src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 105     real             *vftab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     vftab            = kernel_data->table_elec->data;
 128     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_ps();
 171         fiy0             = _mm_setzero_ps();
 172         fiz0             = _mm_setzero_ps();
 173         fix1             = _mm_setzero_ps();
 174         fiy1             = _mm_setzero_ps();
 175         fiz1             = _mm_setzero_ps();
 176         fix2             = _mm_setzero_ps();
 177         fiy2             = _mm_setzero_ps();
 178         fiz2             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               x+j_coord_offsetC,x+j_coord_offsetD,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_ps(ix0,jx0);
 205             dy00             = _mm_sub_ps(iy0,jy0);
 206             dz00             = _mm_sub_ps(iz0,jz0);
 207             dx10             = _mm_sub_ps(ix1,jx0);
 208             dy10             = _mm_sub_ps(iy1,jy0);
 209             dz10             = _mm_sub_ps(iz1,jz0);
 210             dx20             = _mm_sub_ps(ix2,jx0);
 211             dy20             = _mm_sub_ps(iy2,jy0);
 212             dz20             = _mm_sub_ps(iz2,jz0);
 213
 214             /* Calculate squared distance and things based on it */
 215             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 216             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 217             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 218
 219             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 220             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 221             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 222
 223             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                               charge+jnrC+0,charge+jnrD+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232
 233             fjx0             = _mm_setzero_ps();
 234             fjy0             = _mm_setzero_ps();
 235             fjz0             = _mm_setzero_ps();
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             r00              = _mm_mul_ps(rsq00,rinv00);
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             qq00             = _mm_mul_ps(iq0,jq0);
 245             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 246                                          vdwparam+vdwioffset0+vdwjidx0B,
 247                                          vdwparam+vdwioffset0+vdwjidx0C,
 248                                          vdwparam+vdwioffset0+vdwjidx0D,
 249                                          &c6_00,&c12_00);
 250
 251             /* Calculate table index by multiplying r with table scale and truncate to integer */
 252             rt               = _mm_mul_ps(r00,vftabscale);
 253             vfitab           = _mm_cvttps_epi32(rt);
 254 #ifdef __XOP__
 255             vfeps            = _mm_frcz_ps(rt);
 256 #else
 257             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 258 #endif
 259             twovfeps         = _mm_add_ps(vfeps,vfeps);
 260             vfitab           = _mm_slli_epi32(vfitab,2);
 261
 262             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 263             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 264             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 265             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 266             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 267             _MM_TRANSPOSE4_PS(Y,F,G,H);
 268             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 269             VV               = _mm_macc_ps(vfeps,Fp,Y);
 270             velec            = _mm_mul_ps(qq00,VV);
 271             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 272             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 273
 274             /* LENNARD-JONES DISPERSION/REPULSION */
 275
 276             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 277             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 278             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 279             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 280             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velecsum         = _mm_add_ps(velecsum,velec);
 284             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 285
 286             fscal            = _mm_add_ps(felec,fvdw);
 287
 288              /* Update vectorial force */
 289             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 290             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 291             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 292
 293             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 294             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 295             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             r10              = _mm_mul_ps(rsq10,rinv10);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _mm_mul_ps(iq1,jq0);
 305
 306             /* Calculate table index by multiplying r with table scale and truncate to integer */
 307             rt               = _mm_mul_ps(r10,vftabscale);
 308             vfitab           = _mm_cvttps_epi32(rt);
 309 #ifdef __XOP__
 310             vfeps            = _mm_frcz_ps(rt);
 311 #else
 312             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 313 #endif
 314             twovfeps         = _mm_add_ps(vfeps,vfeps);
 315             vfitab           = _mm_slli_epi32(vfitab,2);
 316
 317             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 318             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 319             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 320             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 321             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 322             _MM_TRANSPOSE4_PS(Y,F,G,H);
 323             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 324             VV               = _mm_macc_ps(vfeps,Fp,Y);
 325             velec            = _mm_mul_ps(qq10,VV);
 326             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 327             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velecsum         = _mm_add_ps(velecsum,velec);
 331
 332             fscal            = felec;
 333
 334              /* Update vectorial force */
 335             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 336             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 337             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 338
 339             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 340             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 341             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             r20              = _mm_mul_ps(rsq20,rinv20);
 348
 349             /* Compute parameters for interactions between i and j atoms */
 350             qq20             = _mm_mul_ps(iq2,jq0);
 351
 352             /* Calculate table index by multiplying r with table scale and truncate to integer */
 353             rt               = _mm_mul_ps(r20,vftabscale);
 354             vfitab           = _mm_cvttps_epi32(rt);
 355 #ifdef __XOP__
 356             vfeps            = _mm_frcz_ps(rt);
 357 #else
 358             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 359 #endif
