sub_pd(iy1,jy0);
 211             dz10             = _mm_sub_pd(iz1,jz0);
 212             dx20             = _mm_sub_pd(ix2,jx0);
 213             dy20             = _mm_sub_pd(iy2,jy0);
 214             dz20             = _mm_sub_pd(iz2,jz0);
 215             dx30             = _mm_sub_pd(ix3,jx0);
 216             dy30             = _mm_sub_pd(iy3,jy0);
 217             dz30             = _mm_sub_pd(iz3,jz0);
 218
 219             /* Calculate squared distance and things based on it */
 220             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 221             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 222             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 223             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 224
 225             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 226             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 227             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 228
 229             rinvsq00         = gmx_mm_inv_pd(rsq00);
 230             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 231             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 232             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238
 239             fjx0             = _mm_setzero_pd();
 240             fjy0             = _mm_setzero_pd();
 241             fjz0             = _mm_setzero_pd();
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             if (gmx_mm_any_lt(rsq00,rcutoff2))
 248             {
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 253
 254             /* LENNARD-JONES DISPERSION/REPULSION */
 255
 256             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 257             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 258             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 259             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 260                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 261             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 262
 263             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 267             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 268
 269             fscal            = fvdw;
 270
 271             fscal            = _mm_and_pd(fscal,cutoff_mask);
 272
 273             /* Update vectorial force */
 274             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 275             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 276             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 277
 278             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 279             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 280             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 281
 282             }
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             if (gmx_mm_any_lt(rsq10,rcutoff2))
 289             {
 290
 291             r10              = _mm_mul_pd(rsq10,rinv10);
 292
 293             /* Compute parameters for interactions between i and j atoms */
 294             qq10             = _mm_mul_pd(iq1,jq0);
 295
 296             /* EWALD ELECTROSTATICS */
 297
 298             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 299             ewrt             = _mm_mul_pd(r10,ewtabscale);
 300             ewitab           = _mm_cvttpd_epi32(ewrt);
 301 #ifdef __XOP__
 302             eweps            = _mm_frcz_pd(ewrt);
 303 #else
 304             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 305 #endif
 306             twoeweps         = _mm_add_pd(eweps,eweps);
 307             ewitab           = _mm_slli_epi32(ewitab,2);
 308             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 309             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 310             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 311             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 312             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 313             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 314             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 315             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 316             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 317             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 318
 319             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             velec            = _mm_and_pd(velec,cutoff_mask);
 323             velecsum         = _mm_add_pd(velecsum,velec);
 324
 325             fscal            = felec;
 326
 327             fscal            = _mm_and_pd(fscal,cutoff_mask);
 328
 329             /* Update vectorial force */
 330             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 331             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 332             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 333
 334             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 335             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 336             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 337
 338             }
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             if (gmx_mm_any_lt(rsq20,rcutoff2))
 345             {
 346
 347             r20              = _mm_mul_pd(rsq20,rinv20);
 348
 349             /* Compute parameters for interactions between i and j atoms */
 350             qq20             = _mm_mul_pd(iq2,jq0);
 351
 352             /* EWALD ELECTROSTATICS */
 353
 354             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 355             ewrt             = _mm_mul_pd(r20,ewtabscale);
 356             ewitab           = _mm_cvttpd_epi32(ewrt);
 357 #ifdef __XOP__
 358             eweps            = _mm_frcz_pd(ewrt);
 359 #else
 360             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 361 #endif
 362             twoeweps         = _mm_add_pd(eweps,eweps);
 363             ewitab           = _mm_slli_epi32(ewitab,2);
 364             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 365             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 366             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 367             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 368             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 369             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 370             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 371             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 372             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 373             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 374
 375             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velec            = _mm_and_pd(velec,cutoff_mask);
 379             velecsum         = _mm_add_pd(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             fscal            = _mm_and_pd(fscal,cutoff_mask);
 384
 385             /* Update vectorial force */
 386             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 387             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 388             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 389
 390             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 391             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 392             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 393
 394             }
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             if (gmx_mm_any_lt(rsq30,rcutoff2))
 401             {
 402
 403             r30              = _mm_mul_pd(rsq30,rinv30);
 404
 405             /* Compute parameters for interactions between i and j atoms */
 406             qq30             = _mm_mul_pd(iq3,jq0);
 407
 408             /* EWALD ELECTROSTATICS */
