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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128i          vfitab;
 108     __m128i          ifour       = _mm_set1_epi32(4);
 109     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 110     real             *vftab;
 111     __m128           dummy_mask,cutoff_mask;
 112     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 113     __m128           one     = _mm_set1_ps(1.0);
 114     __m128           two     = _mm_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_ps(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_elec_vdw->data;
 133     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 138     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 139     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 174
 175         fix0             = _mm_setzero_ps();
 176         fiy0             = _mm_setzero_ps();
 177         fiz0             = _mm_setzero_ps();
 178         fix1             = _mm_setzero_ps();
 179         fiy1             = _mm_setzero_ps();
 180         fiz1             = _mm_setzero_ps();
 181         fix2             = _mm_setzero_ps();
 182         fiy2             = _mm_setzero_ps();
 183         fiz2             = _mm_setzero_ps();
 184         fix3             = _mm_setzero_ps();
 185         fiy3             = _mm_setzero_ps();
 186         fiz3             = _mm_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm_setzero_ps();
 190         vvdwsum          = _mm_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               x+j_coord_offsetC,x+j_coord_offsetD,
 209                                               &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm_sub_ps(ix0,jx0);
 213             dy00             = _mm_sub_ps(iy0,jy0);
 214             dz00             = _mm_sub_ps(iz0,jz0);
 215             dx10             = _mm_sub_ps(ix1,jx0);
 216             dy10             = _mm_sub_ps(iy1,jy0);
 217             dz10             = _mm_sub_ps(iz1,jz0);
 218             dx20             = _mm_sub_ps(ix2,jx0);
 219             dy20             = _mm_sub_ps(iy2,jy0);
 220             dz20             = _mm_sub_ps(iz2,jz0);
 221             dx30             = _mm_sub_ps(ix3,jx0);
 222             dy30             = _mm_sub_ps(iy3,jy0);
 223             dz30             = _mm_sub_ps(iz3,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 227             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 228             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 229             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 230
 231             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 232             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 233             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 234             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 235
 236             /* Load parameters for j particles */
 237             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 238                                                               charge+jnrC+0,charge+jnrD+0);
 239             vdwjidx0A        = 2*vdwtype[jnrA+0];
 240             vdwjidx0B        = 2*vdwtype[jnrB+0];
 241             vdwjidx0C        = 2*vdwtype[jnrC+0];
 242             vdwjidx0D        = 2*vdwtype[jnrD+0];
 243
 244             fjx0             = _mm_setzero_ps();
 245             fjy0             = _mm_setzero_ps();
 246             fjz0             = _mm_setzero_ps();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             r00              = _mm_mul_ps(rsq00,rinv00);
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 256                                          vdwparam+vdwioffset0+vdwjidx0B,
 257                                          vdwparam+vdwioffset0+vdwjidx0C,
 258                                          vdwparam+vdwioffset0+vdwjidx0D,
 259                                          &c6_00,&c12_00);
 260
 261             /* Calculate table index by multiplying r with table scale and truncate to integer */
 262             rt               = _mm_mul_ps(r00,vftabscale);
 263             vfitab           = _mm_cvttps_epi32(rt);
 264 #ifdef __XOP__
 265             vfeps            = _mm_frcz_ps(rt);
 266 #else
 267             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 268 #endif
 269             twovfeps         = _mm_add_ps(vfeps,vfeps);
 270             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 271
 272             /* CUBIC SPLINE TABLE DISPERSION */
 273             vfitab           = _mm_add_epi32(vfitab,ifour);
 274             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 275             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 276             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 277             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 278             _MM_TRANSPOSE4_PS(Y,F,G,H);
 279             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 280             VV               = _mm_macc_ps(vfeps,Fp,Y);
 281             vvdw6            = _mm_mul_ps(c6_00,VV);
 282             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 283             fvdw6            = _mm_mul_ps(c6_00,FF);
 284
 285             /* CUBIC SPLINE TABLE REPULSION */
 286             vfitab           = _mm_add_epi32(vfitab,ifour);
 287             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 288             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 289             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 290             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 291             _MM_TRANSPOSE4_PS(Y,F,G,H);
 292             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 293             VV               = _mm_macc_ps(vfeps,Fp,Y);
 294             vvdw12           = _mm_mul_ps(c12_00,VV);
 295             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 296             fvdw12           = _mm_mul_ps(c12_00,FF);
 297             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 298             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 302
 303             fscal            = fvdw;
 304
 305              /* Update vectorial force */
 306             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 307             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 308             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 309
 310             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 311             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 312             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r10              = _mm_mul_ps(rsq10,rinv10);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq10             = _mm_mul_ps(iq1,jq0);
