src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128d           c6grid_00;
 102     __m128d           c6grid_10;
 103     __m128d           c6grid_20;
 104     real             *vdwgridparam;
 105     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 106     __m128d           one_half  = _mm_set1_pd(0.5);
 107     __m128d           minus_one = _mm_set1_pd(-1.0);
 108     __m128i          ewitab;
 109     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     real             *ewtab;
 111     __m128d          dummy_mask,cutoff_mask;
 112     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 113     __m128d          one     = _mm_set1_pd(1.0);
 114     __m128d          two     = _mm_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131     vdwgridparam     = fr->ljpme_c6grid;
 132     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 133     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 134     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 135
 136     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+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     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 173
 174         fix0             = _mm_setzero_pd();
 175         fiy0             = _mm_setzero_pd();
 176         fiz0             = _mm_setzero_pd();
 177         fix1             = _mm_setzero_pd();
 178         fiy1             = _mm_setzero_pd();
 179         fiz1             = _mm_setzero_pd();
 180         fix2             = _mm_setzero_pd();
 181         fiy2             = _mm_setzero_pd();
 182         fiz2             = _mm_setzero_pd();
 183
 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
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 217
 218             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 219             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 220             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 221
 222             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 223             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 224             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230
 231             fjx0             = _mm_setzero_pd();
 232             fjy0             = _mm_setzero_pd();
 233             fjz0             = _mm_setzero_pd();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             r00              = _mm_mul_pd(rsq00,rinv00);
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             qq00             = _mm_mul_pd(iq0,jq0);
 243             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 244                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 245             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 246                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 247
 248             /* EWALD ELECTROSTATICS */
 249
 250             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 251             ewrt             = _mm_mul_pd(r00,ewtabscale);
 252             ewitab           = _mm_cvttpd_epi32(ewrt);
 253 #ifdef __XOP__
 254             eweps            = _mm_frcz_pd(ewrt);
 255 #else
 256             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 257 #endif
 258             twoeweps         = _mm_add_pd(eweps,eweps);
 259             ewitab           = _mm_slli_epi32(ewitab,2);
 260             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 261             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 262             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 263             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 264             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 265             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 266             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 267             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 268             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 269             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 270
 271             /* Analytical LJ-PME */
 272             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 273             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 274             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 275             exponent         = gmx_simd_exp_d(ewcljrsq);
 276             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 277             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 278             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 279             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 280             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 281             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 282             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 283             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);
 284
 285             /* Update potential sum for this i atom from the interaction with this j atom. */
 286             velecsum         = _mm_add_pd(velecsum,velec);
 287             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 288
 289             fscal            = _mm_add_pd(felec,fvdw);
 290
 291             /* Update vectorial force */
 292             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 293             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 294             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 295
 296             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 297             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 298             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r10              = _mm_mul_pd(rsq10,rinv10);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq10             = _mm_mul_pd(iq1,jq0);
 308
 309             /* EWALD ELECTROSTATICS */
 310
 311             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 312             ewrt             = _mm_mul_pd(r10,ewtabscale);
 313             ewitab           = _mm_cvttpd_epi32(ewrt);
 314 #ifdef __XOP__
 315             eweps            = _mm_frcz_pd(ewrt);
 316 #else
 317             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 318 #endif
 319             twoeweps         = _mm_add_pd(eweps,eweps);
 320             ewitab           = _mm_slli_epi32(ewitab,2);
 321             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 322             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 323             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 324             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 325             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 326             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 327             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 328             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 329             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 330             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velecsum         = _mm_add_pd(velecsum,velec);
 334
 335             fscal            = felec;
 336
 337             /* Update vectorial force */
