src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 104     real             *vftab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_elec->data;
 127     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 132     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 133     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 161
 162         fix0             = _mm_setzero_pd();
 163         fiy0             = _mm_setzero_pd();
 164         fiz0             = _mm_setzero_pd();
 165         fix1             = _mm_setzero_pd();
 166         fiy1             = _mm_setzero_pd();
 167         fiz1             = _mm_setzero_pd();
 168         fix2             = _mm_setzero_pd();
 169         fiy2             = _mm_setzero_pd();
 170         fiz2             = _mm_setzero_pd();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_pd();
 174         vvdwsum          = _mm_setzero_pd();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185
 186             /* load j atom coordinates */
 187             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm_sub_pd(ix0,jx0);
 192             dy00             = _mm_sub_pd(iy0,jy0);
 193             dz00             = _mm_sub_pd(iz0,jz0);
 194             dx10             = _mm_sub_pd(ix1,jx0);
 195             dy10             = _mm_sub_pd(iy1,jy0);
 196             dz10             = _mm_sub_pd(iz1,jz0);
 197             dx20             = _mm_sub_pd(ix2,jx0);
 198             dy20             = _mm_sub_pd(iy2,jy0);
 199             dz20             = _mm_sub_pd(iz2,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 203             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 204             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 205
 206             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 207             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 208             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 209
 210             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216
 217             fjx0             = _mm_setzero_pd();
 218             fjy0             = _mm_setzero_pd();
 219             fjz0             = _mm_setzero_pd();
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             r00              = _mm_mul_pd(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             qq00             = _mm_mul_pd(iq0,jq0);
 229             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 230                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 231
 232             /* Calculate table index by multiplying r with table scale and truncate to integer */
 233             rt               = _mm_mul_pd(r00,vftabscale);
 234             vfitab           = _mm_cvttpd_epi32(rt);
 235 #ifdef __XOP__
 236             vfeps            = _mm_frcz_pd(rt);
 237 #else
 238             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 239 #endif
 240             twovfeps         = _mm_add_pd(vfeps,vfeps);
 241             vfitab           = _mm_slli_epi32(vfitab,2);
 242
 243             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 244             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 245             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 246             GMX_MM_TRANSPOSE2_PD(Y,F);
 247             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 248             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 249             GMX_MM_TRANSPOSE2_PD(G,H);
 250             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 251             VV               = _mm_macc_pd(vfeps,Fp,Y);
 252             velec            = _mm_mul_pd(qq00,VV);
 253             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 254             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 255
 256             /* LENNARD-JONES DISPERSION/REPULSION */
 257
 258             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 259             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 260             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 261             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 262             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             velecsum         = _mm_add_pd(velecsum,velec);
 266             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 267
 268             fscal            = _mm_add_pd(felec,fvdw);
 269
 270             /* Update vectorial force */
 271             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 272             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 273             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 274
 275             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 276             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 277             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             r10              = _mm_mul_pd(rsq10,rinv10);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq10             = _mm_mul_pd(iq1,jq0);
 287
 288             /* Calculate table index by multiplying r with table scale and truncate to integer */
 289             rt               = _mm_mul_pd(r10,vftabscale);
 290             vfitab           = _mm_cvttpd_epi32(rt);
 291 #ifdef __XOP__
 292             vfeps            = _mm_frcz_pd(rt);
 293 #else
 294             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 295 #endif
 296             twovfeps         = _mm_add_pd(vfeps,vfeps);
 297             vfitab           = _mm_slli_epi32(vfitab,2);
 298
 299             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 300             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 301             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 302             GMX_MM_TRANSPOSE2_PD(Y,F);
 303             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 304             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 305             GMX_MM_TRANSPOSE2_PD(G,H);
 306             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 307             VV               = _mm_macc_pd(vfeps,Fp,Y);
 308             velec            = _mm_mul_pd(qq10,VV);
 309             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 310             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm_add_pd(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317             /* Update vectorial force */
 318             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 319             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 320             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 321
 322             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 323             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 324             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             r20              = _mm_mul_pd(rsq20,rinv20);
