src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_avx_128_fma_single.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_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  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,C,D refer to j loop unrolling done with AVX_128, e.g. for the four 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,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128i          vfitab;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 108     real             *vftab;
 109     __m128           dummy_mask,cutoff_mask;
 110     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 111     __m128           one     = _mm_set1_ps(1.0);
 112     __m128           two     = _mm_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_ps(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 136     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 137     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = jnrC = jnrD = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144     j_coord_offsetC = 0;
 145     j_coord_offsetD = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 172
 173         fix0             = _mm_setzero_ps();
 174         fiy0             = _mm_setzero_ps();
 175         fiz0             = _mm_setzero_ps();
 176         fix1             = _mm_setzero_ps();
 177         fiy1             = _mm_setzero_ps();
 178         fiz1             = _mm_setzero_ps();
 179         fix2             = _mm_setzero_ps();
 180         fiy2             = _mm_setzero_ps();
 181         fiz2             = _mm_setzero_ps();
 182         fix3             = _mm_setzero_ps();
 183         fiy3             = _mm_setzero_ps();
 184         fiz3             = _mm_setzero_ps();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm_setzero_ps();
 188         vvdwsum          = _mm_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               x+j_coord_offsetC,x+j_coord_offsetD,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_ps(ix0,jx0);
 211             dy00             = _mm_sub_ps(iy0,jy0);
 212             dz00             = _mm_sub_ps(iz0,jz0);
 213             dx10             = _mm_sub_ps(ix1,jx0);
 214             dy10             = _mm_sub_ps(iy1,jy0);
 215             dz10             = _mm_sub_ps(iz1,jz0);
 216             dx20             = _mm_sub_ps(ix2,jx0);
 217             dy20             = _mm_sub_ps(iy2,jy0);
 218             dz20             = _mm_sub_ps(iz2,jz0);
 219             dx30             = _mm_sub_ps(ix3,jx0);
 220             dy30             = _mm_sub_ps(iy3,jy0);
 221             dz30             = _mm_sub_ps(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 225             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 226             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 227             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 228
 229             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 230             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 231             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 232             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 233
 234             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 235             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 236             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 237
 238             /* Load parameters for j particles */
 239             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 240                                                               charge+jnrC+0,charge+jnrD+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243             vdwjidx0C        = 2*vdwtype[jnrC+0];
 244             vdwjidx0D        = 2*vdwtype[jnrD+0];
 245
 246             fjx0             = _mm_setzero_ps();
 247             fjy0             = _mm_setzero_ps();
 248             fjz0             = _mm_setzero_ps();
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             r00              = _mm_mul_ps(rsq00,rinv00);
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 258                                          vdwparam+vdwioffset0+vdwjidx0B,
 259                                          vdwparam+vdwioffset0+vdwjidx0C,
 260                                          vdwparam+vdwioffset0+vdwjidx0D,
 261                                          &c6_00,&c12_00);
 262
 263             /* Calculate table index by multiplying r with table scale and truncate to integer */
 264             rt               = _mm_mul_ps(r00,vftabscale);
 265             vfitab           = _mm_cvttps_epi32(rt);
 266 #ifdef __XOP__
 267             vfeps            = _mm_frcz_ps(rt);
 268 #else
 269             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 270 #endif
 271             twovfeps         = _mm_add_ps(vfeps,vfeps);
 272             vfitab           = _mm_slli_epi32(vfitab,3);
 273
 274             /* CUBIC SPLINE TABLE DISPERSION */
 275             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 276             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 277             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 278             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 279             _MM_TRANSPOSE4_PS(Y,F,G,H);
 280             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 281             VV               = _mm_macc_ps(vfeps,Fp,Y);
 282             vvdw6            = _mm_mul_ps(c6_00,VV);
 283             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 284             fvdw6            = _mm_mul_ps(c6_00,FF);
 285
 286             /* CUBIC SPLINE TABLE REPULSION */
 287             vfitab           = _mm_add_epi32(vfitab,ifour);
