src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 107     real             *vftab;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_elec->data;
 130     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 135     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 136     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 164
 165         fix0             = _mm_setzero_pd();
 166         fiy0             = _mm_setzero_pd();
 167         fiz0             = _mm_setzero_pd();
 168         fix1             = _mm_setzero_pd();
 169         fiy1             = _mm_setzero_pd();
 170         fiz1             = _mm_setzero_pd();
 171         fix2             = _mm_setzero_pd();
 172         fiy2             = _mm_setzero_pd();
 173         fiz2             = _mm_setzero_pd();
 174         fix3             = _mm_setzero_pd();
 175         fiy3             = _mm_setzero_pd();
 176         fiz3             = _mm_setzero_pd();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_pd();
 180         vvdwsum          = _mm_setzero_pd();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_pd(ix0,jx0);
 198             dy00             = _mm_sub_pd(iy0,jy0);
 199             dz00             = _mm_sub_pd(iz0,jz0);
 200             dx10             = _mm_sub_pd(ix1,jx0);
 201             dy10             = _mm_sub_pd(iy1,jy0);
 202             dz10             = _mm_sub_pd(iz1,jz0);
 203             dx20             = _mm_sub_pd(ix2,jx0);
 204             dy20             = _mm_sub_pd(iy2,jy0);
 205             dz20             = _mm_sub_pd(iz2,jz0);
 206             dx30             = _mm_sub_pd(ix3,jx0);
 207             dy30             = _mm_sub_pd(iy3,jy0);
 208             dz30             = _mm_sub_pd(iz3,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 212             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 213             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 214             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 215
 216             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 217             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 218             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 219
 220             rinvsq00         = gmx_mm_inv_pd(rsq00);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226
 227             fjx0             = _mm_setzero_pd();
 228             fjy0             = _mm_setzero_pd();
 229             fjz0             = _mm_setzero_pd();
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             /* Compute parameters for interactions between i and j atoms */
 236             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 237                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 238
 239             /* LENNARD-JONES DISPERSION/REPULSION */
 240
 241             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 242             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 243             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 244             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 245             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 249
 250             fscal            = fvdw;
 251
 252             /* Update vectorial force */
 253             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 254             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 255             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 256
 257             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 258             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 259             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 260
 261             /**************************
 262              * CALCULATE INTERACTIONS *
 263              **************************/
 264
 265             r10              = _mm_mul_pd(rsq10,rinv10);
 266
 267             /* Compute parameters for interactions between i and j atoms */
 268             qq10             = _mm_mul_pd(iq1,jq0);
 269
 270             /* Calculate table index by multiplying r with table scale and truncate to integer */
 271             rt               = _mm_mul_pd(r10,vftabscale);
 272             vfitab           = _mm_cvttpd_epi32(rt);
 273 #ifdef __XOP__
 274             vfeps            = _mm_frcz_pd(rt);
 275 #else
 276             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 277 #endif
 278             twovfeps         = _mm_add_pd(vfeps,vfeps);
 279             vfitab           = _mm_slli_epi32(vfitab,2);
 280
 281             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 282             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 283             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 284             GMX_MM_TRANSPOSE2_PD(Y,F);
 285             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 286             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 287             GMX_MM_TRANSPOSE2_PD(G,H);
 288             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 289             VV               = _mm_macc_pd(vfeps,Fp,Y);
 290             velec            = _mm_mul_pd(qq10,VV);
 291             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 292             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velecsum         = _mm_add_pd(velecsum,velec);
 296
 297             fscal            = felec;
 298
 299             /* Update vectorial force */
 300             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 301             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 302             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 303
 304             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 305             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 306             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             r20              = _mm_mul_pd(rsq20,rinv20);
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq20             = _mm_mul_pd(iq2,jq0);
 316
 317             /* Calculate table index by multiplying r with table scale and truncate to integer */
 318             rt               = _mm_mul_pd(r20,vftabscale);
 319             vfitab           = _mm_cvttpd_epi32(rt);
 320 #ifdef __XOP__
 321             vfeps            = _mm_frcz_pd(rt);
 322 #else
 323             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 324 #endif
 325             twovfeps         = _mm_add_pd(vfeps,vfeps);
 326             vfitab           = _mm_slli_epi32(vfitab,2);
 327
 328             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 329             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 330             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 331             GMX_MM_TRANSPOSE2_PD(Y,F);
 332             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 333             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 334             GMX_MM_TRANSPOSE2_PD(G,H);
 335             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 336             VV               = _mm_macc_pd(vfeps,Fp,Y);
 337             velec            = _mm_mul_pd(qq20,VV);
 338             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 339             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Update vectorial force */
