src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_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, 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_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_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     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 135     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 136     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 164
 165         fix0             = _mm_setzero_pd();
 166         fiy0             = _mm_setzero_pd();
 167         fiz0             = _mm_setzero_pd();
 168         fix1             = _mm_setzero_pd();
 169         fiy1             = _mm_setzero_pd();
 170         fiz1             = _mm_setzero_pd();
 171         fix2             = _mm_setzero_pd();
 172         fiy2             = _mm_setzero_pd();
 173         fiz2             = _mm_setzero_pd();
 174         fix3             = _mm_setzero_pd();
 175         fiy3             = _mm_setzero_pd();
 176         fiz3             = _mm_setzero_pd();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_pd();
 180         vvdwsum          = _mm_setzero_pd();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_pd(ix0,jx0);
 198             dy00             = _mm_sub_pd(iy0,jy0);
 199             dz00             = _mm_sub_pd(iz0,jz0);
 200             dx10             = _mm_sub_pd(ix1,jx0);
 201             dy10             = _mm_sub_pd(iy1,jy0);
 202             dz10             = _mm_sub_pd(iz1,jz0);
 203             dx20             = _mm_sub_pd(ix2,jx0);
 204             dy20             = _mm_sub_pd(iy2,jy0);
 205             dz20             = _mm_sub_pd(iz2,jz0);
 206             dx30             = _mm_sub_pd(ix3,jx0);
 207             dy30             = _mm_sub_pd(iy3,jy0);
 208             dz30             = _mm_sub_pd(iz3,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             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 215
 216             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 217             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 218             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 219             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 220
 221             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 223             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 227             vdwjidx0A        = 2*vdwtype[jnrA+0];
 228             vdwjidx0B        = 2*vdwtype[jnrB+0];
 229
 230             fjx0             = _mm_setzero_pd();
 231             fjy0             = _mm_setzero_pd();
 232             fjz0             = _mm_setzero_pd();
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             r00              = _mm_mul_pd(rsq00,rinv00);
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 242                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 243
 244             /* Calculate table index by multiplying r with table scale and truncate to integer */
 245             rt               = _mm_mul_pd(r00,vftabscale);
 246             vfitab           = _mm_cvttpd_epi32(rt);
 247 #ifdef __XOP__
 248             vfeps            = _mm_frcz_pd(rt);
 249 #else
 250             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 251 #endif
 252             twovfeps         = _mm_add_pd(vfeps,vfeps);
 253             vfitab           = _mm_slli_epi32(vfitab,3);
 254
 255             /* CUBIC SPLINE TABLE DISPERSION */
 256             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 257             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 258             GMX_MM_TRANSPOSE2_PD(Y,F);
 259             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 260             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 261             GMX_MM_TRANSPOSE2_PD(G,H);
 262             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 263             VV               = _mm_macc_pd(vfeps,Fp,Y);
 264             vvdw6            = _mm_mul_pd(c6_00,VV);
 265             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 266             fvdw6            = _mm_mul_pd(c6_00,FF);
 267
 268             /* CUBIC SPLINE TABLE REPULSION */
 269             vfitab           = _mm_add_epi32(vfitab,ifour);
 270             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 271             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 272             GMX_MM_TRANSPOSE2_PD(Y,F);
 273             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 274             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 275             GMX_MM_TRANSPOSE2_PD(G,H);
 276             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 277             VV               = _mm_macc_pd(vfeps,Fp,Y);
 278             vvdw12           = _mm_mul_pd(c12_00,VV);
 279             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 280             fvdw12           = _mm_mul_pd(c12_00,FF);
 281             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 282             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 286
 287             fscal            = fvdw;
 288
 289             /* Update vectorial force */
 290             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 291             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 292             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 293
 294             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 295             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 296             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq10             = _mm_mul_pd(iq1,jq0);
 304
 305             /* COULOMB ELECTROSTATICS */
 306             velec            = _mm_mul_pd(qq10,rinv10);
 307             felec            = _mm_mul_pd(velec,rinvsq10);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velecsum         = _mm_add_pd(velecsum,velec);
 311
 312             fscal            = felec;
 313
 314             /* Update vectorial force */
 315             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 316             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 317             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 318
 319             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 320             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 321             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 322
