src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_avx_128_fma_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 107     real             *vftab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->ic->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_elec_vdw->data;
 130     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 135     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 136     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 171
 172         fix0             = _mm_setzero_ps();
 173         fiy0             = _mm_setzero_ps();
 174         fiz0             = _mm_setzero_ps();
 175         fix1             = _mm_setzero_ps();
 176         fiy1             = _mm_setzero_ps();
 177         fiz1             = _mm_setzero_ps();
 178         fix2             = _mm_setzero_ps();
 179         fiy2             = _mm_setzero_ps();
 180         fiz2             = _mm_setzero_ps();
 181         fix3             = _mm_setzero_ps();
 182         fiy3             = _mm_setzero_ps();
 183         fiz3             = _mm_setzero_ps();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm_setzero_ps();
 187         vvdwsum          = _mm_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200             j_coord_offsetC  = DIM*jnrC;
 201             j_coord_offsetD  = DIM*jnrD;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               x+j_coord_offsetC,x+j_coord_offsetD,
 206                                               &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm_sub_ps(ix0,jx0);
 210             dy00             = _mm_sub_ps(iy0,jy0);
 211             dz00             = _mm_sub_ps(iz0,jz0);
 212             dx10             = _mm_sub_ps(ix1,jx0);
 213             dy10             = _mm_sub_ps(iy1,jy0);
 214             dz10             = _mm_sub_ps(iz1,jz0);
 215             dx20             = _mm_sub_ps(ix2,jx0);
 216             dy20             = _mm_sub_ps(iy2,jy0);
 217             dz20             = _mm_sub_ps(iz2,jz0);
 218             dx30             = _mm_sub_ps(ix3,jx0);
 219             dy30             = _mm_sub_ps(iy3,jy0);
 220             dz30             = _mm_sub_ps(iz3,jz0);
 221
 222             /* Calculate squared distance and things based on it */
 223             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 224             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 225             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 226             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 227
 228             rinv00           = avx128fma_invsqrt_f(rsq00);
 229             rinv10           = avx128fma_invsqrt_f(rsq10);
 230             rinv20           = avx128fma_invsqrt_f(rsq20);
 231             rinv30           = avx128fma_invsqrt_f(rsq30);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 235                                                               charge+jnrC+0,charge+jnrD+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240
 241             fjx0             = _mm_setzero_ps();
 242             fjy0             = _mm_setzero_ps();
 243             fjz0             = _mm_setzero_ps();
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             r00              = _mm_mul_ps(rsq00,rinv00);
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 253                                          vdwparam+vdwioffset0+vdwjidx0B,
 254                                          vdwparam+vdwioffset0+vdwjidx0C,
 255                                          vdwparam+vdwioffset0+vdwjidx0D,
 256                                          &c6_00,&c12_00);
 257
 258             /* Calculate table index by multiplying r with table scale and truncate to integer */
 259             rt               = _mm_mul_ps(r00,vftabscale);
 260             vfitab           = _mm_cvttps_epi32(rt);
 261 #ifdef __XOP__
 262             vfeps            = _mm_frcz_ps(rt);
 263 #else
 264             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 265 #endif
 266             twovfeps         = _mm_add_ps(vfeps,vfeps);
 267             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 268
 269             /* CUBIC SPLINE TABLE DISPERSION */
 270             vfitab           = _mm_add_epi32(vfitab,ifour);
 271             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 272             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 273             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 274             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 275             _MM_TRANSPOSE4_PS(Y,F,G,H);
 276             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 277             VV               = _mm_macc_ps(vfeps,Fp,Y);
 278             vvdw6            = _mm_mul_ps(c6_00,VV);
 279             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 280             fvdw6            = _mm_mul_ps(c6_00,FF);
 281
 282             /* CUBIC SPLINE TABLE REPULSION */
 283             vfitab           = _mm_add_epi32(vfitab,ifour);
 284             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 285             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 286             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 287             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 288             _MM_TRANSPOSE4_PS(Y,F,G,H);
 289             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 290             VV               = _mm_macc_ps(vfeps,Fp,Y);
 291             vvdw12           = _mm_mul_ps(c12_00,VV);
 292             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 293             fvdw12           = _mm_mul_ps(c12_00,FF);
 294             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 295             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 299
 300             fscal            = fvdw;
 301
 302              /* Update vectorial force */
 303             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 304             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 305             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 306
 307             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 308             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 309             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             r10              = _mm_mul_ps(rsq10,rinv10);
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq10             = _mm_mul_ps(iq1,jq0);
 319
 320             /* Calculate table index by multiplying r with table scale and truncate to integer */
 321             rt               = _mm_mul_ps(r10,vftabscale);
