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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 105     real             *vftab;
 106     __m128d          dummy_mask,cutoff_mask;
 107     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 108     __m128d          one     = _mm_set1_pd(1.0);
 109     __m128d          two     = _mm_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm_set1_pd(fr->ic->k_rf);
 124     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 125     crf              = _mm_set1_pd(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 136     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 137     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 164                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 165
 166         fix0             = _mm_setzero_pd();
 167         fiy0             = _mm_setzero_pd();
 168         fiz0             = _mm_setzero_pd();
 169         fix1             = _mm_setzero_pd();
 170         fiy1             = _mm_setzero_pd();
 171         fiz1             = _mm_setzero_pd();
 172         fix2             = _mm_setzero_pd();
 173         fiy2             = _mm_setzero_pd();
 174         fiz2             = _mm_setzero_pd();
 175         fix3             = _mm_setzero_pd();
 176         fiy3             = _mm_setzero_pd();
 177         fiz3             = _mm_setzero_pd();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_pd();
 181         vvdwsum          = _mm_setzero_pd();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             j_coord_offsetA  = DIM*jnrA;
 191             j_coord_offsetB  = DIM*jnrB;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_pd(ix0,jx0);
 199             dy00             = _mm_sub_pd(iy0,jy0);
 200             dz00             = _mm_sub_pd(iz0,jz0);
 201             dx10             = _mm_sub_pd(ix1,jx0);
 202             dy10             = _mm_sub_pd(iy1,jy0);
 203             dz10             = _mm_sub_pd(iz1,jz0);
 204             dx20             = _mm_sub_pd(ix2,jx0);
 205             dy20             = _mm_sub_pd(iy2,jy0);
 206             dz20             = _mm_sub_pd(iz2,jz0);
 207             dx30             = _mm_sub_pd(ix3,jx0);
 208             dy30             = _mm_sub_pd(iy3,jy0);
 209             dz30             = _mm_sub_pd(iz3,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 213             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 215             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 216
 217             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 218             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 219             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 220             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 221
 222             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 223             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 224             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230
 231             fjx0             = _mm_setzero_pd();
 232             fjy0             = _mm_setzero_pd();
 233             fjz0             = _mm_setzero_pd();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             r00              = _mm_mul_pd(rsq00,rinv00);
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 243                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 244
 245             /* Calculate table index by multiplying r with table scale and truncate to integer */
 246             rt               = _mm_mul_pd(r00,vftabscale);
 247             vfitab           = _mm_cvttpd_epi32(rt);
 248 #ifdef __XOP__
 249             vfeps            = _mm_frcz_pd(rt);
 250 #else
 251             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 252 #endif
 253             twovfeps         = _mm_add_pd(vfeps,vfeps);
 254             vfitab           = _mm_slli_epi32(vfitab,3);
 255
 256             /* CUBIC SPLINE TABLE DISPERSION */
 257             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 258             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 259             GMX_MM_TRANSPOSE2_PD(Y,F);
 260             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 261             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 262             GMX_MM_TRANSPOSE2_PD(G,H);
 263             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 264             VV               = _mm_macc_pd(vfeps,Fp,Y);
 265             vvdw6            = _mm_mul_pd(c6_00,VV);
 266             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 267             fvdw6            = _mm_mul_pd(c6_00,FF);
 268
 269             /* CUBIC SPLINE TABLE REPULSION */
 270             vfitab           = _mm_add_epi32(vfitab,ifour);
 271             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 272             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 273             GMX_MM_TRANSPOSE2_PD(Y,F);
 274             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 275             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 276             GMX_MM_TRANSPOSE2_PD(G,H);
 277             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 278             VV               = _mm_macc_pd(vfeps,Fp,Y);
 279             vvdw12           = _mm_mul_pd(c12_00,VV);
 280             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 281             fvdw12           = _mm_mul_pd(c12_00,FF);
 282             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 283             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 284
 285             /* Update potential sum for this i atom from the interaction with this j atom. */
 286             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 287
 288             fscal            = fvdw;
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 292             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 293             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 294
 295             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 296             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 297             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _mm_mul_pd(iq1,jq0);
 305
 306             /* REACTION-FIELD ELECTROSTATICS */
 307             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 308             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             velecsum         = _mm_add_pd(velecsum,velec);
 312
 313             fscal            = felec;
 314
 315             /* Update vectorial force */
