src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJEw_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_ElecEw_VdwLJEw_GeomW4P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_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     __m128d           c6grid_00;
 102     __m128d           c6grid_10;
 103     __m128d           c6grid_20;
 104     __m128d           c6grid_30;
 105     real             *vdwgridparam;
 106     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 107     __m128d           one_half  = _mm_set1_pd(0.5);
 108     __m128d           minus_one = _mm_set1_pd(-1.0);
 109     __m128i          ewitab;
 110     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     real             *ewtab;
 112     __m128d          dummy_mask,cutoff_mask;
 113     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 114     __m128d          one     = _mm_set1_pd(1.0);
 115     __m128d          two     = _mm_set1_pd(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm_set1_pd(fr->ic->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132     vdwgridparam     = fr->ljpme_c6grid;
 133     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 134     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 135     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 136
 137     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 138     ewtab            = fr->ic->tabq_coul_FDV0;
 139     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 140     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 145     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 146     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 147     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149     /* Avoid stupid compiler warnings */
 150     jnrA = jnrB = 0;
 151     j_coord_offsetA = 0;
 152     j_coord_offsetB = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 174
 175         fix0             = _mm_setzero_pd();
 176         fiy0             = _mm_setzero_pd();
 177         fiz0             = _mm_setzero_pd();
 178         fix1             = _mm_setzero_pd();
 179         fiy1             = _mm_setzero_pd();
 180         fiz1             = _mm_setzero_pd();
 181         fix2             = _mm_setzero_pd();
 182         fiy2             = _mm_setzero_pd();
 183         fiz2             = _mm_setzero_pd();
 184         fix3             = _mm_setzero_pd();
 185         fiy3             = _mm_setzero_pd();
 186         fiz3             = _mm_setzero_pd();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm_setzero_pd();
 190         vvdwsum          = _mm_setzero_pd();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201
 202             /* load j atom coordinates */
 203             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                               &jx0,&jy0,&jz0);
 205
 206             /* Calculate displacement vector */
 207             dx00             = _mm_sub_pd(ix0,jx0);
 208             dy00             = _mm_sub_pd(iy0,jy0);
 209             dz00             = _mm_sub_pd(iz0,jz0);
 210             dx10             = _mm_sub_pd(ix1,jx0);
 211             dy10             = _mm_sub_pd(iy1,jy0);
 212             dz10             = _mm_sub_pd(iz1,jz0);
 213             dx20             = _mm_sub_pd(ix2,jx0);
 214             dy20             = _mm_sub_pd(iy2,jy0);
 215             dz20             = _mm_sub_pd(iz2,jz0);
 216             dx30             = _mm_sub_pd(ix3,jx0);
 217             dy30             = _mm_sub_pd(iy3,jy0);
 218             dz30             = _mm_sub_pd(iz3,jz0);
 219
 220             /* Calculate squared distance and things based on it */
 221             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 222             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 223             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 224             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 225
 226             rinv00           = avx128fma_invsqrt_d(rsq00);
 227             rinv10           = avx128fma_invsqrt_d(rsq10);
 228             rinv20           = avx128fma_invsqrt_d(rsq20);
 229             rinv30           = avx128fma_invsqrt_d(rsq30);
 230
 231             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 232             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 233             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 234             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 235
 236             /* Load parameters for j particles */
 237             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 238             vdwjidx0A        = 2*vdwtype[jnrA+0];
 239             vdwjidx0B        = 2*vdwtype[jnrB+0];
 240
 241             fjx0             = _mm_setzero_pd();
 242             fjy0             = _mm_setzero_pd();
 243             fjz0             = _mm_setzero_pd();
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             r00              = _mm_mul_pd(rsq00,rinv00);
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 253                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 254             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 255                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 256
 257             /* Analytical LJ-PME */
 258             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 259             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 260             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 261             exponent         = avx128fma_exp_d(ewcljrsq);
 262             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 263             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 264             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 265             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 266             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 267             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 268             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 269             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 270
 271             /* Update potential sum for this i atom from the interaction with this j atom. */
 272             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 273
 274             fscal            = fvdw;
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 278             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 279             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 280
 281             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 282             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 283             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             r10              = _mm_mul_pd(rsq10,rinv10);
 290
 291             /* Compute parameters for interactions between i and j atoms */
 292             qq10             = _mm_mul_pd(iq1,jq0);
 293
 294             /* EWALD ELECTROSTATICS */
 295
 296             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 297             ewrt             = _mm_mul_pd(r10,ewtabscale);