 331
 332             /* Compute parameters for interactions between i and j atoms */
 333             qq20             = _mm_mul_pd(iq2,jq0);
 334
 335             /* Calculate table index by multiplying r with table scale and truncate to integer */
 336             rt               = _mm_mul_pd(r20,vftabscale);
 337             vfitab           = _mm_cvttpd_epi32(rt);
 338 #ifdef __XOP__
 339             vfeps            = _mm_frcz_pd(rt);
 340 #else
 341             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 342 #endif
 343             twovfeps         = _mm_add_pd(vfeps,vfeps);
 344             vfitab           = _mm_slli_epi32(vfitab,2);
 345
 346             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 347             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 348             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 349             GMX_MM_TRANSPOSE2_PD(Y,F);
 350             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 351             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 352             GMX_MM_TRANSPOSE2_PD(G,H);
 353             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 354             VV               = _mm_macc_pd(vfeps,Fp,Y);
 355             velec            = _mm_mul_pd(qq20,VV);
 356             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 357             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velecsum         = _mm_add_pd(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             /* Update vectorial force */
 365             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 366             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 367             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 368
 369             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 370             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 371             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 372
 373             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 374
 375             /* Inner loop uses 154 flops */
 376         }
 377
 378         if(jidx<j_index_end)
 379         {
 380
 381             jnrA             = jjnr[jidx];
 382             j_coord_offsetA  = DIM*jnrA;
 383
 384             /* load j atom coordinates */
 385             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 386                                               &jx0,&jy0,&jz0);
 387
 388             /* Calculate displacement vector */
 389             dx00             = _mm_sub_pd(ix0,jx0);
 390             dy00             = _mm_sub_pd(iy0,jy0);
 391             dz00             = _mm_sub_pd(iz0,jz0);
 392             dx10             = _mm_sub_pd(ix1,jx0);
 393             dy10             = _mm_sub_pd(iy1,jy0);
 394             dz10             = _mm_sub_pd(iz1,jz0);
 395             dx20             = _mm_sub_pd(ix2,jx0);
 396             dy20             = _mm_sub_pd(iy2,jy0);
 397             dz20             = _mm_sub_pd(iz2,jz0);
 398
 399             /* Calculate squared distance and things based on it */
 400             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 401             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 402             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 403
 404             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 405             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 406             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 407
 408             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 409
 410             /* Load parameters for j particles */
 411             jq0              = _mm_load_sd(charge+jnrA+0);
 412             vdwjidx0A        = 2*vdwtype[jnrA+0];
 413
 414             fjx0             = _mm_setzero_pd();
 415             fjy0             = _mm_setzero_pd();
 416             fjz0             = _mm_setzero_pd();
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             r00              = _mm_mul_pd(rsq00,rinv00);
 423
 424             /* Compute parameters for interactions between i and j atoms */
 425             qq00             = _mm_mul_pd(iq0,jq0);
 426             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 427
 428             /* Calculate table index by multiplying r with table scale and truncate to integer */
 429             rt               = _mm_mul_pd(r00,vftabscale);
 430             vfitab           = _mm_cvttpd_epi32(rt);
 431 #ifdef __XOP__
 432             vfeps            = _mm_frcz_pd(rt);
 433 #else
 434             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 435 #endif
 436             twovfeps         = _mm_add_pd(vfeps,vfeps);
 437             vfitab           = _mm_slli_epi32(vfitab,2);
 438
 439             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 440             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 441             F                = _mm_setzero_pd();
 442             GMX_MM_TRANSPOSE2_PD(Y,F);
 443             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 444             H                = _mm_setzero_pd();
 445             GMX_MM_TRANSPOSE2_PD(G,H);
 446             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 447             VV               = _mm_macc_pd(vfeps,Fp,Y);
 448             velec            = _mm_mul_pd(qq00,VV);
 449             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 450             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 451
 452             /* LENNARD-JONES DISPERSION/REPULSION */
 453
 454             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 455             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 456             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 457             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 458             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 459
 460             /* Update potential sum for this i atom from the interaction with this j atom. */
 461             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 462             velecsum         = _mm_add_pd(velecsum,velec);
 463             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 464             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 465
 466             fscal            = _mm_add_pd(felec,fvdw);
 467
 468             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 469
 470             /* Update vectorial force */
 471             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 472             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 473             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 474