 338             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 339             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 340             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 341
 342             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 343             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 344             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             r20              = _mm_mul_pd(rsq20,rinv20);
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq20             = _mm_mul_pd(iq2,jq0);
 354
 355             /* EWALD ELECTROSTATICS */
 356
 357             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 358             ewrt             = _mm_mul_pd(r20,ewtabscale);
 359             ewitab           = _mm_cvttpd_epi32(ewrt);
 360 #ifdef __XOP__
 361             eweps            = _mm_frcz_pd(ewrt);
 362 #else
 363             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 364 #endif
 365             twoeweps         = _mm_add_pd(eweps,eweps);
 366             ewitab           = _mm_slli_epi32(ewitab,2);
 367             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 368             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 369             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 370             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 371             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 372             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 373             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 374             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 375             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 376             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 377
 378             /* Update potential sum for this i atom from the interaction with this j atom. */
 379             velecsum         = _mm_add_pd(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             /* Update vectorial force */
 384             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 385             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 386             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 387
 388             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 389             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 390             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 391
 392             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 393
 394             /* Inner loop uses 159 flops */
 395         }
 396
 397         if(jidx<j_index_end)
 398         {
 399
 400             jnrA             = jjnr[jidx];
 401             j_coord_offsetA  = DIM*jnrA;
 402
 403             /* load j atom coordinates */
 404             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 405                                               &jx0,&jy0,&jz0);
 406
 407             /* Calculate displacement vector */
 408             dx00             = _mm_sub_pd(ix0,jx0);
 409             dy00             = _mm_sub_pd(iy0,jy0);
 410             dz00             = _mm_sub_pd(iz0,jz0);
 411             dx10             = _mm_sub_pd(ix1,jx0);
 412             dy10             = _mm_sub_pd(iy1,jy0);
 413             dz10             = _mm_sub_pd(iz1,jz0);
 414             dx20             = _mm_sub_pd(ix2,jx0);
 415             dy20             = _mm_sub_pd(iy2,jy0);
 416             dz20             = _mm_sub_pd(iz2,jz0);
 417
 418             /* Calculate squared distance and things based on it */
 419             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 420             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 421             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 422
 423             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 424             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 425             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 426
 427             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 428             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 429             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 430
 431             /* Load parameters for j particles */
 432             jq0              = _mm_load_sd(charge+jnrA+0);
 433             vdwjidx0A        = 2*vdwtype[jnrA+0];
 434
 435             fjx0             = _mm_setzero_pd();
 436             fjy0             = _mm_setzero_pd();
 437             fjz0             = _mm_setzero_pd();
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             r00              = _mm_mul_pd(rsq00,rinv00);
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq00             = _mm_mul_pd(iq0,jq0);
 447             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 448             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 449
 450             /* EWALD ELECTROSTATICS */
 451
 452             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 453             ewrt             = _mm_mul_pd(r00,ewtabscale);
 454             ewitab           = _mm_cvttpd_epi32(ewrt);
 455 #ifdef __XOP__
 456             eweps            = _mm_frcz_pd(ewrt);
 457 #else
 458             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 459 #endif
 460             twoeweps         = _mm_add_pd(eweps,eweps);
 461             ewitab           = _mm_slli_epi32(ewitab,2);
 462             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 463             ewtabD           = _mm_setzero_pd();
 464             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 465             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 466             ewtabFn          = _mm_setzero_pd();
 467             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 468             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 469             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 470             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 471             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 472
 473             /* Analytical LJ-PME */
 474             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 475             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 476             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 477             exponent         = gmx_simd_exp_d(ewcljrsq);
 478             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 479             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 480             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 481             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 482             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 483             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 484             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 485             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);
 486
 487             /* Update potential sum for this i atom from the interaction with this j atom. */
 488             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 489             velecsum         = _mm_add_pd(velecsum,velec);
 490             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 491             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 492
 493             fscal            = _mm_add_pd(felec,fvdw);