 298             ewitab           = _mm_cvttpd_epi32(ewrt);
 299 #ifdef __XOP__
 300             eweps            = _mm_frcz_pd(ewrt);
 301 #else
 302             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 303 #endif
 304             twoeweps         = _mm_add_pd(eweps,eweps);
 305             ewitab           = _mm_slli_epi32(ewitab,2);
 306             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 307             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 308             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 309             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 310             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 311             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 312             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 313             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 314             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 315             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velecsum         = _mm_add_pd(velecsum,velec);
 319
 320             fscal            = felec;
 321
 322             /* Update vectorial force */
 323             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 324             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 325             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 326
 327             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 328             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 329             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             r20              = _mm_mul_pd(rsq20,rinv20);
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq20             = _mm_mul_pd(iq2,jq0);
 339
 340             /* EWALD ELECTROSTATICS */
 341
 342             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 343             ewrt             = _mm_mul_pd(r20,ewtabscale);
 344             ewitab           = _mm_cvttpd_epi32(ewrt);
 345 #ifdef __XOP__
 346             eweps            = _mm_frcz_pd(ewrt);
 347 #else
 348             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 349 #endif
 350             twoeweps         = _mm_add_pd(eweps,eweps);
 351             ewitab           = _mm_slli_epi32(ewitab,2);
 352             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 353             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 354             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 355             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 356             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 357             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 358             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 359             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 360             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 361             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velecsum         = _mm_add_pd(velecsum,velec);
 365
 366             fscal            = felec;
 367
 368             /* Update vectorial force */
 369             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 370             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 371             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 372
 373             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 374             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 375             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             r30              = _mm_mul_pd(rsq30,rinv30);
 382
 383             /* Compute parameters for interactions between i and j atoms */
 384             qq30             = _mm_mul_pd(iq3,jq0);
 385
 386             /* EWALD ELECTROSTATICS */
 387
 388             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 389             ewrt             = _mm_mul_pd(r30,ewtabscale);
 390             ewitab           = _mm_cvttpd_epi32(ewrt);
 391 #ifdef __XOP__
 392             eweps            = _mm_frcz_pd(ewrt);
 393 #else
 394             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 395 #endif
 396             twoeweps         = _mm_add_pd(eweps,eweps);
 397             ewitab           = _mm_slli_epi32(ewitab,2);
 398             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 399             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 400             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 401             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 402             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 403             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 404             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 405             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 406             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 407             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velecsum         = _mm_add_pd(velecsum,velec);
 411
 412             fscal            = felec;
 413
 414             /* Update vectorial force */
 415             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 416             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 417             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 418
 419             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 420             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 421             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 422
 423             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 424
 425             /* Inner loop uses 185 flops */
 426         }
 427
 428         if(jidx<j_index_end)
 429         {
 430
 431             jnrA             = jjnr[jidx];
 432             j_coord_offsetA  = DIM*jnrA;
 433
 434             /* load j atom coordinates */
 435             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 436                                               &jx0,&jy0,&jz0);
 437
 438             /* Calculate displacement vector */
 439             dx00             = _mm_sub_pd(ix0,jx0);
 440             dy00             = _mm_sub_pd(iy0,jy0);
 441             dz00             = _mm_sub_pd(iz0,jz0);
 442             dx10             = _mm_sub_pd(ix1,jx0);
 443             dy10             = _mm_sub_pd(iy1,jy0);
 444             dz10             = _mm_sub_pd(iz1,jz0);
 445             dx20             = _mm_sub_pd(ix2,jx0);
 446             dy20             = _mm_sub_pd(iy2,jy0);
 447             dz20             = _mm_sub_pd(iz2,jz0);
 448             dx30             = _mm_sub_pd(ix3,jx0);
 449             dy30             = _mm_sub_pd(iy3,jy0);
 450             dz30             = _mm_sub_pd(iz3,jz0);
 451
 452             /* Calculate squared distance and things based on it */
 453             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 454             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 455             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 456             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 457
