src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwLJSh_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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128i          ewitab;
 100     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 101     real             *ewtab;
 102     __m128d          dummy_mask,cutoff_mask;
 103     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 104     __m128d          one     = _mm_set1_pd(1.0);
 105     __m128d          two     = _mm_set1_pd(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_pd(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 124     ewtab            = fr->ic->tabq_coul_FDV0;
 125     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 126     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 131     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 132     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 133     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 136     rcutoff_scalar   = fr->rcoulomb;
 137     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 138     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 139
 140     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 141     rvdw             = _mm_set1_pd(fr->rvdw);
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 167                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 168
 169         fix0             = _mm_setzero_pd();
 170         fiy0             = _mm_setzero_pd();
 171         fiz0             = _mm_setzero_pd();
 172         fix1             = _mm_setzero_pd();
 173         fiy1             = _mm_setzero_pd();
 174         fiz1             = _mm_setzero_pd();
 175         fix2             = _mm_setzero_pd();
 176         fiy2             = _mm_setzero_pd();
 177         fiz2             = _mm_setzero_pd();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_pd();
 181         vvdwsum          = _mm_setzero_pd();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             j_coord_offsetA  = DIM*jnrA;
 191             j_coord_offsetB  = DIM*jnrB;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_pd(ix0,jx0);
 199             dy00             = _mm_sub_pd(iy0,jy0);
 200             dz00             = _mm_sub_pd(iz0,jz0);
 201             dx10             = _mm_sub_pd(ix1,jx0);
 202             dy10             = _mm_sub_pd(iy1,jy0);
 203             dz10             = _mm_sub_pd(iz1,jz0);
 204             dx20             = _mm_sub_pd(ix2,jx0);
 205             dy20             = _mm_sub_pd(iy2,jy0);
 206             dz20             = _mm_sub_pd(iz2,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 210             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 211             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 212
 213             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 214             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 215             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 216
 217             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 218             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 219             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 223             vdwjidx0A        = 2*vdwtype[jnrA+0];
 224             vdwjidx0B        = 2*vdwtype[jnrB+0];
 225
 226             fjx0             = _mm_setzero_pd();
 227             fjy0             = _mm_setzero_pd();
 228             fjz0             = _mm_setzero_pd();
 229
 230             /**************************
 231              * CALCULATE INTERACTIONS *
 232              **************************/
 233
 234             if (gmx_mm_any_lt(rsq00,rcutoff2))
 235             {
 236
 237             r00              = _mm_mul_pd(rsq00,rinv00);
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq00             = _mm_mul_pd(iq0,jq0);
 241             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 242                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 243
 244             /* EWALD ELECTROSTATICS */
 245
 246             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 247             ewrt             = _mm_mul_pd(r00,ewtabscale);
 248             ewitab           = _mm_cvttpd_epi32(ewrt);
 249 #ifdef __XOP__
 250             eweps            = _mm_frcz_pd(ewrt);
 251 #else
 252             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 253 #endif
 254             twoeweps         = _mm_add_pd(eweps,eweps);
 255             ewitab           = _mm_slli_epi32(ewitab,2);
 256             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 257             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 258             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 259             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 260             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 261             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 262             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 263             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 264             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 265             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 266
 267             /* LENNARD-JONES DISPERSION/REPULSION */
 268
 269             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 270             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 271             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 272             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 273                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 274             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 275
 276             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velec            = _mm_and_pd(velec,cutoff_mask);
 280             velecsum         = _mm_add_pd(velecsum,velec);
 281             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 282             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 283
 284             fscal            = _mm_add_pd(felec,fvdw);
 285
 286             fscal            = _mm_and_pd(fscal,cutoff_mask);
 287
 288             /* Update vectorial force */
 289             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 290             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 291             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 292