 494
 495             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 496
 497             /* Update vectorial force */
 498             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 499             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 500             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 501
 502             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 503             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 504             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 505
 506             /**************************
 507              * CALCULATE INTERACTIONS *
 508              **************************/
 509
 510             r10              = _mm_mul_pd(rsq10,rinv10);
 511
 512             /* Compute parameters for interactions between i and j atoms */
 513             qq10             = _mm_mul_pd(iq1,jq0);
 514
 515             /* EWALD ELECTROSTATICS */
 516
 517             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 518             ewrt             = _mm_mul_pd(r10,ewtabscale);
 519             ewitab           = _mm_cvttpd_epi32(ewrt);
 520 #ifdef __XOP__
 521             eweps            = _mm_frcz_pd(ewrt);
 522 #else
 523             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 524 #endif
 525             twoeweps         = _mm_add_pd(eweps,eweps);
 526             ewitab           = _mm_slli_epi32(ewitab,2);
 527             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 528             ewtabD           = _mm_setzero_pd();
 529             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 530             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 531             ewtabFn          = _mm_setzero_pd();
 532             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 533             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 534             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 535             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 536             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 537
 538             /* Update potential sum for this i atom from the interaction with this j atom. */
 539             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 540             velecsum         = _mm_add_pd(velecsum,velec);
 541
 542             fscal            = felec;
 543
 544             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 545
 546             /* Update vectorial force */
 547             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 548             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 549             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 550
 551             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 552             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 553             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             r20              = _mm_mul_pd(rsq20,rinv20);
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             qq20             = _mm_mul_pd(iq2,jq0);
 563
 564             /* EWALD ELECTROSTATICS */
 565
 566             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 567             ewrt             = _mm_mul_pd(r20,ewtabscale);
 568             ewitab           = _mm_cvttpd_epi32(ewrt);
 569 #ifdef __XOP__
 570             eweps            = _mm_frcz_pd(ewrt);
 571 #else
 572             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 573 #endif
 574             twoeweps         = _mm_add_pd(eweps,eweps);
 575             ewitab           = _mm_slli_epi32(ewitab,2);
 576             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 577             ewtabD           = _mm_setzero_pd();
 578             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 579             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 580             ewtabFn          = _mm_setzero_pd();
 581             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 582             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 583             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 584             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 585             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 586
 587             /* Update potential sum for this i atom from the interaction with this j atom. */
 588             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 589             velecsum         = _mm_add_pd(velecsum,velec);
 590
 591             fscal            = felec;
 592
 593             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 594
 595             /* Update vectorial force */
 596             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 597             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 598             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 599
 600             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 601             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 602             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 603
 604             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 605
 606             /* Inner loop uses 159 flops */
 607         }
 608
 609         /* End of innermost loop */
 610
 611         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 612                                               f+i_coord_offset,fshift+i_shift_offset);
 613
 614         ggid                        = gid[iidx];
 615         /* Update potential energies */
 616         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 617         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 618
 619         /* Increment number of inner iterations */
 620         inneriter                  += j_index_end - j_index_start;
 621
 622         /* Outer loop uses 20 flops */
 623     }
 624
 625     /* Increment number of outer iterations */
 626     outeriter        += nri;
 627
 628     /* Update outer/inner flops */
 629
 630     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*159);
 631 }
 632 /*
 633  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_128_fma_double
 634  * Electrostatics interaction: Ewald
 635  * VdW interaction:            LJEwald
 636  * Geometry:                   Water3-Particle
 637  * Calculate force/pot:        Force
 638  */
 639 void
 640 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_128_fma_double
 641                     (t_nblist                    * gmx_restrict       nlist,
 642                      rvec                        * gmx_restrict          xx,
 643                      rvec                        * gmx_restrict          ff,
 644                      t_forcerec                  * gmx_restrict          fr,
 645                      t_mdatoms                   * gmx_restrict     mdatoms,
 646                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 647                      t_nrnb                      * gmx_restrict        nrnb)
 648 {
 649     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 650      * just 0 for non-waters.
 651      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 652      * jnr indices corresponding to data put in the four positions in the SIMD register.