 322             vfitab           = _mm_cvttps_epi32(rt);
 323 #ifdef __XOP__
 324             vfeps            = _mm_frcz_ps(rt);
 325 #else
 326             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 327 #endif
 328             twovfeps         = _mm_add_ps(vfeps,vfeps);
 329             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 330
 331             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 332             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 333             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 334             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 335             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 336             _MM_TRANSPOSE4_PS(Y,F,G,H);
 337             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 338             VV               = _mm_macc_ps(vfeps,Fp,Y);
 339             velec            = _mm_mul_ps(qq10,VV);
 340             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 341             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm_add_ps(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348              /* Update vectorial force */
 349             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 350             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 351             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 352
 353             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 354             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 355             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             r20              = _mm_mul_ps(rsq20,rinv20);
 362
 363             /* Compute parameters for interactions between i and j atoms */
 364             qq20             = _mm_mul_ps(iq2,jq0);
 365
 366             /* Calculate table index by multiplying r with table scale and truncate to integer */
 367             rt               = _mm_mul_ps(r20,vftabscale);
 368             vfitab           = _mm_cvttps_epi32(rt);
 369 #ifdef __XOP__
 370             vfeps            = _mm_frcz_ps(rt);
 371 #else
 372             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 373 #endif
 374             twovfeps         = _mm_add_ps(vfeps,vfeps);
 375             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 376
 377             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 378             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 379             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 380             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 381             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 382             _MM_TRANSPOSE4_PS(Y,F,G,H);
 383             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 384             VV               = _mm_macc_ps(vfeps,Fp,Y);
 385             velec            = _mm_mul_ps(qq20,VV);
 386             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 387             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velecsum         = _mm_add_ps(velecsum,velec);
 391
 392             fscal            = felec;
 393
 394              /* Update vectorial force */
 395             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 396             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 397             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 398
 399             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 400             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 401             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 402
 403             /**************************
 404              * CALCULATE INTERACTIONS *
 405              **************************/
 406
 407             r30              = _mm_mul_ps(rsq30,rinv30);
 408
 409             /* Compute parameters for interactions between i and j atoms */
 410             qq30             = _mm_mul_ps(iq3,jq0);
 411
 412             /* Calculate table index by multiplying r with table scale and truncate to integer */
 413             rt               = _mm_mul_ps(r30,vftabscale);
 414             vfitab           = _mm_cvttps_epi32(rt);
 415 #ifdef __XOP__
 416             vfeps            = _mm_frcz_ps(rt);
 417 #else
 418             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 419 #endif
 420             twovfeps         = _mm_add_ps(vfeps,vfeps);
 421             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 422
 423             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 424             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 425             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 426             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 427             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 428             _MM_TRANSPOSE4_PS(Y,F,G,H);
 429             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 430             VV               = _mm_macc_ps(vfeps,Fp,Y);
 431             velec            = _mm_mul_ps(qq30,VV);
 432             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 433             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 434
 435             /* Update potential sum for this i atom from the interaction with this j atom. */
 436             velecsum         = _mm_add_ps(velecsum,velec);
 437
 438             fscal            = felec;
 439
 440              /* Update vectorial force */
 441             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 442             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 443             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 444
 445             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 446             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 447             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 448
 449             fjptrA             = f+j_coord_offsetA;
 450             fjptrB             = f+j_coord_offsetB;
 451             fjptrC             = f+j_coord_offsetC;
 452             fjptrD             = f+j_coord_offsetD;
 453
 454             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 455
 456             /* Inner loop uses 197 flops */
 457         }
 458
 459         if(jidx<j_index_end)
 460         {
 461
 462             /* Get j neighbor index, and coordinate index */
 463             jnrlistA         = jjnr[jidx];
 464             jnrlistB         = jjnr[jidx+1];
 465             jnrlistC         = jjnr[jidx+2];
 466             jnrlistD         = jjnr[jidx+3];
 467             /* Sign of each element will be negative for non-real atoms.