 347             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 348             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 349             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 350
 351             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 352             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 353             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             r30              = _mm_mul_pd(rsq30,rinv30);
 360
 361             /* Compute parameters for interactions between i and j atoms */
 362             qq30             = _mm_mul_pd(iq3,jq0);
 363
 364             /* Calculate table index by multiplying r with table scale and truncate to integer */
 365             rt               = _mm_mul_pd(r30,vftabscale);
 366             vfitab           = _mm_cvttpd_epi32(rt);
 367 #ifdef __XOP__
 368             vfeps            = _mm_frcz_pd(rt);
 369 #else
 370             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 371 #endif
 372             twovfeps         = _mm_add_pd(vfeps,vfeps);
 373             vfitab           = _mm_slli_epi32(vfitab,2);
 374
 375             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 376             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 377             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 378             GMX_MM_TRANSPOSE2_PD(Y,F);
 379             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 380             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 381             GMX_MM_TRANSPOSE2_PD(G,H);
 382             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 383             VV               = _mm_macc_pd(vfeps,Fp,Y);
 384             velec            = _mm_mul_pd(qq30,VV);
 385             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 386             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velecsum         = _mm_add_pd(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             /* Update vectorial force */
 394             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 395             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 396             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 397
 398             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 399             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 400             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 401
 402             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 403
 404             /* Inner loop uses 176 flops */
 405         }
 406
 407         if(jidx<j_index_end)
 408         {
 409
 410             jnrA             = jjnr[jidx];
 411             j_coord_offsetA  = DIM*jnrA;
 412
 413             /* load j atom coordinates */
 414             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 415                                               &jx0,&jy0,&jz0);
 416
 417             /* Calculate displacement vector */
 418             dx00             = _mm_sub_pd(ix0,jx0);
 419             dy00             = _mm_sub_pd(iy0,jy0);
 420             dz00             = _mm_sub_pd(iz0,jz0);
 421             dx10             = _mm_sub_pd(ix1,jx0);
 422             dy10             = _mm_sub_pd(iy1,jy0);
 423             dz10             = _mm_sub_pd(iz1,jz0);
 424             dx20             = _mm_sub_pd(ix2,jx0);
 425             dy20             = _mm_sub_pd(iy2,jy0);
 426             dz20             = _mm_sub_pd(iz2,jz0);
 427             dx30             = _mm_sub_pd(ix3,jx0);
 428             dy30             = _mm_sub_pd(iy3,jy0);
 429             dz30             = _mm_sub_pd(iz3,jz0);
 430
 431             /* Calculate squared distance and things based on it */
 432             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 433             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 434             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 435             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 436
 437             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 438             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 439             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 440
 441             rinvsq00         = gmx_mm_inv_pd(rsq00);
 442
 443             /* Load parameters for j particles */
 444             jq0              = _mm_load_sd(charge+jnrA+0);
 445             vdwjidx0A        = 2*vdwtype[jnrA+0];
 446
 447             fjx0             = _mm_setzero_pd();
 448             fjy0             = _mm_setzero_pd();
 449             fjz0             = _mm_setzero_pd();
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             /* Compute parameters for interactions between i and j atoms */
 456             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 457
 458             /* LENNARD-JONES DISPERSION/REPULSION */
 459
 460             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 461             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 462             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 463             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 464             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 465
 466             /* Update potential sum for this i atom from the interaction with this j atom. */
 467             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 468             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 469
 470             fscal            = fvdw;
 471
 472             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 473
 474             /* Update vectorial force */
 475             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 476             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 477             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 478
 479             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 480             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 481             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             r10              = _mm_mul_pd(rsq10,rinv10);
 488
 489             /* Compute parameters for interactions between i and j atoms */
 490             qq10             = _mm_mul_pd(iq1,jq0);
 491
 492             /* Calculate table index by multiplying r with table scale and truncate to integer */
 493             rt               = _mm_mul_pd(r10,vftabscale);
 494             vfitab           = _mm_cvttpd_epi32(rt);
 495 #ifdef __XOP__
 496             vfeps            = _mm_frcz_pd(rt);
 497 #else
 498             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 499 #endif
 500             twovfeps         = _mm_add_pd(vfeps,vfeps);
 501             vfitab           = _mm_slli_epi32(vfitab,2);
 502
 503             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 504             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 505             F                = _mm_setzero_pd();
 506             GMX_MM_TRANSPOSE2_PD(Y,F);
 507             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 508             H                = _mm_setzero_pd();
 509             GMX_MM_TRANSPOSE2_PD(G,H);
 510             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 511             VV               = _mm_macc_pd(vfeps,Fp,Y);
 512             velec            = _mm_mul_pd(qq10,VV);