 475             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 476             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 477             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 478
 479             /**************************
 480              * CALCULATE INTERACTIONS *
 481              **************************/
 482
 483             r10              = _mm_mul_pd(rsq10,rinv10);
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             qq10             = _mm_mul_pd(iq1,jq0);
 487
 488             /* Calculate table index by multiplying r with table scale and truncate to integer */
 489             rt               = _mm_mul_pd(r10,vftabscale);
 490             vfitab           = _mm_cvttpd_epi32(rt);
 491 #ifdef __XOP__
 492             vfeps            = _mm_frcz_pd(rt);
 493 #else
 494             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 495 #endif
 496             twovfeps         = _mm_add_pd(vfeps,vfeps);
 497             vfitab           = _mm_slli_epi32(vfitab,2);
 498
 499             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 500             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 501             F                = _mm_setzero_pd();
 502             GMX_MM_TRANSPOSE2_PD(Y,F);
 503             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 504             H                = _mm_setzero_pd();
 505             GMX_MM_TRANSPOSE2_PD(G,H);
 506             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 507             VV               = _mm_macc_pd(vfeps,Fp,Y);
 508             velec            = _mm_mul_pd(qq10,VV);
 509             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 510             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 511
 512             /* Update potential sum for this i atom from the interaction with this j atom. */
 513             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 514             velecsum         = _mm_add_pd(velecsum,velec);
 515
 516             fscal            = felec;
 517
 518             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 519
 520             /* Update vectorial force */
 521             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 522             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 523             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 524
 525             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 526             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 527             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             r20              = _mm_mul_pd(rsq20,rinv20);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq20             = _mm_mul_pd(iq2,jq0);
 537
 538             /* Calculate table index by multiplying r with table scale and truncate to integer */
 539             rt               = _mm_mul_pd(r20,vftabscale);
 540             vfitab           = _mm_cvttpd_epi32(rt);
 541 #ifdef __XOP__
 542             vfeps            = _mm_frcz_pd(rt);
 543 #else
 544             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 545 #endif
 546             twovfeps         = _mm_add_pd(vfeps,vfeps);
 547             vfitab           = _mm_slli_epi32(vfitab,2);
 548
 549             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 550             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 551             F                = _mm_setzero_pd();
 552             GMX_MM_TRANSPOSE2_PD(Y,F);
 553             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 554             H                = _mm_setzero_pd();
 555             GMX_MM_TRANSPOSE2_PD(G,H);
 556             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 557             VV               = _mm_macc_pd(vfeps,Fp,Y);
 558             velec            = _mm_mul_pd(qq20,VV);
 559             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 560             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 564             velecsum         = _mm_add_pd(velecsum,velec);
 565
 566             fscal            = felec;
 567
 568             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 569
 570             /* Update vectorial force */
 571             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 572             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 573             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 574
 575             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 576             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 577             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 578
 579             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 580
 581             /* Inner loop uses 154 flops */
 582         }
 583
 584         /* End of innermost loop */
 585
 586         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 587                                               f+i_coord_offset,fshift+i_shift_offset);
 588
 589         ggid                        = gid[iidx];
 590         /* Update potential energies */
 591         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 592         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 593
 594         /* Increment number of inner iterations */
 595         inneriter                  += j_index_end - j_index_start;
 596
 597         /* Outer loop uses 20 flops */
 598     }
 599
 600     /* Increment number of outer iterations */
 601     outeriter        += nri;
 602
 603     /* Update outer/inner flops */
 604
 605     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 606 }
 607 /*
 608  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_double
 609  * Electrostatics interaction: CubicSplineTable
 610  * VdW interaction:            LennardJones
 611  * Geometry:                   Water3-Particle
 612  * Calculate force/pot:        Force
 613  */
 614 void
 615 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_double
 616                     (t_nblist                    * gmx_restrict       nlist,
 617                      rvec                        * gmx_restrict          xx,
 618                      rvec                        * gmx_restrict          ff,
 619                      t_forcerec                  * gmx_restrict          fr,
 620                      t_mdatoms                   * gmx_restrict     mdatoms,
 621                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 622                      t_nrnb                      * gmx_restrict        nrnb)
 623 {
 624     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 625      * just 0 for non-waters.
 626      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 627      * jnr indices corresponding to data put in the four positions in the SIMD register.
