src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwNone_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW3P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwNone_GeomW3P1_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m128i          vfitab;
  94     __m128i          ifour       = _mm_set1_epi32(4);
  95     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  96     real             *vftab;
  97     __m128d          dummy_mask,cutoff_mask;
  98     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  99     __m128d          one     = _mm_set1_pd(1.0);
 100     __m128d          two     = _mm_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_pd(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114
 115     vftab            = kernel_data->table_elec->data;
 116     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 117
 118     /* Setup water-specific parameters */
 119     inr              = nlist->iinr[0];
 120     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 121     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 122     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 148                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 149
 150         fix0             = _mm_setzero_pd();
 151         fiy0             = _mm_setzero_pd();
 152         fiz0             = _mm_setzero_pd();
 153         fix1             = _mm_setzero_pd();
 154         fiy1             = _mm_setzero_pd();
 155         fiz1             = _mm_setzero_pd();
 156         fix2             = _mm_setzero_pd();
 157         fiy2             = _mm_setzero_pd();
 158         fiz2             = _mm_setzero_pd();
 159
 160         /* Reset potential sums */
 161         velecsum         = _mm_setzero_pd();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx00             = _mm_sub_pd(ix0,jx0);
 179             dy00             = _mm_sub_pd(iy0,jy0);
 180             dz00             = _mm_sub_pd(iz0,jz0);
 181             dx10             = _mm_sub_pd(ix1,jx0);
 182             dy10             = _mm_sub_pd(iy1,jy0);
 183             dz10             = _mm_sub_pd(iz1,jz0);
 184             dx20             = _mm_sub_pd(ix2,jx0);
 185             dy20             = _mm_sub_pd(iy2,jy0);
 186             dz20             = _mm_sub_pd(iz2,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 190             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 191             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 192
 193             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 194             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 195             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 196
 197             /* Load parameters for j particles */
 198             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 199
 200             fjx0             = _mm_setzero_pd();
 201             fjy0             = _mm_setzero_pd();
 202             fjz0             = _mm_setzero_pd();
 203
 204             /**************************
 205              * CALCULATE INTERACTIONS *
 206              **************************/
 207
 208             r00              = _mm_mul_pd(rsq00,rinv00);
 209
 210             /* Compute parameters for interactions between i and j atoms */
 211             qq00             = _mm_mul_pd(iq0,jq0);
 212
 213             /* Calculate table index by multiplying r with table scale and truncate to integer */
 214             rt               = _mm_mul_pd(r00,vftabscale);
 215             vfitab           = _mm_cvttpd_epi32(rt);
 216 #ifdef __XOP__
 217             vfeps            = _mm_frcz_pd(rt);
 218 #else
 219             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 220 #endif
 221             twovfeps         = _mm_add_pd(vfeps,vfeps);
 222             vfitab           = _mm_slli_epi32(vfitab,2);
 223
 224             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 225             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 226             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 227             GMX_MM_TRANSPOSE2_PD(Y,F);
 228             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 229             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 230             GMX_MM_TRANSPOSE2_PD(G,H);
 231             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 232             VV               = _mm_macc_pd(vfeps,Fp,Y);
 233             velec            = _mm_mul_pd(qq00,VV);
 234             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 235             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 236
 237             /* Update potential sum for this i atom from the interaction with this j atom. */
 238             velecsum         = _mm_add_pd(velecsum,velec);
 239
 240             fscal            = felec;
 241
 242             /* Update vectorial force */
 243             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 244             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 245             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 246
 247             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 248             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 249             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             r10              = _mm_mul_pd(rsq10,rinv10);
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             qq10             = _mm_mul_pd(iq1,jq0);
 259
 260             /* Calculate table index by multiplying r with table scale and truncate to integer */
 261             rt               = _mm_mul_pd(r10,vftabscale);
 262             vfitab           = _mm_cvttpd_epi32(rt);
 263 #ifdef __XOP__
 264             vfeps            = _mm_frcz_pd(rt);
 265 #else
 266             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 267 #endif
 268             twovfeps         = _mm_add_pd(vfeps,vfeps);
 269             vfitab           = _mm_slli_epi32(vfitab,2);
 270
 271             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 272             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 273             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 274             GMX_MM_TRANSPOSE2_PD(Y,F);
 275             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 276             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 277             GMX_MM_TRANSPOSE2_PD(G,H);
 278             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 279             VV               = _mm_macc_pd(vfeps,Fp,Y);
 280             velec            = _mm_mul_pd(qq10,VV);
 281             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 282             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velecsum         = _mm_add_pd(velecsum,velec);