 513             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 514             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 518             velecsum         = _mm_add_pd(velecsum,velec);
 519
 520             fscal            = felec;
 521
 522             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 523
 524             /* Update vectorial force */
 525             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 526             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 527             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 528
 529             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 530             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 531             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             r20              = _mm_mul_pd(rsq20,rinv20);
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq20             = _mm_mul_pd(iq2,jq0);
 541
 542             /* Calculate table index by multiplying r with table scale and truncate to integer */
 543             rt               = _mm_mul_pd(r20,vftabscale);
 544             vfitab           = _mm_cvttpd_epi32(rt);
 545 #ifdef __XOP__
 546             vfeps            = _mm_frcz_pd(rt);
 547 #else
 548             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 549 #endif
 550             twovfeps         = _mm_add_pd(vfeps,vfeps);
 551             vfitab           = _mm_slli_epi32(vfitab,2);
 552
 553             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 554             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 555             F                = _mm_setzero_pd();
 556             GMX_MM_TRANSPOSE2_PD(Y,F);
 557             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 558             H                = _mm_setzero_pd();
 559             GMX_MM_TRANSPOSE2_PD(G,H);
 560             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 561             VV               = _mm_macc_pd(vfeps,Fp,Y);
 562             velec            = _mm_mul_pd(qq20,VV);
 563             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 564             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 565
 566             /* Update potential sum for this i atom from the interaction with this j atom. */
 567             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 568             velecsum         = _mm_add_pd(velecsum,velec);
 569
 570             fscal            = felec;
 571
 572             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 573
 574             /* Update vectorial force */
 575             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 576             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 577             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 578
 579             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 580             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 581             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 582
 583             /**************************
 584              * CALCULATE INTERACTIONS *
 585              **************************/
 586
 587             r30              = _mm_mul_pd(rsq30,rinv30);
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq30             = _mm_mul_pd(iq3,jq0);
 591
 592             /* Calculate table index by multiplying r with table scale and truncate to integer */
 593             rt               = _mm_mul_pd(r30,vftabscale);
 594             vfitab           = _mm_cvttpd_epi32(rt);
 595 #ifdef __XOP__
 596             vfeps            = _mm_frcz_pd(rt);
 597 #else
 598             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 599 #endif
 600             twovfeps         = _mm_add_pd(vfeps,vfeps);
 601             vfitab           = _mm_slli_epi32(vfitab,2);
 602
 603             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 604             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 605             F                = _mm_setzero_pd();
 606             GMX_MM_TRANSPOSE2_PD(Y,F);
 607             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 608             H                = _mm_setzero_pd();
 609             GMX_MM_TRANSPOSE2_PD(G,H);
 610             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 611             VV               = _mm_macc_pd(vfeps,Fp,Y);
 612             velec            = _mm_mul_pd(qq30,VV);
 613             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 614             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 615
 616             /* Update potential sum for this i atom from the interaction with this j atom. */
 617             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 618             velecsum         = _mm_add_pd(velecsum,velec);
 619
 620             fscal            = felec;
 621
 622             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 623
 624             /* Update vectorial force */
 625             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 626             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 627             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 628
 629             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 630             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 631             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 632
 633             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 634
 635             /* Inner loop uses 176 flops */
 636         }
 637
 638         /* End of innermost loop */
 639
 640         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 641                                               f+i_coord_offset,fshift+i_shift_offset);
 642
 643         ggid                        = gid[iidx];
 644         /* Update potential energies */
 645         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 646         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 647
 648         /* Increment number of inner iterations */
 649         inneriter                  += j_index_end - j_index_start;
 650
 651         /* Outer loop uses 26 flops */
 652     }
 653
 654     /* Increment number of outer iterations */
 655     outeriter        += nri;
 656
 657     /* Update outer/inner flops */
 658
 659     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*176);
 660 }
 661 /*
 662  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_128_fma_double
 663  * Electrostatics interaction: CubicSplineTable
 664  * VdW interaction:            LennardJones
 665  * Geometry:                   Water4-Particle
 666  * Calculate force/pot:        Force
 667  */
 668 void
 669 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_128_fma_double
 670                     (t_nblist                    * gmx_restrict       nlist,
 671                      rvec                        * gmx_restrict          xx,
 672                      rvec                        * gmx_restrict          ff,
 673                      t_forcerec                  * gmx_restrict          fr,
 674                      t_mdatoms                   * gmx_restrict     mdatoms,
 675                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 676                      t_nrnb                      * gmx_restrict        nrnb)
 677 {
 678     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 679      * just 0 for non-waters.