 628      */
 629     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 630     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 631     int              jnrA,jnrB;
 632     int              j_coord_offsetA,j_coord_offsetB;
 633     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 634     real             rcutoff_scalar;
 635     real             *shiftvec,*fshift,*x,*f;
 636     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 637     int              vdwioffset0;
 638     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 639     int              vdwioffset1;
 640     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 641     int              vdwioffset2;
 642     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 643     int              vdwjidx0A,vdwjidx0B;
 644     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 645     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 646     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 647     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 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     nvdwtype         = fr->ntype;
 678     vdwparam         = fr->nbfp;
 679     vdwtype          = mdatoms->typeA;
 680
 681     vftab            = kernel_data->table_elec->data;
 682     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 683
 684     /* Setup water-specific parameters */
 685     inr              = nlist->iinr[0];
 686     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 687     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 688     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 689     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 690
 691     /* Avoid stupid compiler warnings */
 692     jnrA = jnrB = 0;
 693     j_coord_offsetA = 0;
 694     j_coord_offsetB = 0;
 695
 696     outeriter        = 0;
 697     inneriter        = 0;
 698
 699     /* Start outer loop over neighborlists */
 700     for(iidx=0; iidx<nri; iidx++)
 701     {
 702         /* Load shift vector for this list */
 703         i_shift_offset   = DIM*shiftidx[iidx];
 704
 705         /* Load limits for loop over neighbors */
 706         j_index_start    = jindex[iidx];
 707         j_index_end      = jindex[iidx+1];
 708
 709         /* Get outer coordinate index */
 710         inr              = iinr[iidx];
 711         i_coord_offset   = DIM*inr;
 712
 713         /* Load i particle coords and add shift vector */
 714         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 715                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 716
 717         fix0             = _mm_setzero_pd();
 718         fiy0             = _mm_setzero_pd();
 719         fiz0             = _mm_setzero_pd();
 720         fix1             = _mm_setzero_pd();
 721         fiy1             = _mm_setzero_pd();
 722         fiz1             = _mm_setzero_pd();
 723         fix2             = _mm_setzero_pd();
 724         fiy2             = _mm_setzero_pd();
 725         fiz2             = _mm_setzero_pd();
 726
 727         /* Start inner kernel loop */
 728         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 729         {
 730
 731             /* Get j neighbor index, and coordinate index */
 732             jnrA             = jjnr[jidx];
 733             jnrB             = jjnr[jidx+1];
 734             j_coord_offsetA  = DIM*jnrA;
 735             j_coord_offsetB  = DIM*jnrB;
 736
 737             /* load j atom coordinates */
 738             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 739                                               &jx0,&jy0,&jz0);
 740
 741             /* Calculate displacement vector */
 742             dx00             = _mm_sub_pd(ix0,jx0);
 743             dy00             = _mm_sub_pd(iy0,jy0);
 744             dz00             = _mm_sub_pd(iz0,jz0);
 745             dx10             = _mm_sub_pd(ix1,jx0);
 746             dy10             = _mm_sub_pd(iy1,jy0);
 747             dz10             = _mm_sub_pd(iz1,jz0);
 748             dx20             = _mm_sub_pd(ix2,jx0);
 749             dy20             = _mm_sub_pd(iy2,jy0);
 750             dz20             = _mm_sub_pd(iz2,jz0);
 751
 752             /* Calculate squared distance and things based on it */
 753             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 754             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 755             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 756
 757             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 758             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 759             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 760
 761             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 762
 763             /* Load parameters for j particles */
 764             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 765             vdwjidx0A        = 2*vdwtype[jnrA+0];
 766             vdwjidx0B        = 2*vdwtype[jnrB+0];
 767
 768             fjx0             = _mm_setzero_pd();
 769             fjy0             = _mm_setzero_pd();
 770             fjz0             = _mm_setzero_pd();
 771
 772             /**************************
 773              * CALCULATE INTERACTIONS *
 774              **************************/
 775
 776             r00              = _mm_mul_pd(rsq00,rinv00);
 777
 778             /* Compute parameters for interactions between i and j atoms */
 779             qq00             = _mm_mul_pd(iq0,jq0);
 780             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 781                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 782
 783             /* Calculate table index by multiplying r with table scale and truncate to integer */
 784             rt               = _mm_mul_pd(r00,vftabscale);
 785             vfitab           = _mm_cvttpd_epi32(rt);
 786 #ifdef __XOP__
 787             vfeps            = _mm_frcz_pd(rt);
 788 #else
 789             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 790 #endif
 791             twovfeps         = _mm_add_pd(vfeps,vfeps);
 792             vfitab           = _mm_slli_epi32(vfitab,2);
 793
 794             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 795             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 796             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 797             GMX_MM_TRANSPOSE2_PD(Y,F);
 798             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 799             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 800             GMX_MM_TRANSPOSE2_PD(G,H);
 801             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 802             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 803             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 804
 805             /* LENNARD-JONES DISPERSION/REPULSION */
