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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwioffset3;
  86     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  87     int              vdwjidx0A,vdwjidx0B;
  88     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     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->ic->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     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 133     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 134     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->ic->rcoulomb;
 139     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 140     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 141
 142     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 143     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 170
 171         fix0             = _mm_setzero_pd();
 172         fiy0             = _mm_setzero_pd();
 173         fiz0             = _mm_setzero_pd();
 174         fix1             = _mm_setzero_pd();
 175         fiy1             = _mm_setzero_pd();
 176         fiz1             = _mm_setzero_pd();
 177         fix2             = _mm_setzero_pd();
 178         fiy2             = _mm_setzero_pd();
 179         fiz2             = _mm_setzero_pd();
 180         fix3             = _mm_setzero_pd();
 181         fiy3             = _mm_setzero_pd();
 182         fiz3             = _mm_setzero_pd();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_pd();
 186         vvdwsum          = _mm_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_pd(ix0,jx0);
 204             dy00             = _mm_sub_pd(iy0,jy0);
 205             dz00             = _mm_sub_pd(iz0,jz0);
 206             dx10             = _mm_sub_pd(ix1,jx0);
 207             dy10             = _mm_sub_pd(iy1,jy0);
 208             dz10             = _mm_sub_pd(iz1,jz0);
 209             dx20             = _mm_sub_pd(ix2,jx0);
 210             dy20             = _mm_sub_pd(iy2,jy0);
 211             dz20             = _mm_sub_pd(iz2,jz0);
 212             dx30             = _mm_sub_pd(ix3,jx0);
 213             dy30             = _mm_sub_pd(iy3,jy0);
 214             dz30             = _mm_sub_pd(iz3,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 220             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 221
 222             rinv10           = avx128fma_invsqrt_d(rsq10);
 223             rinv20           = avx128fma_invsqrt_d(rsq20);
 224             rinv30           = avx128fma_invsqrt_d(rsq30);
 225
 226             rinvsq00         = avx128fma_inv_d(rsq00);
 227             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 228             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 229             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235
 236             fjx0             = _mm_setzero_pd();
 237             fjy0             = _mm_setzero_pd();
 238             fjz0             = _mm_setzero_pd();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             if (gmx_mm_any_lt(rsq00,rcutoff2))
 245             {
 246
 247             /* Compute parameters for interactions between i and j atoms */
 248             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 249                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 250
 251             /* LENNARD-JONES DISPERSION/REPULSION */
 252
 253             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 254             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 255             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 256             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 257                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 258             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 259
 260             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 264             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 265
 266             fscal            = fvdw;
 267
 268             fscal            = _mm_and_pd(fscal,cutoff_mask);
 269
 270             /* Update vectorial force */
 271             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 272             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 273             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 274
 275             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 276             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 277             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 278
 279             }
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             if (gmx_mm_any_lt(rsq10,rcutoff2))
 286             {
 287
 288             r10              = _mm_mul_pd(rsq10,rinv10);
 289
 290             /* Compute parameters for interactions between i and j atoms */
 291             qq10             = _mm_mul_pd(iq1,jq0);
 292
 293             /* EWALD ELECTROSTATICS */
 294
 295             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 296             ewrt             = _mm_mul_pd(r10,ewtabscale);
 297             ewitab           = _mm_cvttpd_epi32(ewrt);
 298 #ifdef __XOP__
 299             eweps            = _mm_frcz_pd(ewrt);
 300 #else
 301             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 302 #endif
 303             twoeweps         = _mm_add_pd(eweps,eweps);
 304             ewitab           = _mm_slli_epi32(ewitab,2);
 305             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 306             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 307             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 308             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 309             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 310             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 311             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 312             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 313             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 314             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 315
 316             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velec            = _mm_and_pd(velec,cutoff_mask);
 320             velecsum         = _mm_add_pd(velecsum,velec);
 321
 322             fscal            = felec;
 323
 324             fscal            = _mm_and_pd(fscal,cutoff_mask);
 325