 322
 323             /* Calculate table index by multiplying r with table scale and truncate to integer */
 324             rt               = _mm_mul_ps(r10,vftabscale);
 325             vfitab           = _mm_cvttps_epi32(rt);
 326 #ifdef __XOP__
 327             vfeps            = _mm_frcz_ps(rt);
 328 #else
 329             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 330 #endif
 331             twovfeps         = _mm_add_ps(vfeps,vfeps);
 332             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 333
 334             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 335             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 336             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 337             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 338             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 339             _MM_TRANSPOSE4_PS(Y,F,G,H);
 340             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 341             VV               = _mm_macc_ps(vfeps,Fp,Y);
 342             velec            = _mm_mul_ps(qq10,VV);
 343             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 344             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351              /* Update vectorial force */
 352             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 353             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 354             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 355
 356             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 357             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 358             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             r20              = _mm_mul_ps(rsq20,rinv20);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq20             = _mm_mul_ps(iq2,jq0);
 368
 369             /* Calculate table index by multiplying r with table scale and truncate to integer */
 370             rt               = _mm_mul_ps(r20,vftabscale);
 371             vfitab           = _mm_cvttps_epi32(rt);
 372 #ifdef __XOP__
 373             vfeps            = _mm_frcz_ps(rt);
 374 #else
 375             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 376 #endif
 377             twovfeps         = _mm_add_ps(vfeps,vfeps);
 378             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 379
 380             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 381             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 382             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 383             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 384             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 385             _MM_TRANSPOSE4_PS(Y,F,G,H);
 386             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 387             VV               = _mm_macc_ps(vfeps,Fp,Y);
 388             velec            = _mm_mul_ps(qq20,VV);
 389             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 390             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velecsum         = _mm_add_ps(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397              /* Update vectorial force */
 398             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 399             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 400             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 401
 402             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 403             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 404             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             r30              = _mm_mul_ps(rsq30,rinv30);
 411
 412             /* Compute parameters for interactions between i and j atoms */
 413             qq30             = _mm_mul_ps(iq3,jq0);
 414
 415             /* Calculate table index by multiplying r with table scale and truncate to integer */
 416             rt               = _mm_mul_ps(r30,vftabscale);
 417             vfitab           = _mm_cvttps_epi32(rt);
 418 #ifdef __XOP__
 419             vfeps            = _mm_frcz_ps(rt);
 420 #else
 421             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 422 #endif
 423             twovfeps         = _mm_add_ps(vfeps,vfeps);
 424             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 425
 426             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 427             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 428             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 429             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 430             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 431             _MM_TRANSPOSE4_PS(Y,F,G,H);
 432             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 433             VV               = _mm_macc_ps(vfeps,Fp,Y);
 434             velec            = _mm_mul_ps(qq30,VV);
 435             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 436             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 437
 438             /* Update potential sum for this i atom from the interaction with this j atom. */
 439             velecsum         = _mm_add_ps(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443              /* Update vectorial force */
 444             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 445             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 446             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 447
 448             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 449             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 450             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 451
 452             fjptrA             = f+j_coord_offsetA;
 453             fjptrB             = f+j_coord_offsetB;
 454             fjptrC             = f+j_coord_offsetC;
 455             fjptrD             = f+j_coord_offsetD;
 456
 457             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 458
 459             /* Inner loop uses 197 flops */
 460         }
 461
 462         if(jidx<j_index_end)
 463         {
 464
 465             /* Get j neighbor index, and coordinate index */
 466             jnrlistA         = jjnr[jidx];
 467             jnrlistB         = jjnr[jidx+1];
 468             jnrlistC         = jjnr[jidx+2];
 469             jnrlistD         = jjnr[jidx+3];
 470             /* Sign of each element will be negative for non-real atoms.
