src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwNone_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 #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 "gromacs/math/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_ElecCSTab_VdwNone_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwNone_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              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          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  98     real             *vftab;
  99     __m128d          dummy_mask,cutoff_mask;
 100     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 101     __m128d          one     = _mm_set1_pd(1.0);
 102     __m128d          two     = _mm_set1_pd(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_pd(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116
 117     vftab            = kernel_data->table_elec->data;
 118     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 119
 120     /* Setup water-specific parameters */
 121     inr              = nlist->iinr[0];
 122     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 123     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 124     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 150                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 151
 152         fix1             = _mm_setzero_pd();
 153         fiy1             = _mm_setzero_pd();
 154         fiz1             = _mm_setzero_pd();
 155         fix2             = _mm_setzero_pd();
 156         fiy2             = _mm_setzero_pd();
 157         fiz2             = _mm_setzero_pd();
 158         fix3             = _mm_setzero_pd();
 159         fiy3             = _mm_setzero_pd();
 160         fiz3             = _mm_setzero_pd();
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174
 175             /* load j atom coordinates */
 176             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 177                                               &jx0,&jy0,&jz0);
 178
 179             /* Calculate displacement vector */
 180             dx10             = _mm_sub_pd(ix1,jx0);
 181             dy10             = _mm_sub_pd(iy1,jy0);
 182             dz10             = _mm_sub_pd(iz1,jz0);
 183             dx20             = _mm_sub_pd(ix2,jx0);
 184             dy20             = _mm_sub_pd(iy2,jy0);
 185             dz20             = _mm_sub_pd(iz2,jz0);
 186             dx30             = _mm_sub_pd(ix3,jx0);
 187             dy30             = _mm_sub_pd(iy3,jy0);
 188             dz30             = _mm_sub_pd(iz3,jz0);
 189
 190             /* Calculate squared distance and things based on it */
 191             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 192             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 193             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 194
 195             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 196             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 197             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 198
 199             /* Load parameters for j particles */
 200             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 201
 202             fjx0             = _mm_setzero_pd();
 203             fjy0             = _mm_setzero_pd();
 204             fjz0             = _mm_setzero_pd();
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             r10              = _mm_mul_pd(rsq10,rinv10);
 211
 212             /* Compute parameters for interactions between i and j atoms */
 213             qq10             = _mm_mul_pd(iq1,jq0);
 214
 215             /* Calculate table index by multiplying r with table scale and truncate to integer */
 216             rt               = _mm_mul_pd(r10,vftabscale);
 217             vfitab           = _mm_cvttpd_epi32(rt);
 218 #ifdef __XOP__
 219             vfeps            = _mm_frcz_pd(rt);
 220 #else
 221             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 222 #endif
 223             twovfeps         = _mm_add_pd(vfeps,vfeps);
 224             vfitab           = _mm_slli_epi32(vfitab,2);
 225
 226             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 227             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 228             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 229             GMX_MM_TRANSPOSE2_PD(Y,F);
 230             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 231             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 232             GMX_MM_TRANSPOSE2_PD(G,H);
 233             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 234             VV               = _mm_macc_pd(vfeps,Fp,Y);
 235             velec            = _mm_mul_pd(qq10,VV);
 236             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 237             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 238
 239             /* Update potential sum for this i atom from the interaction with this j atom. */
 240             velecsum         = _mm_add_pd(velecsum,velec);
 241
 242             fscal            = felec;
 243
 244             /* Update vectorial force */
 245             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 246             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 247             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 248
 249             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 250             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 251             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             r20              = _mm_mul_pd(rsq20,rinv20);
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             qq20             = _mm_mul_pd(iq2,jq0);
 261
 262             /* Calculate table index by multiplying r with table scale and truncate to integer */
 263             rt               = _mm_mul_pd(r20,vftabscale);
 264             vfitab           = _mm_cvttpd_epi32(rt);
 265 #ifdef __XOP__
 266             vfeps            = _mm_frcz_pd(rt);
 267 #else
 268             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 269 #endif
 270             twovfeps         = _mm_add_pd(vfeps,vfeps);
 271             vfitab           = _mm_slli_epi32(vfitab,2);
 272
 273             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 274             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 275             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 276             GMX_MM_TRANSPOSE2_PD(Y,F);