 653      */
 654     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 655     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 656     int              jnrA,jnrB;
 657     int              j_coord_offsetA,j_coord_offsetB;
 658     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 659     real             rcutoff_scalar;
 660     real             *shiftvec,*fshift,*x,*f;
 661     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 662     int              vdwioffset0;
 663     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 664     int              vdwioffset1;
 665     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 666     int              vdwioffset2;
 667     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 668     int              vdwjidx0A,vdwjidx0B;
 669     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 670     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 671     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 672     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 673     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 674     real             *charge;
 675     int              nvdwtype;
 676     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 677     int              *vdwtype;
 678     real             *vdwparam;
 679     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 680     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 681     __m128d           c6grid_00;
 682     __m128d           c6grid_10;
 683     __m128d           c6grid_20;
 684     real             *vdwgridparam;
 685     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 686     __m128d           one_half  = _mm_set1_pd(0.5);
 687     __m128d           minus_one = _mm_set1_pd(-1.0);
 688     __m128i          ewitab;
 689     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 690     real             *ewtab;
 691     __m128d          dummy_mask,cutoff_mask;
 692     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 693     __m128d          one     = _mm_set1_pd(1.0);
 694     __m128d          two     = _mm_set1_pd(2.0);
 695     x                = xx[0];
 696     f                = ff[0];
 697
 698     nri              = nlist->nri;
 699     iinr             = nlist->iinr;
 700     jindex           = nlist->jindex;
 701     jjnr             = nlist->jjnr;
 702     shiftidx         = nlist->shift;
 703     gid              = nlist->gid;
 704     shiftvec         = fr->shift_vec[0];
 705     fshift           = fr->fshift[0];
 706     facel            = _mm_set1_pd(fr->epsfac);
 707     charge           = mdatoms->chargeA;
 708     nvdwtype         = fr->ntype;
 709     vdwparam         = fr->nbfp;
 710     vdwtype          = mdatoms->typeA;
 711     vdwgridparam     = fr->ljpme_c6grid;
 712     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 713     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 714     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 715
 716     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 717     ewtab            = fr->ic->tabq_coul_F;
 718     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 719     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 720
 721     /* Setup water-specific parameters */
 722     inr              = nlist->iinr[0];
 723     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 724     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 725     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 726     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 727
 728     /* Avoid stupid compiler warnings */
 729     jnrA = jnrB = 0;
 730     j_coord_offsetA = 0;
 731     j_coord_offsetB = 0;
 732
 733     outeriter        = 0;
 734     inneriter        = 0;
 735
 736     /* Start outer loop over neighborlists */
 737     for(iidx=0; iidx<nri; iidx++)
 738     {
 739         /* Load shift vector for this list */
 740         i_shift_offset   = DIM*shiftidx[iidx];
 741
 742         /* Load limits for loop over neighbors */
 743         j_index_start    = jindex[iidx];
 744         j_index_end      = jindex[iidx+1];
 745
 746         /* Get outer coordinate index */
 747         inr              = iinr[iidx];
 748         i_coord_offset   = DIM*inr;
 749
 750         /* Load i particle coords and add shift vector */
 751         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 752                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 753
 754         fix0             = _mm_setzero_pd();
 755         fiy0             = _mm_setzero_pd();
 756         fiz0             = _mm_setzero_pd();
 757         fix1             = _mm_setzero_pd();
 758         fiy1             = _mm_setzero_pd();
 759         fiz1             = _mm_setzero_pd();
 760         fix2             = _mm_setzero_pd();
 761         fiy2             = _mm_setzero_pd();
 762         fiz2             = _mm_setzero_pd();
 763
 764         /* Start inner kernel loop */
 765         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 766         {
 767
 768             /* Get j neighbor index, and coordinate index */
 769             jnrA             = jjnr[jidx];
 770             jnrB             = jjnr[jidx+1];
 771             j_coord_offsetA  = DIM*jnrA;
 772             j_coord_offsetB  = DIM*jnrB;
 773
 774             /* load j atom coordinates */
 775             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 776                                               &jx0,&jy0,&jz0);
 777
 778             /* Calculate displacement vector */
 779             dx00             = _mm_sub_pd(ix0,jx0);
 780             dy00             = _mm_sub_pd(iy0,jy0);
 781             dz00             = _mm_sub_pd(iz0,jz0);
 782             dx10             = _mm_sub_pd(ix1,jx0);
 783             dy10             = _mm_sub_pd(iy1,jy0);
 784             dz10             = _mm_sub_pd(iz1,jz0);
 785             dx20             = _mm_sub_pd(ix2,jx0);
 786             dy20             = _mm_sub_pd(iy2,jy0);
 787             dz20             = _mm_sub_pd(iz2,jz0);
 788
 789             /* Calculate squared distance and things based on it */
 790             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 791             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 792             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 793
 794             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 795             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 796             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 797
 798             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 799             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 800             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 801
 802             /* Load parameters for j particles */
 803             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 804             vdwjidx0A        = 2*vdwtype[jnrA+0];
 805             vdwjidx0B        = 2*vdwtype[jnrB+0];
 806
 807             fjx0             = _mm_setzero_pd();
 808             fjy0             = _mm_setzero_pd();
 809             fjz0             = _mm_setzero_pd();