 360             twovfeps         = _mm_add_ps(vfeps,vfeps);
 361             vfitab           = _mm_slli_epi32(vfitab,2);
 362
 363             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 364             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 365             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 366             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 367             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 368             _MM_TRANSPOSE4_PS(Y,F,G,H);
 369             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 370             VV               = _mm_macc_ps(vfeps,Fp,Y);
 371             velec            = _mm_mul_ps(qq20,VV);
 372             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 373             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velecsum         = _mm_add_ps(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380              /* Update vectorial force */
 381             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 382             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 383             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 384
 385             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 386             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 387             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 388
 389             fjptrA             = f+j_coord_offsetA;
 390             fjptrB             = f+j_coord_offsetB;
 391             fjptrC             = f+j_coord_offsetC;
 392             fjptrD             = f+j_coord_offsetD;
 393
 394             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 395
 396             /* Inner loop uses 151 flops */
 397         }
 398
 399         if(jidx<j_index_end)
 400         {
 401
 402             /* Get j neighbor index, and coordinate index */
 403             jnrlistA         = jjnr[jidx];
 404             jnrlistB         = jjnr[jidx+1];
 405             jnrlistC         = jjnr[jidx+2];
 406             jnrlistD         = jjnr[jidx+3];
 407             /* Sign of each element will be negative for non-real atoms.
 408              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 409              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 410              */
 411             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 412             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 413             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 414             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 415             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 416             j_coord_offsetA  = DIM*jnrA;
 417             j_coord_offsetB  = DIM*jnrB;
 418             j_coord_offsetC  = DIM*jnrC;
 419             j_coord_offsetD  = DIM*jnrD;
 420
 421             /* load j atom coordinates */
 422             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 423                                               x+j_coord_offsetC,x+j_coord_offsetD,
 424                                               &jx0,&jy0,&jz0);
 425
 426             /* Calculate displacement vector */
 427             dx00             = _mm_sub_ps(ix0,jx0);
 428             dy00             = _mm_sub_ps(iy0,jy0);
 429             dz00             = _mm_sub_ps(iz0,jz0);
 430             dx10             = _mm_sub_ps(ix1,jx0);
 431             dy10             = _mm_sub_ps(iy1,jy0);
 432             dz10             = _mm_sub_ps(iz1,jz0);
 433             dx20             = _mm_sub_ps(ix2,jx0);
 434             dy20             = _mm_sub_ps(iy2,jy0);
 435             dz20             = _mm_sub_ps(iz2,jz0);
 436
 437             /* Calculate squared distance and things based on it */
 438             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 439             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 440             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 441
 442             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 443             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 444             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 445
 446             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 447
 448             /* Load parameters for j particles */
 449             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 450                                                               charge+jnrC+0,charge+jnrD+0);
 451             vdwjidx0A        = 2*vdwtype[jnrA+0];
 452             vdwjidx0B        = 2*vdwtype[jnrB+0];
 453             vdwjidx0C        = 2*vdwtype[jnrC+0];
 454             vdwjidx0D        = 2*vdwtype[jnrD+0];
 455
 456             fjx0             = _mm_setzero_ps();
 457             fjy0             = _mm_setzero_ps();
 458             fjz0             = _mm_setzero_ps();
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             r00              = _mm_mul_ps(rsq00,rinv00);
 465             r00              = _mm_andnot_ps(dummy_mask,r00);
 466
 467             /* Compute parameters for interactions between i and j atoms */
 468             qq00             = _mm_mul_ps(iq0,jq0);
 469             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 470                                          vdwparam+vdwioffset0+vdwjidx0B,
 471                                          vdwparam+vdwioffset0+vdwjidx0C,
 472                                          vdwparam+vdwioffset0+vdwjidx0D,
 473                                          &c6_00,&c12_00);
 474
 475             /* Calculate table index by multiplying r with table scale and truncate to integer */
 476             rt               = _mm_mul_ps(r00,vftabscale);
 477             vfitab           = _mm_cvttps_epi32(rt);
 478 #ifdef __XOP__
 479             vfeps            = _mm_frcz_ps(rt);
 480 #else
 481             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 482 #endif
 483             twovfeps         = _mm_add_ps(vfeps,vfeps);
 484             vfitab           = _mm_slli_epi32(vfitab,2);
 485
 486             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 487             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 488             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 489             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 490             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 491             _MM_TRANSPOSE4_PS(Y,F,G,H);
 492             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 493             VV               = _mm_macc_ps(vfeps,Fp,Y);
 494             velec            = _mm_mul_ps(qq00,VV);
 495             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 496             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 497
 498             /* LENNARD-JONES DISPERSION/REPULSION */
 499
 500             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 501             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 502             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 503             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 504             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 505
 506             /* Update potential sum for this i atom from the interaction with this j atom. */
 507             velec            = _mm_andnot_ps(dummy_mask,velec);
 508             velecsum         = _mm_add_ps(velecsum,velec);
