src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_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_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_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_elec_vdw->data;
 131     vftabscale       = _mm_set1_ps(kernel_data->table_elec_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             /* Load parameters for j particles */
 235             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 236                                                               charge+jnrC+0,charge+jnrD+0);
 237             vdwjidx0A        = 2*vdwtype[jnrA+0];
 238             vdwjidx0B        = 2*vdwtype[jnrB+0];
 239             vdwjidx0C        = 2*vdwtype[jnrC+0];
 240             vdwjidx0D        = 2*vdwtype[jnrD+0];
 241
 242             fjx0             = _mm_setzero_ps();
 243             fjy0             = _mm_setzero_ps();
 244             fjz0             = _mm_setzero_ps();
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             r00              = _mm_mul_ps(rsq00,rinv00);
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 254                                          vdwparam+vdwioffset0+vdwjidx0B,
 255                                          vdwparam+vdwioffset0+vdwjidx0C,
 256                                          vdwparam+vdwioffset0+vdwjidx0D,
 257                                          &c6_00,&c12_00);
 258
 259             /* Calculate table index by multiplying r with table scale and truncate to integer */
 260             rt               = _mm_mul_ps(r00,vftabscale);
 261             vfitab           = _mm_cvttps_epi32(rt);
 262 #ifdef __XOP__
 263             vfeps            = _mm_frcz_ps(rt);
 264 #else
 265             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 266 #endif
 267             twovfeps         = _mm_add_ps(vfeps,vfeps);
 268             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 269
 270             /* CUBIC SPLINE TABLE DISPERSION */
 271             vfitab           = _mm_add_epi32(vfitab,ifour);
 272             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 273             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 274             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 275             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 276             _MM_TRANSPOSE4_PS(Y,F,G,H);
 277             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 278             VV               = _mm_macc_ps(vfeps,Fp,Y);
 279             vvdw6            = _mm_mul_ps(c6_00,VV);
 280             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 281             fvdw6            = _mm_mul_ps(c6_00,FF);
 282
 283             /* CUBIC SPLINE TABLE REPULSION */
 284             vfitab           = _mm_add_epi32(vfitab,ifour);
 285             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 286             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 287             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 288             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 289             _MM_TRANSPOSE4_PS(Y,F,G,H);
 290             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 291             VV               = _mm_macc_ps(vfeps,Fp,Y);
 292             vvdw12           = _mm_mul_ps(c12_00,VV);
 293             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 294             fvdw12           = _mm_mul_ps(c12_00,FF);
 295             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 296             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 300
 301             fscal            = fvdw;
 302
 303              /* Update vectorial force */
 304             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 305             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 306             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 307
 308             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 309             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 310             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             r10              = _mm_mul_ps(rsq10,rinv10);
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq10             = _mm_mul_ps(iq1,jq0);
 320
 321             /* Calculate table index by multiplying r with table scale and truncate to integer */
 322             rt               = _mm_mul_ps(r10,vftabscale);
 323             vfitab           = _mm_cvttps_epi32(rt);
 324 #ifdef __XOP__
 325             vfeps            = _mm_frcz_ps(rt);
 326 #else
 327             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 328 #endif
 329             twovfeps         = _mm_add_ps(vfeps,vfeps);
 330             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 331
 332             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 333             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 334             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 335             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 336             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 337             _MM_TRANSPOSE4_PS(Y,F,G,H);
 338             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 339             VV               = _mm_macc_ps(vfeps,Fp,Y);
 340             velec            = _mm_mul_ps(qq10,VV);
 341             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 342             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm_add_ps(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349              /* Update vectorial force */
 350             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 351             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 352             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 353
 354             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 355             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 356             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             r20              = _mm_mul_ps(rsq20,rinv20);
 363
 364             /* Compute parameters for interactions between i and j atoms */
 365             qq20             = _mm_mul_ps(iq2,jq0);
 366
 367             /* Calculate table index by multiplying r with table scale and truncate to integer */
 368             rt               = _mm_mul_ps(r20,vftabscale);
 369             vfitab           = _mm_cvttps_epi32(rt);
 370 #ifdef __XOP__
 371             vfeps            = _mm_frcz_ps(rt);
 372 #else
 373             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 374 #endif
 375             twovfeps         = _mm_add_ps(vfeps,vfeps);
 376             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 377
