src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_ElecRF_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_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              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 107     real             *vftab;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     krf              = _mm_set1_pd(fr->ic->k_rf);
 126     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 127     crf              = _mm_set1_pd(fr->ic->c_rf);
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 138     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 139     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 167
 168         fix0             = _mm_setzero_pd();
 169         fiy0             = _mm_setzero_pd();
 170         fiz0             = _mm_setzero_pd();
 171         fix1             = _mm_setzero_pd();
 172         fiy1             = _mm_setzero_pd();
 173         fiz1             = _mm_setzero_pd();
 174         fix2             = _mm_setzero_pd();
 175         fiy2             = _mm_setzero_pd();
 176         fiz2             = _mm_setzero_pd();
 177         fix3             = _mm_setzero_pd();
 178         fiy3             = _mm_setzero_pd();
 179         fiz3             = _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             dx30             = _mm_sub_pd(ix3,jx0);
 210             dy30             = _mm_sub_pd(iy3,jy0);
 211             dz30             = _mm_sub_pd(iz3,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 217             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 218
 219             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 220             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 221             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 222             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 223
 224             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 225             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 226             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232
 233             fjx0             = _mm_setzero_pd();
 234             fjy0             = _mm_setzero_pd();
 235             fjz0             = _mm_setzero_pd();
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             r00              = _mm_mul_pd(rsq00,rinv00);
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 245                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 246
 247             /* Calculate table index by multiplying r with table scale and truncate to integer */
 248             rt               = _mm_mul_pd(r00,vftabscale);
 249             vfitab           = _mm_cvttpd_epi32(rt);
 250 #ifdef __XOP__
 251             vfeps            = _mm_frcz_pd(rt);
 252 #else
 253             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 254 #endif
 255             twovfeps         = _mm_add_pd(vfeps,vfeps);
 256             vfitab           = _mm_slli_epi32(vfitab,3);
 257
 258             /* CUBIC SPLINE TABLE DISPERSION */
 259             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 260             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 261             GMX_MM_TRANSPOSE2_PD(Y,F);
 262             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 263             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 264             GMX_MM_TRANSPOSE2_PD(G,H);
 265             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 266             VV               = _mm_macc_pd(vfeps,Fp,Y);
 267             vvdw6            = _mm_mul_pd(c6_00,VV);
 268             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 269             fvdw6            = _mm_mul_pd(c6_00,FF);
 270
 271             /* CUBIC SPLINE TABLE REPULSION */
 272             vfitab           = _mm_add_epi32(vfitab,ifour);
 273             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 274             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 275             GMX_MM_TRANSPOSE2_PD(Y,F);
 276             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 277             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 278             GMX_MM_TRANSPOSE2_PD(G,H);
 279             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 280             VV               = _mm_macc_pd(vfeps,Fp,Y);
 281             vvdw12           = _mm_mul_pd(c12_00,VV);
 282             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 283             fvdw12           = _mm_mul_pd(c12_00,FF);
 284             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 285             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 289
 290             fscal            = fvdw;
 291
 292             /* Update vectorial force */
 293             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 294             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 295             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 296
 297             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 298             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 299             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_pd(iq1,jq0);
 307
 308             /* REACTION-FIELD ELECTROSTATICS */
 309             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 310             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm_add_pd(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317             /* Update vectorial force */
 318             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 319             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 320             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 321
 322             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 323             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 324             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             qq20             = _mm_mul_pd(iq2,jq0);
 332
 333             /* REACTION-FIELD ELECTROSTATICS */
 334             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 335             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velecsum         = _mm_add_pd(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342             /* Update vectorial force */
 343             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 344             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 345             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 346
 347             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 348             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 349             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             /* Compute parameters for interactions between i and j atoms */