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             r00              = _mm_mul_pd(rsq00,rinv00);
 816
 817             /* Compute parameters for interactions between i and j atoms */
 818             qq00             = _mm_mul_pd(iq0,jq0);
 819             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 820                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 821             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 822                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 823
 824             /* EWALD ELECTROSTATICS */
 825
 826             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 827             ewrt             = _mm_mul_pd(r00,ewtabscale);
 828             ewitab           = _mm_cvttpd_epi32(ewrt);
 829 #ifdef __XOP__
 830             eweps            = _mm_frcz_pd(ewrt);
 831 #else
 832             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 833 #endif
 834             twoeweps         = _mm_add_pd(eweps,eweps);
 835             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 836                                          &ewtabF,&ewtabFn);
 837             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 838             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 839
 840             /* Analytical LJ-PME */
 841             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 842             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 843             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 844             exponent         = gmx_simd_exp_d(ewcljrsq);
 845             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 846             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 847             /* f6A = 6 * C6grid * (1 - poly) */
 848             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 849             /* f6B = C6grid * exponent * beta^6 */
 850             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 851             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 852             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 853
 854             fscal            = _mm_add_pd(felec,fvdw);
 855
 856             /* Update vectorial force */
 857             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 858             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 859             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 860
 861             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 862             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 863             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 864
 865             /**************************
 866              * CALCULATE INTERACTIONS *
 867              **************************/
 868
 869             r10              = _mm_mul_pd(rsq10,rinv10);
 870
 871             /* Compute parameters for interactions between i and j atoms */
 872             qq10             = _mm_mul_pd(iq1,jq0);
 873
 874             /* EWALD ELECTROSTATICS */
 875
 876             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 877             ewrt             = _mm_mul_pd(r10,ewtabscale);
 878             ewitab           = _mm_cvttpd_epi32(ewrt);
 879 #ifdef __XOP__
 880             eweps            = _mm_frcz_pd(ewrt);
 881 #else
 882             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 883 #endif
 884             twoeweps         = _mm_add_pd(eweps,eweps);
 885             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 886                                          &ewtabF,&ewtabFn);
 887             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 888             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 889
 890             fscal            = felec;
 891
 892             /* Update vectorial force */
 893             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 894             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 895             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 896
 897             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 898             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 899             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             r20              = _mm_mul_pd(rsq20,rinv20);
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             qq20             = _mm_mul_pd(iq2,jq0);
 909
 910             /* EWALD ELECTROSTATICS */
 911
 912             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 913             ewrt             = _mm_mul_pd(r20,ewtabscale);
 914             ewitab           = _mm_cvttpd_epi32(ewrt);
 915 #ifdef __XOP__
 916             eweps            = _mm_frcz_pd(ewrt);
 917 #else
 918             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 919 #endif
 920             twoeweps         = _mm_add_pd(eweps,eweps);
 921             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 922                                          &ewtabF,&ewtabFn);
 923             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 924             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 925
 926             fscal            = felec;
 927
 928             /* Update vectorial force */
 929             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 930             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 931             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 932
 933             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 934             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 935             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 936
 937             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 938
 939             /* Inner loop uses 141 flops */
 940         }
 941
 942         if(jidx<j_index_end)
 943         {
 944
 945             jnrA             = jjnr[jidx];
 946             j_coord_offsetA  = DIM*jnrA;
 947
 948             /* load j atom coordinates */
 949             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 950                                               &jx0,&jy0,&jz0);
 951
 952             /* Calculate displacement vector */
 953             dx00             = _mm_sub_pd(ix0,jx0);
 954             dy00             = _mm_sub_pd(iy0,jy0);
 955             dz00             = _mm_sub_pd(iz0,jz0);
 956             dx10             = _mm_sub_pd(ix1,jx0);
 957             dy10             = _mm_sub_pd(iy1,jy0);
 958             dz10             = _mm_sub_pd(iz1,jz0);
 959             dx20             = _mm_sub_pd(ix2,jx0);
 960             dy20             = _mm_sub_pd(iy2,jy0);
 961             dz20             = _mm_sub_pd(iz2,jz0);
 962
 963             /* Calculate squared distance and things based on it */
 964             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 965             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 966             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 967
