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, 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 "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_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_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     __m128i          ewitab;
 102     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 103     real             *ewtab;
 104     __m128d          dummy_mask,cutoff_mask;
 105     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 106     __m128d          one     = _mm_set1_pd(1.0);
 107     __m128d          two     = _mm_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 126     ewtab            = fr->ic->tabq_coul_FDV0;
 127     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 128     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 133     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 134     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->rcoulomb;
 139     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 140     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 141
 142     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 143     rvdw             = _mm_set1_pd(fr->rvdw);
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 170
 171         fix0             = _mm_setzero_pd();
 172         fiy0             = _mm_setzero_pd();
 173         fiz0             = _mm_setzero_pd();
 174         fix1             = _mm_setzero_pd();
 175         fiy1             = _mm_setzero_pd();
 176         fiz1             = _mm_setzero_pd();
 177         fix2             = _mm_setzero_pd();
 178         fiy2             = _mm_setzero_pd();
 179         fiz2             = _mm_setzero_pd();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm_setzero_pd();
 183         vvdwsum          = _mm_setzero_pd();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194
 195             /* load j atom coordinates */
 196             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm_sub_pd(ix0,jx0);
 201             dy00             = _mm_sub_pd(iy0,jy0);
 202             dz00             = _mm_sub_pd(iz0,jz0);
 203             dx10             = _mm_sub_pd(ix1,jx0);
 204             dy10             = _mm_sub_pd(iy1,jy0);
 205             dz10             = _mm_sub_pd(iz1,jz0);
 206             dx20             = _mm_sub_pd(ix2,jx0);
 207             dy20             = _mm_sub_pd(iy2,jy0);
 208             dz20             = _mm_sub_pd(iz2,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 212             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 213             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 214
 215             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 216             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 217             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 218
 219             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 220             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 221             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227
 228             fjx0             = _mm_setzero_pd();
 229             fjy0             = _mm_setzero_pd();
 230             fjz0             = _mm_setzero_pd();
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             if (gmx_mm_any_lt(rsq00,rcutoff2))
 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
 246             /* EWALD ELECTROSTATICS */
 247
 248             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 249             ewrt             = _mm_mul_pd(r00,ewtabscale);
 250             ewitab           = _mm_cvttpd_epi32(ewrt);
 251 #ifdef __XOP__
 252             eweps            = _mm_frcz_pd(ewrt);
 253 #else
 254             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 255 #endif
 256             twoeweps         = _mm_add_pd(eweps,eweps);
 257             ewitab           = _mm_slli_epi32(ewitab,2);
 258             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 259             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 260             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 261             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 262             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 263             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 264             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 265             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 266             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 267             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 268
 269             /* LENNARD-JONES DISPERSION/REPULSION */
 270
 271             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 272             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 273             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 274             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 275                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 276             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 277
 278             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velec            = _mm_and_pd(velec,cutoff_mask);
 282             velecsum         = _mm_add_pd(velecsum,velec);
 283             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 284             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 285
 286             fscal            = _mm_add_pd(felec,fvdw);
 287
 288             fscal            = _mm_and_pd(fscal,cutoff_mask);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 292             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 293             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 294
 295             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 296             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 297             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 298
 299             }
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             if (gmx_mm_any_lt(rsq10,rcutoff2))
 306             {
 307
 308             r10              = _mm_mul_pd(rsq10,rinv10);
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq10             = _mm_mul_pd(iq1,jq0);
 312
 313             /* EWALD ELECTROSTATICS */
 314