 468              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 469              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 470              */
 471             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 472             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 473             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 474             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 475             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 476             j_coord_offsetA  = DIM*jnrA;
 477             j_coord_offsetB  = DIM*jnrB;
 478             j_coord_offsetC  = DIM*jnrC;
 479             j_coord_offsetD  = DIM*jnrD;
 480
 481             /* load j atom coordinates */
 482             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 483                                               x+j_coord_offsetC,x+j_coord_offsetD,
 484                                               &jx0,&jy0,&jz0);
 485
 486             /* Calculate displacement vector */
 487             dx00             = _mm_sub_ps(ix0,jx0);
 488             dy00             = _mm_sub_ps(iy0,jy0);
 489             dz00             = _mm_sub_ps(iz0,jz0);
 490             dx10             = _mm_sub_ps(ix1,jx0);
 491             dy10             = _mm_sub_ps(iy1,jy0);
 492             dz10             = _mm_sub_ps(iz1,jz0);
 493             dx20             = _mm_sub_ps(ix2,jx0);
 494             dy20             = _mm_sub_ps(iy2,jy0);
 495             dz20             = _mm_sub_ps(iz2,jz0);
 496             dx30             = _mm_sub_ps(ix3,jx0);
 497             dy30             = _mm_sub_ps(iy3,jy0);
 498             dz30             = _mm_sub_ps(iz3,jz0);
 499
 500             /* Calculate squared distance and things based on it */
 501             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 502             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 503             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 504             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 505
 506             rinv00           = avx128fma_invsqrt_f(rsq00);
 507             rinv10           = avx128fma_invsqrt_f(rsq10);
 508             rinv20           = avx128fma_invsqrt_f(rsq20);
 509             rinv30           = avx128fma_invsqrt_f(rsq30);
 510
 511             /* Load parameters for j particles */
 512             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 513                                                               charge+jnrC+0,charge+jnrD+0);
 514             vdwjidx0A        = 2*vdwtype[jnrA+0];
 515             vdwjidx0B        = 2*vdwtype[jnrB+0];
 516             vdwjidx0C        = 2*vdwtype[jnrC+0];
 517             vdwjidx0D        = 2*vdwtype[jnrD+0];
 518
 519             fjx0             = _mm_setzero_ps();
 520             fjy0             = _mm_setzero_ps();
 521             fjz0             = _mm_setzero_ps();
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             r00              = _mm_mul_ps(rsq00,rinv00);
 528             r00              = _mm_andnot_ps(dummy_mask,r00);
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 532                                          vdwparam+vdwioffset0+vdwjidx0B,
 533                                          vdwparam+vdwioffset0+vdwjidx0C,
 534                                          vdwparam+vdwioffset0+vdwjidx0D,
 535                                          &c6_00,&c12_00);
 536
 537             /* Calculate table index by multiplying r with table scale and truncate to integer */
 538             rt               = _mm_mul_ps(r00,vftabscale);
 539             vfitab           = _mm_cvttps_epi32(rt);
 540 #ifdef __XOP__
 541             vfeps            = _mm_frcz_ps(rt);
 542 #else
 543             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 544 #endif
 545             twovfeps         = _mm_add_ps(vfeps,vfeps);
 546             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 547
 548             /* CUBIC SPLINE TABLE DISPERSION */
 549             vfitab           = _mm_add_epi32(vfitab,ifour);
 550             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 551             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 552             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 553             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 554             _MM_TRANSPOSE4_PS(Y,F,G,H);
 555             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 556             VV               = _mm_macc_ps(vfeps,Fp,Y);
 557             vvdw6            = _mm_mul_ps(c6_00,VV);
 558             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 559             fvdw6            = _mm_mul_ps(c6_00,FF);
 560
 561             /* CUBIC SPLINE TABLE REPULSION */
 562             vfitab           = _mm_add_epi32(vfitab,ifour);
 563             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 564             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 565             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 566             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 567             _MM_TRANSPOSE4_PS(Y,F,G,H);
 568             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 569             VV               = _mm_macc_ps(vfeps,Fp,Y);
 570             vvdw12           = _mm_mul_ps(c12_00,VV);
 571             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 572             fvdw12           = _mm_mul_ps(c12_00,FF);
 573             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 574             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 575
 576             /* Update potential sum for this i atom from the interaction with this j atom. */
 577             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 578             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 579
 580             fscal            = fvdw;
 581
 582             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 583
 584              /* Update vectorial force */
 585             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 586             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 587             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 588
 589             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 590             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 591             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 592
 593             /**************************
 594              * CALCULATE INTERACTIONS *
 595              **************************/
 596
 597             r10              = _mm_mul_ps(rsq10,rinv10);
 598             r10              = _mm_andnot_ps(dummy_mask,r10);
 599
 600             /* Compute parameters for interactions between i and j atoms */
 601             qq10             = _mm_mul_ps(iq1,jq0);
 602
 603             /* Calculate table index by multiplying r with table scale and truncate to integer */
 604             rt               = _mm_mul_ps(r10,vftabscale);
 605             vfitab           = _mm_cvttps_epi32(rt);
 606 #ifdef __XOP__
 607             vfeps            = _mm_frcz_ps(rt);
 608 #else
 609             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 610 #endif
 611             twovfeps         = _mm_add_ps(vfeps,vfeps);
 612             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 613
 614             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 615             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 616             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 617             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 618             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 619             _MM_TRANSPOSE4_PS(Y,F,G,H);
 620             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 621             VV               = _mm_macc_ps(vfeps,Fp,Y);
 622             velec            = _mm_mul_ps(qq10,VV);