 286
 287             fscal            = felec;
 288
 289             /* Update vectorial force */
 290             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 291             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 292             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 293
 294             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 295             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 296             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             r20              = _mm_mul_pd(rsq20,rinv20);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq20             = _mm_mul_pd(iq2,jq0);
 306
 307             /* Calculate table index by multiplying r with table scale and truncate to integer */
 308             rt               = _mm_mul_pd(r20,vftabscale);
 309             vfitab           = _mm_cvttpd_epi32(rt);
 310 #ifdef __XOP__
 311             vfeps            = _mm_frcz_pd(rt);
 312 #else
 313             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 314 #endif
 315             twovfeps         = _mm_add_pd(vfeps,vfeps);
 316             vfitab           = _mm_slli_epi32(vfitab,2);
 317
 318             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 319             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 320             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 321             GMX_MM_TRANSPOSE2_PD(Y,F);
 322             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 323             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 324             GMX_MM_TRANSPOSE2_PD(G,H);
 325             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 326             VV               = _mm_macc_pd(vfeps,Fp,Y);
 327             velec            = _mm_mul_pd(qq20,VV);
 328             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 329             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 330
 331             /* Update potential sum for this i atom from the interaction with this j atom. */
 332             velecsum         = _mm_add_pd(velecsum,velec);
 333
 334             fscal            = felec;
 335
 336             /* Update vectorial force */
 337             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 338             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 339             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 340
 341             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 342             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 343             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 344
 345             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 346
 347             /* Inner loop uses 141 flops */
 348         }
 349
 350         if(jidx<j_index_end)
 351         {
 352
 353             jnrA             = jjnr[jidx];
 354             j_coord_offsetA  = DIM*jnrA;
 355
 356             /* load j atom coordinates */
 357             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 358                                               &jx0,&jy0,&jz0);
 359
 360             /* Calculate displacement vector */
 361             dx00             = _mm_sub_pd(ix0,jx0);
 362             dy00             = _mm_sub_pd(iy0,jy0);
 363             dz00             = _mm_sub_pd(iz0,jz0);
 364             dx10             = _mm_sub_pd(ix1,jx0);
 365             dy10             = _mm_sub_pd(iy1,jy0);
 366             dz10             = _mm_sub_pd(iz1,jz0);
 367             dx20             = _mm_sub_pd(ix2,jx0);
 368             dy20             = _mm_sub_pd(iy2,jy0);
 369             dz20             = _mm_sub_pd(iz2,jz0);
 370
 371             /* Calculate squared distance and things based on it */
 372             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 373             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 374             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 375
 376             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 377             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 378             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 379
 380             /* Load parameters for j particles */
 381             jq0              = _mm_load_sd(charge+jnrA+0);
 382
 383             fjx0             = _mm_setzero_pd();
 384             fjy0             = _mm_setzero_pd();
 385             fjz0             = _mm_setzero_pd();
 386
 387             /**************************
 388              * CALCULATE INTERACTIONS *
 389              **************************/
 390
 391             r00              = _mm_mul_pd(rsq00,rinv00);
 392
 393             /* Compute parameters for interactions between i and j atoms */
 394             qq00             = _mm_mul_pd(iq0,jq0);
 395
 396             /* Calculate table index by multiplying r with table scale and truncate to integer */
 397             rt               = _mm_mul_pd(r00,vftabscale);
 398             vfitab           = _mm_cvttpd_epi32(rt);
 399 #ifdef __XOP__
 400             vfeps            = _mm_frcz_pd(rt);
 401 #else
 402             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 403 #endif
 404             twovfeps         = _mm_add_pd(vfeps,vfeps);
 405             vfitab           = _mm_slli_epi32(vfitab,2);
 406
 407             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 408             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 409             F                = _mm_setzero_pd();
 410             GMX_MM_TRANSPOSE2_PD(Y,F);
 411             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 412             H                = _mm_setzero_pd();
 413             GMX_MM_TRANSPOSE2_PD(G,H);
 414             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 415             VV               = _mm_macc_pd(vfeps,Fp,Y);
 416             velec            = _mm_mul_pd(qq00,VV);
 417             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 418             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 422             velecsum         = _mm_add_pd(velecsum,velec);
 423
 424             fscal            = felec;
 425
 426             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 427
 428             /* Update vectorial force */
 429             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 430             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 431             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 432
 433             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 434             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 435             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 436