 323             /**************************
 324              * CALCULATE INTERACTIONS *
 325              **************************/
 326
 327             /* Compute parameters for interactions between i and j atoms */
 328             qq20             = _mm_mul_pd(iq2,jq0);
 329
 330             /* COULOMB ELECTROSTATICS */
 331             velec            = _mm_mul_pd(qq20,rinv20);
 332             felec            = _mm_mul_pd(velec,rinvsq20);
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velecsum         = _mm_add_pd(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             /* Update vectorial force */
 340             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 341             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 342             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 343
 344             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 345             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 346             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq30             = _mm_mul_pd(iq3,jq0);
 354
 355             /* COULOMB ELECTROSTATICS */
 356             velec            = _mm_mul_pd(qq30,rinv30);
 357             felec            = _mm_mul_pd(velec,rinvsq30);
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velecsum         = _mm_add_pd(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             /* Update vectorial force */
 365             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 366             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 367             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 368
 369             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 370             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 371             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 372
 373             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 374
 375             /* Inner loop uses 155 flops */
 376         }
 377
 378         if(jidx<j_index_end)
 379         {
 380
 381             jnrA             = jjnr[jidx];
 382             j_coord_offsetA  = DIM*jnrA;
 383
 384             /* load j atom coordinates */
 385             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 386                                               &jx0,&jy0,&jz0);
 387
 388             /* Calculate displacement vector */
 389             dx00             = _mm_sub_pd(ix0,jx0);
 390             dy00             = _mm_sub_pd(iy0,jy0);
 391             dz00             = _mm_sub_pd(iz0,jz0);
 392             dx10             = _mm_sub_pd(ix1,jx0);
 393             dy10             = _mm_sub_pd(iy1,jy0);
 394             dz10             = _mm_sub_pd(iz1,jz0);
 395             dx20             = _mm_sub_pd(ix2,jx0);
 396             dy20             = _mm_sub_pd(iy2,jy0);
 397             dz20             = _mm_sub_pd(iz2,jz0);
 398             dx30             = _mm_sub_pd(ix3,jx0);
 399             dy30             = _mm_sub_pd(iy3,jy0);
 400             dz30             = _mm_sub_pd(iz3,jz0);
 401
 402             /* Calculate squared distance and things based on it */
 403             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 404             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 405             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 406             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 407
 408             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 409             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 410             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 411             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 412
 413             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 414             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 415             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 416
 417             /* Load parameters for j particles */
 418             jq0              = _mm_load_sd(charge+jnrA+0);
 419             vdwjidx0A        = 2*vdwtype[jnrA+0];
 420
 421             fjx0             = _mm_setzero_pd();
 422             fjy0             = _mm_setzero_pd();
 423             fjz0             = _mm_setzero_pd();
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             r00              = _mm_mul_pd(rsq00,rinv00);
 430
 431             /* Compute parameters for interactions between i and j atoms */
 432             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 433
 434             /* Calculate table index by multiplying r with table scale and truncate to integer */
 435             rt               = _mm_mul_pd(r00,vftabscale);
 436             vfitab           = _mm_cvttpd_epi32(rt);
 437 #ifdef __XOP__
 438             vfeps            = _mm_frcz_pd(rt);
 439 #else
 440             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 441 #endif
 442             twovfeps         = _mm_add_pd(vfeps,vfeps);
 443             vfitab           = _mm_slli_epi32(vfitab,3);
 444
 445             /* CUBIC SPLINE TABLE DISPERSION */
 446             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 447             F                = _mm_setzero_pd();
 448             GMX_MM_TRANSPOSE2_PD(Y,F);
 449             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 450             H                = _mm_setzero_pd();
 451             GMX_MM_TRANSPOSE2_PD(G,H);
 452             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 453             VV               = _mm_macc_pd(vfeps,Fp,Y);
 454             vvdw6            = _mm_mul_pd(c6_00,VV);
 455             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 456             fvdw6            = _mm_mul_pd(c6_00,FF);
 457
 458             /* CUBIC SPLINE TABLE REPULSION */
 459             vfitab           = _mm_add_epi32(vfitab,ifour);
 460             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 461             F                = _mm_setzero_pd();
 462             GMX_MM_TRANSPOSE2_PD(Y,F);
 463             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 464             H                = _mm_setzero_pd();
 465             GMX_MM_TRANSPOSE2_PD(G,H);
 466             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 467             VV               = _mm_macc_pd(vfeps,Fp,Y);
 468             vvdw12           = _mm_mul_pd(c12_00,VV);