 623             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 624             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm_andnot_ps(dummy_mask,velec);
 628             velecsum         = _mm_add_ps(velecsum,velec);
 629
 630             fscal            = felec;
 631
 632             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 633
 634              /* Update vectorial force */
 635             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 636             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 637             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 638
 639             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 640             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 641             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             r20              = _mm_mul_ps(rsq20,rinv20);
 648             r20              = _mm_andnot_ps(dummy_mask,r20);
 649
 650             /* Compute parameters for interactions between i and j atoms */
 651             qq20             = _mm_mul_ps(iq2,jq0);
 652
 653             /* Calculate table index by multiplying r with table scale and truncate to integer */
 654             rt               = _mm_mul_ps(r20,vftabscale);
 655             vfitab           = _mm_cvttps_epi32(rt);
 656 #ifdef __XOP__
 657             vfeps            = _mm_frcz_ps(rt);
 658 #else
 659             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 660 #endif
 661             twovfeps         = _mm_add_ps(vfeps,vfeps);
 662             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 663
 664             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 665             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 666             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 667             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 668             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 669             _MM_TRANSPOSE4_PS(Y,F,G,H);
 670             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 671             VV               = _mm_macc_ps(vfeps,Fp,Y);
 672             velec            = _mm_mul_ps(qq20,VV);
 673             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 674             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 675
 676             /* Update potential sum for this i atom from the interaction with this j atom. */
 677             velec            = _mm_andnot_ps(dummy_mask,velec);
 678             velecsum         = _mm_add_ps(velecsum,velec);
 679
 680             fscal            = felec;
 681
 682             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 683
 684              /* Update vectorial force */
 685             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 686             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 687             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 688
 689             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 690             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 691             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 692
 693             /**************************
 694              * CALCULATE INTERACTIONS *
 695              **************************/
 696
 697             r30              = _mm_mul_ps(rsq30,rinv30);
 698             r30              = _mm_andnot_ps(dummy_mask,r30);
 699
 700             /* Compute parameters for interactions between i and j atoms */
 701             qq30             = _mm_mul_ps(iq3,jq0);
 702
 703             /* Calculate table index by multiplying r with table scale and truncate to integer */
 704             rt               = _mm_mul_ps(r30,vftabscale);
 705             vfitab           = _mm_cvttps_epi32(rt);
 706 #ifdef __XOP__
 707             vfeps            = _mm_frcz_ps(rt);
 708 #else
 709             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 710 #endif
 711             twovfeps         = _mm_add_ps(vfeps,vfeps);
 712             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 713
 714             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 715             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 716             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 717             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 718             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 719             _MM_TRANSPOSE4_PS(Y,F,G,H);
 720             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 721             VV               = _mm_macc_ps(vfeps,Fp,Y);
 722             velec            = _mm_mul_ps(qq30,VV);
 723             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 724             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 725
 726             /* Update potential sum for this i atom from the interaction with this j atom. */
 727             velec            = _mm_andnot_ps(dummy_mask,velec);
 728             velecsum         = _mm_add_ps(velecsum,velec);
 729
 730             fscal            = felec;
 731
 732             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 733
 734              /* Update vectorial force */
 735             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 736             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 737             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 738
 739             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 740             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 741             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 742
 743             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 744             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 745             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 746             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 747
 748             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 749
 750             /* Inner loop uses 201 flops */
 751         }
 752
 753         /* End of innermost loop */
 754
 755         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 756                                               f+i_coord_offset,fshift+i_shift_offset);
 757
 758         ggid                        = gid[iidx];
 759         /* Update potential energies */
 760         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 761         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 762
 763         /* Increment number of inner iterations */
 764         inneriter                  += j_index_end - j_index_start;
 765
 766         /* Outer loop uses 26 flops */
 767     }
 768
 769     /* Increment number of outer iterations */
 770     outeriter        += nri;
 771
 772     /* Update outer/inner flops */
 773
 774     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*201);
 775 }
 776 /*
 777  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 778  * Electrostatics interaction: CubicSplineTable
 779  * VdW interaction:            CubicSplineTable
 780  * Geometry:                   Water4-Particle
 781  * Calculate force/pot:        Force
 782  */
 783 void
 784 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 785                     (t_nblist                    * gmx_restrict       nlist,
 786                      rvec                        * gmx_restrict          xx,
 787                      rvec                        * gmx_restrict          ff,
 788                      struct t_forcerec           * gmx_restrict          fr,
 789                      t_mdatoms                   * gmx_restrict     mdatoms,
 790                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 791                      t_nrnb                      * gmx_restrict        nrnb)
 792 {
 793     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 794      * just 0 for non-waters.
 795      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 796      * jnr indices corresponding to data put in the four positions in the SIMD register.