 293             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 294             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 295             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 296
 297             }
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             if (gmx_mm_any_lt(rsq10,rcutoff2))
 304             {
 305
 306             r10              = _mm_mul_pd(rsq10,rinv10);
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq10             = _mm_mul_pd(iq1,jq0);
 310
 311             /* EWALD ELECTROSTATICS */
 312
 313             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 314             ewrt             = _mm_mul_pd(r10,ewtabscale);
 315             ewitab           = _mm_cvttpd_epi32(ewrt);
 316 #ifdef __XOP__
 317             eweps            = _mm_frcz_pd(ewrt);
 318 #else
 319             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 320 #endif
 321             twoeweps         = _mm_add_pd(eweps,eweps);
 322             ewitab           = _mm_slli_epi32(ewitab,2);
 323             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 324             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 325             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 326             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 327             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 328             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 329             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 330             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 331             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 332             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 333
 334             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velec            = _mm_and_pd(velec,cutoff_mask);
 338             velecsum         = _mm_add_pd(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342             fscal            = _mm_and_pd(fscal,cutoff_mask);
 343
 344             /* Update vectorial force */
 345             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 346             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 347             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 348
 349             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 350             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 351             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 352
 353             }
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             if (gmx_mm_any_lt(rsq20,rcutoff2))
 360             {
 361
 362             r20              = _mm_mul_pd(rsq20,rinv20);
 363
 364             /* Compute parameters for interactions between i and j atoms */
 365             qq20             = _mm_mul_pd(iq2,jq0);
 366
 367             /* EWALD ELECTROSTATICS */
 368
 369             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 370             ewrt             = _mm_mul_pd(r20,ewtabscale);
 371             ewitab           = _mm_cvttpd_epi32(ewrt);
 372 #ifdef __XOP__
 373             eweps            = _mm_frcz_pd(ewrt);
 374 #else
 375             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 376 #endif
 377             twoeweps         = _mm_add_pd(eweps,eweps);
 378             ewitab           = _mm_slli_epi32(ewitab,2);
 379             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 380             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 381             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 382             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 383             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 384             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 385             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 386             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 387             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 388             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 389
 390             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velec            = _mm_and_pd(velec,cutoff_mask);
 394             velecsum         = _mm_add_pd(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             fscal            = _mm_and_pd(fscal,cutoff_mask);
 399
 400             /* Update vectorial force */
 401             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 402             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 403             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 404
 405             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 406             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 407             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 408
 409             }
 410
 411             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 412
 413             /* Inner loop uses 168 flops */
 414         }
 415
 416         if(jidx<j_index_end)
 417         {
 418
 419             jnrA             = jjnr[jidx];
 420             j_coord_offsetA  = DIM*jnrA;
 421
 422             /* load j atom coordinates */
 423             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 424                                               &jx0,&jy0,&jz0);
 425
 426             /* Calculate displacement vector */
 427             dx00             = _mm_sub_pd(ix0,jx0);
 428             dy00             = _mm_sub_pd(iy0,jy0);
 429             dz00             = _mm_sub_pd(iz0,jz0);
 430             dx10             = _mm_sub_pd(ix1,jx0);
 431             dy10             = _mm_sub_pd(iy1,jy0);
 432             dz10             = _mm_sub_pd(iz1,jz0);
 433             dx20             = _mm_sub_pd(ix2,jx0);
 434             dy20             = _mm_sub_pd(iy2,jy0);
 435             dz20             = _mm_sub_pd(iz2,jz0);
 436
 437             /* Calculate squared distance and things based on it */
 438             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 439             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 440             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 441
 442             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 443             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 444             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 445