 471              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 472              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 473              */
 474             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 475             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 476             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 477             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 478             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 479             j_coord_offsetA  = DIM*jnrA;
 480             j_coord_offsetB  = DIM*jnrB;
 481             j_coord_offsetC  = DIM*jnrC;
 482             j_coord_offsetD  = DIM*jnrD;
 483
 484             /* load j atom coordinates */
 485             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 486                                               x+j_coord_offsetC,x+j_coord_offsetD,
 487                                               &jx0,&jy0,&jz0);
 488
 489             /* Calculate displacement vector */
 490             dx00             = _mm_sub_ps(ix0,jx0);
 491             dy00             = _mm_sub_ps(iy0,jy0);
 492             dz00             = _mm_sub_ps(iz0,jz0);
 493             dx10             = _mm_sub_ps(ix1,jx0);
 494             dy10             = _mm_sub_ps(iy1,jy0);
 495             dz10             = _mm_sub_ps(iz1,jz0);
 496             dx20             = _mm_sub_ps(ix2,jx0);
 497             dy20             = _mm_sub_ps(iy2,jy0);
 498             dz20             = _mm_sub_ps(iz2,jz0);
 499             dx30             = _mm_sub_ps(ix3,jx0);
 500             dy30             = _mm_sub_ps(iy3,jy0);
 501             dz30             = _mm_sub_ps(iz3,jz0);
 502
 503             /* Calculate squared distance and things based on it */
 504             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 505             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 506             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 507             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 508
 509             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 510             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 511             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 512             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 513
 514             /* Load parameters for j particles */
 515             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 516                                                               charge+jnrC+0,charge+jnrD+0);
 517             vdwjidx0A        = 2*vdwtype[jnrA+0];
 518             vdwjidx0B        = 2*vdwtype[jnrB+0];
 519             vdwjidx0C        = 2*vdwtype[jnrC+0];
 520             vdwjidx0D        = 2*vdwtype[jnrD+0];
 521
 522             fjx0             = _mm_setzero_ps();
 523             fjy0             = _mm_setzero_ps();
 524             fjz0             = _mm_setzero_ps();
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             r00              = _mm_mul_ps(rsq00,rinv00);
 531             r00              = _mm_andnot_ps(dummy_mask,r00);
 532
 533             /* Compute parameters for interactions between i and j atoms */
 534             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 535                                          vdwparam+vdwioffset0+vdwjidx0B,
 536                                          vdwparam+vdwioffset0+vdwjidx0C,
 537                                          vdwparam+vdwioffset0+vdwjidx0D,
 538                                          &c6_00,&c12_00);
 539
 540             /* Calculate table index by multiplying r with table scale and truncate to integer */
 541             rt               = _mm_mul_ps(r00,vftabscale);
 542             vfitab           = _mm_cvttps_epi32(rt);
 543 #ifdef __XOP__
 544             vfeps            = _mm_frcz_ps(rt);
 545 #else
 546             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 547 #endif
 548             twovfeps         = _mm_add_ps(vfeps,vfeps);
 549             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 550
 551             /* CUBIC SPLINE TABLE DISPERSION */
 552             vfitab           = _mm_add_epi32(vfitab,ifour);
 553             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 554             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 555             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 556             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 557             _MM_TRANSPOSE4_PS(Y,F,G,H);
 558             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 559             VV               = _mm_macc_ps(vfeps,Fp,Y);
 560             vvdw6            = _mm_mul_ps(c6_00,VV);
 561             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 562             fvdw6            = _mm_mul_ps(c6_00,FF);
 563
 564             /* CUBIC SPLINE TABLE REPULSION */
 565             vfitab           = _mm_add_epi32(vfitab,ifour);
 566             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 567             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 568             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 569             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 570             _MM_TRANSPOSE4_PS(Y,F,G,H);
 571             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 572             VV               = _mm_macc_ps(vfeps,Fp,Y);
 573             vvdw12           = _mm_mul_ps(c12_00,VV);
 574             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 575             fvdw12           = _mm_mul_ps(c12_00,FF);
 576             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 577             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 581             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 582
 583             fscal            = fvdw;
 584
 585             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 586
 587              /* Update vectorial force */
 588             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 589             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 590             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 591
 592             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 593             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 594             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             r10              = _mm_mul_ps(rsq10,rinv10);
 601             r10              = _mm_andnot_ps(dummy_mask,r10);
 602
 603             /* Compute parameters for interactions between i and j atoms */
 604             qq10             = _mm_mul_ps(iq1,jq0);
 605
 606             /* Calculate table index by multiplying r with table scale and truncate to integer */
 607             rt               = _mm_mul_ps(r10,vftabscale);
 608             vfitab           = _mm_cvttps_epi32(rt);
 609 #ifdef __XOP__
 610             vfeps            = _mm_frcz_ps(rt);
 611 #else
 612             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 613 #endif
 614             twovfeps         = _mm_add_ps(vfeps,vfeps);
 615             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 616
 617             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 618             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 619             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 620             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 621             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 622             _MM_TRANSPOSE4_PS(Y,F,G,H);
 623             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 624             VV               = _mm_macc_ps(vfeps,Fp,Y);
 625             velec            = _mm_mul_ps(qq10,VV);
 626             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 627             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 628
 629             /* Update potential sum for this i atom from the interaction with this j atom. */
 630             velec            = _mm_andnot_ps(dummy_mask,velec);