 277             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 278             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 279             GMX_MM_TRANSPOSE2_PD(G,H);
 280             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 281             VV               = _mm_macc_pd(vfeps,Fp,Y);
 282             velec            = _mm_mul_pd(qq20,VV);
 283             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 284             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velecsum         = _mm_add_pd(velecsum,velec);
 288
 289             fscal            = felec;
 290
 291             /* Update vectorial force */
 292             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 293             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 294             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 295
 296             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 297             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 298             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r30              = _mm_mul_pd(rsq30,rinv30);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq30             = _mm_mul_pd(iq3,jq0);
 308
 309             /* Calculate table index by multiplying r with table scale and truncate to integer */
 310             rt               = _mm_mul_pd(r30,vftabscale);
 311             vfitab           = _mm_cvttpd_epi32(rt);
 312 #ifdef __XOP__
 313             vfeps            = _mm_frcz_pd(rt);
 314 #else
 315             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 316 #endif
 317             twovfeps         = _mm_add_pd(vfeps,vfeps);
 318             vfitab           = _mm_slli_epi32(vfitab,2);
 319
 320             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 321             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 322             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 323             GMX_MM_TRANSPOSE2_PD(Y,F);
 324             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 325             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 326             GMX_MM_TRANSPOSE2_PD(G,H);
 327             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 328             VV               = _mm_macc_pd(vfeps,Fp,Y);
 329             velec            = _mm_mul_pd(qq30,VV);
 330             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 331             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velecsum         = _mm_add_pd(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             /* Update vectorial force */
 339             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 340             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 341             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 342
 343             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 344             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 345             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 346
 347             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 348
 349             /* Inner loop uses 141 flops */
 350         }
 351
 352         if(jidx<j_index_end)
 353         {
 354
 355             jnrA             = jjnr[jidx];
 356             j_coord_offsetA  = DIM*jnrA;
 357
 358             /* load j atom coordinates */
 359             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 360                                               &jx0,&jy0,&jz0);
 361
 362             /* Calculate displacement vector */
 363             dx10             = _mm_sub_pd(ix1,jx0);
 364             dy10             = _mm_sub_pd(iy1,jy0);
 365             dz10             = _mm_sub_pd(iz1,jz0);
 366             dx20             = _mm_sub_pd(ix2,jx0);
 367             dy20             = _mm_sub_pd(iy2,jy0);
 368             dz20             = _mm_sub_pd(iz2,jz0);
 369             dx30             = _mm_sub_pd(ix3,jx0);
 370             dy30             = _mm_sub_pd(iy3,jy0);
 371             dz30             = _mm_sub_pd(iz3,jz0);
 372
 373             /* Calculate squared distance and things based on it */
 374             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 375             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 376             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 377
 378             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 379             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 380             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 381
 382             /* Load parameters for j particles */
 383             jq0              = _mm_load_sd(charge+jnrA+0);
 384
 385             fjx0             = _mm_setzero_pd();
 386             fjy0             = _mm_setzero_pd();
 387             fjz0             = _mm_setzero_pd();
 388
 389             /**************************
 390              * CALCULATE INTERACTIONS *
 391              **************************/
 392
 393             r10              = _mm_mul_pd(rsq10,rinv10);
 394
 395             /* Compute parameters for interactions between i and j atoms */
 396             qq10             = _mm_mul_pd(iq1,jq0);
 397
 398             /* Calculate table index by multiplying r with table scale and truncate to integer */
 399             rt               = _mm_mul_pd(r10,vftabscale);
 400             vfitab           = _mm_cvttpd_epi32(rt);
 401 #ifdef __XOP__
 402             vfeps            = _mm_frcz_pd(rt);
 403 #else
 404             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 405 #endif
 406             twovfeps         = _mm_add_pd(vfeps,vfeps);
 407             vfitab           = _mm_slli_epi32(vfitab,2);
 408
 409             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 410             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 411             F                = _mm_setzero_pd();
 412             GMX_MM_TRANSPOSE2_PD(Y,F);
 413             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 414             H                = _mm_setzero_pd();
 415             GMX_MM_TRANSPOSE2_PD(G,H);
 416             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 417             VV               = _mm_macc_pd(vfeps,Fp,Y);
 418             velec            = _mm_mul_pd(qq10,VV);
 419             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 420             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 424             velecsum         = _mm_add_pd(velecsum,velec);
 425
 426             fscal            = felec;
 427
 428             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 429