 326             /* Update vectorial force */
 327             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 328             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 329             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 330
 331             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 332             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 333             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 334
 335             }
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             if (gmx_mm_any_lt(rsq20,rcutoff2))
 342             {
 343
 344             r20              = _mm_mul_pd(rsq20,rinv20);
 345
 346             /* Compute parameters for interactions between i and j atoms */
 347             qq20             = _mm_mul_pd(iq2,jq0);
 348
 349             /* EWALD ELECTROSTATICS */
 350
 351             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 352             ewrt             = _mm_mul_pd(r20,ewtabscale);
 353             ewitab           = _mm_cvttpd_epi32(ewrt);
 354 #ifdef __XOP__
 355             eweps            = _mm_frcz_pd(ewrt);
 356 #else
 357             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 358 #endif
 359             twoeweps         = _mm_add_pd(eweps,eweps);
 360             ewitab           = _mm_slli_epi32(ewitab,2);
 361             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 362             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 363             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 364             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 365             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 366             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 367             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 368             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 369             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 370             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 371
 372             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velec            = _mm_and_pd(velec,cutoff_mask);
 376             velecsum         = _mm_add_pd(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             fscal            = _mm_and_pd(fscal,cutoff_mask);
 381
 382             /* Update vectorial force */
 383             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 384             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 385             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 386
 387             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 388             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 389             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 390
 391             }
 392
 393             /**************************
 394              * CALCULATE INTERACTIONS *
 395              **************************/
 396
 397             if (gmx_mm_any_lt(rsq30,rcutoff2))
 398             {
 399
 400             r30              = _mm_mul_pd(rsq30,rinv30);
 401
 402             /* Compute parameters for interactions between i and j atoms */
 403             qq30             = _mm_mul_pd(iq3,jq0);
 404
 405             /* EWALD ELECTROSTATICS */
 406
 407             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 408             ewrt             = _mm_mul_pd(r30,ewtabscale);
 409             ewitab           = _mm_cvttpd_epi32(ewrt);
 410 #ifdef __XOP__
 411             eweps            = _mm_frcz_pd(ewrt);
 412 #else
 413             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 414 #endif
 415             twoeweps         = _mm_add_pd(eweps,eweps);
 416             ewitab           = _mm_slli_epi32(ewitab,2);
 417             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 418             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 419             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 420             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 421             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 422             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 423             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 424             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 425             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 426             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 427
 428             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 429
 430             /* Update potential sum for this i atom from the interaction with this j atom. */
 431             velec            = _mm_and_pd(velec,cutoff_mask);
 432             velecsum         = _mm_add_pd(velecsum,velec);
 433
 434             fscal            = felec;
 435
 436             fscal            = _mm_and_pd(fscal,cutoff_mask);
 437
 438             /* Update vectorial force */
 439             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 440             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 441             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 442
 443             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 444             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 445             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 446
 447             }
 448
 449             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 450
 451             /* Inner loop uses 194 flops */
 452         }
 453
 454         if(jidx<j_index_end)
 455         {
 456
 457             jnrA             = jjnr[jidx];
 458             j_coord_offsetA  = DIM*jnrA;
 459
 460             /* load j atom coordinates */
 461             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 462                                               &jx0,&jy0,&jz0);
 463
 464             /* Calculate displacement vector */
 465             dx00             = _mm_sub_pd(ix0,jx0);
 466             dy00             = _mm_sub_pd(iy0,jy0);
 467             dz00             = _mm_sub_pd(iz0,jz0);
 468             dx10             = _mm_sub_pd(ix1,jx0);
 469             dy10             = _mm_sub_pd(iy1,jy0);
 470             dz10             = _mm_sub_pd(iz1,jz0);
 471             dx20             = _mm_sub_pd(ix2,jx0);
 472             dy20             = _mm_sub_pd(iy2,jy0);
 473             dz20             = _mm_sub_pd(iz2,jz0);