 509             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 510             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 511
 512             fscal            = _mm_add_ps(felec,fvdw);
 513
 514             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 515
 516              /* Update vectorial force */
 517             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 518             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 519             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 520
 521             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 522             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 523             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             r10              = _mm_mul_ps(rsq10,rinv10);
 530             r10              = _mm_andnot_ps(dummy_mask,r10);
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq10             = _mm_mul_ps(iq1,jq0);
 534
 535             /* Calculate table index by multiplying r with table scale and truncate to integer */
 536             rt               = _mm_mul_ps(r10,vftabscale);
 537             vfitab           = _mm_cvttps_epi32(rt);
 538 #ifdef __XOP__
 539             vfeps            = _mm_frcz_ps(rt);
 540 #else
 541             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 542 #endif
 543             twovfeps         = _mm_add_ps(vfeps,vfeps);
 544             vfitab           = _mm_slli_epi32(vfitab,2);
 545
 546             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 547             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 548             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 549             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 550             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 551             _MM_TRANSPOSE4_PS(Y,F,G,H);
 552             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 553             VV               = _mm_macc_ps(vfeps,Fp,Y);
 554             velec            = _mm_mul_ps(qq10,VV);
 555             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 556             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             velec            = _mm_andnot_ps(dummy_mask,velec);
 560             velecsum         = _mm_add_ps(velecsum,velec);
 561
 562             fscal            = felec;
 563
 564             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 565
 566              /* Update vectorial force */
 567             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 568             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 569             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 570
 571             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 572             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 573             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             r20              = _mm_mul_ps(rsq20,rinv20);
 580             r20              = _mm_andnot_ps(dummy_mask,r20);
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq20             = _mm_mul_ps(iq2,jq0);
 584
 585             /* Calculate table index by multiplying r with table scale and truncate to integer */
 586             rt               = _mm_mul_ps(r20,vftabscale);
 587             vfitab           = _mm_cvttps_epi32(rt);
 588 #ifdef __XOP__
 589             vfeps            = _mm_frcz_ps(rt);
 590 #else
 591             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 592 #endif
 593             twovfeps         = _mm_add_ps(vfeps,vfeps);
 594             vfitab           = _mm_slli_epi32(vfitab,2);
 595
 596             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 597             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 598             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 599             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 600             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 601             _MM_TRANSPOSE4_PS(Y,F,G,H);
 602             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 603             VV               = _mm_macc_ps(vfeps,Fp,Y);
 604             velec            = _mm_mul_ps(qq20,VV);
 605             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 606             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 607
 608             /* Update potential sum for this i atom from the interaction with this j atom. */
 609             velec            = _mm_andnot_ps(dummy_mask,velec);
 610             velecsum         = _mm_add_ps(velecsum,velec);
 611
 612             fscal            = felec;
 613
 614             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 615
 616              /* Update vectorial force */
 617             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 618             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 619             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 620
 621             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 622             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 623             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 624
 625             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 626             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 627             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 628             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 629
 630             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 631
 632             /* Inner loop uses 154 flops */
 633         }
 634
 635         /* End of innermost loop */
 636
 637         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 638                                               f+i_coord_offset,fshift+i_shift_offset);
 639
 640         ggid                        = gid[iidx];
 641         /* Update potential energies */
 642         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 643         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 644
 645         /* Increment number of inner iterations */
 646         inneriter                  += j_index_end - j_index_start;
 647
 648         /* Outer loop uses 20 flops */
 649     }
 650
 651     /* Increment number of outer iterations */
 652     outeriter        += nri;
 653
 654     /* Update outer/inner flops */
 655
 656     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 657 }
 658 /*
 659  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_single
 660  * Electrostatics interaction: CubicSplineTable
 661  * VdW interaction:            LennardJones
 662  * Geometry:                   Water3-Particle
 663  * Calculate force/pot:        Force
 664  */
 665 void
 666 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_single
 667                     (t_nblist                    * gmx_restrict       nlist,
 668                      rvec                        * gmx_restrict          xx,
 669                      rvec                        * gmx_restrict          ff,
 670                      t_forcerec                  * gmx_restrict          fr,
 671                      t_mdatoms                   * gmx_restrict     mdatoms,
 672                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 673                      t_nrnb                      * gmx_restrict        nrnb)
 674 {
 675     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 676      * just 0 for non-waters.