 378             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 379             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 380             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 381             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 382             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 383             _MM_TRANSPOSE4_PS(Y,F,G,H);
 384             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 385             VV               = _mm_macc_ps(vfeps,Fp,Y);
 386             velec            = _mm_mul_ps(qq20,VV);
 387             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 388             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velecsum         = _mm_add_ps(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395              /* Update vectorial force */
 396             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 397             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 398             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 399
 400             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 401             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 402             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 403
 404             /**************************
 405              * CALCULATE INTERACTIONS *
 406              **************************/
 407
 408             r30              = _mm_mul_ps(rsq30,rinv30);
 409
 410             /* Compute parameters for interactions between i and j atoms */
 411             qq30             = _mm_mul_ps(iq3,jq0);
 412
 413             /* Calculate table index by multiplying r with table scale and truncate to integer */
 414             rt               = _mm_mul_ps(r30,vftabscale);
 415             vfitab           = _mm_cvttps_epi32(rt);
 416 #ifdef __XOP__
 417             vfeps            = _mm_frcz_ps(rt);
 418 #else
 419             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 420 #endif
 421             twovfeps         = _mm_add_ps(vfeps,vfeps);
 422             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 423
 424             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 425             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 426             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 427             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 428             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 429             _MM_TRANSPOSE4_PS(Y,F,G,H);
 430             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 431             VV               = _mm_macc_ps(vfeps,Fp,Y);
 432             velec            = _mm_mul_ps(qq30,VV);
 433             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 434             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 435
 436             /* Update potential sum for this i atom from the interaction with this j atom. */
 437             velecsum         = _mm_add_ps(velecsum,velec);
 438
 439             fscal            = felec;
 440
 441              /* Update vectorial force */
 442             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 443             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 444             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 445
 446             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 447             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 448             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 449
 450             fjptrA             = f+j_coord_offsetA;
 451             fjptrB             = f+j_coord_offsetB;
 452             fjptrC             = f+j_coord_offsetC;
 453             fjptrD             = f+j_coord_offsetD;
 454
 455             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 456
 457             /* Inner loop uses 197 flops */
 458         }
 459
 460         if(jidx<j_index_end)
 461         {
 462
 463             /* Get j neighbor index, and coordinate index */
 464             jnrlistA         = jjnr[jidx];
 465             jnrlistB         = jjnr[jidx+1];
 466             jnrlistC         = jjnr[jidx+2];
 467             jnrlistD         = jjnr[jidx+3];
 468             /* Sign of each element will be negative for non-real atoms.
 469              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 470              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 471              */
 472             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 473             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 474             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 475             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 476             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 477             j_coord_offsetA  = DIM*jnrA;
 478             j_coord_offsetB  = DIM*jnrB;
 479             j_coord_offsetC  = DIM*jnrC;
 480             j_coord_offsetD  = DIM*jnrD;
 481
 482             /* load j atom coordinates */
 483             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 484                                               x+j_coord_offsetC,x+j_coord_offsetD,
 485                                               &jx0,&jy0,&jz0);
 486
 487             /* Calculate displacement vector */
 488             dx00             = _mm_sub_ps(ix0,jx0);
 489             dy00             = _mm_sub_ps(iy0,jy0);
 490             dz00             = _mm_sub_ps(iz0,jz0);
 491             dx10             = _mm_sub_ps(ix1,jx0);
 492             dy10             = _mm_sub_ps(iy1,jy0);
 493             dz10             = _mm_sub_ps(iz1,jz0);
 494             dx20             = _mm_sub_ps(ix2,jx0);
 495             dy20             = _mm_sub_ps(iy2,jy0);
 496             dz20             = _mm_sub_ps(iz2,jz0);
 497             dx30             = _mm_sub_ps(ix3,jx0);
 498             dy30             = _mm_sub_ps(iy3,jy0);
 499             dz30             = _mm_sub_ps(iz3,jz0);
 500
 501             /* Calculate squared distance and things based on it */
 502             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 503             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 504             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 505             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 506
 507             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 508             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 509             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 510             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 511
 512             /* Load parameters for j particles */
 513             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 514                                                               charge+jnrC+0,charge+jnrD+0);
 515             vdwjidx0A        = 2*vdwtype[jnrA+0];