 315             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 316             ewrt             = _mm_mul_pd(r10,ewtabscale);
 317             ewitab           = _mm_cvttpd_epi32(ewrt);
 318 #ifdef __XOP__
 319             eweps            = _mm_frcz_pd(ewrt);
 320 #else
 321             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 322 #endif
 323             twoeweps         = _mm_add_pd(eweps,eweps);
 324             ewitab           = _mm_slli_epi32(ewitab,2);
 325             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 326             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 327             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 328             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 329             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 330             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 331             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 332             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 333             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 334             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 335
 336             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm_and_pd(velec,cutoff_mask);
 340             velecsum         = _mm_add_pd(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             fscal            = _mm_and_pd(fscal,cutoff_mask);
 345
 346             /* Update vectorial force */
 347             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 348             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 349             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 350
 351             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 352             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 353             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 354
 355             }
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             if (gmx_mm_any_lt(rsq20,rcutoff2))
 362             {
 363
 364             r20              = _mm_mul_pd(rsq20,rinv20);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq20             = _mm_mul_pd(iq2,jq0);
 368
 369             /* EWALD ELECTROSTATICS */
 370
 371             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 372             ewrt             = _mm_mul_pd(r20,ewtabscale);
 373             ewitab           = _mm_cvttpd_epi32(ewrt);
 374 #ifdef __XOP__
 375             eweps            = _mm_frcz_pd(ewrt);
 376 #else
 377             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 378 #endif
 379             twoeweps         = _mm_add_pd(eweps,eweps);
 380             ewitab           = _mm_slli_epi32(ewitab,2);
 381             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 382             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 383             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 384             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 385             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 386             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 387             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 388             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 389             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 390             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 391
 392             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velec            = _mm_and_pd(velec,cutoff_mask);
 396             velecsum         = _mm_add_pd(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             fscal            = _mm_and_pd(fscal,cutoff_mask);
 401
 402             /* Update vectorial force */
 403             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 404             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 405             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 406
 407             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 408             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 409             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 410
 411             }
 412
 413             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 414
 415             /* Inner loop uses 168 flops */
 416         }
 417
 418         if(jidx<j_index_end)
 419         {
 420
 421             jnrA             = jjnr[jidx];
 422             j_coord_offsetA  = DIM*jnrA;
 423
 424             /* load j atom coordinates */
 425             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 426                                               &jx0,&jy0,&jz0);
 427
 428             /* Calculate displacement vector */
 429             dx00             = _mm_sub_pd(ix0,jx0);
 430             dy00             = _mm_sub_pd(iy0,jy0);
 431             dz00             = _mm_sub_pd(iz0,jz0);
 432             dx10             = _mm_sub_pd(ix1,jx0);
 433             dy10             = _mm_sub_pd(iy1,jy0);
 434             dz10             = _mm_sub_pd(iz1,jz0);
 435             dx20             = _mm_sub_pd(ix2,jx0);
 436             dy20             = _mm_sub_pd(iy2,jy0);
 437             dz20             = _mm_sub_pd(iz2,jz0);
 438
 439             /* Calculate squared distance and things based on it */
 440             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 441             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 442             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 443
 444             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 445             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 446             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 447
 448             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 449             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 450             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 451
 452             /* Load parameters for j particles */
 453             jq0              = _mm_load_sd(charge+jnrA+0);
 454             vdwjidx0A        = 2*vdwtype[jnrA+0];
 455
 456             fjx0             = _mm_setzero_pd();
 457             fjy0             = _mm_setzero_pd();
 458             fjz0             = _mm_setzero_pd();
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             if (gmx_mm_any_lt(rsq00,rcutoff2))
 465             {
 466