 797      */
 798     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 799     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 800     int              jnrA,jnrB,jnrC,jnrD;
 801     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 802     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 803     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 804     real             rcutoff_scalar;
 805     real             *shiftvec,*fshift,*x,*f;
 806     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 807     real             scratch[4*DIM];
 808     __m128           fscal,rcutoff,rcutoff2,jidxall;
 809     int              vdwioffset0;
 810     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 811     int              vdwioffset1;
 812     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 813     int              vdwioffset2;
 814     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 815     int              vdwioffset3;
 816     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 817     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 818     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 819     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 820     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 821     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 822     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 823     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 824     real             *charge;
 825     int              nvdwtype;
 826     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 827     int              *vdwtype;
 828     real             *vdwparam;
 829     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 830     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 831     __m128i          vfitab;
 832     __m128i          ifour       = _mm_set1_epi32(4);
 833     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 834     real             *vftab;
 835     __m128           dummy_mask,cutoff_mask;
 836     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 837     __m128           one     = _mm_set1_ps(1.0);
 838     __m128           two     = _mm_set1_ps(2.0);
 839     x                = xx[0];
 840     f                = ff[0];
 841
 842     nri              = nlist->nri;
 843     iinr             = nlist->iinr;
 844     jindex           = nlist->jindex;
 845     jjnr             = nlist->jjnr;
 846     shiftidx         = nlist->shift;
 847     gid              = nlist->gid;
 848     shiftvec         = fr->shift_vec[0];
 849     fshift           = fr->fshift[0];
 850     facel            = _mm_set1_ps(fr->ic->epsfac);
 851     charge           = mdatoms->chargeA;
 852     nvdwtype         = fr->ntype;
 853     vdwparam         = fr->nbfp;
 854     vdwtype          = mdatoms->typeA;
 855
 856     vftab            = kernel_data->table_elec_vdw->data;
 857     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 858
 859     /* Setup water-specific parameters */
 860     inr              = nlist->iinr[0];
 861     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 862     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 863     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 864     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 865
 866     /* Avoid stupid compiler warnings */
 867     jnrA = jnrB = jnrC = jnrD = 0;
 868     j_coord_offsetA = 0;
 869     j_coord_offsetB = 0;
 870     j_coord_offsetC = 0;
 871     j_coord_offsetD = 0;
 872
 873     outeriter        = 0;
 874     inneriter        = 0;
 875
 876     for(iidx=0;iidx<4*DIM;iidx++)
 877     {
 878         scratch[iidx] = 0.0;
 879     }
 880
 881     /* Start outer loop over neighborlists */
 882     for(iidx=0; iidx<nri; iidx++)
 883     {
 884         /* Load shift vector for this list */
 885         i_shift_offset   = DIM*shiftidx[iidx];
 886
 887         /* Load limits for loop over neighbors */
 888         j_index_start    = jindex[iidx];
 889         j_index_end      = jindex[iidx+1];
 890
 891         /* Get outer coordinate index */
 892         inr              = iinr[iidx];
 893         i_coord_offset   = DIM*inr;
 894
 895         /* Load i particle coords and add shift vector */
 896         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 897                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 898
 899         fix0             = _mm_setzero_ps();
 900         fiy0             = _mm_setzero_ps();
 901         fiz0             = _mm_setzero_ps();
 902         fix1             = _mm_setzero_ps();
 903         fiy1             = _mm_setzero_ps();
 904         fiz1             = _mm_setzero_ps();
 905         fix2             = _mm_setzero_ps();
 906         fiy2             = _mm_setzero_ps();
 907         fiz2             = _mm_setzero_ps();
 908         fix3             = _mm_setzero_ps();
 909         fiy3             = _mm_setzero_ps();
 910         fiz3             = _mm_setzero_ps();
 911
 912         /* Start inner kernel loop */
 913         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 914         {
 915
 916             /* Get j neighbor index, and coordinate index */
 917             jnrA             = jjnr[jidx];
 918             jnrB             = jjnr[jidx+1];
 919             jnrC             = jjnr[jidx+2];
 920             jnrD             = jjnr[jidx+3];
 921             j_coord_offsetA  = DIM*jnrA;
 922             j_coord_offsetB  = DIM*jnrB;
 923             j_coord_offsetC  = DIM*jnrC;
 924             j_coord_offsetD  = DIM*jnrD;
 925
 926             /* load j atom coordinates */
 927             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 928                                               x+j_coord_offsetC,x+j_coord_offsetD,
 929                                               &jx0,&jy0,&jz0);
 930
 931             /* Calculate displacement vector */
 932             dx00             = _mm_sub_ps(ix0,jx0);
 933             dy00             = _mm_sub_ps(iy0,jy0);
 934             dz00             = _mm_sub_ps(iz0,jz0);
 935             dx10             = _mm_sub_ps(ix1,jx0);
 936             dy10             = _mm_sub_ps(iy1,jy0);
 937             dz10             = _mm_sub_ps(iz1,jz0);
 938             dx20             = _mm_sub_ps(ix2,jx0);
 939             dy20             = _mm_sub_ps(iy2,jy0);
 940             dz20             = _mm_sub_ps(iz2,jz0);
 941             dx30             = _mm_sub_ps(ix3,jx0);
 942             dy30             = _mm_sub_ps(iy3,jy0);
 943             dz30             = _mm_sub_ps(iz3,jz0);
 944
 945             /* Calculate squared distance and things based on it */
 946             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 947             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 948             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 949             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 950
 951             rinv00           = avx128fma_invsqrt_f(rsq00);
 952             rinv10           = avx128fma_invsqrt_f(rsq10);
 953             rinv20           = avx128fma_invsqrt_f(rsq20);
 954             rinv30           = avx128fma_invsqrt_f(rsq30);
 955
 956             /* Load parameters for j particles */
 957             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 958                                                               charge+jnrC+0,charge+jnrD+0);
 959             vdwjidx0A        = 2*vdwtype[jnrA+0];
 960             vdwjidx0B        = 2*vdwtype[jnrB+0];
 961             vdwjidx0C        = 2*vdwtype[jnrC+0];
 962             vdwjidx0D        = 2*vdwtype[jnrD+0];
 963
 964             fjx0             = _mm_setzero_ps();
 965             fjy0             = _mm_setzero_ps();
 966             fjz0             = _mm_setzero_ps();
 967
 968             /**************************
 969              * CALCULATE INTERACTIONS *
 970              **************************/
 971
 972             r00              = _mm_mul_ps(rsq00,rinv00);
 973
 974             /* Compute parameters for interactions between i and j atoms */
 975             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 976                                          vdwparam+vdwioffset0+vdwjidx0B,
 977                                          vdwparam+vdwioffset0+vdwjidx0C,
 978                                          vdwparam+vdwioffset0+vdwjidx0D,
 979                                          &c6_00,&c12_00);