 316             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 317             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 318             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 319
 320             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 321             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 322             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq20             = _mm_mul_pd(iq2,jq0);
 330
 331             /* REACTION-FIELD ELECTROSTATICS */
 332             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 333             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velecsum         = _mm_add_pd(velecsum,velec);
 337
 338             fscal            = felec;
 339
 340             /* Update vectorial force */
 341             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 342             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 343             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 344
 345             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 346             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 347             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq30             = _mm_mul_pd(iq3,jq0);
 355
 356             /* REACTION-FIELD ELECTROSTATICS */
 357             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 358             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velecsum         = _mm_add_pd(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             /* Update vectorial force */
 366             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 367             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 368             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 369
 370             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 371             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 372             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 373
 374             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 375
 376             /* Inner loop uses 167 flops */
 377         }
 378
 379         if(jidx<j_index_end)
 380         {
 381
 382             jnrA             = jjnr[jidx];
 383             j_coord_offsetA  = DIM*jnrA;
 384
 385             /* load j atom coordinates */
 386             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 387                                               &jx0,&jy0,&jz0);
 388
 389             /* Calculate displacement vector */
 390             dx00             = _mm_sub_pd(ix0,jx0);
 391             dy00             = _mm_sub_pd(iy0,jy0);
 392             dz00             = _mm_sub_pd(iz0,jz0);
 393             dx10             = _mm_sub_pd(ix1,jx0);
 394             dy10             = _mm_sub_pd(iy1,jy0);
 395             dz10             = _mm_sub_pd(iz1,jz0);
 396             dx20             = _mm_sub_pd(ix2,jx0);
 397             dy20             = _mm_sub_pd(iy2,jy0);
 398             dz20             = _mm_sub_pd(iz2,jz0);
 399             dx30             = _mm_sub_pd(ix3,jx0);
 400             dy30             = _mm_sub_pd(iy3,jy0);
 401             dz30             = _mm_sub_pd(iz3,jz0);
 402
 403             /* Calculate squared distance and things based on it */
 404             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 405             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 406             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 407             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 408
 409             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 410             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 411             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 412             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 413
 414             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 415             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 416             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 417
 418             /* Load parameters for j particles */
 419             jq0              = _mm_load_sd(charge+jnrA+0);
 420             vdwjidx0A        = 2*vdwtype[jnrA+0];
 421
 422             fjx0             = _mm_setzero_pd();
 423             fjy0             = _mm_setzero_pd();
 424             fjz0             = _mm_setzero_pd();
 425
 426             /**************************
 427              * CALCULATE INTERACTIONS *
 428              **************************/
 429
 430             r00              = _mm_mul_pd(rsq00,rinv00);
 431
 432             /* Compute parameters for interactions between i and j atoms */
 433             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 434
 435             /* Calculate table index by multiplying r with table scale and truncate to integer */
 436             rt               = _mm_mul_pd(r00,vftabscale);
 437             vfitab           = _mm_cvttpd_epi32(rt);
 438 #ifdef __XOP__
 439             vfeps            = _mm_frcz_pd(rt);
 440 #else
 441             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 442 #endif
 443             twovfeps         = _mm_add_pd(vfeps,vfeps);
 444             vfitab           = _mm_slli_epi32(vfitab,3);
 445
 446             /* CUBIC SPLINE TABLE DISPERSION */
 447             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 448             F                = _mm_setzero_pd();
 449             GMX_MM_TRANSPOSE2_PD(Y,F);
 450             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 451             H                = _mm_setzero_pd();
 452             GMX_MM_TRANSPOSE2_PD(G,H);
 453             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 454             VV               = _mm_macc_pd(vfeps,Fp,Y);
 455             vvdw6            = _mm_mul_pd(c6_00,VV);
 456             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 457             fvdw6            = _mm_mul_pd(c6_00,FF);
 458
 459             /* CUBIC SPLINE TABLE REPULSION */
 460             vfitab           = _mm_add_epi32(vfitab,ifour);
 461             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 462             F                = _mm_setzero_pd();
 463             GMX_MM_TRANSPOSE2_PD(Y,F);
 464             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 465             H                = _mm_setzero_pd();
 466             GMX_MM_TRANSPOSE2_PD(G,H);
 467             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 468             VV               = _mm_macc_pd(vfeps,Fp,Y);
 469             vvdw12           = _mm_mul_pd(c12_00,VV);