 680      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 681      * jnr indices corresponding to data put in the four positions in the SIMD register.
 682      */
 683     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 684     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 685     int              jnrA,jnrB;
 686     int              j_coord_offsetA,j_coord_offsetB;
 687     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 688     real             rcutoff_scalar;
 689     real             *shiftvec,*fshift,*x,*f;
 690     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 691     int              vdwioffset0;
 692     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 693     int              vdwioffset1;
 694     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 695     int              vdwioffset2;
 696     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 697     int              vdwioffset3;
 698     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 699     int              vdwjidx0A,vdwjidx0B;
 700     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 701     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 702     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 703     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 704     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 705     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 706     real             *charge;
 707     int              nvdwtype;
 708     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 709     int              *vdwtype;
 710     real             *vdwparam;
 711     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 712     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 713     __m128i          vfitab;
 714     __m128i          ifour       = _mm_set1_epi32(4);
 715     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 716     real             *vftab;
 717     __m128d          dummy_mask,cutoff_mask;
 718     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 719     __m128d          one     = _mm_set1_pd(1.0);
 720     __m128d          two     = _mm_set1_pd(2.0);
 721     x                = xx[0];
 722     f                = ff[0];
 723
 724     nri              = nlist->nri;
 725     iinr             = nlist->iinr;
 726     jindex           = nlist->jindex;
 727     jjnr             = nlist->jjnr;
 728     shiftidx         = nlist->shift;
 729     gid              = nlist->gid;
 730     shiftvec         = fr->shift_vec[0];
 731     fshift           = fr->fshift[0];
 732     facel            = _mm_set1_pd(fr->epsfac);
 733     charge           = mdatoms->chargeA;
 734     nvdwtype         = fr->ntype;
 735     vdwparam         = fr->nbfp;
 736     vdwtype          = mdatoms->typeA;
 737
 738     vftab            = kernel_data->table_elec->data;
 739     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 740
 741     /* Setup water-specific parameters */
 742     inr              = nlist->iinr[0];
 743     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 744     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 745     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 746     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 747
 748     /* Avoid stupid compiler warnings */
 749     jnrA = jnrB = 0;
 750     j_coord_offsetA = 0;
 751     j_coord_offsetB = 0;
 752
 753     outeriter        = 0;
 754     inneriter        = 0;
 755
 756     /* Start outer loop over neighborlists */
 757     for(iidx=0; iidx<nri; iidx++)
 758     {
 759         /* Load shift vector for this list */
 760         i_shift_offset   = DIM*shiftidx[iidx];
 761
 762         /* Load limits for loop over neighbors */
 763         j_index_start    = jindex[iidx];
 764         j_index_end      = jindex[iidx+1];
 765
 766         /* Get outer coordinate index */
 767         inr              = iinr[iidx];
 768         i_coord_offset   = DIM*inr;
 769
 770         /* Load i particle coords and add shift vector */
 771         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 772                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 773
 774         fix0             = _mm_setzero_pd();
 775         fiy0             = _mm_setzero_pd();
 776         fiz0             = _mm_setzero_pd();
 777         fix1             = _mm_setzero_pd();
 778         fiy1             = _mm_setzero_pd();
 779         fiz1             = _mm_setzero_pd();
 780         fix2             = _mm_setzero_pd();
 781         fiy2             = _mm_setzero_pd();
 782         fiz2             = _mm_setzero_pd();
 783         fix3             = _mm_setzero_pd();
 784         fiy3             = _mm_setzero_pd();
 785         fiz3             = _mm_setzero_pd();
 786
 787         /* Start inner kernel loop */
 788         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 789         {
 790
 791             /* Get j neighbor index, and coordinate index */
 792             jnrA             = jjnr[jidx];
 793             jnrB             = jjnr[jidx+1];
 794             j_coord_offsetA  = DIM*jnrA;
 795             j_coord_offsetB  = DIM*jnrB;
 796
 797             /* load j atom coordinates */
 798             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 799                                               &jx0,&jy0,&jz0);
 800
 801             /* Calculate displacement vector */
 802             dx00             = _mm_sub_pd(ix0,jx0);
 803             dy00             = _mm_sub_pd(iy0,jy0);
 804             dz00             = _mm_sub_pd(iz0,jz0);
 805             dx10             = _mm_sub_pd(ix1,jx0);
 806             dy10             = _mm_sub_pd(iy1,jy0);
 807             dz10             = _mm_sub_pd(iz1,jz0);
 808             dx20             = _mm_sub_pd(ix2,jx0);
 809             dy20             = _mm_sub_pd(iy2,jy0);
 810             dz20             = _mm_sub_pd(iz2,jz0);
 811             dx30             = _mm_sub_pd(ix3,jx0);
 812             dy30             = _mm_sub_pd(iy3,jy0);
 813             dz30             = _mm_sub_pd(iz3,jz0);
 814