 437             /**************************
 438              * CALCULATE INTERACTIONS *
 439              **************************/
 440
 441             r10              = _mm_mul_pd(rsq10,rinv10);
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             qq10             = _mm_mul_pd(iq1,jq0);
 445
 446             /* Calculate table index by multiplying r with table scale and truncate to integer */
 447             rt               = _mm_mul_pd(r10,vftabscale);
 448             vfitab           = _mm_cvttpd_epi32(rt);
 449 #ifdef __XOP__
 450             vfeps            = _mm_frcz_pd(rt);
 451 #else
 452             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 453 #endif
 454             twovfeps         = _mm_add_pd(vfeps,vfeps);
 455             vfitab           = _mm_slli_epi32(vfitab,2);
 456
 457             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 458             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 459             F                = _mm_setzero_pd();
 460             GMX_MM_TRANSPOSE2_PD(Y,F);
 461             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 462             H                = _mm_setzero_pd();
 463             GMX_MM_TRANSPOSE2_PD(G,H);
 464             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 465             VV               = _mm_macc_pd(vfeps,Fp,Y);
 466             velec            = _mm_mul_pd(qq10,VV);
 467             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 468             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 469
 470             /* Update potential sum for this i atom from the interaction with this j atom. */
 471             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 472             velecsum         = _mm_add_pd(velecsum,velec);
 473
 474             fscal            = felec;
 475
 476             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 477
 478             /* Update vectorial force */
 479             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 480             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 481             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 482
 483             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 484             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 485             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             r20              = _mm_mul_pd(rsq20,rinv20);
 492
 493             /* Compute parameters for interactions between i and j atoms */
 494             qq20             = _mm_mul_pd(iq2,jq0);
 495
 496             /* Calculate table index by multiplying r with table scale and truncate to integer */
 497             rt               = _mm_mul_pd(r20,vftabscale);
 498             vfitab           = _mm_cvttpd_epi32(rt);
 499 #ifdef __XOP__
 500             vfeps            = _mm_frcz_pd(rt);
 501 #else
 502             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 503 #endif
 504             twovfeps         = _mm_add_pd(vfeps,vfeps);
 505             vfitab           = _mm_slli_epi32(vfitab,2);
 506
 507             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 508             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 509             F                = _mm_setzero_pd();
 510             GMX_MM_TRANSPOSE2_PD(Y,F);
 511             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 512             H                = _mm_setzero_pd();
 513             GMX_MM_TRANSPOSE2_PD(G,H);
 514             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 515             VV               = _mm_macc_pd(vfeps,Fp,Y);
 516             velec            = _mm_mul_pd(qq20,VV);
 517             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 518             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 522             velecsum         = _mm_add_pd(velecsum,velec);
 523
 524             fscal            = felec;
 525
 526             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 527
 528             /* Update vectorial force */
 529             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 530             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 531             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 532
 533             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 534             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 535             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 536
 537             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 538
 539             /* Inner loop uses 141 flops */
 540         }
 541
 542         /* End of innermost loop */
 543
 544         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 545                                               f+i_coord_offset,fshift+i_shift_offset);
 546
 547         ggid                        = gid[iidx];
 548         /* Update potential energies */
 549         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 550
 551         /* Increment number of inner iterations */
 552         inneriter                  += j_index_end - j_index_start;
 553
 554         /* Outer loop uses 19 flops */
 555     }
 556
 557     /* Increment number of outer iterations */
 558     outeriter        += nri;
 559
 560     /* Update outer/inner flops */
 561
 562     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*141);
 563 }
 564 /*
 565  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_avx_128_fma_double
 566  * Electrostatics interaction: CubicSplineTable
 567  * VdW interaction:            None
 568  * Geometry:                   Water3-Particle
 569  * Calculate force/pot:        Force
 570  */
 571 void
 572 nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_avx_128_fma_double
 573                     (t_nblist                    * gmx_restrict       nlist,
 574                      rvec                        * gmx_restrict          xx,
 575                      rvec                        * gmx_restrict          ff,
 576                      t_forcerec                  * gmx_restrict          fr,
 577                      t_mdatoms                   * gmx_restrict     mdatoms,
 578                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 579                      t_nrnb                      * gmx_restrict        nrnb)
 580 {
 581     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 582      * just 0 for non-waters.
 583      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 584      * jnr indices corresponding to data put in the four positions in the SIMD register.