 467             r00              = _mm_mul_pd(rsq00,rinv00);
 468
 469             /* Compute parameters for interactions between i and j atoms */
 470             qq00             = _mm_mul_pd(iq0,jq0);
 471             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 472
 473             /* EWALD ELECTROSTATICS */
 474
 475             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 476             ewrt             = _mm_mul_pd(r00,ewtabscale);
 477             ewitab           = _mm_cvttpd_epi32(ewrt);
 478 #ifdef __XOP__
 479             eweps            = _mm_frcz_pd(ewrt);
 480 #else
 481             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 482 #endif
 483             twoeweps         = _mm_add_pd(eweps,eweps);
 484             ewitab           = _mm_slli_epi32(ewitab,2);
 485             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 486             ewtabD           = _mm_setzero_pd();
 487             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 488             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 489             ewtabFn          = _mm_setzero_pd();
 490             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 491             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 492             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 493             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 494             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 495
 496             /* LENNARD-JONES DISPERSION/REPULSION */
 497
 498             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 499             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 500             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 501             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 502                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 503             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 504
 505             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 506
 507             /* Update potential sum for this i atom from the interaction with this j atom. */
 508             velec            = _mm_and_pd(velec,cutoff_mask);
 509             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 510             velecsum         = _mm_add_pd(velecsum,velec);
 511             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 512             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 513             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 514
 515             fscal            = _mm_add_pd(felec,fvdw);
 516
 517             fscal            = _mm_and_pd(fscal,cutoff_mask);
 518
 519             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 520
 521             /* Update vectorial force */
 522             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 523             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 524             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 525
 526             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 527             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 528             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 529
 530             }
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             if (gmx_mm_any_lt(rsq10,rcutoff2))
 537             {
 538
 539             r10              = _mm_mul_pd(rsq10,rinv10);
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq10             = _mm_mul_pd(iq1,jq0);
 543
 544             /* EWALD ELECTROSTATICS */
 545
 546             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 547             ewrt             = _mm_mul_pd(r10,ewtabscale);
 548             ewitab           = _mm_cvttpd_epi32(ewrt);
 549 #ifdef __XOP__
 550             eweps            = _mm_frcz_pd(ewrt);
 551 #else
 552             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 553 #endif
 554             twoeweps         = _mm_add_pd(eweps,eweps);
 555             ewitab           = _mm_slli_epi32(ewitab,2);
 556             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 557             ewtabD           = _mm_setzero_pd();
 558             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 559             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 560             ewtabFn          = _mm_setzero_pd();
 561             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 562             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 563             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 564             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 565             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 566
 567             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             velec            = _mm_and_pd(velec,cutoff_mask);
 571             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 572             velecsum         = _mm_add_pd(velecsum,velec);
 573
 574             fscal            = felec;
 575
 576             fscal            = _mm_and_pd(fscal,cutoff_mask);
 577
 578             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 579
 580             /* Update vectorial force */
 581             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 582             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 583             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 584
 585             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 586             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 587             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 588
 589             }
 590
 591             /**************************
 592              * CALCULATE INTERACTIONS *
 593              **************************/
 594
 595             if (gmx_mm_any_lt(rsq20,rcutoff2))
 596             {
 597
 598             r20              = _mm_mul_pd(rsq20,rinv20);
 599
 600             /* Compute parameters for interactions between i and j atoms */
 601             qq20             = _mm_mul_pd(iq2,jq0);
 602
 603             /* EWALD ELECTROSTATICS */
 604
 605             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 606             ewrt             = _mm_mul_pd(r20,ewtabscale);
 607             ewitab           = _mm_cvttpd_epi32(ewrt);
 608 #ifdef __XOP__
 609             eweps            = _mm_frcz_pd(ewrt);