 430             /* Update vectorial force */
 431             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 432             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 433             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 434
 435             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 436             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 437             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             r20              = _mm_mul_pd(rsq20,rinv20);
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq20             = _mm_mul_pd(iq2,jq0);
 447
 448             /* Calculate table index by multiplying r with table scale and truncate to integer */
 449             rt               = _mm_mul_pd(r20,vftabscale);
 450             vfitab           = _mm_cvttpd_epi32(rt);
 451 #ifdef __XOP__
 452             vfeps            = _mm_frcz_pd(rt);
 453 #else
 454             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 455 #endif
 456             twovfeps         = _mm_add_pd(vfeps,vfeps);
 457             vfitab           = _mm_slli_epi32(vfitab,2);
 458
 459             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 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(vfeps,H,G),F);
 467             VV               = _mm_macc_pd(vfeps,Fp,Y);
 468             velec            = _mm_mul_pd(qq20,VV);
 469             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 470             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 471
 472             /* Update potential sum for this i atom from the interaction with this j atom. */
 473             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 474             velecsum         = _mm_add_pd(velecsum,velec);
 475
 476             fscal            = felec;
 477
 478             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 479
 480             /* Update vectorial force */
 481             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 482             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 483             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 484
 485             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 486             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 487             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             r30              = _mm_mul_pd(rsq30,rinv30);
 494
 495             /* Compute parameters for interactions between i and j atoms */
 496             qq30             = _mm_mul_pd(iq3,jq0);
 497
 498             /* Calculate table index by multiplying r with table scale and truncate to integer */
 499             rt               = _mm_mul_pd(r30,vftabscale);
 500             vfitab           = _mm_cvttpd_epi32(rt);
 501 #ifdef __XOP__
 502             vfeps            = _mm_frcz_pd(rt);
 503 #else
 504             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 505 #endif
 506             twovfeps         = _mm_add_pd(vfeps,vfeps);
 507             vfitab           = _mm_slli_epi32(vfitab,2);
 508
 509             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 510             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 511             F                = _mm_setzero_pd();
 512             GMX_MM_TRANSPOSE2_PD(Y,F);
 513             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 514             H                = _mm_setzero_pd();
 515             GMX_MM_TRANSPOSE2_PD(G,H);
 516             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 517             VV               = _mm_macc_pd(vfeps,Fp,Y);
 518             velec            = _mm_mul_pd(qq30,VV);
 519             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 520             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 524             velecsum         = _mm_add_pd(velecsum,velec);
 525
 526             fscal            = felec;
 527
 528             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 529
 530             /* Update vectorial force */
 531             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 532             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 533             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 534
 535             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 536             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 537             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 538
 539             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 540
 541             /* Inner loop uses 141 flops */
 542         }
 543
 544         /* End of innermost loop */
 545
 546         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 547                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 548
 549         ggid                        = gid[iidx];
 550         /* Update potential energies */
 551         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 552
 553         /* Increment number of inner iterations */
 554         inneriter                  += j_index_end - j_index_start;
 555
 556         /* Outer loop uses 19 flops */
 557     }
 558
 559     /* Increment number of outer iterations */
 560     outeriter        += nri;
 561
 562     /* Update outer/inner flops */
 563
 564     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*141);
 565 }
 566 /*
 567  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_128_fma_double
 568  * Electrostatics interaction: CubicSplineTable
 569  * VdW interaction:            None
 570  * Geometry:                   Water4-Particle
 571  * Calculate force/pot:        Force
 572  */
 573 void
 574 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_128_fma_double
 575                     (t_nblist                    * gmx_restrict       nlist,
 576                      rvec                        * gmx_restrict          xx,
 577                      rvec                        * gmx_restrict          ff,
 578                      t_forcerec                  * gmx_restrict          fr,
 579                      t_mdatoms                   * gmx_restrict     mdatoms,
 580                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 581                      t_nrnb                      * gmx_restrict        nrnb)
 582 {
 583     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 584      * just 0 for non-waters.
 585      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 586      * jnr indices corresponding to data put in the four positions in the SIMD register.