 446             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 447             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 448             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 449
 450             /* Load parameters for j particles */
 451             jq0              = _mm_load_sd(charge+jnrA+0);
 452             vdwjidx0A        = 2*vdwtype[jnrA+0];
 453
 454             fjx0             = _mm_setzero_pd();
 455             fjy0             = _mm_setzero_pd();
 456             fjz0             = _mm_setzero_pd();
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             if (gmx_mm_any_lt(rsq00,rcutoff2))
 463             {
 464
 465             r00              = _mm_mul_pd(rsq00,rinv00);
 466
 467             /* Compute parameters for interactions between i and j atoms */
 468             qq00             = _mm_mul_pd(iq0,jq0);
 469             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 470
 471             /* EWALD ELECTROSTATICS */
 472
 473             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 474             ewrt             = _mm_mul_pd(r00,ewtabscale);
 475             ewitab           = _mm_cvttpd_epi32(ewrt);
 476 #ifdef __XOP__
 477             eweps            = _mm_frcz_pd(ewrt);
 478 #else
 479             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 480 #endif
 481             twoeweps         = _mm_add_pd(eweps,eweps);
 482             ewitab           = _mm_slli_epi32(ewitab,2);
 483             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 484             ewtabD           = _mm_setzero_pd();
 485             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 486             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 487             ewtabFn          = _mm_setzero_pd();
 488             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 489             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 490             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 491             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 492             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 493
 494             /* LENNARD-JONES DISPERSION/REPULSION */
 495
 496             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 497             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 498             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 499             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 500                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 501             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 502
 503             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             velec            = _mm_and_pd(velec,cutoff_mask);
 507             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 508             velecsum         = _mm_add_pd(velecsum,velec);
 509             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 510             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 511             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 512
 513             fscal            = _mm_add_pd(felec,fvdw);
 514
 515             fscal            = _mm_and_pd(fscal,cutoff_mask);
 516
 517             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 518
 519             /* Update vectorial force */
 520             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 521             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 522             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 523
 524             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 525             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 526             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 527
 528             }
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             if (gmx_mm_any_lt(rsq10,rcutoff2))
 535             {
 536
 537             r10              = _mm_mul_pd(rsq10,rinv10);
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq10             = _mm_mul_pd(iq1,jq0);
 541
 542             /* EWALD ELECTROSTATICS */
 543
 544             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 545             ewrt             = _mm_mul_pd(r10,ewtabscale);
 546             ewitab           = _mm_cvttpd_epi32(ewrt);
 547 #ifdef __XOP__
 548             eweps            = _mm_frcz_pd(ewrt);
 549 #else
 550             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 551 #endif
 552             twoeweps         = _mm_add_pd(eweps,eweps);
 553             ewitab           = _mm_slli_epi32(ewitab,2);
 554             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 555             ewtabD           = _mm_setzero_pd();
 556             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 557             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 558             ewtabFn          = _mm_setzero_pd();
 559             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 560             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 561             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 562             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 563             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 564
 565             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 566
 567             /* Update potential sum for this i atom from the interaction with this j atom. */
 568             velec            = _mm_and_pd(velec,cutoff_mask);
 569             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 570             velecsum         = _mm_add_pd(velecsum,velec);
 571
 572             fscal            = felec;
 573
 574             fscal            = _mm_and_pd(fscal,cutoff_mask);
 575
 576             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 577
 578             /* Update vectorial force */
 579             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 580             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 581             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 582
 583             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 584             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 585             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 586
 587             }
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             if (gmx_mm_any_lt(rsq20,rcutoff2))
 594             {
 595
 596             r20              = _mm_mul_pd(rsq20,rinv20);
 597