 409
 410             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 411             ewrt             = _mm_mul_pd(r30,ewtabscale);
 412             ewitab           = _mm_cvttpd_epi32(ewrt);
 413 #ifdef __XOP__
 414             eweps            = _mm_frcz_pd(ewrt);
 415 #else
 416             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 417 #endif
 418             twoeweps         = _mm_add_pd(eweps,eweps);
 419             ewitab           = _mm_slli_epi32(ewitab,2);
 420             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 421             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 422             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 423             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 424             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 425             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 426             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 427             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 428             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 429             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 430
 431             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velec            = _mm_and_pd(velec,cutoff_mask);
 435             velecsum         = _mm_add_pd(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439             fscal            = _mm_and_pd(fscal,cutoff_mask);
 440
 441             /* Update vectorial force */
 442             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 443             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 444             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 445
 446             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 447             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 448             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 449
 450             }
 451
 452             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 453
 454             /* Inner loop uses 194 flops */
 455         }
 456
 457         if(jidx<j_index_end)
 458         {
 459
 460             jnrA             = jjnr[jidx];
 461             j_coord_offsetA  = DIM*jnrA;
 462
 463             /* load j atom coordinates */
 464             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 465                                               &jx0,&jy0,&jz0);
 466
 467             /* Calculate displacement vector */
 468             dx00             = _mm_sub_pd(ix0,jx0);
 469             dy00             = _mm_sub_pd(iy0,jy0);
 470             dz00             = _mm_sub_pd(iz0,jz0);
 471             dx10             = _mm_sub_pd(ix1,jx0);
 472             dy10             = _mm_sub_pd(iy1,jy0);
 473             dz10             = _mm_sub_pd(iz1,jz0);
 474             dx20             = _mm_sub_pd(ix2,jx0);
 475             dy20             = _mm_sub_pd(iy2,jy0);
 476             dz20             = _mm_sub_pd(iz2,jz0);
 477             dx30             = _mm_sub_pd(ix3,jx0);
 478             dy30             = _mm_sub_pd(iy3,jy0);
 479             dz30             = _mm_sub_pd(iz3,jz0);
 480
 481             /* Calculate squared distance and things based on it */
 482             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 483             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 484             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 485             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 486
 487             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 488             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 489             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 490
 491             rinvsq00         = gmx_mm_inv_pd(rsq00);
 492             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 493             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 494             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 495
 496             /* Load parameters for j particles */
 497             jq0              = _mm_load_sd(charge+jnrA+0);
 498             vdwjidx0A        = 2*vdwtype[jnrA+0];
 499
 500             fjx0             = _mm_setzero_pd();
 501             fjy0             = _mm_setzero_pd();
 502             fjz0             = _mm_setzero_pd();
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             if (gmx_mm_any_lt(rsq00,rcutoff2))
 509             {
 510
 511             /* Compute parameters for interactions between i and j atoms */
 512             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 513
 514             /* LENNARD-JONES DISPERSION/REPULSION */
 515
 516             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 517             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 518             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 519             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 520                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 521             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 522
 523             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 527             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 528             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 529
 530             fscal            = fvdw;
 531
 532             fscal            = _mm_and_pd(fscal,cutoff_mask);
 533
 534             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 535
 536             /* Update vectorial force */
 537             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 538             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 539             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 540
 541             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 542             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 543             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 544
 545             }
 546
 547             /**************************
 548              * CALCULATE INTERACTIONS *
 549              **************************/
 550
 551             if (gmx_mm_any_lt(rsq10,rcutoff2))
 552             {
 553
 554             r10              = _mm_mul_pd(rsq10,rinv10);
 555
 556             /* Compute parameters for interactions between i and j atoms */
 557             qq10             = _mm_mul_pd(iq1,jq0);
 558
 559             /* EWALD ELECTROSTATICS */
 560
 561             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 562             ewrt             = _mm_mul_pd(r10,ewtabscale);
 563             ewitab           = _mm_cvttpd_epi32(ewrt);
 564 #ifdef __XOP__
 565             eweps            = _mm_frcz_pd(ewrt);
 566 #else
 567             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 568 #endif
 569             twoeweps         = _mm_add_pd(eweps,eweps);
 570             ewitab           = _mm_slli_epi32(ewitab,2);
 571             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 572             ewtabD           = _mm_setzero_pd();
 573             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 574             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 575             ewtabFn          = _mm_setzero_pd();
 576             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 577             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 578             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 579             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 580             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 581
 582             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm_and_pd(velec,cutoff_mask);
 586             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 587             velecsum         = _mm_add_pd(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm_and_pd(fscal,cutoff_mask);