 806
 807             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 808             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 809
 810             fscal            = _mm_add_pd(felec,fvdw);
 811
 812             /* Update vectorial force */
 813             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 814             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 815             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 816
 817             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 818             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 819             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 820
 821             /**************************
 822              * CALCULATE INTERACTIONS *
 823              **************************/
 824
 825             r10              = _mm_mul_pd(rsq10,rinv10);
 826
 827             /* Compute parameters for interactions between i and j atoms */
 828             qq10             = _mm_mul_pd(iq1,jq0);
 829
 830             /* Calculate table index by multiplying r with table scale and truncate to integer */
 831             rt               = _mm_mul_pd(r10,vftabscale);
 832             vfitab           = _mm_cvttpd_epi32(rt);
 833 #ifdef __XOP__
 834             vfeps            = _mm_frcz_pd(rt);
 835 #else
 836             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 837 #endif
 838             twovfeps         = _mm_add_pd(vfeps,vfeps);
 839             vfitab           = _mm_slli_epi32(vfitab,2);
 840
 841             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 842             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 843             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 844             GMX_MM_TRANSPOSE2_PD(Y,F);
 845             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 846             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 847             GMX_MM_TRANSPOSE2_PD(G,H);
 848             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 849             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 850             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 851
 852             fscal            = felec;
 853
 854             /* Update vectorial force */
 855             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 856             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 857             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 858
 859             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 860             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 861             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 862
 863             /**************************
 864              * CALCULATE INTERACTIONS *
 865              **************************/
 866
 867             r20              = _mm_mul_pd(rsq20,rinv20);
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq20             = _mm_mul_pd(iq2,jq0);
 871
 872             /* Calculate table index by multiplying r with table scale and truncate to integer */
 873             rt               = _mm_mul_pd(r20,vftabscale);
 874             vfitab           = _mm_cvttpd_epi32(rt);
 875 #ifdef __XOP__
 876             vfeps            = _mm_frcz_pd(rt);
 877 #else
 878             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 879 #endif
 880             twovfeps         = _mm_add_pd(vfeps,vfeps);
 881             vfitab           = _mm_slli_epi32(vfitab,2);
 882
 883             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 884             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 885             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 886             GMX_MM_TRANSPOSE2_PD(Y,F);
 887             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 888             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 889             GMX_MM_TRANSPOSE2_PD(G,H);
 890             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 891             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 892             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 893
 894             fscal            = felec;
 895
 896             /* Update vectorial force */
 897             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 898             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 899             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 900
 901             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 902             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 903             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 904
 905             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 906
 907             /* Inner loop uses 137 flops */
 908         }
 909
 910         if(jidx<j_index_end)
 911         {
 912
 913             jnrA             = jjnr[jidx];
 914             j_coord_offsetA  = DIM*jnrA;
 915
 916             /* load j atom coordinates */
 917             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 918                                               &jx0,&jy0,&jz0);
 919
 920             /* Calculate displacement vector */
 921             dx00             = _mm_sub_pd(ix0,jx0);
 922             dy00             = _mm_sub_pd(iy0,jy0);
 923             dz00             = _mm_sub_pd(iz0,jz0);
 924             dx10             = _mm_sub_pd(ix1,jx0);
 925             dy10             = _mm_sub_pd(iy1,jy0);
 926             dz10             = _mm_sub_pd(iz1,jz0);
 927             dx20             = _mm_sub_pd(ix2,jx0);
 928             dy20             = _mm_sub_pd(iy2,jy0);
 929             dz20             = _mm_sub_pd(iz2,jz0);
 930
 931             /* Calculate squared distance and things based on it */
 932             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 933             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 934             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 935
 936             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 937             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 938             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 939
 940             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 941
 942             /* Load parameters for j particles */
 943             jq0              = _mm_load_sd(charge+jnrA+0);
 944             vdwjidx0A        = 2*vdwtype[jnrA+0];
 945
 946             fjx0             = _mm_setzero_pd();
 947             fjy0             = _mm_setzero_pd();
 948             fjz0             = _mm_setzero_pd();
 949
 950             /**************************
 951              * CALCULATE INTERACTIONS *
 952              **************************/
 953
 954             r00              = _mm_mul_pd(rsq00,rinv00);