 458             rinv00           = avx128fma_invsqrt_d(rsq00);
 459             rinv10           = avx128fma_invsqrt_d(rsq10);
 460             rinv20           = avx128fma_invsqrt_d(rsq20);
 461             rinv30           = avx128fma_invsqrt_d(rsq30);
 462
 463             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 464             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 465             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 466             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 467
 468             /* Load parameters for j particles */
 469             jq0              = _mm_load_sd(charge+jnrA+0);
 470             vdwjidx0A        = 2*vdwtype[jnrA+0];
 471
 472             fjx0             = _mm_setzero_pd();
 473             fjy0             = _mm_setzero_pd();
 474             fjz0             = _mm_setzero_pd();
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             r00              = _mm_mul_pd(rsq00,rinv00);
 481
 482             /* Compute parameters for interactions between i and j atoms */
 483             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 484             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 485
 486             /* Analytical LJ-PME */
 487             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 488             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 489             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 490             exponent         = avx128fma_exp_d(ewcljrsq);
 491             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 492             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 493             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 494             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 495             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 496             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 497             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 498             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 499
 500             /* Update potential sum for this i atom from the interaction with this j atom. */
 501             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 502             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 503
 504             fscal            = fvdw;
 505
 506             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 507
 508             /* Update vectorial force */
 509             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 510             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 511             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 512
 513             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 514             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 515             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 516
 517             /**************************
 518              * CALCULATE INTERACTIONS *
 519              **************************/
 520
 521             r10              = _mm_mul_pd(rsq10,rinv10);
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             qq10             = _mm_mul_pd(iq1,jq0);
 525
 526             /* EWALD ELECTROSTATICS */
 527
 528             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 529             ewrt             = _mm_mul_pd(r10,ewtabscale);
 530             ewitab           = _mm_cvttpd_epi32(ewrt);
 531 #ifdef __XOP__
 532             eweps            = _mm_frcz_pd(ewrt);
 533 #else
 534             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 535 #endif
 536             twoeweps         = _mm_add_pd(eweps,eweps);
 537             ewitab           = _mm_slli_epi32(ewitab,2);
 538             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 539             ewtabD           = _mm_setzero_pd();
 540             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 541             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 542             ewtabFn          = _mm_setzero_pd();
 543             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 544             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 545             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 546             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 547             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 548
 549             /* Update potential sum for this i atom from the interaction with this j atom. */
 550             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 551             velecsum         = _mm_add_pd(velecsum,velec);
 552
 553             fscal            = felec;
 554
 555             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 556
 557             /* Update vectorial force */
 558             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 559             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 560             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 561
 562             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 563             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 564             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             r20              = _mm_mul_pd(rsq20,rinv20);
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq20             = _mm_mul_pd(iq2,jq0);
 574
 575             /* EWALD ELECTROSTATICS */
 576
 577             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 578             ewrt             = _mm_mul_pd(r20,ewtabscale);
 579             ewitab           = _mm_cvttpd_epi32(ewrt);
 580 #ifdef __XOP__
 581             eweps            = _mm_frcz_pd(ewrt);
 582 #else
 583             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 584 #endif
 585             twoeweps         = _mm_add_pd(eweps,eweps);
 586             ewitab           = _mm_slli_epi32(ewitab,2);
 587             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 588             ewtabD           = _mm_setzero_pd();
 589             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 590             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 591             ewtabFn          = _mm_setzero_pd();
 592             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 593             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 594             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 595             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 596             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 597
 598             /* Update potential sum for this i atom from the interaction with this j atom. */
 599             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 600             velecsum         = _mm_add_pd(velecsum,velec);
 601
 602             fscal            = felec;
 603
 604             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 605
 606             /* Update vectorial force */
 607             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 608             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 609             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 610