 470             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 471             fvdw12           = _mm_mul_pd(c12_00,FF);
 472             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 473             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 477             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 478
 479             fscal            = fvdw;
 480
 481             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 482
 483             /* Update vectorial force */
 484             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 485             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 486             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 487
 488             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 489             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 490             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             qq10             = _mm_mul_pd(iq1,jq0);
 498
 499             /* REACTION-FIELD ELECTROSTATICS */
 500             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 501             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 505             velecsum         = _mm_add_pd(velecsum,velec);
 506
 507             fscal            = felec;
 508
 509             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 510
 511             /* Update vectorial force */
 512             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 513             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 514             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 515
 516             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 517             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 518             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 519
 520             /**************************
 521              * CALCULATE INTERACTIONS *
 522              **************************/
 523
 524             /* Compute parameters for interactions between i and j atoms */
 525             qq20             = _mm_mul_pd(iq2,jq0);
 526
 527             /* REACTION-FIELD ELECTROSTATICS */
 528             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 529             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 530
 531             /* Update potential sum for this i atom from the interaction with this j atom. */
 532             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 533             velecsum         = _mm_add_pd(velecsum,velec);
 534
 535             fscal            = felec;
 536
 537             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 538
 539             /* Update vectorial force */
 540             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 541             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 542             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 543
 544             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 545             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 546             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 547
 548             /**************************
 549              * CALCULATE INTERACTIONS *
 550              **************************/
 551
 552             /* Compute parameters for interactions between i and j atoms */
 553             qq30             = _mm_mul_pd(iq3,jq0);
 554
 555             /* REACTION-FIELD ELECTROSTATICS */
 556             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 557             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 558
 559             /* Update potential sum for this i atom from the interaction with this j atom. */
 560             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 561             velecsum         = _mm_add_pd(velecsum,velec);
 562
 563             fscal            = felec;
 564
 565             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 566
 567             /* Update vectorial force */
 568             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 569             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 570             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 571
 572             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 573             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 574             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 575
 576             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 577
 578             /* Inner loop uses 167 flops */
 579         }
 580
 581         /* End of innermost loop */
 582
 583         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 584                                               f+i_coord_offset,fshift+i_shift_offset);
 585
 586         ggid                        = gid[iidx];
 587         /* Update potential energies */
 588         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 589         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 590
 591         /* Increment number of inner iterations */
 592         inneriter                  += j_index_end - j_index_start;
 593
 594         /* Outer loop uses 26 flops */
 595     }
 596
 597     /* Increment number of outer iterations */
 598     outeriter        += nri;
 599
 600     /* Update outer/inner flops */
 601
 602     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*167);
 603 }
 604 /*
 605  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 606  * Electrostatics interaction: ReactionField
 607  * VdW interaction:            CubicSplineTable
 608  * Geometry:                   Water4-Particle
 609  * Calculate force/pot:        Force
 610  */
 611 void
 612 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 613                     (t_nblist                    * gmx_restrict       nlist,
 614                      rvec                        * gmx_restrict          xx,
 615                      rvec                        * gmx_restrict          ff,
 616                      t_forcerec                  * gmx_restrict          fr,
 617                      t_mdatoms                   * gmx_restrict     mdatoms,
 618                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 619                      t_nrnb                      * gmx_restrict        nrnb)
 620 {
 621     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 622      * just 0 for non-waters.
 623      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 624      * jnr indices corresponding to data put in the four positions in the SIMD register.