 585      */
 586     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 587     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 588     int              jnrA,jnrB;
 589     int              j_coord_offsetA,j_coord_offsetB;
 590     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 591     real             rcutoff_scalar;
 592     real             *shiftvec,*fshift,*x,*f;
 593     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 594     int              vdwioffset0;
 595     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 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              vdwjidx0A,vdwjidx0B;
 601     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 602     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 603     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 604     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 605     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 606     real             *charge;
 607     __m128i          vfitab;
 608     __m128i          ifour       = _mm_set1_epi32(4);
 609     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 610     real             *vftab;
 611     __m128d          dummy_mask,cutoff_mask;
 612     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 613     __m128d          one     = _mm_set1_pd(1.0);
 614     __m128d          two     = _mm_set1_pd(2.0);
 615     x                = xx[0];
 616     f                = ff[0];
 617
 618     nri              = nlist->nri;
 619     iinr             = nlist->iinr;
 620     jindex           = nlist->jindex;
 621     jjnr             = nlist->jjnr;
 622     shiftidx         = nlist->shift;
 623     gid              = nlist->gid;
 624     shiftvec         = fr->shift_vec[0];
 625     fshift           = fr->fshift[0];
 626     facel            = _mm_set1_pd(fr->epsfac);
 627     charge           = mdatoms->chargeA;
 628
 629     vftab            = kernel_data->table_elec->data;
 630     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 631
 632     /* Setup water-specific parameters */
 633     inr              = nlist->iinr[0];
 634     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 635     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 636     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 637
 638     /* Avoid stupid compiler warnings */
 639     jnrA = jnrB = 0;
 640     j_coord_offsetA = 0;
 641     j_coord_offsetB = 0;
 642
 643     outeriter        = 0;
 644     inneriter        = 0;
 645
 646     /* Start outer loop over neighborlists */
 647     for(iidx=0; iidx<nri; iidx++)
 648     {
 649         /* Load shift vector for this list */
 650         i_shift_offset   = DIM*shiftidx[iidx];
 651
 652         /* Load limits for loop over neighbors */
 653         j_index_start    = jindex[iidx];
 654         j_index_end      = jindex[iidx+1];
 655
 656         /* Get outer coordinate index */
 657         inr              = iinr[iidx];
 658         i_coord_offset   = DIM*inr;
 659
 660         /* Load i particle coords and add shift vector */
 661         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 662                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 663
 664         fix0             = _mm_setzero_pd();
 665         fiy0             = _mm_setzero_pd();
 666         fiz0             = _mm_setzero_pd();
 667         fix1             = _mm_setzero_pd();
 668         fiy1             = _mm_setzero_pd();
 669         fiz1             = _mm_setzero_pd();
 670         fix2             = _mm_setzero_pd();
 671         fiy2             = _mm_setzero_pd();
 672         fiz2             = _mm_setzero_pd();
 673
 674         /* Start inner kernel loop */
 675         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 676         {
 677
 678             /* Get j neighbor index, and coordinate index */
 679             jnrA             = jjnr[jidx];
 680             jnrB             = jjnr[jidx+1];
 681             j_coord_offsetA  = DIM*jnrA;
 682             j_coord_offsetB  = DIM*jnrB;
 683
 684             /* load j atom coordinates */
 685             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 686                                               &jx0,&jy0,&jz0);
 687
 688             /* Calculate displacement vector */
 689             dx00             = _mm_sub_pd(ix0,jx0);
 690             dy00             = _mm_sub_pd(iy0,jy0);
 691             dz00             = _mm_sub_pd(iz0,jz0);
 692             dx10             = _mm_sub_pd(ix1,jx0);
 693             dy10             = _mm_sub_pd(iy1,jy0);
 694             dz10             = _mm_sub_pd(iz1,jz0);
 695             dx20             = _mm_sub_pd(ix2,jx0);
 696             dy20             = _mm_sub_pd(iy2,jy0);
 697             dz20             = _mm_sub_pd(iz2,jz0);
 698
 699             /* Calculate squared distance and things based on it */
 700             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 701             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 702             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 703
 704             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 705             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 706             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 707
 708             /* Load parameters for j particles */
 709             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 710
 711             fjx0             = _mm_setzero_pd();
 712             fjy0             = _mm_setzero_pd();
 713             fjz0             = _mm_setzero_pd();
 714
 715             /**************************
 716              * CALCULATE INTERACTIONS *
 717              **************************/
 718
 719             r00              = _mm_mul_pd(rsq00,rinv00);
 720
 721             /* Compute parameters for interactions between i and j atoms */
 722             qq00             = _mm_mul_pd(iq0,jq0);
 723
 724             /* Calculate table index by multiplying r with table scale and truncate to integer */
 725             rt               = _mm_mul_pd(r00,vftabscale);
 726             vfitab           = _mm_cvttpd_epi32(rt);
 727 #ifdef __XOP__
 728             vfeps            = _mm_frcz_pd(rt);
 729 #else
 730             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 731 #endif
 732             twovfeps         = _mm_add_pd(vfeps,vfeps);
 733             vfitab           = _mm_slli_epi32(vfitab,2);
 734
 735             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 736             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 737             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 738             GMX_MM_TRANSPOSE2_PD(Y,F);