 516             vdwjidx0B        = 2*vdwtype[jnrB+0];
 517             vdwjidx0C        = 2*vdwtype[jnrC+0];
 518             vdwjidx0D        = 2*vdwtype[jnrD+0];
 519
 520             fjx0             = _mm_setzero_ps();
 521             fjy0             = _mm_setzero_ps();
 522             fjz0             = _mm_setzero_ps();
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             r00              = _mm_mul_ps(rsq00,rinv00);
 529             r00              = _mm_andnot_ps(dummy_mask,r00);
 530
 531             /* Compute parameters for interactions between i and j atoms */
 532             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 533                                          vdwparam+vdwioffset0+vdwjidx0B,
 534                                          vdwparam+vdwioffset0+vdwjidx0C,
 535                                          vdwparam+vdwioffset0+vdwjidx0D,
 536                                          &c6_00,&c12_00);
 537
 538             /* Calculate table index by multiplying r with table scale and truncate to integer */
 539             rt               = _mm_mul_ps(r00,vftabscale);
 540             vfitab           = _mm_cvttps_epi32(rt);
 541 #ifdef __XOP__
 542             vfeps            = _mm_frcz_ps(rt);
 543 #else
 544             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 545 #endif
 546             twovfeps         = _mm_add_ps(vfeps,vfeps);
 547             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 548
 549             /* CUBIC SPLINE TABLE DISPERSION */
 550             vfitab           = _mm_add_epi32(vfitab,ifour);
 551             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 552             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 553             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 554             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 555             _MM_TRANSPOSE4_PS(Y,F,G,H);
 556             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 557             VV               = _mm_macc_ps(vfeps,Fp,Y);
 558             vvdw6            = _mm_mul_ps(c6_00,VV);
 559             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 560             fvdw6            = _mm_mul_ps(c6_00,FF);
 561
 562             /* CUBIC SPLINE TABLE REPULSION */
 563             vfitab           = _mm_add_epi32(vfitab,ifour);
 564             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 565             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 566             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 567             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 568             _MM_TRANSPOSE4_PS(Y,F,G,H);
 569             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 570             VV               = _mm_macc_ps(vfeps,Fp,Y);
 571             vvdw12           = _mm_mul_ps(c12_00,VV);
 572             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 573             fvdw12           = _mm_mul_ps(c12_00,FF);
 574             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 575             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 579             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 580
 581             fscal            = fvdw;
 582
 583             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 584
 585              /* Update vectorial force */
 586             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 587             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 588             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 589
 590             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 591             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 592             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 593
 594             /**************************
 595              * CALCULATE INTERACTIONS *
 596              **************************/
 597
 598             r10              = _mm_mul_ps(rsq10,rinv10);
 599             r10              = _mm_andnot_ps(dummy_mask,r10);
 600
 601             /* Compute parameters for interactions between i and j atoms */
 602             qq10             = _mm_mul_ps(iq1,jq0);
 603
 604             /* Calculate table index by multiplying r with table scale and truncate to integer */
 605             rt               = _mm_mul_ps(r10,vftabscale);
 606             vfitab           = _mm_cvttps_epi32(rt);
 607 #ifdef __XOP__
 608             vfeps            = _mm_frcz_ps(rt);
 609 #else
 610             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 611 #endif
 612             twovfeps         = _mm_add_ps(vfeps,vfeps);
 613             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 614
 615             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 616             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 617             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 618             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 619             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 620             _MM_TRANSPOSE4_PS(Y,F,G,H);
 621             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 622             VV               = _mm_macc_ps(vfeps,Fp,Y);
 623             velec            = _mm_mul_ps(qq10,VV);
 624             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 625             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 626
 627             /* Update potential sum for this i atom from the interaction with this j atom. */
 628             velec            = _mm_andnot_ps(dummy_mask,velec);
 629             velecsum         = _mm_add_ps(velecsum,velec);
 630
 631             fscal            = felec;
 632
 633             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 634
 635              /* Update vectorial force */
 636             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 637             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 638             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 639
 640             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 641             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 642             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 643
 644             /**************************
 645              * CALCULATE INTERACTIONS *
 646              **************************/
 647
 648             r20              = _mm_mul_ps(rsq20,rinv20);
 649             r20              = _mm_andnot_ps(dummy_mask,r20);
 650
 651             /* Compute parameters for interactions between i and j atoms */
 652             qq20             = _mm_mul_ps(iq2,jq0);
 653
 654             /* Calculate table index by multiplying r with table scale and truncate to integer */
 655             rt               = _mm_mul_ps(r20,vftabscale);