 288             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 289             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 290             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 291             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 292             _MM_TRANSPOSE4_PS(Y,F,G,H);
 293             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 294             VV               = _mm_macc_ps(vfeps,Fp,Y);
 295             vvdw12           = _mm_mul_ps(c12_00,VV);
 296             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 297             fvdw12           = _mm_mul_ps(c12_00,FF);
 298             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 299             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 303
 304             fscal            = fvdw;
 305
 306              /* Update vectorial force */
 307             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 308             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 309             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 310
 311             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 312             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 313             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq10             = _mm_mul_ps(iq1,jq0);
 321
 322             /* COULOMB ELECTROSTATICS */
 323             velec            = _mm_mul_ps(qq10,rinv10);
 324             felec            = _mm_mul_ps(velec,rinvsq10);
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velecsum         = _mm_add_ps(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331              /* Update vectorial force */
 332             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 333             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 334             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 335
 336             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 337             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 338             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq20             = _mm_mul_ps(iq2,jq0);
 346
 347             /* COULOMB ELECTROSTATICS */
 348             velec            = _mm_mul_ps(qq20,rinv20);
 349             felec            = _mm_mul_ps(velec,rinvsq20);
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velecsum         = _mm_add_ps(velecsum,velec);
 353
 354             fscal            = felec;
 355
 356              /* Update vectorial force */
 357             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 358             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 359             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 360
 361             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 362             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 363             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq30             = _mm_mul_ps(iq3,jq0);
 371
 372             /* COULOMB ELECTROSTATICS */
 373             velec            = _mm_mul_ps(qq30,rinv30);
 374             felec            = _mm_mul_ps(velec,rinvsq30);
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velecsum         = _mm_add_ps(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381              /* Update vectorial force */
 382             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 383             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 384             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 385
 386             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 387             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 388             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 389
 390             fjptrA             = f+j_coord_offsetA;
 391             fjptrB             = f+j_coord_offsetB;
 392             fjptrC             = f+j_coord_offsetC;
 393             fjptrD             = f+j_coord_offsetD;
 394
 395             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 396
 397             /* Inner loop uses 152 flops */
 398         }
 399
 400         if(jidx<j_index_end)
 401         {
 402
 403             /* Get j neighbor index, and coordinate index */
 404             jnrlistA         = jjnr[jidx];
 405             jnrlistB         = jjnr[jidx+1];
 406             jnrlistC         = jjnr[jidx+2];
 407             jnrlistD         = jjnr[jidx+3];
 408             /* Sign of each element will be negative for non-real atoms.
 409              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 410              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 411              */
 412             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 413             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 414             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 415             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 416             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 417             j_coord_offsetA  = DIM*jnrA;
 418             j_coord_offsetB  = DIM*jnrB;
 419             j_coord_offsetC  = DIM*jnrC;
 420             j_coord_offsetD  = DIM*jnrD;
 421
 422             /* load j atom coordinates */
 423             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 424                                               x+j_coord_offsetC,x+j_coord_offsetD,
 425                                               &jx0,&jy0,&jz0);
 426
 427             /* Calculate displacement vector */
 428             dx00             = _mm_sub_ps(ix0,jx0);