 625      */
 626     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 627     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 628     int              jnrA,jnrB;
 629     int              j_coord_offsetA,j_coord_offsetB;
 630     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 631     real             rcutoff_scalar;
 632     real             *shiftvec,*fshift,*x,*f;
 633     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 634     int              vdwioffset0;
 635     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 636     int              vdwioffset1;
 637     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 638     int              vdwioffset2;
 639     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 640     int              vdwioffset3;
 641     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 642     int              vdwjidx0A,vdwjidx0B;
 643     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 644     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 645     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 646     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 647     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 648     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 649     real             *charge;
 650     int              nvdwtype;
 651     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 652     int              *vdwtype;
 653     real             *vdwparam;
 654     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 655     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 656     __m128i          vfitab;
 657     __m128i          ifour       = _mm_set1_epi32(4);
 658     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 659     real             *vftab;
 660     __m128d          dummy_mask,cutoff_mask;
 661     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 662     __m128d          one     = _mm_set1_pd(1.0);
 663     __m128d          two     = _mm_set1_pd(2.0);
 664     x                = xx[0];
 665     f                = ff[0];
 666
 667     nri              = nlist->nri;
 668     iinr             = nlist->iinr;
 669     jindex           = nlist->jindex;
 670     jjnr             = nlist->jjnr;
 671     shiftidx         = nlist->shift;
 672     gid              = nlist->gid;
 673     shiftvec         = fr->shift_vec[0];
 674     fshift           = fr->fshift[0];
 675     facel            = _mm_set1_pd(fr->epsfac);
 676     charge           = mdatoms->chargeA;
 677     krf              = _mm_set1_pd(fr->ic->k_rf);
 678     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 679     crf              = _mm_set1_pd(fr->ic->c_rf);
 680     nvdwtype         = fr->ntype;
 681     vdwparam         = fr->nbfp;
 682     vdwtype          = mdatoms->typeA;
 683
 684     vftab            = kernel_data->table_vdw->data;
 685     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 686
 687     /* Setup water-specific parameters */
 688     inr              = nlist->iinr[0];
 689     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 690     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 691     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 692     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 693
 694     /* Avoid stupid compiler warnings */
 695     jnrA = jnrB = 0;
 696     j_coord_offsetA = 0;
 697     j_coord_offsetB = 0;
 698
 699     outeriter        = 0;
 700     inneriter        = 0;
 701
 702     /* Start outer loop over neighborlists */
 703     for(iidx=0; iidx<nri; iidx++)
 704     {
 705         /* Load shift vector for this list */
 706         i_shift_offset   = DIM*shiftidx[iidx];
 707
 708         /* Load limits for loop over neighbors */
 709         j_index_start    = jindex[iidx];
 710         j_index_end      = jindex[iidx+1];
 711
 712         /* Get outer coordinate index */
 713         inr              = iinr[iidx];
 714         i_coord_offset   = DIM*inr;
 715
 716         /* Load i particle coords and add shift vector */
 717         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 718                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 719
 720         fix0             = _mm_setzero_pd();
 721         fiy0             = _mm_setzero_pd();
 722         fiz0             = _mm_setzero_pd();
 723         fix1             = _mm_setzero_pd();
 724         fiy1             = _mm_setzero_pd();
 725         fiz1             = _mm_setzero_pd();
 726         fix2             = _mm_setzero_pd();
 727         fiy2             = _mm_setzero_pd();
 728         fiz2             = _mm_setzero_pd();
 729         fix3             = _mm_setzero_pd();
 730         fiy3             = _mm_setzero_pd();
 731         fiz3             = _mm_setzero_pd();
 732
 733         /* Start inner kernel loop */
 734         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 735         {
 736
 737             /* Get j neighbor index, and coordinate index */
 738             jnrA             = jjnr[jidx];
 739             jnrB             = jjnr[jidx+1];
 740             j_coord_offsetA  = DIM*jnrA;
 741             j_coord_offsetB  = DIM*jnrB;
 742
 743             /* load j atom coordinates */
 744             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 745                                               &jx0,&jy0,&jz0);
 746
 747             /* Calculate displacement vector */
 748             dx00             = _mm_sub_pd(ix0,jx0);
 749             dy00             = _mm_sub_pd(iy0,jy0);
 750             dz00             = _mm_sub_pd(iz0,jz0);
 751             dx10             = _mm_sub_pd(ix1,jx0);
 752             dy10             = _mm_sub_pd(iy1,jy0);
 753             dz10             = _mm_sub_pd(iz1,jz0);
 754             dx20             = _mm_sub_pd(ix2,jx0);
 755             dy20             = _mm_sub_pd(iy2,jy0);
 756             dz20             = _mm_sub_pd(iz2,jz0);
 757             dx30             = _mm_sub_pd(ix3,jx0);
 758             dy30             = _mm_sub_pd(iy3,jy0);
 759             dz30             = _mm_sub_pd(iz3,jz0);
 760
 761             /* Calculate squared distance and things based on it */
 762             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 763             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 764             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 765             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 766
 767             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 768             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 769             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 770             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 771
 772             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 773             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 774             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 775
 776             /* Load parameters for j particles */