 610 #else
 611             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 612 #endif
 613             twoeweps         = _mm_add_pd(eweps,eweps);
 614             ewitab           = _mm_slli_epi32(ewitab,2);
 615             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 616             ewtabD           = _mm_setzero_pd();
 617             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 618             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 619             ewtabFn          = _mm_setzero_pd();
 620             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 621             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 622             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 623             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 624             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 625
 626             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 627
 628             /* Update potential sum for this i atom from the interaction with this j atom. */
 629             velec            = _mm_and_pd(velec,cutoff_mask);
 630             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 631             velecsum         = _mm_add_pd(velecsum,velec);
 632
 633             fscal            = felec;
 634
 635             fscal            = _mm_and_pd(fscal,cutoff_mask);
 636
 637             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 638
 639             /* Update vectorial force */
 640             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 641             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 642             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 643
 644             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 645             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 646             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 647
 648             }
 649
 650             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 651
 652             /* Inner loop uses 168 flops */
 653         }
 654
 655         /* End of innermost loop */
 656
 657         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 658                                               f+i_coord_offset,fshift+i_shift_offset);
 659
 660         ggid                        = gid[iidx];
 661         /* Update potential energies */
 662         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 663         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 664
 665         /* Increment number of inner iterations */
 666         inneriter                  += j_index_end - j_index_start;
 667
 668         /* Outer loop uses 20 flops */
 669     }
 670
 671     /* Increment number of outer iterations */
 672     outeriter        += nri;
 673
 674     /* Update outer/inner flops */
 675
 676     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*168);
 677 }
 678 /*
 679  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_double
 680  * Electrostatics interaction: Ewald
 681  * VdW interaction:            LennardJones
 682  * Geometry:                   Water3-Particle
 683  * Calculate force/pot:        Force
 684  */
 685 void
 686 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_double
 687                     (t_nblist                    * gmx_restrict       nlist,
 688                      rvec                        * gmx_restrict          xx,
 689                      rvec                        * gmx_restrict          ff,
 690                      t_forcerec                  * gmx_restrict          fr,
 691                      t_mdatoms                   * gmx_restrict     mdatoms,
 692                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 693                      t_nrnb                      * gmx_restrict        nrnb)
 694 {
 695     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 696      * just 0 for non-waters.
 697      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 698      * jnr indices corresponding to data put in the four positions in the SIMD register.
 699      */
 700     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 701     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 702     int              jnrA,jnrB;
 703     int              j_coord_offsetA,j_coord_offsetB;
 704     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 705     real             rcutoff_scalar;
 706     real             *shiftvec,*fshift,*x,*f;
 707     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 708     int              vdwioffset0;
 709     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 710     int              vdwioffset1;
 711     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 712     int              vdwioffset2;
 713     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 714     int              vdwjidx0A,vdwjidx0B;
 715     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 716     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 717     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 718     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 719     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 720     real             *charge;
 721     int              nvdwtype;
 722     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 723     int              *vdwtype;
 724     real             *vdwparam;
 725     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 726     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 727     __m128i          ewitab;
 728     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 729     real             *ewtab;
 730     __m128d          dummy_mask,cutoff_mask;
 731     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 732     __m128d          one     = _mm_set1_pd(1.0);
 733     __m128d          two     = _mm_set1_pd(2.0);
 734     x                = xx[0];
 735     f                = ff[0];
 736
 737     nri              = nlist->nri;
 738     iinr             = nlist->iinr;
 739     jindex           = nlist->jindex;
 740     jjnr             = nlist->jjnr;
 741     shiftidx         = nlist->shift;
 742     gid              = nlist->gid;
 743     shiftvec         = fr->shift_vec[0];
 744     fshift           = fr->fshift[0];
 745     facel            = _mm_set1_pd(fr->epsfac);
 746     charge           = mdatoms->chargeA;