 968             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 969             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 970             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 971
 972             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 973             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 974             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 975
 976             /* Load parameters for j particles */
 977             jq0              = _mm_load_sd(charge+jnrA+0);
 978             vdwjidx0A        = 2*vdwtype[jnrA+0];
 979
 980             fjx0             = _mm_setzero_pd();
 981             fjy0             = _mm_setzero_pd();
 982             fjz0             = _mm_setzero_pd();
 983
 984             /**************************
 985              * CALCULATE INTERACTIONS *
 986              **************************/
 987
 988             r00              = _mm_mul_pd(rsq00,rinv00);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq00             = _mm_mul_pd(iq0,jq0);
 992             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 993             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 994
 995             /* EWALD ELECTROSTATICS */
 996
 997             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 998             ewrt             = _mm_mul_pd(r00,ewtabscale);
 999             ewitab           = _mm_cvttpd_epi32(ewrt);
1000 #ifdef __XOP__
1001             eweps            = _mm_frcz_pd(ewrt);
1002 #else
1003             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1004 #endif
1005             twoeweps         = _mm_add_pd(eweps,eweps);
1006             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1007             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1008             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1009
1010             /* Analytical LJ-PME */
1011             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1012             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1013             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1014             exponent         = gmx_simd_exp_d(ewcljrsq);
1015             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1016             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1017             /* f6A = 6 * C6grid * (1 - poly) */
1018             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1019             /* f6B = C6grid * exponent * beta^6 */
1020             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1021             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1022             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1023
1024             fscal            = _mm_add_pd(felec,fvdw);
1025
1026             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1027
1028             /* Update vectorial force */
1029             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1030             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1031             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1032
1033             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1034             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1035             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             r10              = _mm_mul_pd(rsq10,rinv10);
1042
1043             /* Compute parameters for interactions between i and j atoms */
1044             qq10             = _mm_mul_pd(iq1,jq0);
1045
1046             /* EWALD ELECTROSTATICS */
1047
1048             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1049             ewrt             = _mm_mul_pd(r10,ewtabscale);
1050             ewitab           = _mm_cvttpd_epi32(ewrt);
1051 #ifdef __XOP__
1052             eweps            = _mm_frcz_pd(ewrt);
1053 #else
1054             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1055 #endif
1056             twoeweps         = _mm_add_pd(eweps,eweps);
1057             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1058             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1059             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1060
1061             fscal            = felec;
1062
1063             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1064
1065             /* Update vectorial force */
1066             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1067             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1068             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1069
1070             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1071             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1072             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1073
1074             /**************************
1075              * CALCULATE INTERACTIONS *
1076              **************************/
1077
1078             r20              = _mm_mul_pd(rsq20,rinv20);
1079
1080             /* Compute parameters for interactions between i and j atoms */
1081             qq20             = _mm_mul_pd(iq2,jq0);
1082
1083             /* EWALD ELECTROSTATICS */
1084
1085             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1086             ewrt             = _mm_mul_pd(r20,ewtabscale);
1087             ewitab           = _mm_cvttpd_epi32(ewrt);
1088 #ifdef __XOP__
1089             eweps            = _mm_frcz_pd(ewrt);
1090 #else
1091             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1092 #endif
1093             twoeweps         = _mm_add_pd(eweps,eweps);
1094             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1095             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1096             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1097
1098             fscal            = felec;
1099
1100             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1101
1102             /* Update vectorial force */
1103             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1104             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1105             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1106
1107             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1108             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1109             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1110
1111             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1112
1113             /* Inner loop uses 141 flops */
1114         }
1115
1116         /* End of innermost loop */
1117
1118         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1119                                               f+i_coord_offset,fshift+i_shift_offset);
1120
1121         /* Increment number of inner iterations */
1122         inneriter                  += j_index_end - j_index_start;
1123
1124         /* Outer loop uses 18 flops */
1125     }
1126
1127     /* Increment number of outer iterations */
1128     outeriter        += nri;
1129
1130     /* Update outer/inner flops */
1131
1132     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*141);
1133 }