 815             /* Calculate squared distance and things based on it */
 816             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 817             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 818             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 819             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 820
 821             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 822             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 823             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 824
 825             rinvsq00         = gmx_mm_inv_pd(rsq00);
 826
 827             /* Load parameters for j particles */
 828             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 829             vdwjidx0A        = 2*vdwtype[jnrA+0];
 830             vdwjidx0B        = 2*vdwtype[jnrB+0];
 831
 832             fjx0             = _mm_setzero_pd();
 833             fjy0             = _mm_setzero_pd();
 834             fjz0             = _mm_setzero_pd();
 835
 836             /**************************
 837              * CALCULATE INTERACTIONS *
 838              **************************/
 839
 840             /* Compute parameters for interactions between i and j atoms */
 841             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 842                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 843
 844             /* LENNARD-JONES DISPERSION/REPULSION */
 845
 846             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 847             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 848
 849             fscal            = fvdw;
 850
 851             /* Update vectorial force */
 852             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 853             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 854             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 855
 856             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 857             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 858             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 859
 860             /**************************
 861              * CALCULATE INTERACTIONS *
 862              **************************/
 863
 864             r10              = _mm_mul_pd(rsq10,rinv10);
 865
 866             /* Compute parameters for interactions between i and j atoms */
 867             qq10             = _mm_mul_pd(iq1,jq0);
 868
 869             /* Calculate table index by multiplying r with table scale and truncate to integer */
 870             rt               = _mm_mul_pd(r10,vftabscale);
 871             vfitab           = _mm_cvttpd_epi32(rt);
 872 #ifdef __XOP__
 873             vfeps            = _mm_frcz_pd(rt);
 874 #else
 875             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 876 #endif
 877             twovfeps         = _mm_add_pd(vfeps,vfeps);
 878             vfitab           = _mm_slli_epi32(vfitab,2);
 879
 880             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 881             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 882             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 883             GMX_MM_TRANSPOSE2_PD(Y,F);
 884             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 885             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 886             GMX_MM_TRANSPOSE2_PD(G,H);
 887             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 888             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 889             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 890
 891             fscal            = felec;
 892
 893             /* Update vectorial force */
 894             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 895             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 896             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 897
 898             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 899             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 900             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 901
 902             /**************************
 903              * CALCULATE INTERACTIONS *
 904              **************************/
 905
 906             r20              = _mm_mul_pd(rsq20,rinv20);
 907
 908             /* Compute parameters for interactions between i and j atoms */
 909             qq20             = _mm_mul_pd(iq2,jq0);
 910
 911             /* Calculate table index by multiplying r with table scale and truncate to integer */
 912             rt               = _mm_mul_pd(r20,vftabscale);
 913             vfitab           = _mm_cvttpd_epi32(rt);
 914 #ifdef __XOP__
 915             vfeps            = _mm_frcz_pd(rt);
 916 #else
 917             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 918 #endif
 919             twovfeps         = _mm_add_pd(vfeps,vfeps);
 920             vfitab           = _mm_slli_epi32(vfitab,2);
 921
 922             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 923             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 924             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 925             GMX_MM_TRANSPOSE2_PD(Y,F);
 926             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 927             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 928             GMX_MM_TRANSPOSE2_PD(G,H);
 929             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 930             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 931             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 932
 933             fscal            = felec;
 934
 935             /* Update vectorial force */
 936             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 937             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 938             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 939
 940             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 941             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 942             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             r30              = _mm_mul_pd(rsq30,rinv30);