 980
 981             /* Calculate table index by multiplying r with table scale and truncate to integer */
 982             rt               = _mm_mul_ps(r00,vftabscale);
 983             vfitab           = _mm_cvttps_epi32(rt);
 984 #ifdef __XOP__
 985             vfeps            = _mm_frcz_ps(rt);
 986 #else
 987             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 988 #endif
 989             twovfeps         = _mm_add_ps(vfeps,vfeps);
 990             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 991
 992             /* CUBIC SPLINE TABLE DISPERSION */
 993             vfitab           = _mm_add_epi32(vfitab,ifour);
 994             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 995             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 996             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 997             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 998             _MM_TRANSPOSE4_PS(Y,F,G,H);
 999             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1000             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1001             fvdw6            = _mm_mul_ps(c6_00,FF);
1002
1003             /* CUBIC SPLINE TABLE REPULSION */
1004             vfitab           = _mm_add_epi32(vfitab,ifour);
1005             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1006             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1007             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1008             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1009             _MM_TRANSPOSE4_PS(Y,F,G,H);
1010             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1011             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1012             fvdw12           = _mm_mul_ps(c12_00,FF);
1013             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1014
1015             fscal            = fvdw;
1016
1017              /* Update vectorial force */
1018             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1019             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1020             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1021
1022             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1023             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1024             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1025
1026             /**************************
1027              * CALCULATE INTERACTIONS *
1028              **************************/
1029
1030             r10              = _mm_mul_ps(rsq10,rinv10);
1031
1032             /* Compute parameters for interactions between i and j atoms */
1033             qq10             = _mm_mul_ps(iq1,jq0);
1034
1035             /* Calculate table index by multiplying r with table scale and truncate to integer */
1036             rt               = _mm_mul_ps(r10,vftabscale);
1037             vfitab           = _mm_cvttps_epi32(rt);
1038 #ifdef __XOP__
1039             vfeps            = _mm_frcz_ps(rt);
1040 #else
1041             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1042 #endif
1043             twovfeps         = _mm_add_ps(vfeps,vfeps);
1044             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1045
1046             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1047             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1048             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1049             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1050             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1051             _MM_TRANSPOSE4_PS(Y,F,G,H);
1052             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1053             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1054             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1055
1056             fscal            = felec;
1057
1058              /* Update vectorial force */
1059             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1060             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1061             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1062
1063             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1064             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1065             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             r20              = _mm_mul_ps(rsq20,rinv20);
1072
1073             /* Compute parameters for interactions between i and j atoms */
1074             qq20             = _mm_mul_ps(iq2,jq0);
1075
1076             /* Calculate table index by multiplying r with table scale and truncate to integer */
1077             rt               = _mm_mul_ps(r20,vftabscale);
1078             vfitab           = _mm_cvttps_epi32(rt);
1079 #ifdef __XOP__
1080             vfeps            = _mm_frcz_ps(rt);
1081 #else
1082             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1083 #endif
1084             twovfeps         = _mm_add_ps(vfeps,vfeps);
1085             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1086
1087             /* CUBIC SPLINE TABLE ELECTROSTATICS */
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             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1096
1097             fscal            = felec;
1098
1099              /* Update vectorial force */
1100             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1101             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1102             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1103
1104             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1105             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1106             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             r30              = _mm_mul_ps(rsq30,rinv30);
1113
1114             /* Compute parameters for interactions between i and j atoms */
1115             qq30             = _mm_mul_ps(iq3,jq0);
1116
1117             /* Calculate table index by multiplying r with table scale and truncate to integer */
1118             rt               = _mm_mul_ps(r30,vftabscale);
1119             vfitab           = _mm_cvttps_epi32(rt);
1120 #ifdef __XOP__
1121             vfeps            = _mm_frcz_ps(rt);
1122 #else
1123             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1124 #endif
1125             twovfeps         = _mm_add_ps(vfeps,vfeps);
1126             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1127
1128             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1129             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1130             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1131             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1132             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1133             _MM_TRANSPOSE4_PS(Y,F,G,H);
1134             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1135             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1136             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1137
1138             fscal            = felec;
1139
1140              /* Update vectorial force */
1141             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1142             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1143             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1144
1145             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1146             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1147             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1148
1149             fjptrA             = f+j_coord_offsetA;
1150             fjptrB             = f+j_coord_offsetB;
1151             fjptrC             = f+j_coord_offsetC;
1152             fjptrD             = f+j_coord_offsetD;
1153
1154             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1155
1156             /* Inner loop uses 177 flops */
1157         }
1158
1159         if(jidx<j_index_end)
1160         {
1161
1162             /* Get j neighbor index, and coordinate index */
1163             jnrlistA         = jjnr[jidx];
1164             jnrlistB         = jjnr[jidx+1];
1165             jnrlistC         = jjnr[jidx+2];
1166             jnrlistD         = jjnr[jidx+3];
1167             /* Sign of each element will be negative for non-real atoms.