 469             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 470             fvdw12           = _mm_mul_pd(c12_00,FF);
 471             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 472             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 473
 474             /* Update potential sum for this i atom from the interaction with this j atom. */
 475             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 476             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 477
 478             fscal            = fvdw;
 479
 480             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 481
 482             /* Update vectorial force */
 483             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 484             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 485             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 486
 487             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 488             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 489             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             /* Compute parameters for interactions between i and j atoms */
 496             qq10             = _mm_mul_pd(iq1,jq0);
 497
 498             /* COULOMB ELECTROSTATICS */
 499             velec            = _mm_mul_pd(qq10,rinv10);
 500             felec            = _mm_mul_pd(velec,rinvsq10);
 501
 502             /* Update potential sum for this i atom from the interaction with this j atom. */
 503             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 504             velecsum         = _mm_add_pd(velecsum,velec);
 505
 506             fscal            = felec;
 507
 508             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 509
 510             /* Update vectorial force */
 511             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 512             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 513             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 514
 515             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 516             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 517             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             qq20             = _mm_mul_pd(iq2,jq0);
 525
 526             /* COULOMB ELECTROSTATICS */
 527             velec            = _mm_mul_pd(qq20,rinv20);
 528             felec            = _mm_mul_pd(velec,rinvsq20);
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 532             velecsum         = _mm_add_pd(velecsum,velec);
 533
 534             fscal            = felec;
 535
 536             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 537
 538             /* Update vectorial force */
 539             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 540             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 541             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 542
 543             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 544             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 545             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 546
 547             /**************************
 548              * CALCULATE INTERACTIONS *
 549              **************************/
 550
 551             /* Compute parameters for interactions between i and j atoms */
 552             qq30             = _mm_mul_pd(iq3,jq0);
 553
 554             /* COULOMB ELECTROSTATICS */
 555             velec            = _mm_mul_pd(qq30,rinv30);
 556             felec            = _mm_mul_pd(velec,rinvsq30);
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 560             velecsum         = _mm_add_pd(velecsum,velec);
 561
 562             fscal            = felec;
 563
 564             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 565
 566             /* Update vectorial force */
 567             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 568             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 569             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 570
 571             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 572             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 573             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 574
 575             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 576
 577             /* Inner loop uses 155 flops */
 578         }
 579
 580         /* End of innermost loop */
 581
 582         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 583                                               f+i_coord_offset,fshift+i_shift_offset);
 584
 585         ggid                        = gid[iidx];
 586         /* Update potential energies */
 587         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 588         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 589
 590         /* Increment number of inner iterations */
 591         inneriter                  += j_index_end - j_index_start;
 592
 593         /* Outer loop uses 26 flops */
 594     }
 595
 596     /* Increment number of outer iterations */
 597     outeriter        += nri;
 598
 599     /* Update outer/inner flops */
 600
 601     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*155);
 602 }
 603 /*
 604  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 605  * Electrostatics interaction: Coulomb
 606  * VdW interaction:            CubicSplineTable
 607  * Geometry:                   Water4-Particle
 608  * Calculate force/pot:        Force
 609  */
 610 void
 611 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 612                     (t_nblist                    * gmx_restrict       nlist,
 613                      rvec                        * gmx_restrict          xx,
 614                      rvec                        * gmx_restrict          ff,
 615                      t_forcerec                  * gmx_restrict          fr,
 616                      t_mdatoms                   * gmx_restrict     mdatoms,
 617                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 618                      t_nrnb                      * gmx_restrict        nrnb)
 619 {
 620     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 621      * just 0 for non-waters.
 622      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 623      * jnr indices corresponding to data put in the four positions in the SIMD register.