 747     nvdwtype         = fr->ntype;
 748     vdwparam         = fr->nbfp;
 749     vdwtype          = mdatoms->typeA;
 750
 751     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 752     ewtab            = fr->ic->tabq_coul_F;
 753     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 754     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 755
 756     /* Setup water-specific parameters */
 757     inr              = nlist->iinr[0];
 758     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 759     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 760     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 761     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 762
 763     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 764     rcutoff_scalar   = fr->rcoulomb;
 765     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 766     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 767
 768     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 769     rvdw             = _mm_set1_pd(fr->rvdw);
 770
 771     /* Avoid stupid compiler warnings */
 772     jnrA = jnrB = 0;
 773     j_coord_offsetA = 0;
 774     j_coord_offsetB = 0;
 775
 776     outeriter        = 0;
 777     inneriter        = 0;
 778
 779     /* Start outer loop over neighborlists */
 780     for(iidx=0; iidx<nri; iidx++)
 781     {
 782         /* Load shift vector for this list */
 783         i_shift_offset   = DIM*shiftidx[iidx];
 784
 785         /* Load limits for loop over neighbors */
 786         j_index_start    = jindex[iidx];
 787         j_index_end      = jindex[iidx+1];
 788
 789         /* Get outer coordinate index */
 790         inr              = iinr[iidx];
 791         i_coord_offset   = DIM*inr;
 792
 793         /* Load i particle coords and add shift vector */
 794         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 795                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 796
 797         fix0             = _mm_setzero_pd();
 798         fiy0             = _mm_setzero_pd();
 799         fiz0             = _mm_setzero_pd();
 800         fix1             = _mm_setzero_pd();
 801         fiy1             = _mm_setzero_pd();
 802         fiz1             = _mm_setzero_pd();
 803         fix2             = _mm_setzero_pd();
 804         fiy2             = _mm_setzero_pd();
 805         fiz2             = _mm_setzero_pd();
 806
 807         /* Start inner kernel loop */
 808         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 809         {
 810
 811             /* Get j neighbor index, and coordinate index */
 812             jnrA             = jjnr[jidx];
 813             jnrB             = jjnr[jidx+1];
 814             j_coord_offsetA  = DIM*jnrA;
 815             j_coord_offsetB  = DIM*jnrB;
 816
 817             /* load j atom coordinates */
 818             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 819                                               &jx0,&jy0,&jz0);
 820
 821             /* Calculate displacement vector */
 822             dx00             = _mm_sub_pd(ix0,jx0);
 823             dy00             = _mm_sub_pd(iy0,jy0);
 824             dz00             = _mm_sub_pd(iz0,jz0);
 825             dx10             = _mm_sub_pd(ix1,jx0);
 826             dy10             = _mm_sub_pd(iy1,jy0);
 827             dz10             = _mm_sub_pd(iz1,jz0);
 828             dx20             = _mm_sub_pd(ix2,jx0);
 829             dy20             = _mm_sub_pd(iy2,jy0);
 830             dz20             = _mm_sub_pd(iz2,jz0);
 831
 832             /* Calculate squared distance and things based on it */
 833             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 834             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 835             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 836
 837             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 838             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 839             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 840
 841             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 842             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 843             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 844
 845             /* Load parameters for j particles */
 846             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 847             vdwjidx0A        = 2*vdwtype[jnrA+0];
 848             vdwjidx0B        = 2*vdwtype[jnrB+0];
 849
 850             fjx0             = _mm_setzero_pd();
 851             fjy0             = _mm_setzero_pd();
 852             fjz0             = _mm_setzero_pd();
 853
 854             /**************************
 855              * CALCULATE INTERACTIONS *
 856              **************************/
 857
 858             if (gmx_mm_any_lt(rsq00,rcutoff2))
 859             {
 860
 861             r00              = _mm_mul_pd(rsq00,rinv00);
 862
 863             /* Compute parameters for interactions between i and j atoms */
 864             qq00             = _mm_mul_pd(iq0,jq0);
 865             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 866                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 867
 868             /* EWALD ELECTROSTATICS */
 869
 870             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 871             ewrt             = _mm_mul_pd(r00,ewtabscale);
 872             ewitab           = _mm_cvttpd_epi32(ewrt);
 873 #ifdef __XOP__
 874             eweps            = _mm_frcz_pd(ewrt);
 875 #else
 876             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 877 #endif
 878             twoeweps         = _mm_add_pd(eweps,eweps);
 879             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 880                                          &ewtabF,&ewtabFn);
 881             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 882             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 883
 884             /* LENNARD-JONES DISPERSION/REPULSION */
 885
 886             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 887             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 888
 889             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 890
 891             fscal            = _mm_add_pd(felec,fvdw);
 892
 893             fscal            = _mm_and_pd(fscal,cutoff_mask);