 474             dx30             = _mm_sub_pd(ix3,jx0);
 475             dy30             = _mm_sub_pd(iy3,jy0);
 476             dz30             = _mm_sub_pd(iz3,jz0);
 477
 478             /* Calculate squared distance and things based on it */
 479             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 480             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 481             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 482             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 483
 484             rinv10           = avx128fma_invsqrt_d(rsq10);
 485             rinv20           = avx128fma_invsqrt_d(rsq20);
 486             rinv30           = avx128fma_invsqrt_d(rsq30);
 487
 488             rinvsq00         = avx128fma_inv_d(rsq00);
 489             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 490             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 491             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 492
 493             /* Load parameters for j particles */
 494             jq0              = _mm_load_sd(charge+jnrA+0);
 495             vdwjidx0A        = 2*vdwtype[jnrA+0];
 496
 497             fjx0             = _mm_setzero_pd();
 498             fjy0             = _mm_setzero_pd();
 499             fjz0             = _mm_setzero_pd();
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             if (gmx_mm_any_lt(rsq00,rcutoff2))
 506             {
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 510
 511             /* LENNARD-JONES DISPERSION/REPULSION */
 512
 513             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 514             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 515             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 516             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 517                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 518             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 519
 520             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 524             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 525             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 526
 527             fscal            = fvdw;
 528
 529             fscal            = _mm_and_pd(fscal,cutoff_mask);
 530
 531             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 532
 533             /* Update vectorial force */
 534             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 535             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 536             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 537
 538             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 539             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 540             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 541
 542             }
 543
 544             /**************************
 545              * CALCULATE INTERACTIONS *
 546              **************************/
 547
 548             if (gmx_mm_any_lt(rsq10,rcutoff2))
 549             {
 550
 551             r10              = _mm_mul_pd(rsq10,rinv10);
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq10             = _mm_mul_pd(iq1,jq0);
 555
 556             /* EWALD ELECTROSTATICS */
 557
 558             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 559             ewrt             = _mm_mul_pd(r10,ewtabscale);
 560             ewitab           = _mm_cvttpd_epi32(ewrt);
 561 #ifdef __XOP__
 562             eweps            = _mm_frcz_pd(ewrt);
 563 #else
 564             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 565 #endif
 566             twoeweps         = _mm_add_pd(eweps,eweps);
 567             ewitab           = _mm_slli_epi32(ewitab,2);
 568             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 569             ewtabD           = _mm_setzero_pd();
 570             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 571             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 572             ewtabFn          = _mm_setzero_pd();
 573             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 574             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 575             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 576             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 577             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 578
 579             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm_and_pd(velec,cutoff_mask);
 583             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 584             velecsum         = _mm_add_pd(velecsum,velec);
 585
 586             fscal            = felec;
 587
 588             fscal            = _mm_and_pd(fscal,cutoff_mask);
 589
 590             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 591
 592             /* Update vectorial force */
 593             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 594             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 595             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 596
 597             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 598             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 599             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 600
 601             }
 602
 603             /**************************
 604              * CALCULATE INTERACTIONS *
 605              **************************/
 606
 607             if (gmx_mm_any_lt(rsq20,rcutoff2))
 608             {
 609
 610             r20              = _mm_mul_pd(rsq20,rinv20);
 611
 612             /* Compute parameters for interactions between i and j atoms */
 613             qq20             = _mm_mul_pd(iq2,jq0);
 614
 615             /* EWALD ELECTROSTATICS */
 616
 617             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 618             ewrt             = _mm_mul_pd(r20,ewtabscale);
 619             ewitab           = _mm_cvttpd_epi32(ewrt);
 620 #ifdef __XOP__
 621             eweps            = _mm_frcz_pd(ewrt);
 622 #else
 623             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 624 #endif
 625             twoeweps         = _mm_add_pd(eweps,eweps);
 626             ewitab           = _mm_slli_epi32(ewitab,2);
 627             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 628             ewtabD           = _mm_setzero_pd();