 429             dy00             = _mm_sub_ps(iy0,jy0);
 430             dz00             = _mm_sub_ps(iz0,jz0);
 431             dx10             = _mm_sub_ps(ix1,jx0);
 432             dy10             = _mm_sub_ps(iy1,jy0);
 433             dz10             = _mm_sub_ps(iz1,jz0);
 434             dx20             = _mm_sub_ps(ix2,jx0);
 435             dy20             = _mm_sub_ps(iy2,jy0);
 436             dz20             = _mm_sub_ps(iz2,jz0);
 437             dx30             = _mm_sub_ps(ix3,jx0);
 438             dy30             = _mm_sub_ps(iy3,jy0);
 439             dz30             = _mm_sub_ps(iz3,jz0);
 440
 441             /* Calculate squared distance and things based on it */
 442             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 443             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 444             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 445             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 446
 447             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 448             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 449             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 450             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 451
 452             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 453             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 454             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 455
 456             /* Load parameters for j particles */
 457             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 458                                                               charge+jnrC+0,charge+jnrD+0);
 459             vdwjidx0A        = 2*vdwtype[jnrA+0];
 460             vdwjidx0B        = 2*vdwtype[jnrB+0];
 461             vdwjidx0C        = 2*vdwtype[jnrC+0];
 462             vdwjidx0D        = 2*vdwtype[jnrD+0];
 463
 464             fjx0             = _mm_setzero_ps();
 465             fjy0             = _mm_setzero_ps();
 466             fjz0             = _mm_setzero_ps();
 467
 468             /**************************
 469              * CALCULATE INTERACTIONS *
 470              **************************/
 471
 472             r00              = _mm_mul_ps(rsq00,rinv00);
 473             r00              = _mm_andnot_ps(dummy_mask,r00);
 474
 475             /* Compute parameters for interactions between i and j atoms */
 476             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 477                                          vdwparam+vdwioffset0+vdwjidx0B,
 478                                          vdwparam+vdwioffset0+vdwjidx0C,
 479                                          vdwparam+vdwioffset0+vdwjidx0D,
 480                                          &c6_00,&c12_00);
 481
 482             /* Calculate table index by multiplying r with table scale and truncate to integer */
 483             rt               = _mm_mul_ps(r00,vftabscale);
 484             vfitab           = _mm_cvttps_epi32(rt);
 485 #ifdef __XOP__
 486             vfeps            = _mm_frcz_ps(rt);
 487 #else
 488             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 489 #endif
 490             twovfeps         = _mm_add_ps(vfeps,vfeps);
 491             vfitab           = _mm_slli_epi32(vfitab,3);
 492
 493             /* CUBIC SPLINE TABLE DISPERSION */
 494             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 495             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 496             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 497             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 498             _MM_TRANSPOSE4_PS(Y,F,G,H);
 499             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 500             VV               = _mm_macc_ps(vfeps,Fp,Y);
 501             vvdw6            = _mm_mul_ps(c6_00,VV);
 502             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 503             fvdw6            = _mm_mul_ps(c6_00,FF);
 504
 505             /* CUBIC SPLINE TABLE REPULSION */
 506             vfitab           = _mm_add_epi32(vfitab,ifour);
 507             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 508             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 509             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 510             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 511             _MM_TRANSPOSE4_PS(Y,F,G,H);
 512             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 513             VV               = _mm_macc_ps(vfeps,Fp,Y);
 514             vvdw12           = _mm_mul_ps(c12_00,VV);
 515             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 516             fvdw12           = _mm_mul_ps(c12_00,FF);
 517             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 518             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 522             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 523
 524             fscal            = fvdw;
 525
 526             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 527
 528              /* Update vectorial force */
 529             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 530             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 531             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 532
 533             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 534             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 535             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq10             = _mm_mul_ps(iq1,jq0);
 543
 544             /* COULOMB ELECTROSTATICS */
 545             velec            = _mm_mul_ps(qq10,rinv10);