 656             vfitab           = _mm_cvttps_epi32(rt);
 657 #ifdef __XOP__
 658             vfeps            = _mm_frcz_ps(rt);
 659 #else
 660             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 661 #endif
 662             twovfeps         = _mm_add_ps(vfeps,vfeps);
 663             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 664
 665             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 666             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 667             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 668             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 669             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 670             _MM_TRANSPOSE4_PS(Y,F,G,H);
 671             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 672             VV               = _mm_macc_ps(vfeps,Fp,Y);
 673             velec            = _mm_mul_ps(qq20,VV);
 674             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 675             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 676
 677             /* Update potential sum for this i atom from the interaction with this j atom. */
 678             velec            = _mm_andnot_ps(dummy_mask,velec);
 679             velecsum         = _mm_add_ps(velecsum,velec);
 680
 681             fscal            = felec;
 682
 683             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 684
 685              /* Update vectorial force */
 686             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 687             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 688             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 689
 690             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 691             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 692             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 693
 694             /**************************
 695              * CALCULATE INTERACTIONS *
 696              **************************/
 697
 698             r30              = _mm_mul_ps(rsq30,rinv30);
 699             r30              = _mm_andnot_ps(dummy_mask,r30);
 700
 701             /* Compute parameters for interactions between i and j atoms */
 702             qq30             = _mm_mul_ps(iq3,jq0);
 703
 704             /* Calculate table index by multiplying r with table scale and truncate to integer */
 705             rt               = _mm_mul_ps(r30,vftabscale);
 706             vfitab           = _mm_cvttps_epi32(rt);
 707 #ifdef __XOP__
 708             vfeps            = _mm_frcz_ps(rt);
 709 #else
 710             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 711 #endif
 712             twovfeps         = _mm_add_ps(vfeps,vfeps);
 713             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 714
 715             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 716             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 717             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 718             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 719             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 720             _MM_TRANSPOSE4_PS(Y,F,G,H);
 721             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 722             VV               = _mm_macc_ps(vfeps,Fp,Y);
 723             velec            = _mm_mul_ps(qq30,VV);
 724             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 725             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 726
 727             /* Update potential sum for this i atom from the interaction with this j atom. */
 728             velec            = _mm_andnot_ps(dummy_mask,velec);
 729             velecsum         = _mm_add_ps(velecsum,velec);
 730
 731             fscal            = felec;
 732
 733             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 734
 735              /* Update vectorial force */
 736             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 737             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 738             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 739
 740             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 741             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 742             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 743
 744             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 745             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 746             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 747             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 748
 749             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 750
 751             /* Inner loop uses 201 flops */
 752         }
 753
 754         /* End of innermost loop */
 755
 756         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 757                                               f+i_coord_offset,fshift+i_shift_offset);
 758
 759         ggid                        = gid[iidx];
 760         /* Update potential energies */
 761         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 762         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 763
 764         /* Increment number of inner iterations */
 765         inneriter                  += j_index_end - j_index_start;
 766
 767         /* Outer loop uses 26 flops */
 768     }
 769
 770     /* Increment number of outer iterations */
 771     outeriter        += nri;
 772
 773     /* Update outer/inner flops */
 774
 775     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*201);
 776 }
 777 /*
 778  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 779  * Electrostatics interaction: CubicSplineTable
 780  * VdW interaction:            CubicSplineTable
 781  * Geometry:                   Water4-Particle
 782  * Calculate force/pot:        Force
 783  */
 784 void
 785 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 786                     (t_nblist                    * gmx_restrict       nlist,
 787                      rvec                        * gmx_restrict          xx,
 788                      rvec                        * gmx_restrict          ff,
 789                      t_forcerec                  * gmx_restrict          fr,
 790                      t_mdatoms                   * gmx_restrict     mdatoms,
 791                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 792                      t_nrnb                      * gmx_restrict        nrnb)
 793 {
 794     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 795      * just 0 for non-waters.
 796      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 797      * jnr indices corresponding to data put in the four positions in the SIMD register.