 356             qq30             = _mm_mul_pd(iq3,jq0);
 357
 358             /* REACTION-FIELD ELECTROSTATICS */
 359             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 360             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velecsum         = _mm_add_pd(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             /* Update vectorial force */
 368             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 369             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 370             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 371
 372             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 373             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 374             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 375
 376             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 377
 378             /* Inner loop uses 167 flops */
 379         }
 380
 381         if(jidx<j_index_end)
 382         {
 383
 384             jnrA             = jjnr[jidx];
 385             j_coord_offsetA  = DIM*jnrA;
 386
 387             /* load j atom coordinates */
 388             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 389                                               &jx0,&jy0,&jz0);
 390
 391             /* Calculate displacement vector */
 392             dx00             = _mm_sub_pd(ix0,jx0);
 393             dy00             = _mm_sub_pd(iy0,jy0);
 394             dz00             = _mm_sub_pd(iz0,jz0);
 395             dx10             = _mm_sub_pd(ix1,jx0);
 396             dy10             = _mm_sub_pd(iy1,jy0);
 397             dz10             = _mm_sub_pd(iz1,jz0);
 398             dx20             = _mm_sub_pd(ix2,jx0);
 399             dy20             = _mm_sub_pd(iy2,jy0);
 400             dz20             = _mm_sub_pd(iz2,jz0);
 401             dx30             = _mm_sub_pd(ix3,jx0);
 402             dy30             = _mm_sub_pd(iy3,jy0);
 403             dz30             = _mm_sub_pd(iz3,jz0);
 404
 405             /* Calculate squared distance and things based on it */
 406             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 407             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 408             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 409             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 410
 411             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 412             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 413             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 414             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 415
 416             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 417             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 418             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 419
 420             /* Load parameters for j particles */
 421             jq0              = _mm_load_sd(charge+jnrA+0);
 422             vdwjidx0A        = 2*vdwtype[jnrA+0];
 423
 424             fjx0             = _mm_setzero_pd();
 425             fjy0             = _mm_setzero_pd();
 426             fjz0             = _mm_setzero_pd();
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             r00              = _mm_mul_pd(rsq00,rinv00);
 433
 434             /* Compute parameters for interactions between i and j atoms */
 435             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 436
 437             /* Calculate table index by multiplying r with table scale and truncate to integer */
 438             rt               = _mm_mul_pd(r00,vftabscale);
 439             vfitab           = _mm_cvttpd_epi32(rt);
 440 #ifdef __XOP__
 441             vfeps            = _mm_frcz_pd(rt);
 442 #else
 443             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 444 #endif
 445             twovfeps         = _mm_add_pd(vfeps,vfeps);
 446             vfitab           = _mm_slli_epi32(vfitab,3);
 447
 448             /* CUBIC SPLINE TABLE DISPERSION */
 449             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 450             F                = _mm_setzero_pd();
 451             GMX_MM_TRANSPOSE2_PD(Y,F);
 452             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 453             H                = _mm_setzero_pd();
 454             GMX_MM_TRANSPOSE2_PD(G,H);
 455             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 456             VV               = _mm_macc_pd(vfeps,Fp,Y);
 457             vvdw6            = _mm_mul_pd(c6_00,VV);
 458             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 459             fvdw6            = _mm_mul_pd(c6_00,FF);
 460
 461             /* CUBIC SPLINE TABLE REPULSION */
 462             vfitab           = _mm_add_epi32(vfitab,ifour);
 463             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 464             F                = _mm_setzero_pd();
 465             GMX_MM_TRANSPOSE2_PD(Y,F);
 466             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 467             H                = _mm_setzero_pd();
 468             GMX_MM_TRANSPOSE2_PD(G,H);
 469             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 470             VV               = _mm_macc_pd(vfeps,Fp,Y);
 471             vvdw12           = _mm_mul_pd(c12_00,VV);
 472             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 473             fvdw12           = _mm_mul_pd(c12_00,FF);
 474             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 475             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 476
 477             /* Update potential sum for this i atom from the interaction with this j atom. */
 478             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 479             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 480
 481             fscal            = fvdw;
 482
 483             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 484
 485             /* Update vectorial force */
 486             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 487             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 488             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 489
 490             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 491             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 492             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             qq10             = _mm_mul_pd(iq1,jq0);
 500
 501             /* REACTION-FIELD ELECTROSTATICS */