 546             felec            = _mm_mul_ps(velec,rinvsq10);
 547
 548             /* Update potential sum for this i atom from the interaction with this j atom. */
 549             velec            = _mm_andnot_ps(dummy_mask,velec);
 550             velecsum         = _mm_add_ps(velecsum,velec);
 551
 552             fscal            = felec;
 553
 554             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 555
 556              /* Update vectorial force */
 557             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 558             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 559             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 560
 561             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 562             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 563             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq20             = _mm_mul_ps(iq2,jq0);
 571
 572             /* COULOMB ELECTROSTATICS */
 573             velec            = _mm_mul_ps(qq20,rinv20);
 574             felec            = _mm_mul_ps(velec,rinvsq20);
 575
 576             /* Update potential sum for this i atom from the interaction with this j atom. */
 577             velec            = _mm_andnot_ps(dummy_mask,velec);
 578             velecsum         = _mm_add_ps(velecsum,velec);
 579
 580             fscal            = felec;
 581
 582             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 583
 584              /* Update vectorial force */
 585             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 586             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 587             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 588
 589             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 590             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 591             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 592
 593             /**************************
 594              * CALCULATE INTERACTIONS *
 595              **************************/
 596
 597             /* Compute parameters for interactions between i and j atoms */
 598             qq30             = _mm_mul_ps(iq3,jq0);
 599
 600             /* COULOMB ELECTROSTATICS */
 601             velec            = _mm_mul_ps(qq30,rinv30);
 602             felec            = _mm_mul_ps(velec,rinvsq30);
 603
 604             /* Update potential sum for this i atom from the interaction with this j atom. */
 605             velec            = _mm_andnot_ps(dummy_mask,velec);
 606             velecsum         = _mm_add_ps(velecsum,velec);
 607
 608             fscal            = felec;
 609
 610             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 611
 612              /* Update vectorial force */
 613             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 614             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 615             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 616
 617             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 618             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 619             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 620
 621             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 622             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 623             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 624             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 625
 626             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 627
 628             /* Inner loop uses 153 flops */
 629         }
 630
 631         /* End of innermost loop */
 632
 633         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 634                                               f+i_coord_offset,fshift+i_shift_offset);
 635
 636         ggid                        = gid[iidx];
 637         /* Update potential energies */
 638         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 639         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 640
 641         /* Increment number of inner iterations */
 642         inneriter                  += j_index_end - j_index_start;
 643
 644         /* Outer loop uses 26 flops */
 645     }
 646
 647     /* Increment number of outer iterations */
 648     outeriter        += nri;
 649
 650     /* Update outer/inner flops */
 651
 652     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*153);
 653 }
 654 /*
 655  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 656  * Electrostatics interaction: Coulomb
 657  * VdW interaction:            CubicSplineTable
 658  * Geometry:                   Water4-Particle
 659  * Calculate force/pot:        Force
 660  */
 661 void
 662 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 663                     (t_nblist                    * gmx_restrict       nlist,
 664                      rvec                        * gmx_restrict          xx,
 665                      rvec                        * gmx_restrict          ff,
 666                      t_forcerec                  * gmx_restrict          fr,
 667                      t_mdatoms                   * gmx_restrict     mdatoms,
 668                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 669                      t_nrnb                      * gmx_restrict        nrnb)
 670 {
 671     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 672      * just 0 for non-waters.
 673      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 674      * jnr indices corresponding to data put in the four positions in the SIMD register.