 624      */
 625     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 626     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 627     int              jnrA,jnrB;
 628     int              j_coord_offsetA,j_coord_offsetB;
 629     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 630     real             rcutoff_scalar;
 631     real             *shiftvec,*fshift,*x,*f;
 632     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 633     int              vdwioffset0;
 634     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 635     int              vdwioffset1;
 636     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 637     int              vdwioffset2;
 638     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 639     int              vdwioffset3;
 640     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 641     int              vdwjidx0A,vdwjidx0B;
 642     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 643     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 644     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 645     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 646     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 647     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 648     real             *charge;
 649     int              nvdwtype;
 650     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 651     int              *vdwtype;
 652     real             *vdwparam;
 653     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 654     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 655     __m128i          vfitab;
 656     __m128i          ifour       = _mm_set1_epi32(4);
 657     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 658     real             *vftab;
 659     __m128d          dummy_mask,cutoff_mask;
 660     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 661     __m128d          one     = _mm_set1_pd(1.0);
 662     __m128d          two     = _mm_set1_pd(2.0);
 663     x                = xx[0];
 664     f                = ff[0];
 665
 666     nri              = nlist->nri;
 667     iinr             = nlist->iinr;
 668     jindex           = nlist->jindex;
 669     jjnr             = nlist->jjnr;
 670     shiftidx         = nlist->shift;
 671     gid              = nlist->gid;
 672     shiftvec         = fr->shift_vec[0];
 673     fshift           = fr->fshift[0];
 674     facel            = _mm_set1_pd(fr->epsfac);
 675     charge           = mdatoms->chargeA;
 676     nvdwtype         = fr->ntype;
 677     vdwparam         = fr->nbfp;
 678     vdwtype          = mdatoms->typeA;
 679
 680     vftab            = kernel_data->table_vdw->data;
 681     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 682
 683     /* Setup water-specific parameters */
 684     inr              = nlist->iinr[0];
 685     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 686     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 687     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 688     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 689
 690     /* Avoid stupid compiler warnings */
 691     jnrA = jnrB = 0;
 692     j_coord_offsetA = 0;
 693     j_coord_offsetB = 0;
 694
 695     outeriter        = 0;
 696     inneriter        = 0;
 697
 698     /* Start outer loop over neighborlists */
 699     for(iidx=0; iidx<nri; iidx++)
 700     {
 701         /* Load shift vector for this list */
 702         i_shift_offset   = DIM*shiftidx[iidx];
 703
 704         /* Load limits for loop over neighbors */
 705         j_index_start    = jindex[iidx];
 706         j_index_end      = jindex[iidx+1];
 707
 708         /* Get outer coordinate index */
 709         inr              = iinr[iidx];
 710         i_coord_offset   = DIM*inr;
 711
 712         /* Load i particle coords and add shift vector */
 713         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 714                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 715
 716         fix0             = _mm_setzero_pd();
 717         fiy0             = _mm_setzero_pd();
 718         fiz0             = _mm_setzero_pd();
 719         fix1             = _mm_setzero_pd();
 720         fiy1             = _mm_setzero_pd();
 721         fiz1             = _mm_setzero_pd();
 722         fix2             = _mm_setzero_pd();
 723         fiy2             = _mm_setzero_pd();
 724         fiz2             = _mm_setzero_pd();
 725         fix3             = _mm_setzero_pd();
 726         fiy3             = _mm_setzero_pd();
 727         fiz3             = _mm_setzero_pd();
 728
 729         /* Start inner kernel loop */
 730         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 731         {
 732
 733             /* Get j neighbor index, and coordinate index */
 734             jnrA             = jjnr[jidx];
 735             jnrB             = jjnr[jidx+1];
 736             j_coord_offsetA  = DIM*jnrA;
 737             j_coord_offsetB  = DIM*jnrB;
 738
 739             /* load j atom coordinates */
 740             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 741                                               &jx0,&jy0,&jz0);
 742
 743             /* Calculate displacement vector */
 744             dx00             = _mm_sub_pd(ix0,jx0);
 745             dy00             = _mm_sub_pd(iy0,jy0);
 746             dz00             = _mm_sub_pd(iz0,jz0);
 747             dx10             = _mm_sub_pd(ix1,jx0);
 748             dy10             = _mm_sub_pd(iy1,jy0);
 749             dz10             = _mm_sub_pd(iz1,jz0);
 750             dx20             = _mm_sub_pd(ix2,jx0);
 751             dy20             = _mm_sub_pd(iy2,jy0);
 752             dz20             = _mm_sub_pd(iz2,jz0);
 753             dx30             = _mm_sub_pd(ix3,jx0);
 754             dy30             = _mm_sub_pd(iy3,jy0);
 755             dz30             = _mm_sub_pd(iz3,jz0);
 756
 757             /* Calculate squared distance and things based on it */
 758             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 759             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 760             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 761             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 762
 763             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 764             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 765             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 766             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 767
 768             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 769             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 770             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 771
 772             /* Load parameters for j particles */