 592
 593             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 594
 595             /* Update vectorial force */
 596             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 597             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 598             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 599
 600             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 601             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 602             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 603
 604             }
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             if (gmx_mm_any_lt(rsq20,rcutoff2))
 611             {
 612
 613             r20              = _mm_mul_pd(rsq20,rinv20);
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             qq20             = _mm_mul_pd(iq2,jq0);
 617
 618             /* EWALD ELECTROSTATICS */
 619
 620             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 621             ewrt             = _mm_mul_pd(r20,ewtabscale);
 622             ewitab           = _mm_cvttpd_epi32(ewrt);
 623 #ifdef __XOP__
 624             eweps            = _mm_frcz_pd(ewrt);
 625 #else
 626             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 627 #endif
 628             twoeweps         = _mm_add_pd(eweps,eweps);
 629             ewitab           = _mm_slli_epi32(ewitab,2);
 630             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 631             ewtabD           = _mm_setzero_pd();
 632             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 633             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 634             ewtabFn          = _mm_setzero_pd();
 635             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 636             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 637             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 638             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 639             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 640
 641             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 642
 643             /* Update potential sum for this i atom from the interaction with this j atom. */
 644             velec            = _mm_and_pd(velec,cutoff_mask);
 645             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 646             velecsum         = _mm_add_pd(velecsum,velec);
 647
 648             fscal            = felec;
 649
 650             fscal            = _mm_and_pd(fscal,cutoff_mask);
 651
 652             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 653
 654             /* Update vectorial force */
 655             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 656             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 657             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 658
 659             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 660             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 661             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 662
 663             }
 664
 665             /**************************
 666              * CALCULATE INTERACTIONS *
 667              **************************/
 668
 669             if (gmx_mm_any_lt(rsq30,rcutoff2))
 670             {
 671
 672             r30              = _mm_mul_pd(rsq30,rinv30);
 673
 674             /* Compute parameters for interactions between i and j atoms */
 675             qq30             = _mm_mul_pd(iq3,jq0);
 676
 677             /* EWALD ELECTROSTATICS */
 678
 679             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 680             ewrt             = _mm_mul_pd(r30,ewtabscale);
 681             ewitab           = _mm_cvttpd_epi32(ewrt);
 682 #ifdef __XOP__
 683             eweps            = _mm_frcz_pd(ewrt);
 684 #else
 685             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 686 #endif
 687             twoeweps         = _mm_add_pd(eweps,eweps);
 688             ewitab           = _mm_slli_epi32(ewitab,2);
 689             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 690             ewtabD           = _mm_setzero_pd();
 691             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 692             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 693             ewtabFn          = _mm_setzero_pd();
 694             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 695             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 696             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 697             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 698             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 699
 700             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 701
 702             /* Update potential sum for this i atom from the interaction with this j atom. */
 703             velec            = _mm_and_pd(velec,cutoff_mask);
 704             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 705             velecsum         = _mm_add_pd(velecsum,velec);
 706
 707             fscal            = felec;
 708
 709             fscal            = _mm_and_pd(fscal,cutoff_mask);
 710
 711             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 712
 713             /* Update vectorial force */
 714             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 715             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 716             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 717
 718             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 719             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 720             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 721
 722             }
 723
 724             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 725
 726             /* Inner loop uses 194 flops */
 727         }
 728
 729         /* End of innermost loop */
 730
 731         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 732                                               f+i_coord_offset,fshift+i_shift_offset);
 733
 734         ggid                        = gid[iidx];
 735         /* Update potential energies */
 736         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 737         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 738
 739         /* Increment number of inner iterations */
 740         inneriter                  += j_index_end - j_index_start;
 741
 742         /* Outer loop uses 26 flops */
 743     }
 744
 745     /* Increment number of outer iterations */
 746     outeriter        += nri;
 747
 748     /* Update outer/inner flops */
 749
 750     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*194);
 751 }
 752 /*
 753  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_128_fma_double
 754  * Electrostatics interaction: Ewald
 755  * VdW interaction:            LennardJones
 756  * Geometry:                   Water4-Particle
 757  * Calculate force/pot:        Force
 758  */
 759 void
 760 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_128_fma_double
 761                     (t_nblist                    * gmx_restrict       nlist,
 762                      rvec                        * gmx_restrict          xx,
 763                      rvec                        * gmx_restrict          ff,
 764                      t_forcerec                  * gmx_restrict          fr,
 765                      t_mdatoms                   * gmx_restrict     mdatoms,
 766                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 767                      t_nrnb                      * gmx_restrict        nrnb)
 768 {
 769     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 770      * just 0 for non-waters.