 629             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 630             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 631             ewtabFn          = _mm_setzero_pd();
 632             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 633             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 634             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 635             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 636             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 637
 638             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 639
 640             /* Update potential sum for this i atom from the interaction with this j atom. */
 641             velec            = _mm_and_pd(velec,cutoff_mask);
 642             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 643             velecsum         = _mm_add_pd(velecsum,velec);
 644
 645             fscal            = felec;
 646
 647             fscal            = _mm_and_pd(fscal,cutoff_mask);
 648
 649             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 650
 651             /* Update vectorial force */
 652             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 653             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 654             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 655
 656             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 657             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 658             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 659
 660             }
 661
 662             /**************************
 663              * CALCULATE INTERACTIONS *
 664              **************************/
 665
 666             if (gmx_mm_any_lt(rsq30,rcutoff2))
 667             {
 668
 669             r30              = _mm_mul_pd(rsq30,rinv30);
 670
 671             /* Compute parameters for interactions between i and j atoms */
 672             qq30             = _mm_mul_pd(iq3,jq0);
 673
 674             /* EWALD ELECTROSTATICS */
 675
 676             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 677             ewrt             = _mm_mul_pd(r30,ewtabscale);
 678             ewitab           = _mm_cvttpd_epi32(ewrt);
 679 #ifdef __XOP__
 680             eweps            = _mm_frcz_pd(ewrt);
 681 #else
 682             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 683 #endif
 684             twoeweps         = _mm_add_pd(eweps,eweps);
 685             ewitab           = _mm_slli_epi32(ewitab,2);
 686             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 687             ewtabD           = _mm_setzero_pd();
 688             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 689             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 690             ewtabFn          = _mm_setzero_pd();
 691             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 692             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 693             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 694             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 695             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 696
 697             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 698
 699             /* Update potential sum for this i atom from the interaction with this j atom. */
 700             velec            = _mm_and_pd(velec,cutoff_mask);
 701             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 702             velecsum         = _mm_add_pd(velecsum,velec);
 703
 704             fscal            = felec;
 705
 706             fscal            = _mm_and_pd(fscal,cutoff_mask);
 707
 708             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 709
 710             /* Update vectorial force */
 711             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 712             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 713             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 714
 715             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 716             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 717             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 718
 719             }
 720
 721             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 722
 723             /* Inner loop uses 194 flops */
 724         }
 725
 726         /* End of innermost loop */
 727
 728         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 729                                               f+i_coord_offset,fshift+i_shift_offset);
 730
 731         ggid                        = gid[iidx];
 732         /* Update potential energies */
 733         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 734         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 735
 736         /* Increment number of inner iterations */
 737         inneriter                  += j_index_end - j_index_start;
 738
 739         /* Outer loop uses 26 flops */
 740     }
 741
 742     /* Increment number of outer iterations */
 743     outeriter        += nri;
 744
 745     /* Update outer/inner flops */
 746
 747     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*194);
 748 }
 749 /*
 750  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_128_fma_double
 751  * Electrostatics interaction: Ewald
 752  * VdW interaction:            LennardJones
 753  * Geometry:                   Water4-Particle
 754  * Calculate force/pot:        Force
 755  */
 756 void
 757 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_128_fma_double
 758                     (t_nblist                    * gmx_restrict       nlist,
 759                      rvec                        * gmx_restrict          xx,
 760                      rvec                        * gmx_restrict          ff,
 761                      struct t_forcerec           * gmx_restrict          fr,
 762                      t_mdatoms                   * gmx_restrict     mdatoms,
 763                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 764                      t_nrnb                      * gmx_restrict        nrnb)
 765 {
 766     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 767      * just 0 for non-waters.
 768      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 769      * jnr indices corresponding to data put in the four positions in the SIMD register.