 598             /* Compute parameters for interactions between i and j atoms */
 599             qq20             = _mm_mul_pd(iq2,jq0);
 600
 601             /* EWALD ELECTROSTATICS */
 602
 603             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 604             ewrt             = _mm_mul_pd(r20,ewtabscale);
 605             ewitab           = _mm_cvttpd_epi32(ewrt);
 606 #ifdef __XOP__
 607             eweps            = _mm_frcz_pd(ewrt);
 608 #else
 609             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 610 #endif
 611             twoeweps         = _mm_add_pd(eweps,eweps);
 612             ewitab           = _mm_slli_epi32(ewitab,2);
 613             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 614             ewtabD           = _mm_setzero_pd();
 615             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 616             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 617             ewtabFn          = _mm_setzero_pd();
 618             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 619             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 620             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 621             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 622             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 623
 624             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm_and_pd(velec,cutoff_mask);
 628             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 629             velecsum         = _mm_add_pd(velecsum,velec);
 630
 631             fscal            = felec;
 632
 633             fscal            = _mm_and_pd(fscal,cutoff_mask);
 634
 635             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 636
 637             /* Update vectorial force */
 638             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 639             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 640             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 641
 642             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 643             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 644             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 645
 646             }
 647
 648             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 649
 650             /* Inner loop uses 168 flops */
 651         }
 652
 653         /* End of innermost loop */
 654
 655         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 656                                               f+i_coord_offset,fshift+i_shift_offset);
 657
 658         ggid                        = gid[iidx];
 659         /* Update potential energies */
 660         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 661         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 662
 663         /* Increment number of inner iterations */
 664         inneriter                  += j_index_end - j_index_start;
 665
 666         /* Outer loop uses 20 flops */
 667     }
 668
 669     /* Increment number of outer iterations */
 670     outeriter        += nri;
 671
 672     /* Update outer/inner flops */
 673
 674     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*168);
 675 }
 676 /*
 677  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_double
 678  * Electrostatics interaction: Ewald
 679  * VdW interaction:            LennardJones
 680  * Geometry:                   Water3-Particle
 681  * Calculate force/pot:        Force
 682  */
 683 void
 684 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_double
 685                     (t_nblist                    * gmx_restrict       nlist,
 686                      rvec                        * gmx_restrict          xx,
 687                      rvec                        * gmx_restrict          ff,
 688                      t_forcerec                  * gmx_restrict          fr,
 689                      t_mdatoms                   * gmx_restrict     mdatoms,
 690                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 691                      t_nrnb                      * gmx_restrict        nrnb)
 692 {
 693     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 694      * just 0 for non-waters.
 695      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 696      * jnr indices corresponding to data put in the four positions in the SIMD register.
 697      */
 698     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 699     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 700     int              jnrA,jnrB;
 701     int              j_coord_offsetA,j_coord_offsetB;
 702     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 703     real             rcutoff_scalar;
 704     real             *shiftvec,*fshift,*x,*f;
 705     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 706     int              vdwioffset0;
 707     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 708     int              vdwioffset1;
 709     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 710     int              vdwioffset2;
 711     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 712     int              vdwjidx0A,vdwjidx0B;
 713     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 714     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 715     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 716     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 717     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 718     real             *charge;
 719     int              nvdwtype;
 720     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 721     int              *vdwtype;
 722     real             *vdwparam;
 723     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 724     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 725     __m128i          ewitab;
 726     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 727     real             *ewtab;
 728     __m128d          dummy_mask,cutoff_mask;
 729     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 730     __m128d          one     = _mm_set1_pd(1.0);
 731     __m128d          two     = _mm_set1_pd(2.0);
 732     x                = xx[0];
 733     f                = ff[0];
 734
 735     nri              = nlist->nri;
 736     iinr             = nlist->iinr;
 737     jindex           = nlist->jindex;
 738     jjnr             = nlist->jjnr;
 739     shiftidx         = nlist->shift;