 771      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 772      * jnr indices corresponding to data put in the four positions in the SIMD register.
 773      */
 774     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 775     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 776     int              jnrA,jnrB;
 777     int              j_coord_offsetA,j_coord_offsetB;
 778     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 779     real             rcutoff_scalar;
 780     real             *shiftvec,*fshift,*x,*f;
 781     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 782     int              vdwioffset0;
 783     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 784     int              vdwioffset1;
 785     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 786     int              vdwioffset2;
 787     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 788     int              vdwioffset3;
 789     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 790     int              vdwjidx0A,vdwjidx0B;
 791     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 792     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 793     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 794     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 795     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 796     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 797     real             *charge;
 798     int              nvdwtype;
 799     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 800     int              *vdwtype;
 801     real             *vdwparam;
 802     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 803     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 804     __m128i          ewitab;
 805     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 806     real             *ewtab;
 807     __m128d          dummy_mask,cutoff_mask;
 808     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 809     __m128d          one     = _mm_set1_pd(1.0);
 810     __m128d          two     = _mm_set1_pd(2.0);
 811     x                = xx[0];
 812     f                = ff[0];
 813
 814     nri              = nlist->nri;
 815     iinr             = nlist->iinr;
 816     jindex           = nlist->jindex;
 817     jjnr             = nlist->jjnr;
 818     shiftidx         = nlist->shift;
 819     gid              = nlist->gid;
 820     shiftvec         = fr->shift_vec[0];
 821     fshift           = fr->fshift[0];
 822     facel            = _mm_set1_pd(fr->epsfac);
 823     charge           = mdatoms->chargeA;
 824     nvdwtype         = fr->ntype;
 825     vdwparam         = fr->nbfp;
 826     vdwtype          = mdatoms->typeA;
 827
 828     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 829     ewtab            = fr->ic->tabq_coul_F;
 830     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 831     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 832
 833     /* Setup water-specific parameters */
 834     inr              = nlist->iinr[0];
 835     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 836     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 837     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 838     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 839
 840     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 841     rcutoff_scalar   = fr->rcoulomb;
 842     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 843     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 844
 845     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 846     rvdw             = _mm_set1_pd(fr->rvdw);
 847
 848     /* Avoid stupid compiler warnings */
 849     jnrA = jnrB = 0;
 850     j_coord_offsetA = 0;
 851     j_coord_offsetB = 0;
 852
 853     outeriter        = 0;
 854     inneriter        = 0;
 855
 856     /* Start outer loop over neighborlists */
 857     for(iidx=0; iidx<nri; iidx++)
 858     {
 859         /* Load shift vector for this list */
 860         i_shift_offset   = DIM*shiftidx[iidx];
 861
 862         /* Load limits for loop over neighbors */
 863         j_index_start    = jindex[iidx];
 864         j_index_end      = jindex[iidx+1];
 865
 866         /* Get outer coordinate index */
 867         inr              = iinr[iidx];
 868         i_coord_offset   = DIM*inr;
 869
 870         /* Load i particle coords and add shift vector */
 871         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 872                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 873
 874         fix0             = _mm_setzero_pd();
 875         fiy0             = _mm_setzero_pd();
 876         fiz0             = _mm_setzero_pd();
 877         fix1             = _mm_setzero_pd();
 878         fiy1             = _mm_setzero_pd();
 879         fiz1             = _mm_setzero_pd();
 880         fix2             = _mm_setzero_pd();
 881         fiy2             = _mm_setzero_pd();
 882         fiz2             = _mm_setzero_pd();
 883         fix3             = _mm_setzero_pd();
 884         fiy3             = _mm_setzero_pd();
 885         fiz3             = _mm_setzero_pd();
 886
 887         /* Start inner kernel loop */
 888         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 889         {
 890
 891             /* Get j neighbor index, and coordinate index */
 892             jnrA             = jjnr[jidx];
 893             jnrB             = jjnr[jidx+1];
 894             j_coord_offsetA  = DIM*jnrA;
 895             j_coord_offsetB  = DIM*jnrB;
 896
 897             /* load j atom coordinates */
 898             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 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             dx30             = _mm_sub_pd(ix3,jx0);
 912             dy30             = _mm_sub_pd(iy3,jy0);
 913             dz30             = _mm_sub_pd(iz3,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             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 920