 502             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 503             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 507             velecsum         = _mm_add_pd(velecsum,velec);
 508
 509             fscal            = felec;
 510
 511             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 512
 513             /* Update vectorial force */
 514             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 515             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 516             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 517
 518             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 519             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 520             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             /* Compute parameters for interactions between i and j atoms */
 527             qq20             = _mm_mul_pd(iq2,jq0);
 528
 529             /* REACTION-FIELD ELECTROSTATICS */
 530             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 531             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 532
 533             /* Update potential sum for this i atom from the interaction with this j atom. */
 534             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 535             velecsum         = _mm_add_pd(velecsum,velec);
 536
 537             fscal            = felec;
 538
 539             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 540
 541             /* Update vectorial force */
 542             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 543             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 544             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 545
 546             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 547             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 548             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             /* Compute parameters for interactions between i and j atoms */
 555             qq30             = _mm_mul_pd(iq3,jq0);
 556
 557             /* REACTION-FIELD ELECTROSTATICS */
 558             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 559             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 560
 561             /* Update potential sum for this i atom from the interaction with this j atom. */
 562             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 563             velecsum         = _mm_add_pd(velecsum,velec);
 564
 565             fscal            = felec;
 566
 567             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 568
 569             /* Update vectorial force */
 570             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 571             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 572             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 573
 574             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 575             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 576             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 577
 578             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 579
 580             /* Inner loop uses 167 flops */
 581         }
 582
 583         /* End of innermost loop */
 584
 585         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 586                                               f+i_coord_offset,fshift+i_shift_offset);
 587
 588         ggid                        = gid[iidx];
 589         /* Update potential energies */
 590         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 591         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 592
 593         /* Increment number of inner iterations */
 594         inneriter                  += j_index_end - j_index_start;
 595
 596         /* Outer loop uses 26 flops */
 597     }
 598
 599     /* Increment number of outer iterations */
 600     outeriter        += nri;
 601
 602     /* Update outer/inner flops */
 603
 604     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*167);
 605 }
 606 /*
 607  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 608  * Electrostatics interaction: ReactionField
 609  * VdW interaction:            CubicSplineTable
 610  * Geometry:                   Water4-Particle
 611  * Calculate force/pot:        Force
 612  */
 613 void
 614 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 615                     (t_nblist                    * gmx_restrict       nlist,
 616                      rvec                        * gmx_restrict          xx,
 617                      rvec                        * gmx_restrict          ff,
 618                      t_forcerec                  * gmx_restrict          fr,
 619                      t_mdatoms                   * gmx_restrict     mdatoms,
 620                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 621                      t_nrnb                      * gmx_restrict        nrnb)
 622 {
 623     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 624      * just 0 for non-waters.
 625      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 626      * jnr indices corresponding to data put in the four positions in the SIMD register.
 627      */
 628     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 629     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 630     int              jnrA,jnrB;
 631     int              j_coord_offsetA,j_coord_offsetB;
 632     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 633     real             rcutoff_scalar;
 634     real             *shiftvec,*fshift,*x,*f;
 635     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 636     int              vdwioffset0;
 637     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 638     int              vdwioffset1;
 639     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 640     int              vdwioffset2;
 641     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 642     int              vdwioffset3;
 643     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 644     int              vdwjidx0A,vdwjidx0B;
 645     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 646     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 647     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 648     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 649     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 650     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 651     real             *charge;
 652     int              nvdwtype;
 653     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 654     int              *vdwtype;