 798      */
 799     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 800     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 801     int              jnrA,jnrB,jnrC,jnrD;
 802     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 803     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 804     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 805     real             rcutoff_scalar;
 806     real             *shiftvec,*fshift,*x,*f;
 807     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 808     real             scratch[4*DIM];
 809     __m128           fscal,rcutoff,rcutoff2,jidxall;
 810     int              vdwioffset0;
 811     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 812     int              vdwioffset1;
 813     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 814     int              vdwioffset2;
 815     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 816     int              vdwioffset3;
 817     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 818     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 819     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 820     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 821     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 822     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 823     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 824     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 825     real             *charge;
 826     int              nvdwtype;
 827     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 828     int              *vdwtype;
 829     real             *vdwparam;
 830     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 831     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 832     __m128i          vfitab;
 833     __m128i          ifour       = _mm_set1_epi32(4);
 834     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 835     real             *vftab;
 836     __m128           dummy_mask,cutoff_mask;
 837     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 838     __m128           one     = _mm_set1_ps(1.0);
 839     __m128           two     = _mm_set1_ps(2.0);
 840     x                = xx[0];
 841     f                = ff[0];
 842
 843     nri              = nlist->nri;
 844     iinr             = nlist->iinr;
 845     jindex           = nlist->jindex;
 846     jjnr             = nlist->jjnr;
 847     shiftidx         = nlist->shift;
 848     gid              = nlist->gid;
 849     shiftvec         = fr->shift_vec[0];
 850     fshift           = fr->fshift[0];
 851     facel            = _mm_set1_ps(fr->epsfac);
 852     charge           = mdatoms->chargeA;
 853     nvdwtype         = fr->ntype;
 854     vdwparam         = fr->nbfp;
 855     vdwtype          = mdatoms->typeA;
 856
 857     vftab            = kernel_data->table_elec_vdw->data;
 858     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 859
 860     /* Setup water-specific parameters */
 861     inr              = nlist->iinr[0];
 862     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 863     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 864     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 865     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 866
 867     /* Avoid stupid compiler warnings */
 868     jnrA = jnrB = jnrC = jnrD = 0;
 869     j_coord_offsetA = 0;
 870     j_coord_offsetB = 0;
 871     j_coord_offsetC = 0;
 872     j_coord_offsetD = 0;
 873
 874     outeriter        = 0;
 875     inneriter        = 0;
 876
 877     for(iidx=0;iidx<4*DIM;iidx++)
 878     {
 879         scratch[iidx] = 0.0;
 880     }
 881
 882     /* Start outer loop over neighborlists */
 883     for(iidx=0; iidx<nri; iidx++)
 884     {
 885         /* Load shift vector for this list */
 886         i_shift_offset   = DIM*shiftidx[iidx];
 887
 888         /* Load limits for loop over neighbors */
 889         j_index_start    = jindex[iidx];
 890         j_index_end      = jindex[iidx+1];
 891
 892         /* Get outer coordinate index */
 893         inr              = iinr[iidx];
 894         i_coord_offset   = DIM*inr;
 895
 896         /* Load i particle coords and add shift vector */
 897         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 898                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 899
 900         fix0             = _mm_setzero_ps();
 901         fiy0             = _mm_setzero_ps();
 902         fiz0             = _mm_setzero_ps();
 903         fix1             = _mm_setzero_ps();
 904         fiy1             = _mm_setzero_ps();
 905         fiz1             = _mm_setzero_ps();
 906         fix2             = _mm_setzero_ps();
 907         fiy2             = _mm_setzero_ps();
 908         fiz2             = _mm_setzero_ps();
 909         fix3             = _mm_setzero_ps();
 910         fiy3             = _mm_setzero_ps();
 911         fiz3             = _mm_setzero_ps();
 912
 913         /* Start inner kernel loop */
 914         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 915         {
 916
 917             /* Get j neighbor index, and coordinate index */
 918             jnrA             = jjnr[jidx];
 919             jnrB             = jjnr[jidx+1];
 920             jnrC             = jjnr[jidx+2];
 921             jnrD             = jjnr[jidx+3];
 922             j_coord_offsetA  = DIM*jnrA;
 923             j_coord_offsetB  = DIM*jnrB;
 924             j_coord_offsetC  = DIM*jnrC;
 925             j_coord_offsetD  = DIM*jnrD;
 926
 927             /* load j atom coordinates */
 928             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 929                                               x+j_coord_offsetC,x+j_coord_offsetD,
 930                                               &jx0,&jy0,&jz0);
 931
 932             /* Calculate displacement vector */
 933             dx00             = _mm_sub_ps(ix0,jx0);
 934             dy00             = _mm_sub_ps(iy0,jy0);
 935             dz00             = _mm_sub_ps(iz0,jz0);
 936             dx10             = _mm_sub_ps(ix1,jx0);
 937             dy10             = _mm_sub_ps(iy1,jy0);
 938             dz10             = _mm_sub_ps(iz1,jz0);
 939             dx20             = _mm_sub_ps(ix2,jx0);
 940             dy20             = _mm_sub_ps(iy2,jy0);
 941             dz20             = _mm_sub_ps(iz2,jz0);
 942             dx30             = _mm_sub_ps(ix3,jx0);
 943             dy30             = _mm_sub_ps(iy3,jy0);
 944             dz30             = _mm_sub_ps(iz3,jz0);
 945
 946             /* Calculate squared distance and things based on it */
 947             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 948             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 949             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 950             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 951
 952             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 953             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 954             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 955             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 956
 957             /* Load parameters for j particles */
 958             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 959                                                               charge+jnrC+0,charge+jnrD+0);
 960             vdwjidx0A        = 2*vdwtype[jnrA+0];
 961             vdwjidx0B        = 2*vdwtype[jnrB+0];
 962             vdwjidx0C        = 2*vdwtype[jnrC+0];
 963             vdwjidx0D        = 2*vdwtype[jnrD+0];
 964
 965             fjx0             = _mm_setzero_ps();
 966             fjy0             = _mm_setzero_ps();
 967             fjz0             = _mm_setzero_ps();
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             r00              = _mm_mul_ps(rsq00,rinv00);
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 977                                          vdwparam+vdwioffset0+vdwjidx0B,
 978                                          vdwparam+vdwioffset0+vdwjidx0C,
 979                                          vdwparam+vdwioffset0+vdwjidx0D,
 980                                          &c6_00,&c12_00);
 981
 982             /* Calculate table index by multiplying r with table scale and truncate to integer */
 983             rt               = _mm_mul_ps(r00,vftabscale);
 984             vfitab           = _mm_cvttps_epi32(rt);
 985 #ifdef __XOP__
 986             vfeps            = _mm_frcz_ps(rt);
 987 #else
 988             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 989 #endif