 677      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 678      * jnr indices corresponding to data put in the four positions in the SIMD register.
 679      */
 680     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 681     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 682     int              jnrA,jnrB,jnrC,jnrD;
 683     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 684     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 685     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 686     real             rcutoff_scalar;
 687     real             *shiftvec,*fshift,*x,*f;
 688     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 689     real             scratch[4*DIM];
 690     __m128           fscal,rcutoff,rcutoff2,jidxall;
 691     int              vdwioffset0;
 692     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 693     int              vdwioffset1;
 694     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 695     int              vdwioffset2;
 696     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 697     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 698     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 699     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 700     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 701     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 702     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 703     real             *charge;
 704     int              nvdwtype;
 705     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 706     int              *vdwtype;
 707     real             *vdwparam;
 708     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 709     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 710     __m128i          vfitab;
 711     __m128i          ifour       = _mm_set1_epi32(4);
 712     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 713     real             *vftab;
 714     __m128           dummy_mask,cutoff_mask;
 715     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 716     __m128           one     = _mm_set1_ps(1.0);
 717     __m128           two     = _mm_set1_ps(2.0);
 718     x                = xx[0];
 719     f                = ff[0];
 720
 721     nri              = nlist->nri;
 722     iinr             = nlist->iinr;
 723     jindex           = nlist->jindex;
 724     jjnr             = nlist->jjnr;
 725     shiftidx         = nlist->shift;
 726     gid              = nlist->gid;
 727     shiftvec         = fr->shift_vec[0];
 728     fshift           = fr->fshift[0];
 729     facel            = _mm_set1_ps(fr->epsfac);
 730     charge           = mdatoms->chargeA;
 731     nvdwtype         = fr->ntype;
 732     vdwparam         = fr->nbfp;
 733     vdwtype          = mdatoms->typeA;
 734
 735     vftab            = kernel_data->table_elec->data;
 736     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 737
 738     /* Setup water-specific parameters */
 739     inr              = nlist->iinr[0];
 740     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 741     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 742     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 743     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 744
 745     /* Avoid stupid compiler warnings */
 746     jnrA = jnrB = jnrC = jnrD = 0;
 747     j_coord_offsetA = 0;
 748     j_coord_offsetB = 0;
 749     j_coord_offsetC = 0;
 750     j_coord_offsetD = 0;
 751
 752     outeriter        = 0;
 753     inneriter        = 0;
 754
 755     for(iidx=0;iidx<4*DIM;iidx++)
 756     {
 757         scratch[iidx] = 0.0;
 758     }
 759
 760     /* Start outer loop over neighborlists */
 761     for(iidx=0; iidx<nri; iidx++)
 762     {
 763         /* Load shift vector for this list */
 764         i_shift_offset   = DIM*shiftidx[iidx];
 765
 766         /* Load limits for loop over neighbors */
 767         j_index_start    = jindex[iidx];
 768         j_index_end      = jindex[iidx+1];
 769
 770         /* Get outer coordinate index */
 771         inr              = iinr[iidx];
 772         i_coord_offset   = DIM*inr;
 773
 774         /* Load i particle coords and add shift vector */
 775         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 776                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 777
 778         fix0             = _mm_setzero_ps();
 779         fiy0             = _mm_setzero_ps();
 780         fiz0             = _mm_setzero_ps();
 781         fix1             = _mm_setzero_ps();
 782         fiy1             = _mm_setzero_ps();
 783         fiz1             = _mm_setzero_ps();
 784         fix2             = _mm_setzero_ps();
 785         fiy2             = _mm_setzero_ps();
 786         fiz2             = _mm_setzero_ps();
 787
 788         /* Start inner kernel loop */
 789         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 790         {
 791
 792             /* Get j neighbor index, and coordinate index */
 793             jnrA             = jjnr[jidx];
 794             jnrB             = jjnr[jidx+1];
 795             jnrC             = jjnr[jidx+2];
 796             jnrD             = jjnr[jidx+3];
 797             j_coord_offsetA  = DIM*jnrA;
 798             j_coord_offsetB  = DIM*jnrB;
 799             j_coord_offsetC  = DIM*jnrC;
 800             j_coord_offsetD  = DIM*jnrD;
 801
 802             /* load j atom coordinates */
 803             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 804                                               x+j_coord_offsetC,x+j_coord_offsetD,
 805                                               &jx0,&jy0,&jz0);
 806
 807             /* Calculate displacement vector */
 808             dx00             = _mm_sub_ps(ix0,jx0);
 809             dy00             = _mm_sub_ps(iy0,jy0);
 810             dz00             = _mm_sub_ps(iz0,jz0);
 811             dx10             = _mm_sub_ps(ix1,jx0);