 631             velecsum         = _mm_add_ps(velecsum,velec);
 632
 633             fscal            = felec;
 634
 635             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 636
 637              /* Update vectorial force */
 638             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 639             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 640             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 641
 642             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 643             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 644             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 645
 646             /**************************
 647              * CALCULATE INTERACTIONS *
 648              **************************/
 649
 650             r20              = _mm_mul_ps(rsq20,rinv20);
 651             r20              = _mm_andnot_ps(dummy_mask,r20);
 652
 653             /* Compute parameters for interactions between i and j atoms */
 654             qq20             = _mm_mul_ps(iq2,jq0);
 655
 656             /* Calculate table index by multiplying r with table scale and truncate to integer */
 657             rt               = _mm_mul_ps(r20,vftabscale);
 658             vfitab           = _mm_cvttps_epi32(rt);
 659 #ifdef __XOP__
 660             vfeps            = _mm_frcz_ps(rt);
 661 #else
 662             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 663 #endif
 664             twovfeps         = _mm_add_ps(vfeps,vfeps);
 665             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 666
 667             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 668             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 669             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 670             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 671             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 672             _MM_TRANSPOSE4_PS(Y,F,G,H);
 673             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 674             VV               = _mm_macc_ps(vfeps,Fp,Y);
 675             velec            = _mm_mul_ps(qq20,VV);
 676             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 677             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 678
 679             /* Update potential sum for this i atom from the interaction with this j atom. */
 680             velec            = _mm_andnot_ps(dummy_mask,velec);
 681             velecsum         = _mm_add_ps(velecsum,velec);
 682
 683             fscal            = felec;
 684
 685             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 686
 687              /* Update vectorial force */
 688             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 689             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 690             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 691
 692             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 693             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 694             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 695
 696             /**************************
 697              * CALCULATE INTERACTIONS *
 698              **************************/
 699
 700             r30              = _mm_mul_ps(rsq30,rinv30);
 701             r30              = _mm_andnot_ps(dummy_mask,r30);
 702
 703             /* Compute parameters for interactions between i and j atoms */
 704             qq30             = _mm_mul_ps(iq3,jq0);
 705
 706             /* Calculate table index by multiplying r with table scale and truncate to integer */
 707             rt               = _mm_mul_ps(r30,vftabscale);
 708             vfitab           = _mm_cvttps_epi32(rt);
 709 #ifdef __XOP__
 710             vfeps            = _mm_frcz_ps(rt);
 711 #else
 712             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 713 #endif
 714             twovfeps         = _mm_add_ps(vfeps,vfeps);
 715             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 716
 717             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 718             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 719             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 720             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 721             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 722             _MM_TRANSPOSE4_PS(Y,F,G,H);
 723             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 724             VV               = _mm_macc_ps(vfeps,Fp,Y);
 725             velec            = _mm_mul_ps(qq30,VV);
 726             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 727             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 728
 729             /* Update potential sum for this i atom from the interaction with this j atom. */
 730             velec            = _mm_andnot_ps(dummy_mask,velec);
 731             velecsum         = _mm_add_ps(velecsum,velec);
 732
 733             fscal            = felec;
 734
 735             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 736
 737              /* Update vectorial force */
 738             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 739             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 740             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 741
 742             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 743             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 744             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 745
 746             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 747             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 748             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 749             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 750
 751             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 752
 753             /* Inner loop uses 201 flops */
 754         }
 755
 756         /* End of innermost loop */
 757
 758         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 759                                               f+i_coord_offset,fshift+i_shift_offset);
 760
 761         ggid                        = gid[iidx];
 762         /* Update potential energies */
 763         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 764         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 765
 766         /* Increment number of inner iterations */
 767         inneriter                  += j_index_end - j_index_start;
 768
 769         /* Outer loop uses 26 flops */
 770     }
 771
 772     /* Increment number of outer iterations */
 773     outeriter        += nri;
 774
 775     /* Update outer/inner flops */
 776
 777     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*201);
 778 }
 779 /*
 780  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 781  * Electrostatics interaction: CubicSplineTable
 782  * VdW interaction:            CubicSplineTable
 783  * Geometry:                   Water4-Particle
 784  * Calculate force/pot:        Force
 785  */
 786 void
 787 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 788                     (t_nblist                    * gmx_restrict       nlist,
 789                      rvec                        * gmx_restrict          xx,
 790                      rvec                        * gmx_restrict          ff,
 791                      t_forcerec                  * gmx_restrict          fr,
 792                      t_mdatoms                   * gmx_restrict     mdatoms,
 793                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 794                      t_nrnb                      * gmx_restrict        nrnb)
 795 {
 796     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 797      * just 0 for non-waters.
 798      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 799      * jnr indices corresponding to data put in the four positions in the SIMD register.