 894
 895             /* Update vectorial force */
 896             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 897             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 898             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 899
 900             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 901             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 902             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 903
 904             }
 905
 906             /**************************
 907              * CALCULATE INTERACTIONS *
 908              **************************/
 909
 910             if (gmx_mm_any_lt(rsq10,rcutoff2))
 911             {
 912
 913             r10              = _mm_mul_pd(rsq10,rinv10);
 914
 915             /* Compute parameters for interactions between i and j atoms */
 916             qq10             = _mm_mul_pd(iq1,jq0);
 917
 918             /* EWALD ELECTROSTATICS */
 919
 920             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 921             ewrt             = _mm_mul_pd(r10,ewtabscale);
 922             ewitab           = _mm_cvttpd_epi32(ewrt);
 923 #ifdef __XOP__
 924             eweps            = _mm_frcz_pd(ewrt);
 925 #else
 926             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 927 #endif
 928             twoeweps         = _mm_add_pd(eweps,eweps);
 929             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 930                                          &ewtabF,&ewtabFn);
 931             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 932             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 933
 934             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 935
 936             fscal            = felec;
 937
 938             fscal            = _mm_and_pd(fscal,cutoff_mask);
 939
 940             /* Update vectorial force */
 941             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 942             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 943             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 944
 945             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 946             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 947             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 948
 949             }
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             if (gmx_mm_any_lt(rsq20,rcutoff2))
 956             {
 957
 958             r20              = _mm_mul_pd(rsq20,rinv20);
 959
 960             /* Compute parameters for interactions between i and j atoms */
 961             qq20             = _mm_mul_pd(iq2,jq0);
 962
 963             /* EWALD ELECTROSTATICS */
 964
 965             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 966             ewrt             = _mm_mul_pd(r20,ewtabscale);
 967             ewitab           = _mm_cvttpd_epi32(ewrt);
 968 #ifdef __XOP__
 969             eweps            = _mm_frcz_pd(ewrt);
 970 #else
 971             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 972 #endif
 973             twoeweps         = _mm_add_pd(eweps,eweps);
 974             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 975                                          &ewtabF,&ewtabFn);
 976             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 977             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 978
 979             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 980
 981             fscal            = felec;
 982
 983             fscal            = _mm_and_pd(fscal,cutoff_mask);
 984
 985             /* Update vectorial force */
 986             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 987             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 988             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 989
 990             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 991             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 992             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 993
 994             }
 995
 996             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 997
 998             /* Inner loop uses 136 flops */
 999         }
1000
1001         if(jidx<j_index_end)
1002         {
1003
1004             jnrA             = jjnr[jidx];
1005             j_coord_offsetA  = DIM*jnrA;
1006
1007             /* load j atom coordinates */
1008             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1009                                               &jx0,&jy0,&jz0);
1010
1011             /* Calculate displacement vector */
1012             dx00             = _mm_sub_pd(ix0,jx0);
1013             dy00             = _mm_sub_pd(iy0,jy0);
1014             dz00             = _mm_sub_pd(iz0,jz0);
1015             dx10             = _mm_sub_pd(ix1,jx0);
1016             dy10             = _mm_sub_pd(iy1,jy0);
1017             dz10             = _mm_sub_pd(iz1,jz0);
1018             dx20             = _mm_sub_pd(ix2,jx0);
1019             dy20             = _mm_sub_pd(iy2,jy0);
1020             dz20             = _mm_sub_pd(iz2,jz0);
1021
1022             /* Calculate squared distance and things based on it */
1023             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1024             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1025             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1026
1027             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1028             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1029             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1030
1031             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1032             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1033             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1034
1035             /* Load parameters for j particles */
1036             jq0              = _mm_load_sd(charge+jnrA+0);
1037             vdwjidx0A        = 2*vdwtype[jnrA+0];
1038
1039             fjx0             = _mm_setzero_pd();
1040             fjy0             = _mm_setzero_pd();
1041             fjz0             = _mm_setzero_pd();
1042
1043             /**************************
1044              * CALCULATE INTERACTIONS *
1045              **************************/
1046