 655     real             *vdwparam;
 656     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 657     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 658     __m128i          vfitab;
 659     __m128i          ifour       = _mm_set1_epi32(4);
 660     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 661     real             *vftab;
 662     __m128d          dummy_mask,cutoff_mask;
 663     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 664     __m128d          one     = _mm_set1_pd(1.0);
 665     __m128d          two     = _mm_set1_pd(2.0);
 666     x                = xx[0];
 667     f                = ff[0];
 668
 669     nri              = nlist->nri;
 670     iinr             = nlist->iinr;
 671     jindex           = nlist->jindex;
 672     jjnr             = nlist->jjnr;
 673     shiftidx         = nlist->shift;
 674     gid              = nlist->gid;
 675     shiftvec         = fr->shift_vec[0];
 676     fshift           = fr->fshift[0];
 677     facel            = _mm_set1_pd(fr->epsfac);
 678     charge           = mdatoms->chargeA;
 679     krf              = _mm_set1_pd(fr->ic->k_rf);
 680     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 681     crf              = _mm_set1_pd(fr->ic->c_rf);
 682     nvdwtype         = fr->ntype;
 683     vdwparam         = fr->nbfp;
 684     vdwtype          = mdatoms->typeA;
 685
 686     vftab            = kernel_data->table_vdw->data;
 687     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 688
 689     /* Setup water-specific parameters */
 690     inr              = nlist->iinr[0];
 691     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 692     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 693     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 694     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 695
 696     /* Avoid stupid compiler warnings */
 697     jnrA = jnrB = 0;
 698     j_coord_offsetA = 0;
 699     j_coord_offsetB = 0;
 700
 701     outeriter        = 0;
 702     inneriter        = 0;
 703
 704     /* Start outer loop over neighborlists */
 705     for(iidx=0; iidx<nri; iidx++)
 706     {
 707         /* Load shift vector for this list */
 708         i_shift_offset   = DIM*shiftidx[iidx];
 709
 710         /* Load limits for loop over neighbors */
 711         j_index_start    = jindex[iidx];
 712         j_index_end      = jindex[iidx+1];
 713
 714         /* Get outer coordinate index */
 715         inr              = iinr[iidx];
 716         i_coord_offset   = DIM*inr;
 717
 718         /* Load i particle coords and add shift vector */
 719         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 720                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 721
 722         fix0             = _mm_setzero_pd();
 723         fiy0             = _mm_setzero_pd();
 724         fiz0             = _mm_setzero_pd();
 725         fix1             = _mm_setzero_pd();
 726         fiy1             = _mm_setzero_pd();
 727         fiz1             = _mm_setzero_pd();
 728         fix2             = _mm_setzero_pd();
 729         fiy2             = _mm_setzero_pd();
 730         fiz2             = _mm_setzero_pd();
 731         fix3             = _mm_setzero_pd();
 732         fiy3             = _mm_setzero_pd();
 733         fiz3             = _mm_setzero_pd();
 734
 735         /* Start inner kernel loop */
 736         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 737         {
 738
 739             /* Get j neighbor index, and coordinate index */
 740             jnrA             = jjnr[jidx];
 741             jnrB             = jjnr[jidx+1];
 742             j_coord_offsetA  = DIM*jnrA;
 743             j_coord_offsetB  = DIM*jnrB;
 744
 745             /* load j atom coordinates */
 746             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 747                                               &jx0,&jy0,&jz0);
 748
 749             /* Calculate displacement vector */
 750             dx00             = _mm_sub_pd(ix0,jx0);
 751             dy00             = _mm_sub_pd(iy0,jy0);
 752             dz00             = _mm_sub_pd(iz0,jz0);
 753             dx10             = _mm_sub_pd(ix1,jx0);
 754             dy10             = _mm_sub_pd(iy1,jy0);
 755             dz10             = _mm_sub_pd(iz1,jz0);
 756             dx20             = _mm_sub_pd(ix2,jx0);
 757             dy20             = _mm_sub_pd(iy2,jy0);
 758             dz20             = _mm_sub_pd(iz2,jz0);
 759             dx30             = _mm_sub_pd(ix3,jx0);
 760             dy30             = _mm_sub_pd(iy3,jy0);
 761             dz30             = _mm_sub_pd(iz3,jz0);
 762
 763             /* Calculate squared distance and things based on it */
 764             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 765             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 766             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 767             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 768
 769             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 770             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 771             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 772             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 773
 774             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 775             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 776             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 777
 778             /* Load parameters for j particles */
 779             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 780             vdwjidx0A        = 2*vdwtype[jnrA+0];
 781             vdwjidx0B        = 2*vdwtype[jnrB+0];
 782
 783             fjx0             = _mm_setzero_pd();
 784             fjy0             = _mm_setzero_pd();
 785             fjz0             = _mm_setzero_pd();
 786
 787             /**************************
 788              * CALCULATE INTERACTIONS *
 789              **************************/
 790
 791             r00              = _mm_mul_pd(rsq00,rinv00);
 792
 793             /* Compute parameters for interactions between i and j atoms */
 794             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 795                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 796
 797             /* Calculate table index by multiplying r with table scale and truncate to integer */
 798             rt               = _mm_mul_pd(r00,vftabscale);
 799             vfitab           = _mm_cvttpd_epi32(rt);
 800 #ifdef __XOP__
 801             vfeps            = _mm_frcz_pd(rt);
 802 #else