 921             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 922             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 923             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 924
 925             rinvsq00         = gmx_mm_inv_pd(rsq00);
 926             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 927             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 928             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 929
 930             /* Load parameters for j particles */
 931             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 932             vdwjidx0A        = 2*vdwtype[jnrA+0];
 933             vdwjidx0B        = 2*vdwtype[jnrB+0];
 934
 935             fjx0             = _mm_setzero_pd();
 936             fjy0             = _mm_setzero_pd();
 937             fjz0             = _mm_setzero_pd();
 938
 939             /**************************
 940              * CALCULATE INTERACTIONS *
 941              **************************/
 942
 943             if (gmx_mm_any_lt(rsq00,rcutoff2))
 944             {
 945
 946             /* Compute parameters for interactions between i and j atoms */
 947             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 948                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 949
 950             /* LENNARD-JONES DISPERSION/REPULSION */
 951
 952             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 953             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 954
 955             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 956
 957             fscal            = fvdw;
 958
 959             fscal            = _mm_and_pd(fscal,cutoff_mask);
 960
 961             /* Update vectorial force */
 962             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 963             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 964             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 965
 966             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 967             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 968             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 969
 970             }
 971
 972             /**************************
 973              * CALCULATE INTERACTIONS *
 974              **************************/
 975
 976             if (gmx_mm_any_lt(rsq10,rcutoff2))
 977             {
 978
 979             r10              = _mm_mul_pd(rsq10,rinv10);
 980
 981             /* Compute parameters for interactions between i and j atoms */
 982             qq10             = _mm_mul_pd(iq1,jq0);
 983
 984             /* EWALD ELECTROSTATICS */
 985
 986             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 987             ewrt             = _mm_mul_pd(r10,ewtabscale);
 988             ewitab           = _mm_cvttpd_epi32(ewrt);
 989 #ifdef __XOP__
 990             eweps            = _mm_frcz_pd(ewrt);
 991 #else
 992             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 993 #endif
 994             twoeweps         = _mm_add_pd(eweps,eweps);
 995             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 996                                          &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             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1001
1002             fscal            = felec;
1003
1004             fscal            = _mm_and_pd(fscal,cutoff_mask);
1005
1006             /* Update vectorial force */
1007             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1008             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1009             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1010
1011             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1012             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1013             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1014
1015             }
1016
1017             /**************************
1018              * CALCULATE INTERACTIONS *
1019              **************************/
1020
1021             if (gmx_mm_any_lt(rsq20,rcutoff2))
1022             {
1023
1024             r20              = _mm_mul_pd(rsq20,rinv20);
1025
1026             /* Compute parameters for interactions between i and j atoms */
1027             qq20             = _mm_mul_pd(iq2,jq0);
1028
1029             /* EWALD ELECTROSTATICS */
1030
1031             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1032             ewrt             = _mm_mul_pd(r20,ewtabscale);
1033             ewitab           = _mm_cvttpd_epi32(ewrt);
1034 #ifdef __XOP__
1035             eweps            = _mm_frcz_pd(ewrt);
1036 #else
1037             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1038 #endif
1039             twoeweps         = _mm_add_pd(eweps,eweps);
1040             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1041                                          &ewtabF,&ewtabFn);
1042             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1043             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1044
1045             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1046
1047             fscal            = felec;
1048
1049             fscal            = _mm_and_pd(fscal,cutoff_mask);
1050
1051             /* Update vectorial force */
1052             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1053             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1054             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1055
1056             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1057             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1058             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1059
1060             }
1061
1062             /**************************