 740     gid              = nlist->gid;
 741     shiftvec         = fr->shift_vec[0];
 742     fshift           = fr->fshift[0];
 743     facel            = _mm_set1_pd(fr->epsfac);
 744     charge           = mdatoms->chargeA;
 745     nvdwtype         = fr->ntype;
 746     vdwparam         = fr->nbfp;
 747     vdwtype          = mdatoms->typeA;
 748
 749     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 750     ewtab            = fr->ic->tabq_coul_F;
 751     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 752     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 753
 754     /* Setup water-specific parameters */
 755     inr              = nlist->iinr[0];
 756     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 757     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 758     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 759     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 760
 761     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 762     rcutoff_scalar   = fr->rcoulomb;
 763     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 764     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 765
 766     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 767     rvdw             = _mm_set1_pd(fr->rvdw);
 768
 769     /* Avoid stupid compiler warnings */
 770     jnrA = jnrB = 0;
 771     j_coord_offsetA = 0;
 772     j_coord_offsetB = 0;
 773
 774     outeriter        = 0;
 775     inneriter        = 0;
 776
 777     /* Start outer loop over neighborlists */
 778     for(iidx=0; iidx<nri; iidx++)
 779     {
 780         /* Load shift vector for this list */
 781         i_shift_offset   = DIM*shiftidx[iidx];
 782
 783         /* Load limits for loop over neighbors */
 784         j_index_start    = jindex[iidx];
 785         j_index_end      = jindex[iidx+1];
 786
 787         /* Get outer coordinate index */
 788         inr              = iinr[iidx];
 789         i_coord_offset   = DIM*inr;
 790
 791         /* Load i particle coords and add shift vector */
 792         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 793                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 794
 795         fix0             = _mm_setzero_pd();
 796         fiy0             = _mm_setzero_pd();
 797         fiz0             = _mm_setzero_pd();
 798         fix1             = _mm_setzero_pd();
 799         fiy1             = _mm_setzero_pd();
 800         fiz1             = _mm_setzero_pd();
 801         fix2             = _mm_setzero_pd();
 802         fiy2             = _mm_setzero_pd();
 803         fiz2             = _mm_setzero_pd();
 804
 805         /* Start inner kernel loop */
 806         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 807         {
 808
 809             /* Get j neighbor index, and coordinate index */
 810             jnrA             = jjnr[jidx];
 811             jnrB             = jjnr[jidx+1];
 812             j_coord_offsetA  = DIM*jnrA;
 813             j_coord_offsetB  = DIM*jnrB;
 814
 815             /* load j atom coordinates */
 816             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 817                                               &jx0,&jy0,&jz0);
 818
 819             /* Calculate displacement vector */
 820             dx00             = _mm_sub_pd(ix0,jx0);
 821             dy00             = _mm_sub_pd(iy0,jy0);
 822             dz00             = _mm_sub_pd(iz0,jz0);
 823             dx10             = _mm_sub_pd(ix1,jx0);
 824             dy10             = _mm_sub_pd(iy1,jy0);
 825             dz10             = _mm_sub_pd(iz1,jz0);
 826             dx20             = _mm_sub_pd(ix2,jx0);
 827             dy20             = _mm_sub_pd(iy2,jy0);
 828             dz20             = _mm_sub_pd(iz2,jz0);
 829
 830             /* Calculate squared distance and things based on it */
 831             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 832             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 833             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 834
 835             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 836             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 837             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 838
 839             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 840             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 841             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 842
 843             /* Load parameters for j particles */
 844             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 845             vdwjidx0A        = 2*vdwtype[jnrA+0];
 846             vdwjidx0B        = 2*vdwtype[jnrB+0];
 847
 848             fjx0             = _mm_setzero_pd();
 849             fjy0             = _mm_setzero_pd();
 850             fjz0             = _mm_setzero_pd();
 851
 852             /**************************
 853              * CALCULATE INTERACTIONS *
 854              **************************/
 855
 856             if (gmx_mm_any_lt(rsq00,rcutoff2))
 857             {
 858
 859             r00              = _mm_mul_pd(rsq00,rinv00);
 860
 861             /* Compute parameters for interactions between i and j atoms */
 862             qq00             = _mm_mul_pd(iq0,jq0);
 863             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 864                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 865
 866             /* EWALD ELECTROSTATICS */
 867
 868             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 869             ewrt             = _mm_mul_pd(r00,ewtabscale);
 870             ewitab           = _mm_cvttpd_epi32(ewrt);
 871 #ifdef __XOP__
 872             eweps            = _mm_frcz_pd(ewrt);
 873 #else
 874             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 875 #endif
 876             twoeweps         = _mm_add_pd(eweps,eweps);
 877             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 878                                          &ewtabF,&ewtabFn);
 879             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 880             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 881
 882             /* LENNARD-JONES DISPERSION/REPULSION */
 883
 884             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 885             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 886
 887             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 888