 770      */
 771     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 772     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 773     int              jnrA,jnrB;
 774     int              j_coord_offsetA,j_coord_offsetB;
 775     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 776     real             rcutoff_scalar;
 777     real             *shiftvec,*fshift,*x,*f;
 778     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 779     int              vdwioffset0;
 780     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 781     int              vdwioffset1;
 782     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 783     int              vdwioffset2;
 784     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 785     int              vdwioffset3;
 786     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 787     int              vdwjidx0A,vdwjidx0B;
 788     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 789     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 790     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 791     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 792     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 793     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 794     real             *charge;
 795     int              nvdwtype;
 796     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 797     int              *vdwtype;
 798     real             *vdwparam;
 799     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 800     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 801     __m128i          ewitab;
 802     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 803     real             *ewtab;
 804     __m128d          dummy_mask,cutoff_mask;
 805     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 806     __m128d          one     = _mm_set1_pd(1.0);
 807     __m128d          two     = _mm_set1_pd(2.0);
 808     x                = xx[0];
 809     f                = ff[0];
 810
 811     nri              = nlist->nri;
 812     iinr             = nlist->iinr;
 813     jindex           = nlist->jindex;
 814     jjnr             = nlist->jjnr;
 815     shiftidx         = nlist->shift;
 816     gid              = nlist->gid;
 817     shiftvec         = fr->shift_vec[0];
 818     fshift           = fr->fshift[0];
 819     facel            = _mm_set1_pd(fr->ic->epsfac);
 820     charge           = mdatoms->chargeA;
 821     nvdwtype         = fr->ntype;
 822     vdwparam         = fr->nbfp;
 823     vdwtype          = mdatoms->typeA;
 824
 825     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 826     ewtab            = fr->ic->tabq_coul_F;
 827     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 828     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 829
 830     /* Setup water-specific parameters */
 831     inr              = nlist->iinr[0];
 832     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 833     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 834     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 835     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 836
 837     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 838     rcutoff_scalar   = fr->ic->rcoulomb;
 839     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 840     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 841
 842     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 843     rvdw             = _mm_set1_pd(fr->ic->rvdw);
 844
 845     /* Avoid stupid compiler warnings */
 846     jnrA = jnrB = 0;
 847     j_coord_offsetA = 0;
 848     j_coord_offsetB = 0;
 849
 850     outeriter        = 0;
 851     inneriter        = 0;
 852
 853     /* Start outer loop over neighborlists */
 854     for(iidx=0; iidx<nri; iidx++)
 855     {
 856         /* Load shift vector for this list */
 857         i_shift_offset   = DIM*shiftidx[iidx];
 858
 859         /* Load limits for loop over neighbors */
 860         j_index_start    = jindex[iidx];
 861         j_index_end      = jindex[iidx+1];
 862
 863         /* Get outer coordinate index */
 864         inr              = iinr[iidx];
 865         i_coord_offset   = DIM*inr;
 866
 867         /* Load i particle coords and add shift vector */
 868         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 869                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 870
 871         fix0             = _mm_setzero_pd();
 872         fiy0             = _mm_setzero_pd();
 873         fiz0             = _mm_setzero_pd();
 874         fix1             = _mm_setzero_pd();
 875         fiy1             = _mm_setzero_pd();
 876         fiz1             = _mm_setzero_pd();
 877         fix2             = _mm_setzero_pd();
 878         fiy2             = _mm_setzero_pd();
 879         fiz2             = _mm_setzero_pd();
 880         fix3             = _mm_setzero_pd();
 881         fiy3             = _mm_setzero_pd();
 882         fiz3             = _mm_setzero_pd();
 883
 884         /* Start inner kernel loop */
 885         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 886         {
 887
 888             /* Get j neighbor index, and coordinate index */
 889             jnrA             = jjnr[jidx];
 890             jnrB             = jjnr[jidx+1];
 891             j_coord_offsetA  = DIM*jnrA;
 892             j_coord_offsetB  = DIM*jnrB;
 893
 894             /* load j atom coordinates */
 895             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 896                                               &jx0,&jy0,&jz0);
 897
 898             /* Calculate displacement vector */
 899             dx00             = _mm_sub_pd(ix0,jx0);
 900             dy00             = _mm_sub_pd(iy0,jy0);
 901             dz00             = _mm_sub_pd(iz0,jz0);
 902             dx10             = _mm_sub_pd(ix1,jx0);
 903             dy10             = _mm_sub_pd(iy1,jy0);
 904             dz10             = _mm_sub_pd(iz1,jz0);
 905             dx20             = _mm_sub_pd(ix2,jx0);