 611             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 612             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 613             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             r30              = _mm_mul_pd(rsq30,rinv30);
 620
 621             /* Compute parameters for interactions between i and j atoms */
 622             qq30             = _mm_mul_pd(iq3,jq0);
 623
 624             /* EWALD ELECTROSTATICS */
 625
 626             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 627             ewrt             = _mm_mul_pd(r30,ewtabscale);
 628             ewitab           = _mm_cvttpd_epi32(ewrt);
 629 #ifdef __XOP__
 630             eweps            = _mm_frcz_pd(ewrt);
 631 #else
 632             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 633 #endif
 634             twoeweps         = _mm_add_pd(eweps,eweps);
 635             ewitab           = _mm_slli_epi32(ewitab,2);
 636             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 637             ewtabD           = _mm_setzero_pd();
 638             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 639             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 640             ewtabFn          = _mm_setzero_pd();
 641             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 642             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 643             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 644             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 645             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 646
 647             /* Update potential sum for this i atom from the interaction with this j atom. */
 648             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 649             velecsum         = _mm_add_pd(velecsum,velec);
 650
 651             fscal            = felec;
 652
 653             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 654
 655             /* Update vectorial force */
 656             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 657             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 658             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 659
 660             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 661             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 662             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 663
 664             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 665
 666             /* Inner loop uses 185 flops */
 667         }
 668
 669         /* End of innermost loop */
 670
 671         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 672                                               f+i_coord_offset,fshift+i_shift_offset);
 673
 674         ggid                        = gid[iidx];
 675         /* Update potential energies */
 676         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 677         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 678
 679         /* Increment number of inner iterations */
 680         inneriter                  += j_index_end - j_index_start;
 681
 682         /* Outer loop uses 26 flops */
 683     }
 684
 685     /* Increment number of outer iterations */
 686     outeriter        += nri;
 687
 688     /* Update outer/inner flops */
 689
 690     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*185);
 691 }
 692 /*
 693  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_double
 694  * Electrostatics interaction: Ewald
 695  * VdW interaction:            LJEwald
 696  * Geometry:                   Water4-Particle
 697  * Calculate force/pot:        Force
 698  */
 699 void
 700 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_double
 701                     (t_nblist                    * gmx_restrict       nlist,
 702                      rvec                        * gmx_restrict          xx,
 703                      rvec                        * gmx_restrict          ff,
 704                      struct t_forcerec           * gmx_restrict          fr,
 705                      t_mdatoms                   * gmx_restrict     mdatoms,
 706                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 707                      t_nrnb                      * gmx_restrict        nrnb)
 708 {
 709     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 710      * just 0 for non-waters.
 711      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 712      * jnr indices corresponding to data put in the four positions in the SIMD register.
 713      */
 714     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 715     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 716     int              jnrA,jnrB;
 717     int              j_coord_offsetA,j_coord_offsetB;
 718     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 719     real             rcutoff_scalar;
 720     real             *shiftvec,*fshift,*x,*f;
 721     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 722     int              vdwioffset0;
 723     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 724     int              vdwioffset1;
 725     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 726     int              vdwioffset2;
 727     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 728     int              vdwioffset3;
 729     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 730     int              vdwjidx0A,vdwjidx0B;
 731     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 732     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 733     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 734     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 735     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 736     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 737     real             *charge;
 738     int              nvdwtype;
 739     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 740     int              *vdwtype;
 741     real             *vdwparam;
 742     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 743     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 744     __m128d           c6grid_00;
 745     __m128d           c6grid_10;
 746     __m128d           c6grid_20;
 747     __m128d           c6grid_30;
 748     real             *vdwgridparam;
 749     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 750     __m128d           one_half  = _mm_set1_pd(0.5);
 751     __m128d           minus_one = _mm_set1_pd(-1.0);
 752     __m128i          ewitab;
 753     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 754     real             *ewtab;
 755     __m128d          dummy_mask,cutoff_mask;
 756     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 757     __m128d          one     = _mm_set1_pd(1.0);
 758     __m128d          two     = _mm_set1_pd(2.0);
 759     x                = xx[0];
 760     f                = ff[0];
 761
 762     nri              = nlist->nri;
 763     iinr             = nlist->iinr;