 889             fscal            = _mm_add_pd(felec,fvdw);
 890
 891             fscal            = _mm_and_pd(fscal,cutoff_mask);
 892
 893             /* Update vectorial force */
 894             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 895             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 896             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 897
 898             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 899             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 900             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 901
 902             }
 903
 904             /**************************
 905              * CALCULATE INTERACTIONS *
 906              **************************/
 907
 908             if (gmx_mm_any_lt(rsq10,rcutoff2))
 909             {
 910
 911             r10              = _mm_mul_pd(rsq10,rinv10);
 912
 913             /* Compute parameters for interactions between i and j atoms */
 914             qq10             = _mm_mul_pd(iq1,jq0);
 915
 916             /* EWALD ELECTROSTATICS */
 917
 918             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 919             ewrt             = _mm_mul_pd(r10,ewtabscale);
 920             ewitab           = _mm_cvttpd_epi32(ewrt);
 921 #ifdef __XOP__
 922             eweps            = _mm_frcz_pd(ewrt);
 923 #else
 924             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 925 #endif
 926             twoeweps         = _mm_add_pd(eweps,eweps);
 927             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 928                                          &ewtabF,&ewtabFn);
 929             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 930             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 931
 932             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 933
 934             fscal            = felec;
 935
 936             fscal            = _mm_and_pd(fscal,cutoff_mask);
 937
 938             /* Update vectorial force */
 939             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 940             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 941             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 942
 943             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 944             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 945             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 946
 947             }
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             if (gmx_mm_any_lt(rsq20,rcutoff2))
 954             {
 955
 956             r20              = _mm_mul_pd(rsq20,rinv20);
 957
 958             /* Compute parameters for interactions between i and j atoms */
 959             qq20             = _mm_mul_pd(iq2,jq0);
 960
 961             /* EWALD ELECTROSTATICS */
 962
 963             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 964             ewrt             = _mm_mul_pd(r20,ewtabscale);
 965             ewitab           = _mm_cvttpd_epi32(ewrt);
 966 #ifdef __XOP__
 967             eweps            = _mm_frcz_pd(ewrt);
 968 #else
 969             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 970 #endif
 971             twoeweps         = _mm_add_pd(eweps,eweps);
 972             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 973                                          &ewtabF,&ewtabFn);
 974             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 975             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 976
 977             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 978
 979             fscal            = felec;
 980
 981             fscal            = _mm_and_pd(fscal,cutoff_mask);
 982
 983             /* Update vectorial force */
 984             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 985             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 986             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 987
 988             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 989             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 990             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 991
 992             }
 993
 994             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 995
 996             /* Inner loop uses 136 flops */
 997         }
 998
 999         if(jidx<j_index_end)
1000         {
1001
1002             jnrA             = jjnr[jidx];
1003             j_coord_offsetA  = DIM*jnrA;
1004
1005             /* load j atom coordinates */
1006             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1007                                               &jx0,&jy0,&jz0);
1008
1009             /* Calculate displacement vector */
1010             dx00             = _mm_sub_pd(ix0,jx0);
1011             dy00             = _mm_sub_pd(iy0,jy0);
1012             dz00             = _mm_sub_pd(iz0,jz0);
1013             dx10             = _mm_sub_pd(ix1,jx0);
1014             dy10             = _mm_sub_pd(iy1,jy0);
1015             dz10             = _mm_sub_pd(iz1,jz0);
1016             dx20             = _mm_sub_pd(ix2,jx0);
1017             dy20             = _mm_sub_pd(iy2,jy0);
1018             dz20             = _mm_sub_pd(iz2,jz0);
1019
1020             /* Calculate squared distance and things based on it */
1021             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1022             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1023             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1024
1025             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1026             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1027             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1028
1029             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1030             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1031             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1032
1033             /* Load parameters for j particles */
1034             jq0              = _mm_load_sd(charge+jnrA+0);
1035             vdwjidx0A        = 2*vdwtype[jnrA+0];
1036
1037             fjx0             = _mm_setzero_pd();
1038             fjy0             = _mm_setzero_pd();
1039             fjz0             = _mm_setzero_pd();
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             if (gmx_mm_any_lt(rsq00,rcutoff2))