 777             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 778             vdwjidx0A        = 2*vdwtype[jnrA+0];
 779             vdwjidx0B        = 2*vdwtype[jnrB+0];
 780
 781             fjx0             = _mm_setzero_pd();
 782             fjy0             = _mm_setzero_pd();
 783             fjz0             = _mm_setzero_pd();
 784
 785             /**************************
 786              * CALCULATE INTERACTIONS *
 787              **************************/
 788
 789             r00              = _mm_mul_pd(rsq00,rinv00);
 790
 791             /* Compute parameters for interactions between i and j atoms */
 792             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 793                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 794
 795             /* Calculate table index by multiplying r with table scale and truncate to integer */
 796             rt               = _mm_mul_pd(r00,vftabscale);
 797             vfitab           = _mm_cvttpd_epi32(rt);
 798 #ifdef __XOP__
 799             vfeps            = _mm_frcz_pd(rt);
 800 #else
 801             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 802 #endif
 803             twovfeps         = _mm_add_pd(vfeps,vfeps);
 804             vfitab           = _mm_slli_epi32(vfitab,3);
 805
 806             /* CUBIC SPLINE TABLE DISPERSION */
 807             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 808             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 809             GMX_MM_TRANSPOSE2_PD(Y,F);
 810             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 811             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 812             GMX_MM_TRANSPOSE2_PD(G,H);
 813             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 814             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 815             fvdw6            = _mm_mul_pd(c6_00,FF);
 816
 817             /* CUBIC SPLINE TABLE REPULSION */
 818             vfitab           = _mm_add_epi32(vfitab,ifour);
 819             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 820             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 821             GMX_MM_TRANSPOSE2_PD(Y,F);
 822             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 823             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 824             GMX_MM_TRANSPOSE2_PD(G,H);
 825             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 826             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 827             fvdw12           = _mm_mul_pd(c12_00,FF);
 828             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 829
 830             fscal            = fvdw;
 831
 832             /* Update vectorial force */
 833             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 834             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 835             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 836
 837             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 838             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 839             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 840
 841             /**************************
 842              * CALCULATE INTERACTIONS *
 843              **************************/
 844
 845             /* Compute parameters for interactions between i and j atoms */
 846             qq10             = _mm_mul_pd(iq1,jq0);
 847
 848             /* REACTION-FIELD ELECTROSTATICS */
 849             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 850
 851             fscal            = felec;
 852
 853             /* Update vectorial force */
 854             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 855             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 856             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 857
 858             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 859             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 860             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 861
 862             /**************************
 863              * CALCULATE INTERACTIONS *
 864              **************************/
 865
 866             /* Compute parameters for interactions between i and j atoms */
 867             qq20             = _mm_mul_pd(iq2,jq0);
 868
 869             /* REACTION-FIELD ELECTROSTATICS */
 870             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 871
 872             fscal            = felec;
 873
 874             /* Update vectorial force */
 875             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 876             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 877             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 878
 879             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 880             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 881             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 882
 883             /**************************
 884              * CALCULATE INTERACTIONS *
 885              **************************/
 886
 887             /* Compute parameters for interactions between i and j atoms */
 888             qq30             = _mm_mul_pd(iq3,jq0);
 889
 890             /* REACTION-FIELD ELECTROSTATICS */
 891             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 892
 893             fscal            = felec;
 894
 895             /* Update vectorial force */
 896             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 897             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 898             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 899
 900             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 901             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 902             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 903
 904             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 905
 906             /* Inner loop uses 144 flops */
 907         }
 908
 909         if(jidx<j_index_end)
 910         {
 911
 912             jnrA             = jjnr[jidx];
 913             j_coord_offsetA  = DIM*jnrA;
 914
 915             /* load j atom coordinates */
 916             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 917                                               &jx0,&jy0,&jz0);
 918
 919             /* Calculate displacement vector */
 920             dx00             = _mm_sub_pd(ix0,jx0);
 921             dy00             = _mm_sub_pd(iy0,jy0);
 922             dz00             = _mm_sub_pd(iz0,jz0);
 923             dx10             = _mm_sub_pd(ix1,jx0);
 924             dy10             = _mm_sub_pd(iy1,jy0);
 925             dz10             = _mm_sub_pd(iz1,jz0);
 926             dx20             = _mm_sub_pd(ix2,jx0);
 927             dy20             = _mm_sub_pd(iy2,jy0);