1063              * CALCULATE INTERACTIONS *
1064              **************************/
1065
1066             if (gmx_mm_any_lt(rsq30,rcutoff2))
1067             {
1068
1069             r30              = _mm_mul_pd(rsq30,rinv30);
1070
1071             /* Compute parameters for interactions between i and j atoms */
1072             qq30             = _mm_mul_pd(iq3,jq0);
1073
1074             /* EWALD ELECTROSTATICS */
1075
1076             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1077             ewrt             = _mm_mul_pd(r30,ewtabscale);
1078             ewitab           = _mm_cvttpd_epi32(ewrt);
1079 #ifdef __XOP__
1080             eweps            = _mm_frcz_pd(ewrt);
1081 #else
1082             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1083 #endif
1084             twoeweps         = _mm_add_pd(eweps,eweps);
1085             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1086                                          &ewtabF,&ewtabFn);
1087             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1088             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1089
1090             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1091
1092             fscal            = felec;
1093
1094             fscal            = _mm_and_pd(fscal,cutoff_mask);
1095
1096             /* Update vectorial force */
1097             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1098             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1099             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1100
1101             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1102             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1103             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1104
1105             }
1106
1107             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1108
1109             /* Inner loop uses 162 flops */
1110         }
1111
1112         if(jidx<j_index_end)
1113         {
1114
1115             jnrA             = jjnr[jidx];
1116             j_coord_offsetA  = DIM*jnrA;
1117
1118             /* load j atom coordinates */
1119             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1120                                               &jx0,&jy0,&jz0);
1121
1122             /* Calculate displacement vector */
1123             dx00             = _mm_sub_pd(ix0,jx0);
1124             dy00             = _mm_sub_pd(iy0,jy0);
1125             dz00             = _mm_sub_pd(iz0,jz0);
1126             dx10             = _mm_sub_pd(ix1,jx0);
1127             dy10             = _mm_sub_pd(iy1,jy0);
1128             dz10             = _mm_sub_pd(iz1,jz0);
1129             dx20             = _mm_sub_pd(ix2,jx0);
1130             dy20             = _mm_sub_pd(iy2,jy0);
1131             dz20             = _mm_sub_pd(iz2,jz0);
1132             dx30             = _mm_sub_pd(ix3,jx0);
1133             dy30             = _mm_sub_pd(iy3,jy0);
1134             dz30             = _mm_sub_pd(iz3,jz0);
1135
1136             /* Calculate squared distance and things based on it */
1137             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1138             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1139             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1140             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1141
1142             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1143             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1144             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1145
1146             rinvsq00         = gmx_mm_inv_pd(rsq00);
1147             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1148             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1149             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1150
1151             /* Load parameters for j particles */
1152             jq0              = _mm_load_sd(charge+jnrA+0);
1153             vdwjidx0A        = 2*vdwtype[jnrA+0];
1154
1155             fjx0             = _mm_setzero_pd();
1156             fjy0             = _mm_setzero_pd();
1157             fjz0             = _mm_setzero_pd();
1158
1159             /**************************
1160              * CALCULATE INTERACTIONS *
1161              **************************/
1162
1163             if (gmx_mm_any_lt(rsq00,rcutoff2))
1164             {
1165
1166             /* Compute parameters for interactions between i and j atoms */
1167             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1168
1169             /* LENNARD-JONES DISPERSION/REPULSION */
1170
1171             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1172             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1173
1174             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1175
1176             fscal            = fvdw;
1177
1178             fscal            = _mm_and_pd(fscal,cutoff_mask);
1179
1180             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1181
1182             /* Update vectorial force */
1183             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1184             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1185             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1186
1187             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1188             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1189             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1190
1191             }
1192
1193             /**************************
1194              * CALCULATE INTERACTIONS *
1195              **************************/
1196
1197             if (gmx_mm_any_lt(rsq10,rcutoff2))
1198             {
1199
1200             r10              = _mm_mul_pd(rsq10,rinv10);
1201
1202             /* Compute parameters for interactions between i and j atoms */
1203             qq10             = _mm_mul_pd(iq1,jq0);