 949
 950             /* Compute parameters for interactions between i and j atoms */
 951             qq30             = _mm_mul_pd(iq3,jq0);
 952
 953             /* Calculate table index by multiplying r with table scale and truncate to integer */
 954             rt               = _mm_mul_pd(r30,vftabscale);
 955             vfitab           = _mm_cvttpd_epi32(rt);
 956 #ifdef __XOP__
 957             vfeps            = _mm_frcz_pd(rt);
 958 #else
 959             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 960 #endif
 961             twovfeps         = _mm_add_pd(vfeps,vfeps);
 962             vfitab           = _mm_slli_epi32(vfitab,2);
 963
 964             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 965             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 966             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 967             GMX_MM_TRANSPOSE2_PD(Y,F);
 968             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 969             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 970             GMX_MM_TRANSPOSE2_PD(G,H);
 971             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 972             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 973             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 974
 975             fscal            = felec;
 976
 977             /* Update vectorial force */
 978             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 979             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 980             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 981
 982             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 983             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 984             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 985
 986             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 987
 988             /* Inner loop uses 159 flops */
 989         }
 990
 991         if(jidx<j_index_end)
 992         {
 993
 994             jnrA             = jjnr[jidx];
 995             j_coord_offsetA  = DIM*jnrA;
 996
 997             /* load j atom coordinates */
 998             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 999                                               &jx0,&jy0,&jz0);
1000
1001             /* Calculate displacement vector */
1002             dx00             = _mm_sub_pd(ix0,jx0);
1003             dy00             = _mm_sub_pd(iy0,jy0);
1004             dz00             = _mm_sub_pd(iz0,jz0);
1005             dx10             = _mm_sub_pd(ix1,jx0);
1006             dy10             = _mm_sub_pd(iy1,jy0);
1007             dz10             = _mm_sub_pd(iz1,jz0);
1008             dx20             = _mm_sub_pd(ix2,jx0);
1009             dy20             = _mm_sub_pd(iy2,jy0);
1010             dz20             = _mm_sub_pd(iz2,jz0);
1011             dx30             = _mm_sub_pd(ix3,jx0);
1012             dy30             = _mm_sub_pd(iy3,jy0);
1013             dz30             = _mm_sub_pd(iz3,jz0);
1014
1015             /* Calculate squared distance and things based on it */
1016             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1017             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1018             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1019             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1020
1021             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1022             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1023             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1024
1025             rinvsq00         = gmx_mm_inv_pd(rsq00);
1026
1027             /* Load parameters for j particles */
1028             jq0              = _mm_load_sd(charge+jnrA+0);
1029             vdwjidx0A        = 2*vdwtype[jnrA+0];
1030
1031             fjx0             = _mm_setzero_pd();
1032             fjy0             = _mm_setzero_pd();
1033             fjz0             = _mm_setzero_pd();
1034
1035             /**************************
1036              * CALCULATE INTERACTIONS *
1037              **************************/
1038
1039             /* Compute parameters for interactions between i and j atoms */
1040             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1041
1042             /* LENNARD-JONES DISPERSION/REPULSION */
1043
1044             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1045             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1046
1047             fscal            = fvdw;
1048
1049             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1050
1051             /* Update vectorial force */
1052             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1053             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1054             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1055
1056             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1057             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1058             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1059
1060             /**************************
1061              * CALCULATE INTERACTIONS *
1062              **************************/
1063
1064             r10              = _mm_mul_pd(rsq10,rinv10);
1065
1066             /* Compute parameters for interactions between i and j atoms */
1067             qq10             = _mm_mul_pd(iq1,jq0);
1068
1069             /* Calculate table index by multiplying r with table scale and truncate to integer */
1070             rt               = _mm_mul_pd(r10,vftabscale);
1071             vfitab           = _mm_cvttpd_epi32(rt);
1072 #ifdef __XOP__
1073             vfeps            = _mm_frcz_pd(rt);
1074 #else
1075             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1076 #endif
1077             twovfeps         = _mm_add_pd(vfeps,vfeps);
1078             vfitab           = _mm_slli_epi32(vfitab,2);
1079