1046             {
1047
1048             r00              = _mm_mul_pd(rsq00,rinv00);
1049
1050             /* Compute parameters for interactions between i and j atoms */
1051             qq00             = _mm_mul_pd(iq0,jq0);
1052             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1053
1054             /* EWALD ELECTROSTATICS */
1055
1056             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1057             ewrt             = _mm_mul_pd(r00,ewtabscale);
1058             ewitab           = _mm_cvttpd_epi32(ewrt);
1059 #ifdef __XOP__
1060             eweps            = _mm_frcz_pd(ewrt);
1061 #else
1062             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1063 #endif
1064             twoeweps         = _mm_add_pd(eweps,eweps);
1065             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1066             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1067             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1068
1069             /* LENNARD-JONES DISPERSION/REPULSION */
1070
1071             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1072             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1073
1074             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1075
1076             fscal            = _mm_add_pd(felec,fvdw);
1077
1078             fscal            = _mm_and_pd(fscal,cutoff_mask);
1079
1080             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1081
1082             /* Update vectorial force */
1083             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1084             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1085             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1086
1087             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1088             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1089             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1090
1091             }
1092
1093             /**************************
1094              * CALCULATE INTERACTIONS *
1095              **************************/
1096
1097             if (gmx_mm_any_lt(rsq10,rcutoff2))
1098             {
1099
1100             r10              = _mm_mul_pd(rsq10,rinv10);
1101
1102             /* Compute parameters for interactions between i and j atoms */
1103             qq10             = _mm_mul_pd(iq1,jq0);
1104
1105             /* EWALD ELECTROSTATICS */
1106
1107             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1108             ewrt             = _mm_mul_pd(r10,ewtabscale);
1109             ewitab           = _mm_cvttpd_epi32(ewrt);
1110 #ifdef __XOP__
1111             eweps            = _mm_frcz_pd(ewrt);
1112 #else
1113             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1114 #endif
1115             twoeweps         = _mm_add_pd(eweps,eweps);
1116             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1117             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1118             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1119
1120             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1121
1122             fscal            = felec;
1123
1124             fscal            = _mm_and_pd(fscal,cutoff_mask);
1125
1126             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1127
1128             /* Update vectorial force */
1129             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1130             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1131             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1132
1133             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1134             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1135             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1136
1137             }
1138
1139             /**************************
1140              * CALCULATE INTERACTIONS *
1141              **************************/
1142
1143             if (gmx_mm_any_lt(rsq20,rcutoff2))
1144             {
1145
1146             r20              = _mm_mul_pd(rsq20,rinv20);
1147
1148             /* Compute parameters for interactions between i and j atoms */
1149             qq20             = _mm_mul_pd(iq2,jq0);
1150
1151             /* EWALD ELECTROSTATICS */
1152
1153             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1154             ewrt             = _mm_mul_pd(r20,ewtabscale);
1155             ewitab           = _mm_cvttpd_epi32(ewrt);
1156 #ifdef __XOP__
1157             eweps            = _mm_frcz_pd(ewrt);
1158 #else
1159             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1160 #endif
1161             twoeweps         = _mm_add_pd(eweps,eweps);
1162             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1163             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1164             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1165
1166             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1167
1168             fscal            = felec;
1169
1170             fscal            = _mm_and_pd(fscal,cutoff_mask);
1171
1172             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1173
1174             /* Update vectorial force */
1175             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1176             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1177             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1178
1179             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1180             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1181             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1182
1183             }
1184
1185             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1186
1187             /* Inner loop uses 136 flops */
1188         }
1189
1190         /* End of innermost loop */
1191
1192         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1193                                               f+i_coord_offset,fshift+i_shift_offset);
1194
1195         /* Increment number of inner iterations */
1196         inneriter                  += j_index_end - j_index_start;
1197
1198         /* Outer loop uses 18 flops */
1199     }
1200
1201     /* Increment number of outer iterations */
1202     outeriter        += nri;
1203
1204     /* Update outer/inner flops */
1205
1206     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*136);
1207 }