 773             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 774             vdwjidx0A        = 2*vdwtype[jnrA+0];
 775             vdwjidx0B        = 2*vdwtype[jnrB+0];
 776
 777             fjx0             = _mm_setzero_pd();
 778             fjy0             = _mm_setzero_pd();
 779             fjz0             = _mm_setzero_pd();
 780
 781             /**************************
 782              * CALCULATE INTERACTIONS *
 783              **************************/
 784
 785             r00              = _mm_mul_pd(rsq00,rinv00);
 786
 787             /* Compute parameters for interactions between i and j atoms */
 788             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 789                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 790
 791             /* Calculate table index by multiplying r with table scale and truncate to integer */
 792             rt               = _mm_mul_pd(r00,vftabscale);
 793             vfitab           = _mm_cvttpd_epi32(rt);
 794 #ifdef __XOP__
 795             vfeps            = _mm_frcz_pd(rt);
 796 #else
 797             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 798 #endif
 799             twovfeps         = _mm_add_pd(vfeps,vfeps);
 800             vfitab           = _mm_slli_epi32(vfitab,3);
 801
 802             /* CUBIC SPLINE TABLE DISPERSION */
 803             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 804             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 805             GMX_MM_TRANSPOSE2_PD(Y,F);
 806             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 807             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 808             GMX_MM_TRANSPOSE2_PD(G,H);
 809             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 810             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 811             fvdw6            = _mm_mul_pd(c6_00,FF);
 812
 813             /* CUBIC SPLINE TABLE REPULSION */
 814             vfitab           = _mm_add_epi32(vfitab,ifour);
 815             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 816             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 817             GMX_MM_TRANSPOSE2_PD(Y,F);
 818             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 819             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 820             GMX_MM_TRANSPOSE2_PD(G,H);
 821             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 822             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 823             fvdw12           = _mm_mul_pd(c12_00,FF);
 824             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 825
 826             fscal            = fvdw;
 827
 828             /* Update vectorial force */
 829             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 830             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 831             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 832
 833             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 834             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 835             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 836
 837             /**************************
 838              * CALCULATE INTERACTIONS *
 839              **************************/
 840
 841             /* Compute parameters for interactions between i and j atoms */
 842             qq10             = _mm_mul_pd(iq1,jq0);
 843
 844             /* COULOMB ELECTROSTATICS */
 845             velec            = _mm_mul_pd(qq10,rinv10);
 846             felec            = _mm_mul_pd(velec,rinvsq10);
 847
 848             fscal            = felec;
 849
 850             /* Update vectorial force */
 851             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 852             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 853             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 854
 855             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 856             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 857             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 858
 859             /**************************
 860              * CALCULATE INTERACTIONS *
 861              **************************/
 862
 863             /* Compute parameters for interactions between i and j atoms */
 864             qq20             = _mm_mul_pd(iq2,jq0);
 865
 866             /* COULOMB ELECTROSTATICS */
 867             velec            = _mm_mul_pd(qq20,rinv20);
 868             felec            = _mm_mul_pd(velec,rinvsq20);
 869
 870             fscal            = felec;
 871
 872             /* Update vectorial force */
 873             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 874             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 875             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 876
 877             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 878             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 879             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 880
 881             /**************************
 882              * CALCULATE INTERACTIONS *
 883              **************************/
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             qq30             = _mm_mul_pd(iq3,jq0);
 887
 888             /* COULOMB ELECTROSTATICS */
 889             velec            = _mm_mul_pd(qq30,rinv30);
 890             felec            = _mm_mul_pd(velec,rinvsq30);
 891
 892             fscal            = felec;
 893
 894             /* Update vectorial force */
 895             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 896             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 897             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 898
 899             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 900             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 901             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 902
 903             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 904
 905             /* Inner loop uses 144 flops */
 906         }
 907
 908         if(jidx<j_index_end)
 909         {
 910
 911             jnrA             = jjnr[jidx];
 912             j_coord_offsetA  = DIM*jnrA;
 913
 914             /* load j atom coordinates */
 915             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 916                                               &jx0,&jy0,&jz0);
 917
 918             /* Calculate displacement vector */
 919             dx00             = _mm_sub_pd(ix0,jx0);
 920             dy00             = _mm_sub_pd(iy0,jy0);
 921             dz00             = _mm_sub_pd(iz0,jz0);
 922             dx10             = _mm_sub_pd(ix1,jx0);