 800      */
 801     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 802     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 803     int              jnrA,jnrB,jnrC,jnrD;
 804     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 805     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 806     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 807     real             rcutoff_scalar;
 808     real             *shiftvec,*fshift,*x,*f;
 809     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 810     real             scratch[4*DIM];
 811     __m128           fscal,rcutoff,rcutoff2,jidxall;
 812     int              vdwioffset0;
 813     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 814     int              vdwioffset1;
 815     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 816     int              vdwioffset2;
 817     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 818     int              vdwioffset3;
 819     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 820     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 821     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 822     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 823     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 824     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 825     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 826     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 827     real             *charge;
 828     int              nvdwtype;
 829     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 830     int              *vdwtype;
 831     real             *vdwparam;
 832     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 833     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 834     __m128i          vfitab;
 835     __m128i          ifour       = _mm_set1_epi32(4);
 836     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 837     real             *vftab;
 838     __m128           dummy_mask,cutoff_mask;
 839     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 840     __m128           one     = _mm_set1_ps(1.0);
 841     __m128           two     = _mm_set1_ps(2.0);
 842     x                = xx[0];
 843     f                = ff[0];
 844
 845     nri              = nlist->nri;
 846     iinr             = nlist->iinr;
 847     jindex           = nlist->jindex;
 848     jjnr             = nlist->jjnr;
 849     shiftidx         = nlist->shift;
 850     gid              = nlist->gid;
 851     shiftvec         = fr->shift_vec[0];
 852     fshift           = fr->fshift[0];
 853     facel            = _mm_set1_ps(fr->epsfac);
 854     charge           = mdatoms->chargeA;
 855     nvdwtype         = fr->ntype;
 856     vdwparam         = fr->nbfp;
 857     vdwtype          = mdatoms->typeA;
 858
 859     vftab            = kernel_data->table_elec_vdw->data;
 860     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 861
 862     /* Setup water-specific parameters */
 863     inr              = nlist->iinr[0];
 864     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 865     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 866     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 867     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 868
 869     /* Avoid stupid compiler warnings */
 870     jnrA = jnrB = jnrC = jnrD = 0;
 871     j_coord_offsetA = 0;
 872     j_coord_offsetB = 0;
 873     j_coord_offsetC = 0;
 874     j_coord_offsetD = 0;
 875
 876     outeriter        = 0;
 877     inneriter        = 0;
 878
 879     for(iidx=0;iidx<4*DIM;iidx++)
 880     {
 881         scratch[iidx] = 0.0;
 882     }
 883
 884     /* Start outer loop over neighborlists */
 885     for(iidx=0; iidx<nri; iidx++)
 886     {
 887         /* Load shift vector for this list */
 888         i_shift_offset   = DIM*shiftidx[iidx];
 889
 890         /* Load limits for loop over neighbors */
 891         j_index_start    = jindex[iidx];
 892         j_index_end      = jindex[iidx+1];
 893
 894         /* Get outer coordinate index */
 895         inr              = iinr[iidx];
 896         i_coord_offset   = DIM*inr;
 897
 898         /* Load i particle coords and add shift vector */
 899         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 900                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 901
 902         fix0             = _mm_setzero_ps();
 903         fiy0             = _mm_setzero_ps();
 904         fiz0             = _mm_setzero_ps();
 905         fix1             = _mm_setzero_ps();
 906         fiy1             = _mm_setzero_ps();
 907         fiz1             = _mm_setzero_ps();
 908         fix2             = _mm_setzero_ps();
 909         fiy2             = _mm_setzero_ps();
 910         fiz2             = _mm_setzero_ps();
 911         fix3             = _mm_setzero_ps();
 912         fiy3             = _mm_setzero_ps();
 913         fiz3             = _mm_setzero_ps();
 914
 915         /* Start inner kernel loop */
 916         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 917         {
 918
 919             /* Get j neighbor index, and coordinate index */
 920             jnrA             = jjnr[jidx];
 921             jnrB             = jjnr[jidx+1];
 922             jnrC             = jjnr[jidx+2];
 923             jnrD             = jjnr[jidx+3];
 924             j_coord_offsetA  = DIM*jnrA;
 925             j_coord_offsetB  = DIM*jnrB;
 926             j_coord_offsetC  = DIM*jnrC;
 927             j_coord_offsetD  = DIM*jnrD;
 928
 929             /* load j atom coordinates */
 930             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 931                                               x+j_coord_offsetC,x+j_coord_offsetD,
 932                                               &jx0,&jy0,&jz0);
 933
 934             /* Calculate displacement vector */
 935             dx00             = _mm_sub_ps(ix0,jx0);
 936             dy00             = _mm_sub_ps(iy0,jy0);
 937             dz00             = _mm_sub_ps(iz0,jz0);
 938             dx10             = _mm_sub_ps(ix1,jx0);
 939             dy10             = _mm_sub_ps(iy1,jy0);
 940             dz10             = _mm_sub_ps(iz1,jz0);
 941             dx20             = _mm_sub_ps(ix2,jx0);
 942             dy20             = _mm_sub_ps(iy2,jy0);
 943             dz20             = _mm_sub_ps(iz2,jz0);
 944             dx30             = _mm_sub_ps(ix3,jx0);
 945             dy30             = _mm_sub_ps(iy3,jy0);
 946             dz30             = _mm_sub_ps(iz3,jz0);
 947
 948             /* Calculate squared distance and things based on it */
 949             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 950             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 951             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 952             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 953
 954             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 955             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 956             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 957             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 958
 959             /* Load parameters for j particles */
 960             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 961                                                               charge+jnrC+0,charge+jnrD+0);
 962             vdwjidx0A        = 2*vdwtype[jnrA+0];
 963             vdwjidx0B        = 2*vdwtype[jnrB+0];
 964             vdwjidx0C        = 2*vdwtype[jnrC+0];
 965             vdwjidx0D        = 2*vdwtype[jnrD+0];
 966
 967             fjx0             = _mm_setzero_ps();
 968             fjy0             = _mm_setzero_ps();