1047             if (gmx_mm_any_lt(rsq00,rcutoff2))
1048             {
1049
1050             r00              = _mm_mul_pd(rsq00,rinv00);
1051
1052             /* Compute parameters for interactions between i and j atoms */
1053             qq00             = _mm_mul_pd(iq0,jq0);
1054             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1055
1056             /* EWALD ELECTROSTATICS */
1057
1058             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1059             ewrt             = _mm_mul_pd(r00,ewtabscale);
1060             ewitab           = _mm_cvttpd_epi32(ewrt);
1061 #ifdef __XOP__
1062             eweps            = _mm_frcz_pd(ewrt);
1063 #else
1064             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1065 #endif
1066             twoeweps         = _mm_add_pd(eweps,eweps);
1067             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1068             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1069             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1070
1071             /* LENNARD-JONES DISPERSION/REPULSION */
1072
1073             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1074             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1075
1076             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1077
1078             fscal            = _mm_add_pd(felec,fvdw);
1079
1080             fscal            = _mm_and_pd(fscal,cutoff_mask);
1081
1082             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1083
1084             /* Update vectorial force */
1085             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1086             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1087             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1088
1089             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1090             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1091             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1092
1093             }
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             if (gmx_mm_any_lt(rsq10,rcutoff2))
1100             {
1101
1102             r10              = _mm_mul_pd(rsq10,rinv10);
1103
1104             /* Compute parameters for interactions between i and j atoms */
1105             qq10             = _mm_mul_pd(iq1,jq0);
1106
1107             /* EWALD ELECTROSTATICS */
1108
1109             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1110             ewrt             = _mm_mul_pd(r10,ewtabscale);
1111             ewitab           = _mm_cvttpd_epi32(ewrt);
1112 #ifdef __XOP__
1113             eweps            = _mm_frcz_pd(ewrt);
1114 #else
1115             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1116 #endif
1117             twoeweps         = _mm_add_pd(eweps,eweps);
1118             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1119             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1120             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1121
1122             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1123
1124             fscal            = felec;
1125
1126             fscal            = _mm_and_pd(fscal,cutoff_mask);
1127
1128             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1129
1130             /* Update vectorial force */
1131             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1132             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1133             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1134
1135             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1136             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1137             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1138
1139             }
1140
1141             /**************************
1142              * CALCULATE INTERACTIONS *
1143              **************************/
1144
1145             if (gmx_mm_any_lt(rsq20,rcutoff2))
1146             {
1147
1148             r20              = _mm_mul_pd(rsq20,rinv20);
1149
1150             /* Compute parameters for interactions between i and j atoms */
1151             qq20             = _mm_mul_pd(iq2,jq0);
1152
1153             /* EWALD ELECTROSTATICS */
1154
1155             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1156             ewrt             = _mm_mul_pd(r20,ewtabscale);
1157             ewitab           = _mm_cvttpd_epi32(ewrt);
1158 #ifdef __XOP__
1159             eweps            = _mm_frcz_pd(ewrt);
1160 #else
1161             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1162 #endif
1163             twoeweps         = _mm_add_pd(eweps,eweps);
1164             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1165             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1166             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1167
1168             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1169
1170             fscal            = felec;
1171
1172             fscal            = _mm_and_pd(fscal,cutoff_mask);
1173
1174             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1175
1176             /* Update vectorial force */
1177             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1178             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1179             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1180
1181             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1182             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1183             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1184
1185             }
1186
1187             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1188
1189             /* Inner loop uses 136 flops */
1190         }
1191
1192         /* End of innermost loop */
1193
1194         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1195                                               f+i_coord_offset,fshift+i_shift_offset);
1196
1197         /* Increment number of inner iterations */
1198         inneriter                  += j_index_end - j_index_start;
1199
1200         /* Outer loop uses 18 flops */
1201     }
1202
1203     /* Increment number of outer iterations */
1204     outeriter        += nri;
1205
1206     /* Update outer/inner flops */
1207
1208     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*136);
1209 }