 803             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 804 #endif
 805             twovfeps         = _mm_add_pd(vfeps,vfeps);
 806             vfitab           = _mm_slli_epi32(vfitab,3);
 807
 808             /* CUBIC SPLINE TABLE DISPERSION */
 809             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 810             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 811             GMX_MM_TRANSPOSE2_PD(Y,F);
 812             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 813             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 814             GMX_MM_TRANSPOSE2_PD(G,H);
 815             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 816             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 817             fvdw6            = _mm_mul_pd(c6_00,FF);
 818
 819             /* CUBIC SPLINE TABLE REPULSION */
 820             vfitab           = _mm_add_epi32(vfitab,ifour);
 821             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 822             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 823             GMX_MM_TRANSPOSE2_PD(Y,F);
 824             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 825             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 826             GMX_MM_TRANSPOSE2_PD(G,H);
 827             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 828             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 829             fvdw12           = _mm_mul_pd(c12_00,FF);
 830             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 831
 832             fscal            = fvdw;
 833
 834             /* Update vectorial force */
 835             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 836             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 837             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 838
 839             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 840             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 841             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 842
 843             /**************************
 844              * CALCULATE INTERACTIONS *
 845              **************************/
 846
 847             /* Compute parameters for interactions between i and j atoms */
 848             qq10             = _mm_mul_pd(iq1,jq0);
 849
 850             /* REACTION-FIELD ELECTROSTATICS */
 851             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 852
 853             fscal            = felec;
 854
 855             /* Update vectorial force */
 856             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 857             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 858             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 859
 860             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 861             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 862             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 863
 864             /**************************
 865              * CALCULATE INTERACTIONS *
 866              **************************/
 867
 868             /* Compute parameters for interactions between i and j atoms */
 869             qq20             = _mm_mul_pd(iq2,jq0);
 870
 871             /* REACTION-FIELD ELECTROSTATICS */
 872             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 873
 874             fscal            = felec;
 875
 876             /* Update vectorial force */
 877             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 878             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 879             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 880
 881             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 882             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 883             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             /* Compute parameters for interactions between i and j atoms */
 890             qq30             = _mm_mul_pd(iq3,jq0);
 891
 892             /* REACTION-FIELD ELECTROSTATICS */
 893             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 894
 895             fscal            = felec;
 896
 897             /* Update vectorial force */
 898             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 899             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 900             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 901
 902             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 903             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 904             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 905
 906             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 907
 908             /* Inner loop uses 144 flops */
 909         }
 910
 911         if(jidx<j_index_end)
 912         {
 913
 914             jnrA             = jjnr[jidx];
 915             j_coord_offsetA  = DIM*jnrA;
 916
 917             /* load j atom coordinates */
 918             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 919                                               &jx0,&jy0,&jz0);
 920
 921             /* Calculate displacement vector */
 922             dx00             = _mm_sub_pd(ix0,jx0);
 923             dy00             = _mm_sub_pd(iy0,jy0);
 924             dz00             = _mm_sub_pd(iz0,jz0);
 925             dx10             = _mm_sub_pd(ix1,jx0);
 926             dy10             = _mm_sub_pd(iy1,jy0);
 927             dz10             = _mm_sub_pd(iz1,jz0);
 928             dx20             = _mm_sub_pd(ix2,jx0);
 929             dy20             = _mm_sub_pd(iy2,jy0);
 930             dz20             = _mm_sub_pd(iz2,jz0);
 931             dx30             = _mm_sub_pd(ix3,jx0);
 932             dy30             = _mm_sub_pd(iy3,jy0);
 933             dz30             = _mm_sub_pd(iz3,jz0);
 934
 935             /* Calculate squared distance and things based on it */
 936             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 937             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 938             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 939             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 940
 941             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 942             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 943             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 944             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 945
 946             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 947             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 948             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 949