 969             fjz0             = _mm_setzero_ps();
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             r00              = _mm_mul_ps(rsq00,rinv00);
 976
 977             /* Compute parameters for interactions between i and j atoms */
 978             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 979                                          vdwparam+vdwioffset0+vdwjidx0B,
 980                                          vdwparam+vdwioffset0+vdwjidx0C,
 981                                          vdwparam+vdwioffset0+vdwjidx0D,
 982                                          &c6_00,&c12_00);
 983
 984             /* Calculate table index by multiplying r with table scale and truncate to integer */
 985             rt               = _mm_mul_ps(r00,vftabscale);
 986             vfitab           = _mm_cvttps_epi32(rt);
 987 #ifdef __XOP__
 988             vfeps            = _mm_frcz_ps(rt);
 989 #else
 990             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 991 #endif
 992             twovfeps         = _mm_add_ps(vfeps,vfeps);
 993             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 994
 995             /* CUBIC SPLINE TABLE DISPERSION */
 996             vfitab           = _mm_add_epi32(vfitab,ifour);
 997             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 998             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 999             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1000             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1001             _MM_TRANSPOSE4_PS(Y,F,G,H);
1002             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1003             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1004             fvdw6            = _mm_mul_ps(c6_00,FF);
1005
1006             /* CUBIC SPLINE TABLE REPULSION */
1007             vfitab           = _mm_add_epi32(vfitab,ifour);
1008             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1009             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1010             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1011             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1012             _MM_TRANSPOSE4_PS(Y,F,G,H);
1013             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1014             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1015             fvdw12           = _mm_mul_ps(c12_00,FF);
1016             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1017
1018             fscal            = fvdw;
1019
1020              /* Update vectorial force */
1021             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1022             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1023             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1024
1025             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1026             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1027             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1028
1029             /**************************
1030              * CALCULATE INTERACTIONS *
1031              **************************/
1032
1033             r10              = _mm_mul_ps(rsq10,rinv10);
1034
1035             /* Compute parameters for interactions between i and j atoms */
1036             qq10             = _mm_mul_ps(iq1,jq0);
1037
1038             /* Calculate table index by multiplying r with table scale and truncate to integer */
1039             rt               = _mm_mul_ps(r10,vftabscale);
1040             vfitab           = _mm_cvttps_epi32(rt);
1041 #ifdef __XOP__
1042             vfeps            = _mm_frcz_ps(rt);
1043 #else
1044             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1045 #endif
1046             twovfeps         = _mm_add_ps(vfeps,vfeps);
1047             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1048
1049             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1050             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1051             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1052             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1053             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1054             _MM_TRANSPOSE4_PS(Y,F,G,H);
1055             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1056             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1057             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1058
1059             fscal            = felec;
1060
1061              /* Update vectorial force */
1062             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1063             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1064             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1065
1066             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1067             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1068             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1069
1070             /**************************
1071              * CALCULATE INTERACTIONS *
1072              **************************/
1073
1074             r20              = _mm_mul_ps(rsq20,rinv20);
1075
1076             /* Compute parameters for interactions between i and j atoms */
1077             qq20             = _mm_mul_ps(iq2,jq0);
1078
1079             /* Calculate table index by multiplying r with table scale and truncate to integer */
1080             rt               = _mm_mul_ps(r20,vftabscale);
1081             vfitab           = _mm_cvttps_epi32(rt);
1082 #ifdef __XOP__
1083             vfeps            = _mm_frcz_ps(rt);
1084 #else
1085             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1086 #endif
1087             twovfeps         = _mm_add_ps(vfeps,vfeps);
1088             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1089
1090             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1091             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1092             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1093             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1094             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1095             _MM_TRANSPOSE4_PS(Y,F,G,H);
1096             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1097             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1098             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1099
1100             fscal            = felec;
1101
1102              /* Update vectorial force */
1103             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1104             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1105             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1106
1107             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1108             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1109             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             r30              = _mm_mul_ps(rsq30,rinv30);
1116
1117             /* Compute parameters for interactions between i and j atoms */
1118             qq30             = _mm_mul_ps(iq3,jq0);
1119
1120             /* Calculate table index by multiplying r with table scale and truncate to integer */
1121             rt               = _mm_mul_ps(r30,vftabscale);
1122             vfitab           = _mm_cvttps_epi32(rt);
1123 #ifdef __XOP__
1124             vfeps            = _mm_frcz_ps(rt);
1125 #else
1126             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1127 #endif
1128             twovfeps         = _mm_add_ps(vfeps,vfeps);
1129             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1130