 675      */
 676     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 677     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 678     int              jnrA,jnrB,jnrC,jnrD;
 679     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 680     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 681     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 682     real             rcutoff_scalar;
 683     real             *shiftvec,*fshift,*x,*f;
 684     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 685     real             scratch[4*DIM];
 686     __m128           fscal,rcutoff,rcutoff2,jidxall;
 687     int              vdwioffset0;
 688     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 689     int              vdwioffset1;
 690     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 691     int              vdwioffset2;
 692     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 693     int              vdwioffset3;
 694     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 695     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 696     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 697     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 698     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 699     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 700     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 701     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 702     real             *charge;
 703     int              nvdwtype;
 704     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 705     int              *vdwtype;
 706     real             *vdwparam;
 707     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 708     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 709     __m128i          vfitab;
 710     __m128i          ifour       = _mm_set1_epi32(4);
 711     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 712     real             *vftab;
 713     __m128           dummy_mask,cutoff_mask;
 714     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 715     __m128           one     = _mm_set1_ps(1.0);
 716     __m128           two     = _mm_set1_ps(2.0);
 717     x                = xx[0];
 718     f                = ff[0];
 719
 720     nri              = nlist->nri;
 721     iinr             = nlist->iinr;
 722     jindex           = nlist->jindex;
 723     jjnr             = nlist->jjnr;
 724     shiftidx         = nlist->shift;
 725     gid              = nlist->gid;
 726     shiftvec         = fr->shift_vec[0];
 727     fshift           = fr->fshift[0];
 728     facel            = _mm_set1_ps(fr->epsfac);
 729     charge           = mdatoms->chargeA;
 730     nvdwtype         = fr->ntype;
 731     vdwparam         = fr->nbfp;
 732     vdwtype          = mdatoms->typeA;
 733
 734     vftab            = kernel_data->table_vdw->data;
 735     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 736
 737     /* Setup water-specific parameters */
 738     inr              = nlist->iinr[0];
 739     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 740     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 741     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 742     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 743
 744     /* Avoid stupid compiler warnings */
 745     jnrA = jnrB = jnrC = jnrD = 0;
 746     j_coord_offsetA = 0;
 747     j_coord_offsetB = 0;
 748     j_coord_offsetC = 0;
 749     j_coord_offsetD = 0;
 750
 751     outeriter        = 0;
 752     inneriter        = 0;
 753
 754     for(iidx=0;iidx<4*DIM;iidx++)
 755     {
 756         scratch[iidx] = 0.0;
 757     }
 758
 759     /* Start outer loop over neighborlists */
 760     for(iidx=0; iidx<nri; iidx++)
 761     {
 762         /* Load shift vector for this list */
 763         i_shift_offset   = DIM*shiftidx[iidx];
 764
 765         /* Load limits for loop over neighbors */
 766         j_index_start    = jindex[iidx];
 767         j_index_end      = jindex[iidx+1];
 768
 769         /* Get outer coordinate index */
 770         inr              = iinr[iidx];
 771         i_coord_offset   = DIM*inr;
 772
 773         /* Load i particle coords and add shift vector */
 774         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 775                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 776
 777         fix0             = _mm_setzero_ps();
 778         fiy0             = _mm_setzero_ps();
 779         fiz0             = _mm_setzero_ps();
 780         fix1             = _mm_setzero_ps();
 781         fiy1             = _mm_setzero_ps();
 782         fiz1             = _mm_setzero_ps();
 783         fix2             = _mm_setzero_ps();
 784         fiy2             = _mm_setzero_ps();
 785         fiz2             = _mm_setzero_ps();
 786         fix3             = _mm_setzero_ps();
 787         fiy3             = _mm_setzero_ps();
 788         fiz3             = _mm_setzero_ps();
 789
 790         /* Start inner kernel loop */
 791         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 792         {
 793
 794             /* Get j neighbor index, and coordinate index */
 795             jnrA             = jjnr[jidx];
 796             jnrB             = jjnr[jidx+1];
 797             jnrC             = jjnr[jidx+2];
 798             jnrD             = jjnr[jidx+3];
 799             j_coord_offsetA  = DIM*jnrA;
 800             j_coord_offsetB  = DIM*jnrB;
 801             j_coord_offsetC  = DIM*jnrC;
 802             j_coord_offsetD  = DIM*jnrD;
 803
 804             /* load j atom coordinates */
 805             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 806                                               x+j_coord_offsetC,x+j_coord_offsetD,
 807                                               &jx0,&jy0,&jz0);
 808
 809             /* Calculate displacement vector */
 810             dx00             = _mm_sub_ps(ix0,jx0);
 811             dy00             = _mm_sub_ps(iy0,jy0);
 812             dz00             = _mm_sub_ps(iz0,jz0);
 813             dx10             = _mm_sub_ps(ix1,jx0);
 814             dy10             = _mm_sub_ps(iy1,jy0);