 906             dy20             = _mm_sub_pd(iy2,jy0);
 907             dz20             = _mm_sub_pd(iz2,jz0);
 908             dx30             = _mm_sub_pd(ix3,jx0);
 909             dy30             = _mm_sub_pd(iy3,jy0);
 910             dz30             = _mm_sub_pd(iz3,jz0);
 911
 912             /* Calculate squared distance and things based on it */
 913             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 914             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 915             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 916             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 917
 918             rinv10           = avx128fma_invsqrt_d(rsq10);
 919             rinv20           = avx128fma_invsqrt_d(rsq20);
 920             rinv30           = avx128fma_invsqrt_d(rsq30);
 921
 922             rinvsq00         = avx128fma_inv_d(rsq00);
 923             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 924             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 925             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 926
 927             /* Load parameters for j particles */
 928             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 929             vdwjidx0A        = 2*vdwtype[jnrA+0];
 930             vdwjidx0B        = 2*vdwtype[jnrB+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             if (gmx_mm_any_lt(rsq00,rcutoff2))
 941             {
 942
 943             /* Compute parameters for interactions between i and j atoms */
 944             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 945                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 946
 947             /* LENNARD-JONES DISPERSION/REPULSION */
 948
 949             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 950             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 951
 952             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 953
 954             fscal            = fvdw;
 955
 956             fscal            = _mm_and_pd(fscal,cutoff_mask);
 957
 958             /* Update vectorial force */
 959             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 960             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 961             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 962
 963             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 964             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 965             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 966
 967             }
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             if (gmx_mm_any_lt(rsq10,rcutoff2))
 974             {
 975
 976             r10              = _mm_mul_pd(rsq10,rinv10);
 977
 978             /* Compute parameters for interactions between i and j atoms */
 979             qq10             = _mm_mul_pd(iq1,jq0);
 980
 981             /* EWALD ELECTROSTATICS */
 982
 983             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 984             ewrt             = _mm_mul_pd(r10,ewtabscale);
 985             ewitab           = _mm_cvttpd_epi32(ewrt);
 986 #ifdef __XOP__
 987             eweps            = _mm_frcz_pd(ewrt);
 988 #else
 989             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 990 #endif
 991             twoeweps         = _mm_add_pd(eweps,eweps);
 992             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 993                                          &ewtabF,&ewtabFn);
 994             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 995             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 996
 997             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 998
 999             fscal            = felec;
1000
1001             fscal            = _mm_and_pd(fscal,cutoff_mask);
1002
1003             /* Update vectorial force */
1004             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1005             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1006             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1007
1008             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1009             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1010             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1011
1012             }
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             if (gmx_mm_any_lt(rsq20,rcutoff2))
1019             {
1020
1021             r20              = _mm_mul_pd(rsq20,rinv20);
1022
1023             /* Compute parameters for interactions between i and j atoms */
1024             qq20             = _mm_mul_pd(iq2,jq0);
1025
1026             /* EWALD ELECTROSTATICS */
1027
1028             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1029             ewrt             = _mm_mul_pd(r20,ewtabscale);
1030             ewitab           = _mm_cvttpd_epi32(ewrt);
1031 #ifdef __XOP__
1032             eweps            = _mm_frcz_pd(ewrt);
1033 #else
1034             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1035 #endif
1036             twoeweps         = _mm_add_pd(eweps,eweps);
1037             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1038                                          &ewtabF,&ewtabFn);
1039             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1040             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1041
1042             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1043
1044             fscal            = felec;
1045
1046             fscal            = _mm_and_pd(fscal,cutoff_mask);
1047
1048             /* Update vectorial force */
1049             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1050             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1051             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1052
1053             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1054             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1055             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1056