1131             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1132             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1133             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1134             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1135             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1136             _MM_TRANSPOSE4_PS(Y,F,G,H);
1137             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1138             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1139             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1140
1141             fscal            = felec;
1142
1143              /* Update vectorial force */
1144             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1145             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1146             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1147
1148             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1149             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1150             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1151
1152             fjptrA             = f+j_coord_offsetA;
1153             fjptrB             = f+j_coord_offsetB;
1154             fjptrC             = f+j_coord_offsetC;
1155             fjptrD             = f+j_coord_offsetD;
1156
1157             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1158
1159             /* Inner loop uses 177 flops */
1160         }
1161
1162         if(jidx<j_index_end)
1163         {
1164
1165             /* Get j neighbor index, and coordinate index */
1166             jnrlistA         = jjnr[jidx];
1167             jnrlistB         = jjnr[jidx+1];
1168             jnrlistC         = jjnr[jidx+2];
1169             jnrlistD         = jjnr[jidx+3];
1170             /* Sign of each element will be negative for non-real atoms.
1171              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1172              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1173              */
1174             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1175             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1176             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1177             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1178             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1179             j_coord_offsetA  = DIM*jnrA;
1180             j_coord_offsetB  = DIM*jnrB;
1181             j_coord_offsetC  = DIM*jnrC;
1182             j_coord_offsetD  = DIM*jnrD;
1183
1184             /* load j atom coordinates */
1185             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1186                                               x+j_coord_offsetC,x+j_coord_offsetD,
1187                                               &jx0,&jy0,&jz0);
1188
1189             /* Calculate displacement vector */
1190             dx00             = _mm_sub_ps(ix0,jx0);
1191             dy00             = _mm_sub_ps(iy0,jy0);
1192             dz00             = _mm_sub_ps(iz0,jz0);
1193             dx10             = _mm_sub_ps(ix1,jx0);
1194             dy10             = _mm_sub_ps(iy1,jy0);
1195             dz10             = _mm_sub_ps(iz1,jz0);
1196             dx20             = _mm_sub_ps(ix2,jx0);
1197             dy20             = _mm_sub_ps(iy2,jy0);
1198             dz20             = _mm_sub_ps(iz2,jz0);
1199             dx30             = _mm_sub_ps(ix3,jx0);
1200             dy30             = _mm_sub_ps(iy3,jy0);
1201             dz30             = _mm_sub_ps(iz3,jz0);
1202
1203             /* Calculate squared distance and things based on it */
1204             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1205             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1206             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1207             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1208
1209             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1210             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1211             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1212             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1213
1214             /* Load parameters for j particles */
1215             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1216                                                               charge+jnrC+0,charge+jnrD+0);
1217             vdwjidx0A        = 2*vdwtype[jnrA+0];
1218             vdwjidx0B        = 2*vdwtype[jnrB+0];
1219             vdwjidx0C        = 2*vdwtype[jnrC+0];
1220             vdwjidx0D        = 2*vdwtype[jnrD+0];
1221
1222             fjx0             = _mm_setzero_ps();
1223             fjy0             = _mm_setzero_ps();
1224             fjz0             = _mm_setzero_ps();
1225
1226             /**************************
1227              * CALCULATE INTERACTIONS *
1228              **************************/
1229
1230             r00              = _mm_mul_ps(rsq00,rinv00);
1231             r00              = _mm_andnot_ps(dummy_mask,r00);
1232
1233             /* Compute parameters for interactions between i and j atoms */
1234             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1235                                          vdwparam+vdwioffset0+vdwjidx0B,
1236                                          vdwparam+vdwioffset0+vdwjidx0C,
1237                                          vdwparam+vdwioffset0+vdwjidx0D,
1238                                          &c6_00,&c12_00);
1239
1240             /* Calculate table index by multiplying r with table scale and truncate to integer */
1241             rt               = _mm_mul_ps(r00,vftabscale);
1242             vfitab           = _mm_cvttps_epi32(rt);
1243 #ifdef __XOP__
1244             vfeps            = _mm_frcz_ps(rt);
1245 #else
1246             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1247 #endif
1248             twovfeps         = _mm_add_ps(vfeps,vfeps);
1249             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1250
1251             /* CUBIC SPLINE TABLE DISPERSION */
1252             vfitab           = _mm_add_epi32(vfitab,ifour);
1253             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1254             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1255             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1256             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1257             _MM_TRANSPOSE4_PS(Y,F,G,H);
1258             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1259             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1260             fvdw6            = _mm_mul_ps(c6_00,FF);
1261
1262             /* CUBIC SPLINE TABLE REPULSION */
1263             vfitab           = _mm_add_epi32(vfitab,ifour);
1264             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1265             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1266             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1267             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1268             _MM_TRANSPOSE4_PS(Y,F,G,H);
1269             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1270             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1271             fvdw12           = _mm_mul_ps(c12_00,FF);
1272             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1273
1274             fscal            = fvdw;
1275
1276             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1277
1278              /* Update vectorial force */
1279             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1280             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1281             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1282