 815             dz10             = _mm_sub_ps(iz1,jz0);
 816             dx20             = _mm_sub_ps(ix2,jx0);
 817             dy20             = _mm_sub_ps(iy2,jy0);
 818             dz20             = _mm_sub_ps(iz2,jz0);
 819             dx30             = _mm_sub_ps(ix3,jx0);
 820             dy30             = _mm_sub_ps(iy3,jy0);
 821             dz30             = _mm_sub_ps(iz3,jz0);
 822
 823             /* Calculate squared distance and things based on it */
 824             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 825             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 826             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 827             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 828
 829             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 830             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 831             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 832             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 833
 834             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 835             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 836             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 837
 838             /* Load parameters for j particles */
 839             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 840                                                               charge+jnrC+0,charge+jnrD+0);
 841             vdwjidx0A        = 2*vdwtype[jnrA+0];
 842             vdwjidx0B        = 2*vdwtype[jnrB+0];
 843             vdwjidx0C        = 2*vdwtype[jnrC+0];
 844             vdwjidx0D        = 2*vdwtype[jnrD+0];
 845
 846             fjx0             = _mm_setzero_ps();
 847             fjy0             = _mm_setzero_ps();
 848             fjz0             = _mm_setzero_ps();
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             r00              = _mm_mul_ps(rsq00,rinv00);
 855
 856             /* Compute parameters for interactions between i and j atoms */
 857             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 858                                          vdwparam+vdwioffset0+vdwjidx0B,
 859                                          vdwparam+vdwioffset0+vdwjidx0C,
 860                                          vdwparam+vdwioffset0+vdwjidx0D,
 861                                          &c6_00,&c12_00);
 862
 863             /* Calculate table index by multiplying r with table scale and truncate to integer */
 864             rt               = _mm_mul_ps(r00,vftabscale);
 865             vfitab           = _mm_cvttps_epi32(rt);
 866 #ifdef __XOP__
 867             vfeps            = _mm_frcz_ps(rt);
 868 #else
 869             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 870 #endif
 871             twovfeps         = _mm_add_ps(vfeps,vfeps);
 872             vfitab           = _mm_slli_epi32(vfitab,3);
 873
 874             /* CUBIC SPLINE TABLE DISPERSION */
 875             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 876             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 877             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 878             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 879             _MM_TRANSPOSE4_PS(Y,F,G,H);
 880             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 881             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 882             fvdw6            = _mm_mul_ps(c6_00,FF);
 883
 884             /* CUBIC SPLINE TABLE REPULSION */
 885             vfitab           = _mm_add_epi32(vfitab,ifour);
 886             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 887             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 888             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 889             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 890             _MM_TRANSPOSE4_PS(Y,F,G,H);
 891             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 892             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 893             fvdw12           = _mm_mul_ps(c12_00,FF);
 894             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 895
 896             fscal            = fvdw;
 897
 898              /* Update vectorial force */
 899             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 900             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 901             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 902
 903             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 904             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 905             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 906
 907             /**************************
 908              * CALCULATE INTERACTIONS *
 909              **************************/
 910
 911             /* Compute parameters for interactions between i and j atoms */
 912             qq10             = _mm_mul_ps(iq1,jq0);
 913
 914             /* COULOMB ELECTROSTATICS */
 915             velec            = _mm_mul_ps(qq10,rinv10);
 916             felec            = _mm_mul_ps(velec,rinvsq10);
 917
 918             fscal            = felec;
 919
 920              /* Update vectorial force */
 921             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 922             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 923             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 924
 925             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 926             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 927             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 928
 929             /**************************
 930              * CALCULATE INTERACTIONS *
 931              **************************/
 932
 933             /* Compute parameters for interactions between i and j atoms */
 934             qq20             = _mm_mul_ps(iq2,jq0);
 935
 936             /* COULOMB ELECTROSTATICS */
 937             velec            = _mm_mul_ps(qq20,rinv20);
 938             felec            = _mm_mul_ps(velec,rinvsq20);
 939