1057             }
1058
1059             /**************************
1060              * CALCULATE INTERACTIONS *
1061              **************************/
1062
1063             if (gmx_mm_any_lt(rsq30,rcutoff2))
1064             {
1065
1066             r30              = _mm_mul_pd(rsq30,rinv30);
1067
1068             /* Compute parameters for interactions between i and j atoms */
1069             qq30             = _mm_mul_pd(iq3,jq0);
1070
1071             /* EWALD ELECTROSTATICS */
1072
1073             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1074             ewrt             = _mm_mul_pd(r30,ewtabscale);
1075             ewitab           = _mm_cvttpd_epi32(ewrt);
1076 #ifdef __XOP__
1077             eweps            = _mm_frcz_pd(ewrt);
1078 #else
1079             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1080 #endif
1081             twoeweps         = _mm_add_pd(eweps,eweps);
1082             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1083                                          &ewtabF,&ewtabFn);
1084             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1085             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1086
1087             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1088
1089             fscal            = felec;
1090
1091             fscal            = _mm_and_pd(fscal,cutoff_mask);
1092
1093             /* Update vectorial force */
1094             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1095             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1096             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1097
1098             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1099             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1100             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1101
1102             }
1103
1104             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1105
1106             /* Inner loop uses 162 flops */
1107         }
1108
1109         if(jidx<j_index_end)
1110         {
1111
1112             jnrA             = jjnr[jidx];
1113             j_coord_offsetA  = DIM*jnrA;
1114
1115             /* load j atom coordinates */
1116             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1117                                               &jx0,&jy0,&jz0);
1118
1119             /* Calculate displacement vector */
1120             dx00             = _mm_sub_pd(ix0,jx0);
1121             dy00             = _mm_sub_pd(iy0,jy0);
1122             dz00             = _mm_sub_pd(iz0,jz0);
1123             dx10             = _mm_sub_pd(ix1,jx0);
1124             dy10             = _mm_sub_pd(iy1,jy0);
1125             dz10             = _mm_sub_pd(iz1,jz0);
1126             dx20             = _mm_sub_pd(ix2,jx0);
1127             dy20             = _mm_sub_pd(iy2,jy0);
1128             dz20             = _mm_sub_pd(iz2,jz0);
1129             dx30             = _mm_sub_pd(ix3,jx0);
1130             dy30             = _mm_sub_pd(iy3,jy0);
1131             dz30             = _mm_sub_pd(iz3,jz0);
1132
1133             /* Calculate squared distance and things based on it */
1134             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1135             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1136             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1137             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1138
1139             rinv10           = avx128fma_invsqrt_d(rsq10);
1140             rinv20           = avx128fma_invsqrt_d(rsq20);
1141             rinv30           = avx128fma_invsqrt_d(rsq30);
1142
1143             rinvsq00         = avx128fma_inv_d(rsq00);
1144             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1145             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1146             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1147
1148             /* Load parameters for j particles */
1149             jq0              = _mm_load_sd(charge+jnrA+0);
1150             vdwjidx0A        = 2*vdwtype[jnrA+0];
1151
1152             fjx0             = _mm_setzero_pd();
1153             fjy0             = _mm_setzero_pd();
1154             fjz0             = _mm_setzero_pd();
1155
1156             /**************************
1157              * CALCULATE INTERACTIONS *
1158              **************************/
1159
1160             if (gmx_mm_any_lt(rsq00,rcutoff2))
1161             {
1162
1163             /* Compute parameters for interactions between i and j atoms */
1164             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1165
1166             /* LENNARD-JONES DISPERSION/REPULSION */
1167
1168             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1169             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1170
1171             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1172
1173             fscal            = fvdw;
1174
1175             fscal            = _mm_and_pd(fscal,cutoff_mask);
1176
1177             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1178
1179             /* Update vectorial force */
1180             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1181             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1182             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1183
1184             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1185             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1186             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1187
1188             }
1189
1190             /**************************
1191              * CALCULATE INTERACTIONS *
1192              **************************/
1193
1194             if (gmx_mm_any_lt(rsq10,rcutoff2))
1195             {
1196
1197             r10              = _mm_mul_pd(rsq10,rinv10);
1198
1199             /* Compute parameters for interactions between i and j atoms */