1168              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1169              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1170              */
1171             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1172             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1173             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1174             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1175             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1176             j_coord_offsetA  = DIM*jnrA;
1177             j_coord_offsetB  = DIM*jnrB;
1178             j_coord_offsetC  = DIM*jnrC;
1179             j_coord_offsetD  = DIM*jnrD;
1180
1181             /* load j atom coordinates */
1182             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1183                                               x+j_coord_offsetC,x+j_coord_offsetD,
1184                                               &jx0,&jy0,&jz0);
1185
1186             /* Calculate displacement vector */
1187             dx00             = _mm_sub_ps(ix0,jx0);
1188             dy00             = _mm_sub_ps(iy0,jy0);
1189             dz00             = _mm_sub_ps(iz0,jz0);
1190             dx10             = _mm_sub_ps(ix1,jx0);
1191             dy10             = _mm_sub_ps(iy1,jy0);
1192             dz10             = _mm_sub_ps(iz1,jz0);
1193             dx20             = _mm_sub_ps(ix2,jx0);
1194             dy20             = _mm_sub_ps(iy2,jy0);
1195             dz20             = _mm_sub_ps(iz2,jz0);
1196             dx30             = _mm_sub_ps(ix3,jx0);
1197             dy30             = _mm_sub_ps(iy3,jy0);
1198             dz30             = _mm_sub_ps(iz3,jz0);
1199
1200             /* Calculate squared distance and things based on it */
1201             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1202             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1203             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1204             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1205
1206             rinv00           = avx128fma_invsqrt_f(rsq00);
1207             rinv10           = avx128fma_invsqrt_f(rsq10);
1208             rinv20           = avx128fma_invsqrt_f(rsq20);
1209             rinv30           = avx128fma_invsqrt_f(rsq30);
1210
1211             /* Load parameters for j particles */
1212             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1213                                                               charge+jnrC+0,charge+jnrD+0);
1214             vdwjidx0A        = 2*vdwtype[jnrA+0];
1215             vdwjidx0B        = 2*vdwtype[jnrB+0];
1216             vdwjidx0C        = 2*vdwtype[jnrC+0];
1217             vdwjidx0D        = 2*vdwtype[jnrD+0];
1218
1219             fjx0             = _mm_setzero_ps();
1220             fjy0             = _mm_setzero_ps();
1221             fjz0             = _mm_setzero_ps();
1222
1223             /**************************
1224              * CALCULATE INTERACTIONS *
1225              **************************/
1226
1227             r00              = _mm_mul_ps(rsq00,rinv00);
1228             r00              = _mm_andnot_ps(dummy_mask,r00);
1229
1230             /* Compute parameters for interactions between i and j atoms */
1231             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1232                                          vdwparam+vdwioffset0+vdwjidx0B,
1233                                          vdwparam+vdwioffset0+vdwjidx0C,
1234                                          vdwparam+vdwioffset0+vdwjidx0D,
1235                                          &c6_00,&c12_00);
1236
1237             /* Calculate table index by multiplying r with table scale and truncate to integer */
1238             rt               = _mm_mul_ps(r00,vftabscale);
1239             vfitab           = _mm_cvttps_epi32(rt);
1240 #ifdef __XOP__
1241             vfeps            = _mm_frcz_ps(rt);
1242 #else
1243             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1244 #endif
1245             twovfeps         = _mm_add_ps(vfeps,vfeps);
1246             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1247
1248             /* CUBIC SPLINE TABLE DISPERSION */
1249             vfitab           = _mm_add_epi32(vfitab,ifour);
1250             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1251             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1252             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1253             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1254             _MM_TRANSPOSE4_PS(Y,F,G,H);
1255             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1256             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1257             fvdw6            = _mm_mul_ps(c6_00,FF);
1258
1259             /* CUBIC SPLINE TABLE REPULSION */
1260             vfitab           = _mm_add_epi32(vfitab,ifour);
1261             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1262             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1263             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1264             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1265             _MM_TRANSPOSE4_PS(Y,F,G,H);
1266             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1267             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1268             fvdw12           = _mm_mul_ps(c12_00,FF);
1269             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1270
1271             fscal            = fvdw;
1272
1273             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1274
1275              /* Update vectorial force */
1276             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1277             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1278             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1279
1280             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1281             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1282             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1283
1284             /**************************
1285              * CALCULATE INTERACTIONS *
1286              **************************/
1287
1288             r10              = _mm_mul_ps(rsq10,rinv10);
1289             r10              = _mm_andnot_ps(dummy_mask,r10);
1290
1291             /* Compute parameters for interactions between i and j atoms */
1292             qq10             = _mm_mul_ps(iq1,jq0);
1293
1294             /* Calculate table index by multiplying r with table scale and truncate to integer */
1295             rt               = _mm_mul_ps(r10,vftabscale);
1296             vfitab           = _mm_cvttps_epi32(rt);
1297 #ifdef __XOP__