 940             fscal            = felec;
 941
 942              /* Update vectorial force */
 943             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 944             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 945             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 946
 947             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 948             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 949             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq30             = _mm_mul_ps(iq3,jq0);
 957
 958             /* COULOMB ELECTROSTATICS */
 959             velec            = _mm_mul_ps(qq30,rinv30);
 960             felec            = _mm_mul_ps(velec,rinvsq30);
 961
 962             fscal            = felec;
 963
 964              /* Update vectorial force */
 965             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 966             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 967             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 968
 969             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 970             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 971             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 972
 973             fjptrA             = f+j_coord_offsetA;
 974             fjptrB             = f+j_coord_offsetB;
 975             fjptrC             = f+j_coord_offsetC;
 976             fjptrD             = f+j_coord_offsetD;
 977
 978             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 979
 980             /* Inner loop uses 141 flops */
 981         }
 982
 983         if(jidx<j_index_end)
 984         {
 985
 986             /* Get j neighbor index, and coordinate index */
 987             jnrlistA         = jjnr[jidx];
 988             jnrlistB         = jjnr[jidx+1];
 989             jnrlistC         = jjnr[jidx+2];
 990             jnrlistD         = jjnr[jidx+3];
 991             /* Sign of each element will be negative for non-real atoms.
 992              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 993              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 994              */
 995             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 996             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 997             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 998             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 999             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1000             j_coord_offsetA  = DIM*jnrA;
1001             j_coord_offsetB  = DIM*jnrB;
1002             j_coord_offsetC  = DIM*jnrC;
1003             j_coord_offsetD  = DIM*jnrD;
1004
1005             /* load j atom coordinates */
1006             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1007                                               x+j_coord_offsetC,x+j_coord_offsetD,
1008                                               &jx0,&jy0,&jz0);
1009
1010             /* Calculate displacement vector */
1011             dx00             = _mm_sub_ps(ix0,jx0);
1012             dy00             = _mm_sub_ps(iy0,jy0);
1013             dz00             = _mm_sub_ps(iz0,jz0);
1014             dx10             = _mm_sub_ps(ix1,jx0);
1015             dy10             = _mm_sub_ps(iy1,jy0);
1016             dz10             = _mm_sub_ps(iz1,jz0);
1017             dx20             = _mm_sub_ps(ix2,jx0);
1018             dy20             = _mm_sub_ps(iy2,jy0);
1019             dz20             = _mm_sub_ps(iz2,jz0);
1020             dx30             = _mm_sub_ps(ix3,jx0);
1021             dy30             = _mm_sub_ps(iy3,jy0);
1022             dz30             = _mm_sub_ps(iz3,jz0);
1023
1024             /* Calculate squared distance and things based on it */
1025             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1026             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1027             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1028             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1029
1030             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1031             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1032             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1033             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1034
1035             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1036             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1037             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1038
1039             /* Load parameters for j particles */
1040             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1041                                                               charge+jnrC+0,charge+jnrD+0);
1042             vdwjidx0A        = 2*vdwtype[jnrA+0];
1043             vdwjidx0B        = 2*vdwtype[jnrB+0];
1044             vdwjidx0C        = 2*vdwtype[jnrC+0];
1045             vdwjidx0D        = 2*vdwtype[jnrD+0];
1046
1047             fjx0             = _mm_setzero_ps();
1048             fjy0             = _mm_setzero_ps();
1049             fjz0             = _mm_setzero_ps();
1050
1051             /**************************
1052              * CALCULATE INTERACTIONS *
1053              **************************/
1054
1055             r00              = _mm_mul_ps(rsq00,rinv00);
1056             r00              = _mm_andnot_ps(dummy_mask,r00);
1057
1058             /* Compute parameters for interactions between i and j atoms */
1059             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1060                                          vdwparam+vdwioffset0+vdwjidx0B,
1061                                          vdwparam+vdwioffset0+vdwjidx0C,
1062                                          vdwparam+vdwioffset0+vdwjidx0D,
1063                                          &c6_00,&c12_00);
1064
1065             /* Calculate table index by multiplying r with table scale and truncate to integer */
1066             rt               = _mm_mul_ps(r00,vftabscale);
1067             vfitab           = _mm_cvttps_epi32(rt);
1068 #ifdef __XOP__