 587      */
 588     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 589     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 590     int              jnrA,jnrB;
 591     int              j_coord_offsetA,j_coord_offsetB;
 592     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 593     real             rcutoff_scalar;
 594     real             *shiftvec,*fshift,*x,*f;
 595     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 596     int              vdwioffset1;
 597     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 598     int              vdwioffset2;
 599     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 600     int              vdwioffset3;
 601     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 602     int              vdwjidx0A,vdwjidx0B;
 603     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 604     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 605     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 606     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 607     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 608     real             *charge;
 609     __m128i          vfitab;
 610     __m128i          ifour       = _mm_set1_epi32(4);
 611     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 612     real             *vftab;
 613     __m128d          dummy_mask,cutoff_mask;
 614     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 615     __m128d          one     = _mm_set1_pd(1.0);
 616     __m128d          two     = _mm_set1_pd(2.0);
 617     x                = xx[0];
 618     f                = ff[0];
 619
 620     nri              = nlist->nri;
 621     iinr             = nlist->iinr;
 622     jindex           = nlist->jindex;
 623     jjnr             = nlist->jjnr;
 624     shiftidx         = nlist->shift;
 625     gid              = nlist->gid;
 626     shiftvec         = fr->shift_vec[0];
 627     fshift           = fr->fshift[0];
 628     facel            = _mm_set1_pd(fr->epsfac);
 629     charge           = mdatoms->chargeA;
 630
 631     vftab            = kernel_data->table_elec->data;
 632     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 633
 634     /* Setup water-specific parameters */
 635     inr              = nlist->iinr[0];
 636     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 637     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 638     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 639
 640     /* Avoid stupid compiler warnings */
 641     jnrA = jnrB = 0;
 642     j_coord_offsetA = 0;
 643     j_coord_offsetB = 0;
 644
 645     outeriter        = 0;
 646     inneriter        = 0;
 647
 648     /* Start outer loop over neighborlists */
 649     for(iidx=0; iidx<nri; iidx++)
 650     {
 651         /* Load shift vector for this list */
 652         i_shift_offset   = DIM*shiftidx[iidx];
 653
 654         /* Load limits for loop over neighbors */
 655         j_index_start    = jindex[iidx];
 656         j_index_end      = jindex[iidx+1];
 657
 658         /* Get outer coordinate index */
 659         inr              = iinr[iidx];
 660         i_coord_offset   = DIM*inr;
 661
 662         /* Load i particle coords and add shift vector */
 663         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 664                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 665
 666         fix1             = _mm_setzero_pd();
 667         fiy1             = _mm_setzero_pd();
 668         fiz1             = _mm_setzero_pd();
 669         fix2             = _mm_setzero_pd();
 670         fiy2             = _mm_setzero_pd();
 671         fiz2             = _mm_setzero_pd();
 672         fix3             = _mm_setzero_pd();
 673         fiy3             = _mm_setzero_pd();
 674         fiz3             = _mm_setzero_pd();
 675
 676         /* Start inner kernel loop */
 677         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 678         {
 679
 680             /* Get j neighbor index, and coordinate index */
 681             jnrA             = jjnr[jidx];
 682             jnrB             = jjnr[jidx+1];
 683             j_coord_offsetA  = DIM*jnrA;
 684             j_coord_offsetB  = DIM*jnrB;
 685
 686             /* load j atom coordinates */
 687             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 688                                               &jx0,&jy0,&jz0);
 689
 690             /* Calculate displacement vector */
 691             dx10             = _mm_sub_pd(ix1,jx0);
 692             dy10             = _mm_sub_pd(iy1,jy0);
 693             dz10             = _mm_sub_pd(iz1,jz0);
 694             dx20             = _mm_sub_pd(ix2,jx0);
 695             dy20             = _mm_sub_pd(iy2,jy0);
 696             dz20             = _mm_sub_pd(iz2,jz0);
 697             dx30             = _mm_sub_pd(ix3,jx0);
 698             dy30             = _mm_sub_pd(iy3,jy0);
 699             dz30             = _mm_sub_pd(iz3,jz0);
 700
 701             /* Calculate squared distance and things based on it */
 702             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 703             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 704             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 705
 706             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 707             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 708             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 709
 710             /* Load parameters for j particles */
 711             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 712
 713             fjx0             = _mm_setzero_pd();
 714             fjy0             = _mm_setzero_pd();
 715             fjz0             = _mm_setzero_pd();
 716
 717             /**************************
 718              * CALCULATE INTERACTIONS *
 719              **************************/
 720
 721             r10              = _mm_mul_pd(rsq10,rinv10);
 722
 723             /* Compute parameters for interactions between i and j atoms */
 724             qq10             = _mm_mul_pd(iq1,jq0);
 725
 726             /* Calculate table index by multiplying r with table scale and truncate to integer */
 727             rt               = _mm_mul_pd(r10,vftabscale);
 728             vfitab           = _mm_cvttpd_epi32(rt);
 729 #ifdef __XOP__
 730             vfeps            = _mm_frcz_pd(rt);
 731 #else
 732             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 733 #endif
 734             twovfeps         = _mm_add_pd(vfeps,vfeps);
 735             vfitab           = _mm_slli_epi32(vfitab,2);
 736
 737             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 738             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 739             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 740             GMX_MM_TRANSPOSE2_PD(Y,F);