 990             twovfeps         = _mm_add_ps(vfeps,vfeps);
 991             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 992
 993             /* CUBIC SPLINE TABLE DISPERSION */
 994             vfitab           = _mm_add_epi32(vfitab,ifour);
 995             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 996             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 997             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 998             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 999             _MM_TRANSPOSE4_PS(Y,F,G,H);
1000             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1001             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1002             fvdw6            = _mm_mul_ps(c6_00,FF);
1003
1004             /* CUBIC SPLINE TABLE REPULSION */
1005             vfitab           = _mm_add_epi32(vfitab,ifour);
1006             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1007             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1008             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1009             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1010             _MM_TRANSPOSE4_PS(Y,F,G,H);
1011             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1012             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1013             fvdw12           = _mm_mul_ps(c12_00,FF);
1014             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1015
1016             fscal            = fvdw;
1017
1018              /* Update vectorial force */
1019             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1020             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1021             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1022
1023             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1024             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1025             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1026
1027             /**************************
1028              * CALCULATE INTERACTIONS *
1029              **************************/
1030
1031             r10              = _mm_mul_ps(rsq10,rinv10);
1032
1033             /* Compute parameters for interactions between i and j atoms */
1034             qq10             = _mm_mul_ps(iq1,jq0);
1035
1036             /* Calculate table index by multiplying r with table scale and truncate to integer */
1037             rt               = _mm_mul_ps(r10,vftabscale);
1038             vfitab           = _mm_cvttps_epi32(rt);
1039 #ifdef __XOP__
1040             vfeps            = _mm_frcz_ps(rt);
1041 #else
1042             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1043 #endif
1044             twovfeps         = _mm_add_ps(vfeps,vfeps);
1045             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1046
1047             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1048             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1049             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1050             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1051             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1052             _MM_TRANSPOSE4_PS(Y,F,G,H);
1053             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1054             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1055             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1056
1057             fscal            = felec;
1058
1059              /* Update vectorial force */
1060             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1061             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1062             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1063
1064             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1065             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1066             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1067
1068             /**************************
1069              * CALCULATE INTERACTIONS *
1070              **************************/
1071
1072             r20              = _mm_mul_ps(rsq20,rinv20);
1073
1074             /* Compute parameters for interactions between i and j atoms */
1075             qq20             = _mm_mul_ps(iq2,jq0);
1076
1077             /* Calculate table index by multiplying r with table scale and truncate to integer */
1078             rt               = _mm_mul_ps(r20,vftabscale);
1079             vfitab           = _mm_cvttps_epi32(rt);
1080 #ifdef __XOP__
1081             vfeps            = _mm_frcz_ps(rt);
1082 #else
1083             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1084 #endif
1085             twovfeps         = _mm_add_ps(vfeps,vfeps);
1086             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1087
1088             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1089             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1090             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1091             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1092             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1093             _MM_TRANSPOSE4_PS(Y,F,G,H);
1094             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1095             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1096             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1097
1098             fscal            = felec;
1099
1100              /* Update vectorial force */
1101             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1102             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1103             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1104
1105             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1106             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1107             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             r30              = _mm_mul_ps(rsq30,rinv30);
1114
1115             /* Compute parameters for interactions between i and j atoms */
1116             qq30             = _mm_mul_ps(iq3,jq0);
1117
1118             /* Calculate table index by multiplying r with table scale and truncate to integer */
1119             rt               = _mm_mul_ps(r30,vftabscale);
1120             vfitab           = _mm_cvttps_epi32(rt);
1121 #ifdef __XOP__
1122             vfeps            = _mm_frcz_ps(rt);
1123 #else
1124             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1125 #endif
1126             twovfeps         = _mm_add_ps(vfeps,vfeps);
1127             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1128
1129             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1130             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1131             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1132             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1133             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1134             _MM_TRANSPOSE4_PS(Y,F,G,H);
1135             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1136             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1137             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1138
1139             fscal            = felec;
1140
1141              /* Update vectorial force */
1142             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1143             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1144             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1145
1146             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1147             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1148             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1149
1150             fjptrA             = f+j_coord_offsetA;
1151             fjptrB             = f+j_coord_offsetB;
1152             fjptrC             = f+j_coord_offsetC;
1153             fjptrD             = f+j_coord_offsetD;
1154
1155             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1156
1157             /* Inner loop uses 177 flops */
1158         }
1159
1160         if(jidx<j_index_end)
1161         {
1162
1163             /* Get j neighbor index, and coordinate index */
1164             jnrlistA         = jjnr[jidx];
1165             jnrlistB         = jjnr[jidx+1];
1166             jnrlistC         = jjnr[jidx+2];
1167             jnrlistD         = jjnr[jidx+3];
1168             /* Sign of each element will be negative for non-real atoms.