 739             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 740             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 741             GMX_MM_TRANSPOSE2_PD(G,H);
 742             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 743             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 744             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 745
 746             fscal            = felec;
 747
 748             /* Update vectorial force */
 749             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 750             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 751             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 752
 753             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 754             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 755             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 756
 757             /**************************
 758              * CALCULATE INTERACTIONS *
 759              **************************/
 760
 761             r10              = _mm_mul_pd(rsq10,rinv10);
 762
 763             /* Compute parameters for interactions between i and j atoms */
 764             qq10             = _mm_mul_pd(iq1,jq0);
 765
 766             /* Calculate table index by multiplying r with table scale and truncate to integer */
 767             rt               = _mm_mul_pd(r10,vftabscale);
 768             vfitab           = _mm_cvttpd_epi32(rt);
 769 #ifdef __XOP__
 770             vfeps            = _mm_frcz_pd(rt);
 771 #else
 772             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 773 #endif
 774             twovfeps         = _mm_add_pd(vfeps,vfeps);
 775             vfitab           = _mm_slli_epi32(vfitab,2);
 776
 777             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 778             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 779             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 780             GMX_MM_TRANSPOSE2_PD(Y,F);
 781             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 782             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 783             GMX_MM_TRANSPOSE2_PD(G,H);
 784             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 785             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 786             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 787
 788             fscal            = felec;
 789
 790             /* Update vectorial force */
 791             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 792             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 793             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 794
 795             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 796             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 797             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 798
 799             /**************************
 800              * CALCULATE INTERACTIONS *
 801              **************************/
 802
 803             r20              = _mm_mul_pd(rsq20,rinv20);
 804
 805             /* Compute parameters for interactions between i and j atoms */
 806             qq20             = _mm_mul_pd(iq2,jq0);
 807
 808             /* Calculate table index by multiplying r with table scale and truncate to integer */
 809             rt               = _mm_mul_pd(r20,vftabscale);
 810             vfitab           = _mm_cvttpd_epi32(rt);
 811 #ifdef __XOP__
 812             vfeps            = _mm_frcz_pd(rt);
 813 #else
 814             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 815 #endif
 816             twovfeps         = _mm_add_pd(vfeps,vfeps);
 817             vfitab           = _mm_slli_epi32(vfitab,2);
 818
 819             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 820             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 821             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 822             GMX_MM_TRANSPOSE2_PD(Y,F);
 823             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 824             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 825             GMX_MM_TRANSPOSE2_PD(G,H);
 826             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 827             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 828             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 829
 830             fscal            = felec;
 831
 832             /* Update vectorial force */
 833             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 834             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 835             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 836
 837             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 838             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 839             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 840
 841             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 842
 843             /* Inner loop uses 129 flops */
 844         }
 845
 846         if(jidx<j_index_end)
 847         {
 848
 849             jnrA             = jjnr[jidx];
 850             j_coord_offsetA  = DIM*jnrA;
 851
 852             /* load j atom coordinates */
 853             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 854                                               &jx0,&jy0,&jz0);
 855
 856             /* Calculate displacement vector */
 857             dx00             = _mm_sub_pd(ix0,jx0);
 858             dy00             = _mm_sub_pd(iy0,jy0);
 859             dz00             = _mm_sub_pd(iz0,jz0);
 860             dx10             = _mm_sub_pd(ix1,jx0);
 861             dy10             = _mm_sub_pd(iy1,jy0);
 862             dz10             = _mm_sub_pd(iz1,jz0);
 863             dx20             = _mm_sub_pd(ix2,jx0);
 864             dy20             = _mm_sub_pd(iy2,jy0);
 865             dz20             = _mm_sub_pd(iz2,jz0);
 866
 867             /* Calculate squared distance and things based on it */
 868             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 869             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 870             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 871
 872             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 873             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 874             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 875
 876             /* Load parameters for j particles */
 877             jq0              = _mm_load_sd(charge+jnrA+0);
 878
 879             fjx0             = _mm_setzero_pd();
 880             fjy0             = _mm_setzero_pd();