 923             dy10             = _mm_sub_pd(iy1,jy0);
 924             dz10             = _mm_sub_pd(iz1,jz0);
 925             dx20             = _mm_sub_pd(ix2,jx0);
 926             dy20             = _mm_sub_pd(iy2,jy0);
 927             dz20             = _mm_sub_pd(iz2,jz0);
 928             dx30             = _mm_sub_pd(ix3,jx0);
 929             dy30             = _mm_sub_pd(iy3,jy0);
 930             dz30             = _mm_sub_pd(iz3,jz0);
 931
 932             /* Calculate squared distance and things based on it */
 933             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 934             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 935             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 936             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 937
 938             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 939             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 940             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 941             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 942
 943             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 944             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 945             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 946
 947             /* Load parameters for j particles */
 948             jq0              = _mm_load_sd(charge+jnrA+0);
 949             vdwjidx0A        = 2*vdwtype[jnrA+0];
 950
 951             fjx0             = _mm_setzero_pd();
 952             fjy0             = _mm_setzero_pd();
 953             fjz0             = _mm_setzero_pd();
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             r00              = _mm_mul_pd(rsq00,rinv00);
 960
 961             /* Compute parameters for interactions between i and j atoms */
 962             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 963
 964             /* Calculate table index by multiplying r with table scale and truncate to integer */
 965             rt               = _mm_mul_pd(r00,vftabscale);
 966             vfitab           = _mm_cvttpd_epi32(rt);
 967 #ifdef __XOP__
 968             vfeps            = _mm_frcz_pd(rt);
 969 #else
 970             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 971 #endif
 972             twovfeps         = _mm_add_pd(vfeps,vfeps);
 973             vfitab           = _mm_slli_epi32(vfitab,3);
 974
 975             /* CUBIC SPLINE TABLE DISPERSION */
 976             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 977             F                = _mm_setzero_pd();
 978             GMX_MM_TRANSPOSE2_PD(Y,F);
 979             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 980             H                = _mm_setzero_pd();
 981             GMX_MM_TRANSPOSE2_PD(G,H);
 982             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 983             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 984             fvdw6            = _mm_mul_pd(c6_00,FF);
 985
 986             /* CUBIC SPLINE TABLE REPULSION */
 987             vfitab           = _mm_add_epi32(vfitab,ifour);
 988             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 989             F                = _mm_setzero_pd();
 990             GMX_MM_TRANSPOSE2_PD(Y,F);
 991             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 992             H                = _mm_setzero_pd();
 993             GMX_MM_TRANSPOSE2_PD(G,H);
 994             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 995             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 996             fvdw12           = _mm_mul_pd(c12_00,FF);
 997             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 998
 999             fscal            = fvdw;
1000
1001             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1002
1003             /* Update vectorial force */
1004             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1005             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1006             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1007
1008             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1009             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1010             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             /* Compute parameters for interactions between i and j atoms */
1017             qq10             = _mm_mul_pd(iq1,jq0);
1018
1019             /* COULOMB ELECTROSTATICS */
1020             velec            = _mm_mul_pd(qq10,rinv10);
1021             felec            = _mm_mul_pd(velec,rinvsq10);
1022
1023             fscal            = felec;
1024
1025             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1026
1027             /* Update vectorial force */
1028             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1029             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1030             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1031
1032             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1033             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1034             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1035
1036             /**************************
1037              * CALCULATE INTERACTIONS *
1038              **************************/
1039
1040             /* Compute parameters for interactions between i and j atoms */
1041             qq20             = _mm_mul_pd(iq2,jq0);
1042
1043             /* COULOMB ELECTROSTATICS */
1044             velec            = _mm_mul_pd(qq20,rinv20);
1045             felec            = _mm_mul_pd(velec,rinvsq20);
1046
1047             fscal            = felec;
1048
1049             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1050
1051             /* Update vectorial force */
1052             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1053             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1054             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1055
1056             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1057             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1058             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1059
1060             /**************************
1061              * CALCULATE INTERACTIONS *
1062              **************************/
1063
1064             /* Compute parameters for interactions between i and j atoms */
1065             qq30             = _mm_mul_pd(iq3,jq0);
1066
1067             /* COULOMB ELECTROSTATICS */
1068             velec            = _mm_mul_pd(qq30,rinv30);
1069             felec            = _mm_mul_pd(velec,rinvsq30);
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 }