1283             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1284             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1285             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1286
1287             /**************************
1288              * CALCULATE INTERACTIONS *
1289              **************************/
1290
1291             r10              = _mm_mul_ps(rsq10,rinv10);
1292             r10              = _mm_andnot_ps(dummy_mask,r10);
1293
1294             /* Compute parameters for interactions between i and j atoms */
1295             qq10             = _mm_mul_ps(iq1,jq0);
1296
1297             /* Calculate table index by multiplying r with table scale and truncate to integer */
1298             rt               = _mm_mul_ps(r10,vftabscale);
1299             vfitab           = _mm_cvttps_epi32(rt);
1300 #ifdef __XOP__
1301             vfeps            = _mm_frcz_ps(rt);
1302 #else
1303             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1304 #endif
1305             twovfeps         = _mm_add_ps(vfeps,vfeps);
1306             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1307
1308             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1309             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1310             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1311             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1312             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1313             _MM_TRANSPOSE4_PS(Y,F,G,H);
1314             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1315             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1316             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1317
1318             fscal            = felec;
1319
1320             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1321
1322              /* Update vectorial force */
1323             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1324             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1325             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1326
1327             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1328             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1329             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1330
1331             /**************************
1332              * CALCULATE INTERACTIONS *
1333              **************************/
1334
1335             r20              = _mm_mul_ps(rsq20,rinv20);
1336             r20              = _mm_andnot_ps(dummy_mask,r20);
1337
1338             /* Compute parameters for interactions between i and j atoms */
1339             qq20             = _mm_mul_ps(iq2,jq0);
1340
1341             /* Calculate table index by multiplying r with table scale and truncate to integer */
1342             rt               = _mm_mul_ps(r20,vftabscale);
1343             vfitab           = _mm_cvttps_epi32(rt);
1344 #ifdef __XOP__
1345             vfeps            = _mm_frcz_ps(rt);
1346 #else
1347             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1348 #endif
1349             twovfeps         = _mm_add_ps(vfeps,vfeps);
1350             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1351
1352             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1353             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1354             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1355             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1356             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1357             _MM_TRANSPOSE4_PS(Y,F,G,H);
1358             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1359             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1360             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1361
1362             fscal            = felec;
1363
1364             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1365
1366              /* Update vectorial force */
1367             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1368             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1369             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1370
1371             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1372             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1373             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1374
1375             /**************************
1376              * CALCULATE INTERACTIONS *
1377              **************************/
1378
1379             r30              = _mm_mul_ps(rsq30,rinv30);
1380             r30              = _mm_andnot_ps(dummy_mask,r30);
1381
1382             /* Compute parameters for interactions between i and j atoms */
1383             qq30             = _mm_mul_ps(iq3,jq0);
1384
1385             /* Calculate table index by multiplying r with table scale and truncate to integer */
1386             rt               = _mm_mul_ps(r30,vftabscale);
1387             vfitab           = _mm_cvttps_epi32(rt);
1388 #ifdef __XOP__
1389             vfeps            = _mm_frcz_ps(rt);
1390 #else
1391             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1392 #endif
1393             twovfeps         = _mm_add_ps(vfeps,vfeps);
1394             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1395
1396             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1397             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1398             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1399             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1400             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1401             _MM_TRANSPOSE4_PS(Y,F,G,H);
1402             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1403             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1404             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1405
1406             fscal            = felec;
1407
1408             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1409
1410              /* Update vectorial force */
1411             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1412             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1413             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1414
1415             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1416             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1417             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1418
1419             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1420             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1421             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1422             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1423
1424             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1425
1426             /* Inner loop uses 181 flops */
1427         }
1428
1429         /* End of innermost loop */
1430
1431         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1432                                               f+i_coord_offset,fshift+i_shift_offset);
1433
1434         /* Increment number of inner iterations */
1435         inneriter                  += j_index_end - j_index_start;
1436
1437         /* Outer loop uses 24 flops */
1438     }
1439
1440     /* Increment number of outer iterations */
1441     outeriter        += nri;
1442
1443     /* Update outer/inner flops */
1444
1445     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*181);
1446 }