1069             vfeps            = _mm_frcz_ps(rt);
1070 #else
1071             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1072 #endif
1073             twovfeps         = _mm_add_ps(vfeps,vfeps);
1074             vfitab           = _mm_slli_epi32(vfitab,3);
1075
1076             /* CUBIC SPLINE TABLE DISPERSION */
1077             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1078             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1079             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1080             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1081             _MM_TRANSPOSE4_PS(Y,F,G,H);
1082             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1083             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1084             fvdw6            = _mm_mul_ps(c6_00,FF);
1085
1086             /* CUBIC SPLINE TABLE REPULSION */
1087             vfitab           = _mm_add_epi32(vfitab,ifour);
1088             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1089             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1090             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1091             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1092             _MM_TRANSPOSE4_PS(Y,F,G,H);
1093             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1094             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1095             fvdw12           = _mm_mul_ps(c12_00,FF);
1096             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1097
1098             fscal            = fvdw;
1099
1100             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1101
1102              /* Update vectorial force */
1103             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1104             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1105             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1106
1107             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1108             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1109             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             /* Compute parameters for interactions between i and j atoms */
1116             qq10             = _mm_mul_ps(iq1,jq0);
1117
1118             /* COULOMB ELECTROSTATICS */
1119             velec            = _mm_mul_ps(qq10,rinv10);
1120             felec            = _mm_mul_ps(velec,rinvsq10);
1121
1122             fscal            = felec;
1123
1124             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1125
1126              /* Update vectorial force */
1127             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1128             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1129             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1130
1131             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1132             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1133             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1134
1135             /**************************
1136              * CALCULATE INTERACTIONS *
1137              **************************/
1138
1139             /* Compute parameters for interactions between i and j atoms */
1140             qq20             = _mm_mul_ps(iq2,jq0);
1141
1142             /* COULOMB ELECTROSTATICS */
1143             velec            = _mm_mul_ps(qq20,rinv20);
1144             felec            = _mm_mul_ps(velec,rinvsq20);
1145
1146             fscal            = felec;
1147
1148             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1149
1150              /* Update vectorial force */
1151             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1152             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1153             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1154
1155             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1156             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1157             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1158
1159             /**************************
1160              * CALCULATE INTERACTIONS *
1161              **************************/
1162
1163             /* Compute parameters for interactions between i and j atoms */
1164             qq30             = _mm_mul_ps(iq3,jq0);
1165
1166             /* COULOMB ELECTROSTATICS */
1167             velec            = _mm_mul_ps(qq30,rinv30);
1168             felec            = _mm_mul_ps(velec,rinvsq30);
1169
1170             fscal            = felec;
1171
1172             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1173
1174              /* Update vectorial force */
1175             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1176             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1177             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1178
1179             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1180             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1181             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1182
1183             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1184             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1185             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1186             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1187
1188             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1189
1190             /* Inner loop uses 142 flops */
1191         }
1192
1193         /* End of innermost loop */
1194
1195         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1196                                               f+i_coord_offset,fshift+i_shift_offset);
1197
1198         /* Increment number of inner iterations */
1199         inneriter                  += j_index_end - j_index_start;
1200
1201         /* Outer loop uses 24 flops */
1202     }
1203
1204     /* Increment number of outer iterations */
1205     outeriter        += nri;
1206
1207     /* Update outer/inner flops */
1208
1209     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*142);
1210 }