 764     jindex           = nlist->jindex;
 765     jjnr             = nlist->jjnr;
 766     shiftidx         = nlist->shift;
 767     gid              = nlist->gid;
 768     shiftvec         = fr->shift_vec[0];
 769     fshift           = fr->fshift[0];
 770     facel            = _mm_set1_pd(fr->ic->epsfac);
 771     charge           = mdatoms->chargeA;
 772     nvdwtype         = fr->ntype;
 773     vdwparam         = fr->nbfp;
 774     vdwtype          = mdatoms->typeA;
 775     vdwgridparam     = fr->ljpme_c6grid;
 776     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 777     ewclj            = _mm_set1_pd(fr->ic->ewaldcoeff_lj);
 778     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 779
 780     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 781     ewtab            = fr->ic->tabq_coul_F;
 782     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 783     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 784
 785     /* Setup water-specific parameters */
 786     inr              = nlist->iinr[0];
 787     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 788     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 789     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 790     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 791
 792     /* Avoid stupid compiler warnings */
 793     jnrA = jnrB = 0;
 794     j_coord_offsetA = 0;
 795     j_coord_offsetB = 0;
 796
 797     outeriter        = 0;
 798     inneriter        = 0;
 799
 800     /* Start outer loop over neighborlists */
 801     for(iidx=0; iidx<nri; iidx++)
 802     {
 803         /* Load shift vector for this list */
 804         i_shift_offset   = DIM*shiftidx[iidx];
 805
 806         /* Load limits for loop over neighbors */
 807         j_index_start    = jindex[iidx];
 808         j_index_end      = jindex[iidx+1];
 809
 810         /* Get outer coordinate index */
 811         inr              = iinr[iidx];
 812         i_coord_offset   = DIM*inr;
 813
 814         /* Load i particle coords and add shift vector */
 815         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 816                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 817
 818         fix0             = _mm_setzero_pd();
 819         fiy0             = _mm_setzero_pd();
 820         fiz0             = _mm_setzero_pd();
 821         fix1             = _mm_setzero_pd();
 822         fiy1             = _mm_setzero_pd();
 823         fiz1             = _mm_setzero_pd();
 824         fix2             = _mm_setzero_pd();
 825         fiy2             = _mm_setzero_pd();
 826         fiz2             = _mm_setzero_pd();
 827         fix3             = _mm_setzero_pd();
 828         fiy3             = _mm_setzero_pd();
 829         fiz3             = _mm_setzero_pd();
 830
 831         /* Start inner kernel loop */
 832         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 833         {
 834
 835             /* Get j neighbor index, and coordinate index */
 836             jnrA             = jjnr[jidx];
 837             jnrB             = jjnr[jidx+1];
 838             j_coord_offsetA  = DIM*jnrA;
 839             j_coord_offsetB  = DIM*jnrB;
 840
 841             /* load j atom coordinates */
 842             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 843                                               &jx0,&jy0,&jz0);
 844
 845             /* Calculate displacement vector */
 846             dx00             = _mm_sub_pd(ix0,jx0);
 847             dy00             = _mm_sub_pd(iy0,jy0);
 848             dz00             = _mm_sub_pd(iz0,jz0);
 849             dx10             = _mm_sub_pd(ix1,jx0);
 850             dy10             = _mm_sub_pd(iy1,jy0);
 851             dz10             = _mm_sub_pd(iz1,jz0);
 852             dx20             = _mm_sub_pd(ix2,jx0);
 853             dy20             = _mm_sub_pd(iy2,jy0);
 854             dz20             = _mm_sub_pd(iz2,jz0);
 855             dx30             = _mm_sub_pd(ix3,jx0);
 856             dy30             = _mm_sub_pd(iy3,jy0);
 857             dz30             = _mm_sub_pd(iz3,jz0);
 858
 859             /* Calculate squared distance and things based on it */
 860             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 861             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 862             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 863             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 864
 865             rinv00           = avx128fma_invsqrt_d(rsq00);
 866             rinv10           = avx128fma_invsqrt_d(rsq10);
 867             rinv20           = avx128fma_invsqrt_d(rsq20);
 868             rinv30           = avx128fma_invsqrt_d(rsq30);
 869
 870             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 871             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 872             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 873             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 874
 875             /* Load parameters for j particles */
 876             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 877             vdwjidx0A        = 2*vdwtype[jnrA+0];
 878             vdwjidx0B        = 2*vdwtype[jnrB+0];
 879
 880             fjx0             = _mm_setzero_pd();
 881             fjy0             = _mm_setzero_pd();
 882             fjz0             = _mm_setzero_pd();
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             r00              = _mm_mul_pd(rsq00,rinv00);
 889
 890             /* Compute parameters for interactions between i and j atoms */
 891             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 892                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 893             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 894                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 895
 896             /* Analytical LJ-PME */
 897             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 898             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 899             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 900             exponent         = avx128fma_exp_d(ewcljrsq);
 901             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 902             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 903             /* f6A = 6 * C6grid * (1 - poly) */
 904             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 905             /* f6B = C6grid * exponent * beta^6 */
 906             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 907             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 908             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 909
 910             fscal            = fvdw;
 911
 912             /* Update vectorial force */
 913             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 914             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 915             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 916