1204
1205             /* EWALD ELECTROSTATICS */
1206
1207             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1208             ewrt             = _mm_mul_pd(r10,ewtabscale);
1209             ewitab           = _mm_cvttpd_epi32(ewrt);
1210 #ifdef __XOP__
1211             eweps            = _mm_frcz_pd(ewrt);
1212 #else
1213             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1214 #endif
1215             twoeweps         = _mm_add_pd(eweps,eweps);
1216             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1217             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1218             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1219
1220             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1221
1222             fscal            = felec;
1223
1224             fscal            = _mm_and_pd(fscal,cutoff_mask);
1225
1226             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1227
1228             /* Update vectorial force */
1229             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1230             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1231             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1232
1233             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1234             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1235             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1236
1237             }
1238
1239             /**************************
1240              * CALCULATE INTERACTIONS *
1241              **************************/
1242
1243             if (gmx_mm_any_lt(rsq20,rcutoff2))
1244             {
1245
1246             r20              = _mm_mul_pd(rsq20,rinv20);
1247
1248             /* Compute parameters for interactions between i and j atoms */
1249             qq20             = _mm_mul_pd(iq2,jq0);
1250
1251             /* EWALD ELECTROSTATICS */
1252
1253             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1254             ewrt             = _mm_mul_pd(r20,ewtabscale);
1255             ewitab           = _mm_cvttpd_epi32(ewrt);
1256 #ifdef __XOP__
1257             eweps            = _mm_frcz_pd(ewrt);
1258 #else
1259             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1260 #endif
1261             twoeweps         = _mm_add_pd(eweps,eweps);
1262             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1263             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1264             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1265
1266             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1267
1268             fscal            = felec;
1269
1270             fscal            = _mm_and_pd(fscal,cutoff_mask);
1271
1272             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1273
1274             /* Update vectorial force */
1275             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1276             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1277             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1278
1279             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1280             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1281             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1282
1283             }
1284
1285             /**************************
1286              * CALCULATE INTERACTIONS *
1287              **************************/
1288
1289             if (gmx_mm_any_lt(rsq30,rcutoff2))
1290             {
1291
1292             r30              = _mm_mul_pd(rsq30,rinv30);
1293
1294             /* Compute parameters for interactions between i and j atoms */
1295             qq30             = _mm_mul_pd(iq3,jq0);
1296
1297             /* EWALD ELECTROSTATICS */
1298
1299             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1300             ewrt             = _mm_mul_pd(r30,ewtabscale);
1301             ewitab           = _mm_cvttpd_epi32(ewrt);
1302 #ifdef __XOP__
1303             eweps            = _mm_frcz_pd(ewrt);
1304 #else
1305             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1306 #endif
1307             twoeweps         = _mm_add_pd(eweps,eweps);
1308             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1309             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1310             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1311
1312             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1313
1314             fscal            = felec;
1315
1316             fscal            = _mm_and_pd(fscal,cutoff_mask);
1317
1318             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1319
1320             /* Update vectorial force */
1321             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1322             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1323             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1324
1325             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1326             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1327             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1328
1329             }
1330
1331             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1332
1333             /* Inner loop uses 162 flops */
1334         }
1335
1336         /* End of innermost loop */
1337
1338         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1339                                               f+i_coord_offset,fshift+i_shift_offset);
1340
1341         /* Increment number of inner iterations */
1342         inneriter                  += j_index_end - j_index_start;
1343
1344         /* Outer loop uses 24 flops */
1345     }
1346
1347     /* Increment number of outer iterations */
1348     outeriter        += nri;
1349
1350     /* Update outer/inner flops */
1351
1352     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*162);
1353 }