 881             fjz0             = _mm_setzero_pd();
 882
 883             /**************************
 884              * CALCULATE INTERACTIONS *
 885              **************************/
 886
 887             r00              = _mm_mul_pd(rsq00,rinv00);
 888
 889             /* Compute parameters for interactions between i and j atoms */
 890             qq00             = _mm_mul_pd(iq0,jq0);
 891
 892             /* Calculate table index by multiplying r with table scale and truncate to integer */
 893             rt               = _mm_mul_pd(r00,vftabscale);
 894             vfitab           = _mm_cvttpd_epi32(rt);
 895 #ifdef __XOP__
 896             vfeps            = _mm_frcz_pd(rt);
 897 #else
 898             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 899 #endif
 900             twovfeps         = _mm_add_pd(vfeps,vfeps);
 901             vfitab           = _mm_slli_epi32(vfitab,2);
 902
 903             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 904             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 905             F                = _mm_setzero_pd();
 906             GMX_MM_TRANSPOSE2_PD(Y,F);
 907             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 908             H                = _mm_setzero_pd();
 909             GMX_MM_TRANSPOSE2_PD(G,H);
 910             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 911             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 912             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 913
 914             fscal            = felec;
 915
 916             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 917
 918             /* Update vectorial force */
 919             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 920             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 921             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 922
 923             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 924             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 925             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             r10              = _mm_mul_pd(rsq10,rinv10);
 932
 933             /* Compute parameters for interactions between i and j atoms */
 934             qq10             = _mm_mul_pd(iq1,jq0);
 935
 936             /* Calculate table index by multiplying r with table scale and truncate to integer */
 937             rt               = _mm_mul_pd(r10,vftabscale);
 938             vfitab           = _mm_cvttpd_epi32(rt);
 939 #ifdef __XOP__
 940             vfeps            = _mm_frcz_pd(rt);
 941 #else
 942             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 943 #endif
 944             twovfeps         = _mm_add_pd(vfeps,vfeps);
 945             vfitab           = _mm_slli_epi32(vfitab,2);
 946
 947             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 948             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 949             F                = _mm_setzero_pd();
 950             GMX_MM_TRANSPOSE2_PD(Y,F);
 951             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 952             H                = _mm_setzero_pd();
 953             GMX_MM_TRANSPOSE2_PD(G,H);
 954             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 955             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 956             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 957
 958             fscal            = felec;
 959
 960             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 961
 962             /* Update vectorial force */
 963             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 964             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 965             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 966
 967             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 968             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 969             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             r20              = _mm_mul_pd(rsq20,rinv20);
 976
 977             /* Compute parameters for interactions between i and j atoms */
 978             qq20             = _mm_mul_pd(iq2,jq0);
 979
 980             /* Calculate table index by multiplying r with table scale and truncate to integer */
 981             rt               = _mm_mul_pd(r20,vftabscale);
 982             vfitab           = _mm_cvttpd_epi32(rt);
 983 #ifdef __XOP__
 984             vfeps            = _mm_frcz_pd(rt);
 985 #else
 986             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 987 #endif
 988             twovfeps         = _mm_add_pd(vfeps,vfeps);
 989             vfitab           = _mm_slli_epi32(vfitab,2);
 990
 991             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 992             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 993             F                = _mm_setzero_pd();
 994             GMX_MM_TRANSPOSE2_PD(Y,F);
 995             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 996             H                = _mm_setzero_pd();
 997             GMX_MM_TRANSPOSE2_PD(G,H);
 998             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 999             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1000             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1001
1002             fscal            = felec;
1003
1004             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1005
1006             /* Update vectorial force */
1007             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1008             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1009             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1010
1011             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1012             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1013             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1014
1015             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1016
1017             /* Inner loop uses 129 flops */
1018         }
1019
1020         /* End of innermost loop */
1021
1022         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1023                                               f+i_coord_offset,fshift+i_shift_offset);
1024
1025         /* Increment number of inner iterations */
1026         inneriter                  += j_index_end - j_index_start;
1027
1028         /* Outer loop uses 18 flops */
1029     }
1030
1031     /* Increment number of outer iterations */
1032     outeriter        += nri;
1033
1034     /* Update outer/inner flops */
1035
1036     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*129);
1037 }