1169              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1170              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1171              */
1172             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1173             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1174             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1175             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1176             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1177             j_coord_offsetA  = DIM*jnrA;
1178             j_coord_offsetB  = DIM*jnrB;
1179             j_coord_offsetC  = DIM*jnrC;
1180             j_coord_offsetD  = DIM*jnrD;
1181
1182             /* load j atom coordinates */
1183             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1184                                               x+j_coord_offsetC,x+j_coord_offsetD,
1185                                               &jx0,&jy0,&jz0);
1186
1187             /* Calculate displacement vector */
1188             dx00             = _mm_sub_ps(ix0,jx0);
1189             dy00             = _mm_sub_ps(iy0,jy0);
1190             dz00             = _mm_sub_ps(iz0,jz0);
1191             dx10             = _mm_sub_ps(ix1,jx0);
1192             dy10             = _mm_sub_ps(iy1,jy0);
1193             dz10             = _mm_sub_ps(iz1,jz0);
1194             dx20             = _mm_sub_ps(ix2,jx0);
1195             dy20             = _mm_sub_ps(iy2,jy0);
1196             dz20             = _mm_sub_ps(iz2,jz0);
1197             dx30             = _mm_sub_ps(ix3,jx0);
1198             dy30             = _mm_sub_ps(iy3,jy0);
1199             dz30             = _mm_sub_ps(iz3,jz0);
1200
1201             /* Calculate squared distance and things based on it */
1202             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1203             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1204             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1205             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1206
1207             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1208             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1209             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1210             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1211
1212             /* Load parameters for j particles */
1213             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1214                                                               charge+jnrC+0,charge+jnrD+0);
1215             vdwjidx0A        = 2*vdwtype[jnrA+0];
1216             vdwjidx0B        = 2*vdwtype[jnrB+0];
1217             vdwjidx0C        = 2*vdwtype[jnrC+0];
1218             vdwjidx0D        = 2*vdwtype[jnrD+0];
1219
1220             fjx0             = _mm_setzero_ps();
1221             fjy0             = _mm_setzero_ps();
1222             fjz0             = _mm_setzero_ps();
1223
1224             /**************************
1225              * CALCULATE INTERACTIONS *
1226              **************************/
1227
1228             r00              = _mm_mul_ps(rsq00,rinv00);
1229             r00              = _mm_andnot_ps(dummy_mask,r00);
1230
1231             /* Compute parameters for interactions between i and j atoms */
1232             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1233                                          vdwparam+vdwioffset0+vdwjidx0B,
1234                                          vdwparam+vdwioffset0+vdwjidx0C,
1235                                          vdwparam+vdwioffset0+vdwjidx0D,
1236                                          &c6_00,&c12_00);
1237
1238             /* Calculate table index by multiplying r with table scale and truncate to integer */
1239             rt               = _mm_mul_ps(r00,vftabscale);
1240             vfitab           = _mm_cvttps_epi32(rt);
1241 #ifdef __XOP__
1242             vfeps            = _mm_frcz_ps(rt);
1243 #else
1244             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1245 #endif
1246             twovfeps         = _mm_add_ps(vfeps,vfeps);
1247             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1248
1249             /* CUBIC SPLINE TABLE DISPERSION */
1250             vfitab           = _mm_add_epi32(vfitab,ifour);
1251             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1252             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1253             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1254             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1255             _MM_TRANSPOSE4_PS(Y,F,G,H);
1256             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1257             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1258             fvdw6            = _mm_mul_ps(c6_00,FF);
1259
1260             /* CUBIC SPLINE TABLE REPULSION */
1261             vfitab           = _mm_add_epi32(vfitab,ifour);
1262             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1263             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1264             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1265             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1266             _MM_TRANSPOSE4_PS(Y,F,G,H);
1267             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1268             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1269             fvdw12           = _mm_mul_ps(c12_00,FF);
1270             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1271
1272             fscal            = fvdw;
1273
1274             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1275
1276              /* Update vectorial force */
1277             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1278             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1279             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1280
1281             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1282             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1283             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1284
1285             /**************************
1286              * CALCULATE INTERACTIONS *
1287              **************************/
1288
1289             r10              = _mm_mul_ps(rsq10,rinv10);
1290             r10              = _mm_andnot_ps(dummy_mask,r10);
1291
1292             /* Compute parameters for interactions between i and j atoms */
1293             qq10             = _mm_mul_ps(iq1,jq0);
1294
1295             /* Calculate table index by multiplying r with table scale and truncate to integer */
1296             rt               = _mm_mul_ps(r10,vftabscale);
1297             vfitab           = _mm_cvttps_epi32(rt);
1298 #ifdef __XOP__