1200             qq10             = _mm_mul_pd(iq1,jq0);
1201
1202             /* EWALD ELECTROSTATICS */
1203
1204             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1205             ewrt             = _mm_mul_pd(r10,ewtabscale);
1206             ewitab           = _mm_cvttpd_epi32(ewrt);
1207 #ifdef __XOP__
1208             eweps            = _mm_frcz_pd(ewrt);
1209 #else
1210             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1211 #endif
1212             twoeweps         = _mm_add_pd(eweps,eweps);
1213             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1214             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1215             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1216
1217             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1218
1219             fscal            = felec;
1220
1221             fscal            = _mm_and_pd(fscal,cutoff_mask);
1222
1223             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1224
1225             /* Update vectorial force */
1226             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1227             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1228             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1229
1230             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1231             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1232             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1233
1234             }
1235
1236             /**************************
1237              * CALCULATE INTERACTIONS *
1238              **************************/
1239
1240             if (gmx_mm_any_lt(rsq20,rcutoff2))
1241             {
1242
1243             r20              = _mm_mul_pd(rsq20,rinv20);
1244
1245             /* Compute parameters for interactions between i and j atoms */
1246             qq20             = _mm_mul_pd(iq2,jq0);
1247
1248             /* EWALD ELECTROSTATICS */
1249
1250             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1251             ewrt             = _mm_mul_pd(r20,ewtabscale);
1252             ewitab           = _mm_cvttpd_epi32(ewrt);
1253 #ifdef __XOP__
1254             eweps            = _mm_frcz_pd(ewrt);
1255 #else
1256             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1257 #endif
1258             twoeweps         = _mm_add_pd(eweps,eweps);
1259             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1260             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1261             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1262
1263             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1264
1265             fscal            = felec;
1266
1267             fscal            = _mm_and_pd(fscal,cutoff_mask);
1268
1269             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1270
1271             /* Update vectorial force */
1272             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1273             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1274             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1275
1276             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1277             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1278             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1279
1280             }
1281
1282             /**************************
1283              * CALCULATE INTERACTIONS *
1284              **************************/
1285
1286             if (gmx_mm_any_lt(rsq30,rcutoff2))
1287             {
1288
1289             r30              = _mm_mul_pd(rsq30,rinv30);
1290
1291             /* Compute parameters for interactions between i and j atoms */
1292             qq30             = _mm_mul_pd(iq3,jq0);
1293
1294             /* EWALD ELECTROSTATICS */
1295
1296             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1297             ewrt             = _mm_mul_pd(r30,ewtabscale);
1298             ewitab           = _mm_cvttpd_epi32(ewrt);
1299 #ifdef __XOP__
1300             eweps            = _mm_frcz_pd(ewrt);
1301 #else
1302             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1303 #endif
1304             twoeweps         = _mm_add_pd(eweps,eweps);
1305             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1306             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1307             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1308
1309             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1310
1311             fscal            = felec;
1312
1313             fscal            = _mm_and_pd(fscal,cutoff_mask);
1314
1315             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1316
1317             /* Update vectorial force */
1318             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1319             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1320             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1321
1322             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1323             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1324             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1325
1326             }
1327
1328             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1329
1330             /* Inner loop uses 162 flops */
1331         }
1332
1333         /* End of innermost loop */
1334
1335         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1336                                               f+i_coord_offset,fshift+i_shift_offset);
1337
1338         /* Increment number of inner iterations */
1339         inneriter                  += j_index_end - j_index_start;
1340
1341         /* Outer loop uses 24 flops */
1342     }
1343
1344     /* Increment number of outer iterations */
1345     outeriter        += nri;
1346
1347     /* Update outer/inner flops */
1348
1349     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*162);
1350 }