 950             /* Load parameters for j particles */
 951             jq0              = _mm_load_sd(charge+jnrA+0);
 952             vdwjidx0A        = 2*vdwtype[jnrA+0];
 953
 954             fjx0             = _mm_setzero_pd();
 955             fjy0             = _mm_setzero_pd();
 956             fjz0             = _mm_setzero_pd();
 957
 958             /**************************
 959              * CALCULATE INTERACTIONS *
 960              **************************/
 961
 962             r00              = _mm_mul_pd(rsq00,rinv00);
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 966
 967             /* Calculate table index by multiplying r with table scale and truncate to integer */
 968             rt               = _mm_mul_pd(r00,vftabscale);
 969             vfitab           = _mm_cvttpd_epi32(rt);
 970 #ifdef __XOP__
 971             vfeps            = _mm_frcz_pd(rt);
 972 #else
 973             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 974 #endif
 975             twovfeps         = _mm_add_pd(vfeps,vfeps);
 976             vfitab           = _mm_slli_epi32(vfitab,3);
 977
 978             /* CUBIC SPLINE TABLE DISPERSION */
 979             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 980             F                = _mm_setzero_pd();
 981             GMX_MM_TRANSPOSE2_PD(Y,F);
 982             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 983             H                = _mm_setzero_pd();
 984             GMX_MM_TRANSPOSE2_PD(G,H);
 985             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 986             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 987             fvdw6            = _mm_mul_pd(c6_00,FF);
 988
 989             /* CUBIC SPLINE TABLE REPULSION */
 990             vfitab           = _mm_add_epi32(vfitab,ifour);
 991             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 992             F                = _mm_setzero_pd();
 993             GMX_MM_TRANSPOSE2_PD(Y,F);
 994             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 995             H                = _mm_setzero_pd();
 996             GMX_MM_TRANSPOSE2_PD(G,H);
 997             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 998             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 999             fvdw12           = _mm_mul_pd(c12_00,FF);
1000             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1001
1002             fscal            = fvdw;
1003
1004             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1005
1006             /* Update vectorial force */
1007             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1008             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1009             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1010
1011             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1012             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1013             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1014
1015             /**************************
1016              * CALCULATE INTERACTIONS *
1017              **************************/
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             qq10             = _mm_mul_pd(iq1,jq0);
1021
1022             /* REACTION-FIELD ELECTROSTATICS */
1023             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
1024
1025             fscal            = felec;
1026
1027             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1028
1029             /* Update vectorial force */
1030             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1031             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1032             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1033
1034             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1035             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1036             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             /* Compute parameters for interactions between i and j atoms */
1043             qq20             = _mm_mul_pd(iq2,jq0);
1044
1045             /* REACTION-FIELD ELECTROSTATICS */
1046             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1047
1048             fscal            = felec;
1049
1050             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1051
1052             /* Update vectorial force */
1053             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1054             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1055             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1056
1057             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1058             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1059             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1060
1061             /**************************
1062              * CALCULATE INTERACTIONS *
1063              **************************/
1064
1065             /* Compute parameters for interactions between i and j atoms */
1066             qq30             = _mm_mul_pd(iq3,jq0);
1067
1068             /* REACTION-FIELD ELECTROSTATICS */
1069             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
1070
1071             fscal            = felec;
1072
1073             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1074
1075             /* Update vectorial force */
1076             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1077             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1078             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1079
1080             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1081             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1082             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1083
1084             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1085
1086             /* Inner loop uses 144 flops */
1087         }
1088
1089         /* End of innermost loop */
1090
1091         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1092                                               f+i_coord_offset,fshift+i_shift_offset);
1093
1094         /* Increment number of inner iterations */
1095         inneriter                  += j_index_end - j_index_start;
1096
1097         /* Outer loop uses 24 flops */
1098     }
1099
1100     /* Increment number of outer iterations */
1101     outeriter        += nri;
1102
1103     /* Update outer/inner flops */
1104
1105     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*144);
1106 }