1299             vfeps            = _mm_frcz_ps(rt);
1300 #else
1301             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1302 #endif
1303             twovfeps         = _mm_add_ps(vfeps,vfeps);
1304             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1305
1306             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1307             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1308             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1309             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1310             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1311             _MM_TRANSPOSE4_PS(Y,F,G,H);
1312             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1313             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1314             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1315
1316             fscal            = felec;
1317
1318             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1319
1320              /* Update vectorial force */
1321             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1322             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1323             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1324
1325             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1326             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1327             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1328
1329             /**************************
1330              * CALCULATE INTERACTIONS *
1331              **************************/
1332
1333             r20              = _mm_mul_ps(rsq20,rinv20);
1334             r20              = _mm_andnot_ps(dummy_mask,r20);
1335
1336             /* Compute parameters for interactions between i and j atoms */
1337             qq20             = _mm_mul_ps(iq2,jq0);
1338
1339             /* Calculate table index by multiplying r with table scale and truncate to integer */
1340             rt               = _mm_mul_ps(r20,vftabscale);
1341             vfitab           = _mm_cvttps_epi32(rt);
1342 #ifdef __XOP__
1343             vfeps            = _mm_frcz_ps(rt);
1344 #else
1345             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1346 #endif
1347             twovfeps         = _mm_add_ps(vfeps,vfeps);
1348             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1349
1350             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1351             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1352             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1353             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1354             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1355             _MM_TRANSPOSE4_PS(Y,F,G,H);
1356             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1357             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1358             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1359
1360             fscal            = felec;
1361
1362             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1363
1364              /* Update vectorial force */
1365             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1366             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1367             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1368
1369             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1370             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1371             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1372
1373             /**************************
1374              * CALCULATE INTERACTIONS *
1375              **************************/
1376
1377             r30              = _mm_mul_ps(rsq30,rinv30);
1378             r30              = _mm_andnot_ps(dummy_mask,r30);
1379
1380             /* Compute parameters for interactions between i and j atoms */
1381             qq30             = _mm_mul_ps(iq3,jq0);
1382
1383             /* Calculate table index by multiplying r with table scale and truncate to integer */
1384             rt               = _mm_mul_ps(r30,vftabscale);
1385             vfitab           = _mm_cvttps_epi32(rt);
1386 #ifdef __XOP__
1387             vfeps            = _mm_frcz_ps(rt);
1388 #else
1389             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1390 #endif
1391             twovfeps         = _mm_add_ps(vfeps,vfeps);
1392             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1393
1394             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1395             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1396             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1397             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1398             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1399             _MM_TRANSPOSE4_PS(Y,F,G,H);
1400             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1401             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1402             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1403
1404             fscal            = felec;
1405
1406             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1407
1408              /* Update vectorial force */
1409             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1410             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1411             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1412
1413             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1414             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1415             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1416
1417             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1418             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1419             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1420             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1421
1422             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1423
1424             /* Inner loop uses 181 flops */
1425         }
1426
1427         /* End of innermost loop */
1428
1429         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1430                                               f+i_coord_offset,fshift+i_shift_offset);
1431
1432         /* Increment number of inner iterations */
1433         inneriter                  += j_index_end - j_index_start;
1434
1435         /* Outer loop uses 24 flops */
1436     }
1437
1438     /* Increment number of outer iterations */
1439     outeriter        += nri;
1440
1441     /* Update outer/inner flops */
1442
1443     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*181);
1444 }