1298             vfeps            = _mm_frcz_ps(rt);
1299 #else
1300             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1301 #endif
1302             twovfeps         = _mm_add_ps(vfeps,vfeps);
1303             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1304
1305             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1306             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1307             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1308             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1309             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1310             _MM_TRANSPOSE4_PS(Y,F,G,H);
1311             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1312             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1313             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1314
1315             fscal            = felec;
1316
1317             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1318
1319              /* Update vectorial force */
1320             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1321             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1322             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1323
1324             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1325             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1326             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1327
1328             /**************************
1329              * CALCULATE INTERACTIONS *
1330              **************************/
1331
1332             r20              = _mm_mul_ps(rsq20,rinv20);
1333             r20              = _mm_andnot_ps(dummy_mask,r20);
1334
1335             /* Compute parameters for interactions between i and j atoms */
1336             qq20             = _mm_mul_ps(iq2,jq0);
1337
1338             /* Calculate table index by multiplying r with table scale and truncate to integer */
1339             rt               = _mm_mul_ps(r20,vftabscale);
1340             vfitab           = _mm_cvttps_epi32(rt);
1341 #ifdef __XOP__
1342             vfeps            = _mm_frcz_ps(rt);
1343 #else
1344             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1345 #endif
1346             twovfeps         = _mm_add_ps(vfeps,vfeps);
1347             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1348
1349             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1350             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1351             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1352             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1353             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1354             _MM_TRANSPOSE4_PS(Y,F,G,H);
1355             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1356             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1357             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1358
1359             fscal            = felec;
1360
1361             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1362
1363              /* Update vectorial force */
1364             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1365             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1366             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1367
1368             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1369             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1370             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1371
1372             /**************************
1373              * CALCULATE INTERACTIONS *
1374              **************************/
1375
1376             r30              = _mm_mul_ps(rsq30,rinv30);
1377             r30              = _mm_andnot_ps(dummy_mask,r30);
1378
1379             /* Compute parameters for interactions between i and j atoms */
1380             qq30             = _mm_mul_ps(iq3,jq0);
1381
1382             /* Calculate table index by multiplying r with table scale and truncate to integer */
1383             rt               = _mm_mul_ps(r30,vftabscale);
1384             vfitab           = _mm_cvttps_epi32(rt);
1385 #ifdef __XOP__
1386             vfeps            = _mm_frcz_ps(rt);
1387 #else
1388             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1389 #endif
1390             twovfeps         = _mm_add_ps(vfeps,vfeps);
1391             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1392
1393             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1394             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1395             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1396             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1397             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1398             _MM_TRANSPOSE4_PS(Y,F,G,H);
1399             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1400             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1401             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1402
1403             fscal            = felec;
1404
1405             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1406
1407              /* Update vectorial force */
1408             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1409             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1410             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1411
1412             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1413             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1414             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1415
1416             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1417             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1418             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1419             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1420
1421             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1422
1423             /* Inner loop uses 181 flops */
1424         }
1425
1426         /* End of innermost loop */
1427
1428         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1429                                               f+i_coord_offset,fshift+i_shift_offset);
1430
1431         /* Increment number of inner iterations */
1432         inneriter                  += j_index_end - j_index_start;
1433
1434         /* Outer loop uses 24 flops */
1435     }
1436
1437     /* Increment number of outer iterations */
1438     outeriter        += nri;
1439
1440     /* Update outer/inner flops */
1441
1442     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*181);
1443 }