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq00             = _mm_mul_pd(iq0,jq0);
 958             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 959
 960             /* Calculate table index by multiplying r with table scale and truncate to integer */
 961             rt               = _mm_mul_pd(r00,vftabscale);
 962             vfitab           = _mm_cvttpd_epi32(rt);
 963 #ifdef __XOP__
 964             vfeps            = _mm_frcz_pd(rt);
 965 #else
 966             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 967 #endif
 968             twovfeps         = _mm_add_pd(vfeps,vfeps);
 969             vfitab           = _mm_slli_epi32(vfitab,2);
 970
 971             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 972             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 973             F                = _mm_setzero_pd();
 974             GMX_MM_TRANSPOSE2_PD(Y,F);
 975             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 976             H                = _mm_setzero_pd();
 977             GMX_MM_TRANSPOSE2_PD(G,H);
 978             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 979             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 980             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 981
 982             /* LENNARD-JONES DISPERSION/REPULSION */
 983
 984             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 985             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 986
 987             fscal            = _mm_add_pd(felec,fvdw);
 988
 989             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 990
 991             /* Update vectorial force */
 992             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 993             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 994             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 995
 996             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 997             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 998             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             r10              = _mm_mul_pd(rsq10,rinv10);
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             qq10             = _mm_mul_pd(iq1,jq0);
1008
1009             /* Calculate table index by multiplying r with table scale and truncate to integer */
1010             rt               = _mm_mul_pd(r10,vftabscale);
1011             vfitab           = _mm_cvttpd_epi32(rt);
1012 #ifdef __XOP__
1013             vfeps            = _mm_frcz_pd(rt);
1014 #else
1015             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1016 #endif
1017             twovfeps         = _mm_add_pd(vfeps,vfeps);
1018             vfitab           = _mm_slli_epi32(vfitab,2);
1019
1020             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1021             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1022             F                = _mm_setzero_pd();
1023             GMX_MM_TRANSPOSE2_PD(Y,F);
1024             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1025             H                = _mm_setzero_pd();
1026             GMX_MM_TRANSPOSE2_PD(G,H);
1027             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1028             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1029             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1030
1031             fscal            = felec;
1032
1033             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1034
1035             /* Update vectorial force */
1036             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1037             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1038             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1039
1040             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1041             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1042             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1043
1044             /**************************
1045              * CALCULATE INTERACTIONS *
1046              **************************/
1047
1048             r20              = _mm_mul_pd(rsq20,rinv20);
1049
1050             /* Compute parameters for interactions between i and j atoms */
1051             qq20             = _mm_mul_pd(iq2,jq0);
1052
1053             /* Calculate table index by multiplying r with table scale and truncate to integer */
1054             rt               = _mm_mul_pd(r20,vftabscale);
1055             vfitab           = _mm_cvttpd_epi32(rt);
1056 #ifdef __XOP__
1057             vfeps            = _mm_frcz_pd(rt);
1058 #else
1059             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1060 #endif
1061             twovfeps         = _mm_add_pd(vfeps,vfeps);
1062             vfitab           = _mm_slli_epi32(vfitab,2);
1063
1064             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1065             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1066             F                = _mm_setzero_pd();
1067             GMX_MM_TRANSPOSE2_PD(Y,F);
1068             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1069             H                = _mm_setzero_pd();
1070             GMX_MM_TRANSPOSE2_PD(G,H);
1071             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1072             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1073             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1074
1075             fscal            = felec;
1076
1077             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1078
1079             /* Update vectorial force */
1080             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1081             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1082             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1083
1084             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1085             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1086             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1087
1088             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1089
1090             /* Inner loop uses 137 flops */
1091         }
1092
1093         /* End of innermost loop */
1094
1095         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1096                                               f+i_coord_offset,fshift+i_shift_offset);
1097
1098         /* Increment number of inner iterations */
1099         inneriter                  += j_index_end - j_index_start;
1100
1101         /* Outer loop uses 18 flops */
1102     }
1103
1104     /* Increment number of outer iterations */
1105     outeriter        += nri;
1106
1107     /* Update outer/inner flops */
1108
1109     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1110 }