1080             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1081             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1082             F                = _mm_setzero_pd();
1083             GMX_MM_TRANSPOSE2_PD(Y,F);
1084             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1085             H                = _mm_setzero_pd();
1086             GMX_MM_TRANSPOSE2_PD(G,H);
1087             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1088             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1089             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1090
1091             fscal            = felec;
1092
1093             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1094
1095             /* Update vectorial force */
1096             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1097             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1098             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1099
1100             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1101             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1102             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1103
1104             /**************************
1105              * CALCULATE INTERACTIONS *
1106              **************************/
1107
1108             r20              = _mm_mul_pd(rsq20,rinv20);
1109
1110             /* Compute parameters for interactions between i and j atoms */
1111             qq20             = _mm_mul_pd(iq2,jq0);
1112
1113             /* Calculate table index by multiplying r with table scale and truncate to integer */
1114             rt               = _mm_mul_pd(r20,vftabscale);
1115             vfitab           = _mm_cvttpd_epi32(rt);
1116 #ifdef __XOP__
1117             vfeps            = _mm_frcz_pd(rt);
1118 #else
1119             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1120 #endif
1121             twovfeps         = _mm_add_pd(vfeps,vfeps);
1122             vfitab           = _mm_slli_epi32(vfitab,2);
1123
1124             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1125             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1126             F                = _mm_setzero_pd();
1127             GMX_MM_TRANSPOSE2_PD(Y,F);
1128             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1129             H                = _mm_setzero_pd();
1130             GMX_MM_TRANSPOSE2_PD(G,H);
1131             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1132             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1133             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1134
1135             fscal            = felec;
1136
1137             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1138
1139             /* Update vectorial force */
1140             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1141             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1142             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1143
1144             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1145             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1146             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1147
1148             /**************************
1149              * CALCULATE INTERACTIONS *
1150              **************************/
1151
1152             r30              = _mm_mul_pd(rsq30,rinv30);
1153
1154             /* Compute parameters for interactions between i and j atoms */
1155             qq30             = _mm_mul_pd(iq3,jq0);
1156
1157             /* Calculate table index by multiplying r with table scale and truncate to integer */
1158             rt               = _mm_mul_pd(r30,vftabscale);
1159             vfitab           = _mm_cvttpd_epi32(rt);
1160 #ifdef __XOP__
1161             vfeps            = _mm_frcz_pd(rt);
1162 #else
1163             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1164 #endif
1165             twovfeps         = _mm_add_pd(vfeps,vfeps);
1166             vfitab           = _mm_slli_epi32(vfitab,2);
1167
1168             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1169             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1170             F                = _mm_setzero_pd();
1171             GMX_MM_TRANSPOSE2_PD(Y,F);
1172             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1173             H                = _mm_setzero_pd();
1174             GMX_MM_TRANSPOSE2_PD(G,H);
1175             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1176             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1177             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1178
1179             fscal            = felec;
1180
1181             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1182
1183             /* Update vectorial force */
1184             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1185             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1186             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1187
1188             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1189             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1190             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1191
1192             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1193
1194             /* Inner loop uses 159 flops */
1195         }
1196
1197         /* End of innermost loop */
1198
1199         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1200                                               f+i_coord_offset,fshift+i_shift_offset);
1201
1202         /* Increment number of inner iterations */
1203         inneriter                  += j_index_end - j_index_start;
1204
1205         /* Outer loop uses 24 flops */
1206     }
1207
1208     /* Increment number of outer iterations */
1209     outeriter        += nri;
1210
1211     /* Update outer/inner flops */
1212
1213     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*159);
1214 }