 741             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 742             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 743             GMX_MM_TRANSPOSE2_PD(G,H);
 744             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 745             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 746             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 747
 748             fscal            = felec;
 749
 750             /* Update vectorial force */
 751             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 752             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 753             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 754
 755             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 756             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 757             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 758
 759             /**************************
 760              * CALCULATE INTERACTIONS *
 761              **************************/
 762
 763             r20              = _mm_mul_pd(rsq20,rinv20);
 764
 765             /* Compute parameters for interactions between i and j atoms */
 766             qq20             = _mm_mul_pd(iq2,jq0);
 767
 768             /* Calculate table index by multiplying r with table scale and truncate to integer */
 769             rt               = _mm_mul_pd(r20,vftabscale);
 770             vfitab           = _mm_cvttpd_epi32(rt);
 771 #ifdef __XOP__
 772             vfeps            = _mm_frcz_pd(rt);
 773 #else
 774             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 775 #endif
 776             twovfeps         = _mm_add_pd(vfeps,vfeps);
 777             vfitab           = _mm_slli_epi32(vfitab,2);
 778
 779             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 780             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 781             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 782             GMX_MM_TRANSPOSE2_PD(Y,F);
 783             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 784             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 785             GMX_MM_TRANSPOSE2_PD(G,H);
 786             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 787             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 788             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 789
 790             fscal            = felec;
 791
 792             /* Update vectorial force */
 793             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 794             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 795             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 796
 797             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 798             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 799             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 800
 801             /**************************
 802              * CALCULATE INTERACTIONS *
 803              **************************/
 804
 805             r30              = _mm_mul_pd(rsq30,rinv30);
 806
 807             /* Compute parameters for interactions between i and j atoms */
 808             qq30             = _mm_mul_pd(iq3,jq0);
 809
 810             /* Calculate table index by multiplying r with table scale and truncate to integer */
 811             rt               = _mm_mul_pd(r30,vftabscale);
 812             vfitab           = _mm_cvttpd_epi32(rt);
 813 #ifdef __XOP__
 814             vfeps            = _mm_frcz_pd(rt);
 815 #else
 816             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 817 #endif
 818             twovfeps         = _mm_add_pd(vfeps,vfeps);
 819             vfitab           = _mm_slli_epi32(vfitab,2);
 820
 821             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 822             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 823             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 824             GMX_MM_TRANSPOSE2_PD(Y,F);
 825             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 826             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 827             GMX_MM_TRANSPOSE2_PD(G,H);
 828             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 829             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 830             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 831
 832             fscal            = felec;
 833
 834             /* Update vectorial force */
 835             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 836             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 837             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 838
 839             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 840             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 841             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 842
 843             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 844
 845             /* Inner loop uses 129 flops */
 846         }
 847
 848         if(jidx<j_index_end)
 849         {
 850
 851             jnrA             = jjnr[jidx];
 852             j_coord_offsetA  = DIM*jnrA;
 853
 854             /* load j atom coordinates */
 855             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 856                                               &jx0,&jy0,&jz0);
 857
 858             /* Calculate displacement vector */
 859             dx10             = _mm_sub_pd(ix1,jx0);
 860             dy10             = _mm_sub_pd(iy1,jy0);
 861             dz10             = _mm_sub_pd(iz1,jz0);
 862             dx20             = _mm_sub_pd(ix2,jx0);
 863             dy20             = _mm_sub_pd(iy2,jy0);
 864             dz20             = _mm_sub_pd(iz2,jz0);
 865             dx30             = _mm_sub_pd(ix3,jx0);
 866             dy30             = _mm_sub_pd(iy3,jy0);
 867             dz30             = _mm_sub_pd(iz3,jz0);
 868
 869             /* Calculate squared distance and things based on it */
 870             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 871             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 872             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 873
 874             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 875             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 876             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 877
 878             /* Load parameters for j particles */
 879             jq0              = _mm_load_sd(charge+jnrA+0);
 880
 881             fjx0             = _mm_setzero_pd();
 882             fjy0             = _mm_setzero_pd();
 883             fjz0             = _mm_setzero_pd();