 917             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 918             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 919             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 920
 921             /**************************
 922              * CALCULATE INTERACTIONS *
 923              **************************/
 924
 925             r10              = _mm_mul_pd(rsq10,rinv10);
 926
 927             /* Compute parameters for interactions between i and j atoms */
 928             qq10             = _mm_mul_pd(iq1,jq0);
 929
 930             /* EWALD ELECTROSTATICS */
 931
 932             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 933             ewrt             = _mm_mul_pd(r10,ewtabscale);
 934             ewitab           = _mm_cvttpd_epi32(ewrt);
 935 #ifdef __XOP__
 936             eweps            = _mm_frcz_pd(ewrt);
 937 #else
 938             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 939 #endif
 940             twoeweps         = _mm_add_pd(eweps,eweps);
 941             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 942                                          &ewtabF,&ewtabFn);
 943             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 944             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 945
 946             fscal            = felec;
 947
 948             /* Update vectorial force */
 949             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 950             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 951             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 952
 953             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 954             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 955             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             r20              = _mm_mul_pd(rsq20,rinv20);
 962
 963             /* Compute parameters for interactions between i and j atoms */
 964             qq20             = _mm_mul_pd(iq2,jq0);
 965
 966             /* EWALD ELECTROSTATICS */
 967
 968             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 969             ewrt             = _mm_mul_pd(r20,ewtabscale);
 970             ewitab           = _mm_cvttpd_epi32(ewrt);
 971 #ifdef __XOP__
 972             eweps            = _mm_frcz_pd(ewrt);
 973 #else
 974             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 975 #endif
 976             twoeweps         = _mm_add_pd(eweps,eweps);
 977             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 978                                          &ewtabF,&ewtabFn);
 979             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 980             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 981
 982             fscal            = felec;
 983
 984             /* Update vectorial force */
 985             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 986             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 987             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 988
 989             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 990             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 991             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 992
 993             /**************************
 994              * CALCULATE INTERACTIONS *
 995              **************************/
 996
 997             r30              = _mm_mul_pd(rsq30,rinv30);
 998
 999             /* Compute parameters for interactions between i and j atoms */
1000             qq30             = _mm_mul_pd(iq3,jq0);
1001
1002             /* EWALD ELECTROSTATICS */
1003
1004             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1005             ewrt             = _mm_mul_pd(r30,ewtabscale);
1006             ewitab           = _mm_cvttpd_epi32(ewrt);
1007 #ifdef __XOP__
1008             eweps            = _mm_frcz_pd(ewrt);
1009 #else
1010             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1011 #endif
1012             twoeweps         = _mm_add_pd(eweps,eweps);
1013             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1014                                          &ewtabF,&ewtabFn);
1015             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1016             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1017
1018             fscal            = felec;
1019
1020             /* Update vectorial force */
1021             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1022             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1023             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1024
1025             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1026             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1027             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1028
1029             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1030
1031             /* Inner loop uses 167 flops */
1032         }
1033
1034         if(jidx<j_index_end)
1035         {
1036
1037             jnrA             = jjnr[jidx];
1038             j_coord_offsetA  = DIM*jnrA;
1039
1040             /* load j atom coordinates */
1041             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1042                                               &jx0,&jy0,&jz0);
1043
1044             /* Calculate displacement vector */
1045             dx00             = _mm_sub_pd(ix0,jx0);
1046             dy00             = _mm_sub_pd(iy0,jy0);
1047             dz00             = _mm_sub_pd(iz0,jz0);
1048             dx10             = _mm_sub_pd(ix1,jx0);
1049             dy10             = _mm_sub_pd(iy1,jy0);
1050             dz10             = _mm_sub_pd(iz1,jz0);
1051             dx20             = _mm_sub_pd(ix2,jx0);
1052             dy20             = _mm_sub_pd(iy2,jy0);
1053             dz20             = _mm_sub_pd(iz2,jz0);
1054             dx30             = _mm_sub_pd(ix3,jx0);
1055             dy30             = _mm_sub_pd(iy3,jy0);
1056             dz30             = _mm_sub_pd(iz3,jz0);
1057
1058             /* Calculate squared distance and things based on it */
1059             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1060             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1061             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1062             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1063
1064             rinv00           = avx128fma_invsqrt_d(rsq00);
1065             rinv10           = avx128fma_invsqrt_d(rsq10);
1066             rinv20           = avx128fma_invsqrt_d(rsq20);
1067             rinv30           = avx128fma_invsqrt_d(rsq30);
1068
1069             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1070             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1071             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1072             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1073
1074             /* Load parameters for j particles */
1075             jq0              = _mm_load_sd(charge+jnrA+0);
1076             vdwjidx0A        = 2*vdwtype[jnrA+0];
1077