 928             dz20             = _mm_sub_pd(iz2,jz0);
 929             dx30             = _mm_sub_pd(ix3,jx0);
 930             dy30             = _mm_sub_pd(iy3,jy0);
 931             dz30             = _mm_sub_pd(iz3,jz0);
 932
 933             /* Calculate squared distance and things based on it */
 934             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 935             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 936             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 937             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 938
 939             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 940             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 941             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 942             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 943
 944             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 945             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 946             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 947
 948             /* Load parameters for j particles */
 949             jq0              = _mm_load_sd(charge+jnrA+0);
 950             vdwjidx0A        = 2*vdwtype[jnrA+0];
 951
 952             fjx0             = _mm_setzero_pd();
 953             fjy0             = _mm_setzero_pd();
 954             fjz0             = _mm_setzero_pd();
 955
 956             /**************************
 957              * CALCULATE INTERACTIONS *
 958              **************************/
 959
 960             r00              = _mm_mul_pd(rsq00,rinv00);
 961
 962             /* Compute parameters for interactions between i and j atoms */
 963             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 964
 965             /* Calculate table index by multiplying r with table scale and truncate to integer */
 966             rt               = _mm_mul_pd(r00,vftabscale);
 967             vfitab           = _mm_cvttpd_epi32(rt);
 968 #ifdef __XOP__
 969             vfeps            = _mm_frcz_pd(rt);
 970 #else
 971             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 972 #endif
 973             twovfeps         = _mm_add_pd(vfeps,vfeps);
 974             vfitab           = _mm_slli_epi32(vfitab,3);
 975
 976             /* CUBIC SPLINE TABLE DISPERSION */
 977             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 978             F                = _mm_setzero_pd();
 979             GMX_MM_TRANSPOSE2_PD(Y,F);
 980             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 981             H                = _mm_setzero_pd();
 982             GMX_MM_TRANSPOSE2_PD(G,H);
 983             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 984             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 985             fvdw6            = _mm_mul_pd(c6_00,FF);
 986
 987             /* CUBIC SPLINE TABLE REPULSION */
 988             vfitab           = _mm_add_epi32(vfitab,ifour);
 989             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 990             F                = _mm_setzero_pd();
 991             GMX_MM_TRANSPOSE2_PD(Y,F);
 992             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 993             H                = _mm_setzero_pd();
 994             GMX_MM_TRANSPOSE2_PD(G,H);
 995             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 996             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 997             fvdw12           = _mm_mul_pd(c12_00,FF);
 998             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 999
1000             fscal            = fvdw;
1001
1002             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1003
1004             /* Update vectorial force */
1005             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1006             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1007             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1008
1009             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1010             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1011             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             /* Compute parameters for interactions between i and j atoms */
1018             qq10             = _mm_mul_pd(iq1,jq0);
1019
1020             /* REACTION-FIELD ELECTROSTATICS */
1021             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
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             /* REACTION-FIELD ELECTROSTATICS */
1044             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1045
1046             fscal            = felec;
1047
1048             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1049
1050             /* Update vectorial force */
1051             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1052             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1053             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1054
1055             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1056             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1057             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1058
1059             /**************************
1060              * CALCULATE INTERACTIONS *
1061              **************************/
1062
1063             /* Compute parameters for interactions between i and j atoms */
1064             qq30             = _mm_mul_pd(iq3,jq0);
1065
1066             /* REACTION-FIELD ELECTROSTATICS */
1067             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
1068
1069             fscal            = felec;
1070
1071             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1072
1073             /* Update vectorial force */
1074             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1075             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1076             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1077
1078             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1079             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1080             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1081
1082             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1083
1084             /* Inner loop uses 144 flops */
1085         }
1086
1087         /* End of innermost loop */
1088
1089         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1090                                               f+i_coord_offset,fshift+i_shift_offset);
1091
1092         /* Increment number of inner iterations */
1093         inneriter                  += j_index_end - j_index_start;
1094
1095         /* Outer loop uses 24 flops */
1096     }
1097
1098     /* Increment number of outer iterations */
1099     outeriter        += nri;
1100
1101     /* Update outer/inner flops */
1102
1103     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*144);
1104 }