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             r10              = _mm_mul_pd(rsq10,rinv10);
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq10             = _mm_mul_pd(iq1,jq0);
 893
 894             /* Calculate table index by multiplying r with table scale and truncate to integer */
 895             rt               = _mm_mul_pd(r10,vftabscale);
 896             vfitab           = _mm_cvttpd_epi32(rt);
 897 #ifdef __XOP__
 898             vfeps            = _mm_frcz_pd(rt);
 899 #else
 900             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 901 #endif
 902             twovfeps         = _mm_add_pd(vfeps,vfeps);
 903             vfitab           = _mm_slli_epi32(vfitab,2);
 904
 905             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 906             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 907             F                = _mm_setzero_pd();
 908             GMX_MM_TRANSPOSE2_PD(Y,F);
 909             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 910             H                = _mm_setzero_pd();
 911             GMX_MM_TRANSPOSE2_PD(G,H);
 912             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 913             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 914             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 915
 916             fscal            = felec;
 917
 918             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 919
 920             /* Update vectorial force */
 921             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 922             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 923             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 924
 925             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 926             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 927             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 928
 929             /**************************
 930              * CALCULATE INTERACTIONS *
 931              **************************/
 932
 933             r20              = _mm_mul_pd(rsq20,rinv20);
 934
 935             /* Compute parameters for interactions between i and j atoms */
 936             qq20             = _mm_mul_pd(iq2,jq0);
 937
 938             /* Calculate table index by multiplying r with table scale and truncate to integer */
 939             rt               = _mm_mul_pd(r20,vftabscale);
 940             vfitab           = _mm_cvttpd_epi32(rt);
 941 #ifdef __XOP__
 942             vfeps            = _mm_frcz_pd(rt);
 943 #else
 944             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 945 #endif
 946             twovfeps         = _mm_add_pd(vfeps,vfeps);
 947             vfitab           = _mm_slli_epi32(vfitab,2);
 948
 949             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 950             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 951             F                = _mm_setzero_pd();
 952             GMX_MM_TRANSPOSE2_PD(Y,F);
 953             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 954             H                = _mm_setzero_pd();
 955             GMX_MM_TRANSPOSE2_PD(G,H);
 956             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 957             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 958             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 959
 960             fscal            = felec;
 961
 962             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 963
 964             /* Update vectorial force */
 965             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 966             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 967             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 968
 969             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 970             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 971             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 972
 973             /**************************
 974              * CALCULATE INTERACTIONS *
 975              **************************/
 976
 977             r30              = _mm_mul_pd(rsq30,rinv30);
 978
 979             /* Compute parameters for interactions between i and j atoms */
 980             qq30             = _mm_mul_pd(iq3,jq0);
 981
 982             /* Calculate table index by multiplying r with table scale and truncate to integer */
 983             rt               = _mm_mul_pd(r30,vftabscale);
 984             vfitab           = _mm_cvttpd_epi32(rt);
 985 #ifdef __XOP__
 986             vfeps            = _mm_frcz_pd(rt);
 987 #else
 988             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 989 #endif
 990             twovfeps         = _mm_add_pd(vfeps,vfeps);
 991             vfitab           = _mm_slli_epi32(vfitab,2);
 992
 993             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 994             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 995             F                = _mm_setzero_pd();
 996             GMX_MM_TRANSPOSE2_PD(Y,F);
 997             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 998             H                = _mm_setzero_pd();
 999             GMX_MM_TRANSPOSE2_PD(G,H);
1000             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1001             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1002             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1003
1004             fscal            = felec;
1005
1006             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1007
1008             /* Update vectorial force */
1009             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1010             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1011             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1012
1013             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1014             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1015             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1016
1017             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1018
1019             /* Inner loop uses 129 flops */
1020         }
1021
1022         /* End of innermost loop */
1023
1024         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1025                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1026
1027         /* Increment number of inner iterations */
1028         inneriter                  += j_index_end - j_index_start;
1029
1030         /* Outer loop uses 18 flops */
1031     }
1032
1033     /* Increment number of outer iterations */
1034     outeriter        += nri;
1035
1036     /* Update outer/inner flops */
1037
1038     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*129);
1039 }