1078             fjx0             = _mm_setzero_pd();
1079             fjy0             = _mm_setzero_pd();
1080             fjz0             = _mm_setzero_pd();
1081
1082             /**************************
1083              * CALCULATE INTERACTIONS *
1084              **************************/
1085
1086             r00              = _mm_mul_pd(rsq00,rinv00);
1087
1088             /* Compute parameters for interactions between i and j atoms */
1089             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1090             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1091
1092             /* Analytical LJ-PME */
1093             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1094             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1095             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1096             exponent         = avx128fma_exp_d(ewcljrsq);
1097             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1098             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1099             /* f6A = 6 * C6grid * (1 - poly) */
1100             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1101             /* f6B = C6grid * exponent * beta^6 */
1102             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1103             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1104             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1105
1106             fscal            = fvdw;
1107
1108             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1109
1110             /* Update vectorial force */
1111             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1112             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1113             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1114
1115             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1116             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1117             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1118
1119             /**************************
1120              * CALCULATE INTERACTIONS *
1121              **************************/
1122
1123             r10              = _mm_mul_pd(rsq10,rinv10);
1124
1125             /* Compute parameters for interactions between i and j atoms */
1126             qq10             = _mm_mul_pd(iq1,jq0);
1127
1128             /* EWALD ELECTROSTATICS */
1129
1130             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1131             ewrt             = _mm_mul_pd(r10,ewtabscale);
1132             ewitab           = _mm_cvttpd_epi32(ewrt);
1133 #ifdef __XOP__
1134             eweps            = _mm_frcz_pd(ewrt);
1135 #else
1136             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1137 #endif
1138             twoeweps         = _mm_add_pd(eweps,eweps);
1139             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1140             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1141             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1142
1143             fscal            = felec;
1144
1145             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1146
1147             /* Update vectorial force */
1148             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1149             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1150             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1151
1152             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1153             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1154             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1155
1156             /**************************
1157              * CALCULATE INTERACTIONS *
1158              **************************/
1159
1160             r20              = _mm_mul_pd(rsq20,rinv20);
1161
1162             /* Compute parameters for interactions between i and j atoms */
1163             qq20             = _mm_mul_pd(iq2,jq0);
1164
1165             /* EWALD ELECTROSTATICS */
1166
1167             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1168             ewrt             = _mm_mul_pd(r20,ewtabscale);
1169             ewitab           = _mm_cvttpd_epi32(ewrt);
1170 #ifdef __XOP__
1171             eweps            = _mm_frcz_pd(ewrt);
1172 #else
1173             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1174 #endif
1175             twoeweps         = _mm_add_pd(eweps,eweps);
1176             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1177             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1178             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1179
1180             fscal            = felec;
1181
1182             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1183
1184             /* Update vectorial force */
1185             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1186             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1187             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1188
1189             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1190             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1191             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1192
1193             /**************************
1194              * CALCULATE INTERACTIONS *
1195              **************************/
1196
1197             r30              = _mm_mul_pd(rsq30,rinv30);
1198
1199             /* Compute parameters for interactions between i and j atoms */
1200             qq30             = _mm_mul_pd(iq3,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(r30,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(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1216
1217             fscal            = felec;
1218
1219             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1220
1221             /* Update vectorial force */
1222             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1223             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1224             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1225
1226             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1227             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1228             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1229
1230             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1231
1232             /* Inner loop uses 167 flops */
1233         }
1234
1235         /* End of innermost loop */
1236
1237         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1238                                               f+i_coord_offset,fshift+i_shift_offset);
1239
1240         /* Increment number of inner iterations */
1241         inneriter                  += j_index_end - j_index_start;
1242
1243         /* Outer loop uses 24 flops */
1244     }
1245
1246     /* Increment number of outer iterations */
1247     outeriter        += nri;
1248
1249     /* Update outer/inner flops */
1250
1251     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*167);
1252 }