 812             dy10             = _mm_sub_ps(iy1,jy0);
 813             dz10             = _mm_sub_ps(iz1,jz0);
 814             dx20             = _mm_sub_ps(ix2,jx0);
 815             dy20             = _mm_sub_ps(iy2,jy0);
 816             dz20             = _mm_sub_ps(iz2,jz0);
 817
 818             /* Calculate squared distance and things based on it */
 819             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 820             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 821             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 822
 823             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 824             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 825             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 826
 827             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 828
 829             /* Load parameters for j particles */
 830             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 831                                                               charge+jnrC+0,charge+jnrD+0);
 832             vdwjidx0A        = 2*vdwtype[jnrA+0];
 833             vdwjidx0B        = 2*vdwtype[jnrB+0];
 834             vdwjidx0C        = 2*vdwtype[jnrC+0];
 835             vdwjidx0D        = 2*vdwtype[jnrD+0];
 836
 837             fjx0             = _mm_setzero_ps();
 838             fjy0             = _mm_setzero_ps();
 839             fjz0             = _mm_setzero_ps();
 840
 841             /**************************
 842              * CALCULATE INTERACTIONS *
 843              **************************/
 844
 845             r00              = _mm_mul_ps(rsq00,rinv00);
 846
 847             /* Compute parameters for interactions between i and j atoms */
 848             qq00             = _mm_mul_ps(iq0,jq0);
 849             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 850                                          vdwparam+vdwioffset0+vdwjidx0B,
 851                                          vdwparam+vdwioffset0+vdwjidx0C,
 852                                          vdwparam+vdwioffset0+vdwjidx0D,
 853                                          &c6_00,&c12_00);
 854
 855             /* Calculate table index by multiplying r with table scale and truncate to integer */
 856             rt               = _mm_mul_ps(r00,vftabscale);
 857             vfitab           = _mm_cvttps_epi32(rt);
 858 #ifdef __XOP__
 859             vfeps            = _mm_frcz_ps(rt);
 860 #else
 861             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 862 #endif
 863             twovfeps         = _mm_add_ps(vfeps,vfeps);
 864             vfitab           = _mm_slli_epi32(vfitab,2);
 865
 866             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 867             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 868             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 869             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 870             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 871             _MM_TRANSPOSE4_PS(Y,F,G,H);
 872             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 873             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 874             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 875
 876             /* LENNARD-JONES DISPERSION/REPULSION */
 877
 878             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 879             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 880
 881             fscal            = _mm_add_ps(felec,fvdw);
 882
 883              /* Update vectorial force */
 884             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 885             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 886             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 887
 888             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 889             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 890             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             r10              = _mm_mul_ps(rsq10,rinv10);
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             qq10             = _mm_mul_ps(iq1,jq0);
 900
 901             /* Calculate table index by multiplying r with table scale and truncate to integer */
 902             rt               = _mm_mul_ps(r10,vftabscale);
 903             vfitab           = _mm_cvttps_epi32(rt);
 904 #ifdef __XOP__
 905             vfeps            = _mm_frcz_ps(rt);
 906 #else
 907             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 908 #endif
 909             twovfeps         = _mm_add_ps(vfeps,vfeps);
 910             vfitab           = _mm_slli_epi32(vfitab,2);
 911
 912             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 913             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 914             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 915             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 916             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 917             _MM_TRANSPOSE4_PS(Y,F,G,H);
 918             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 919             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 920             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 921
 922             fscal            = felec;
 923
 924              /* Update vectorial force */
 925             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 926             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 927             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 928
 929             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 930             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 931             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 932
 933             /**************************
 934              * CALCULATE INTERACTIONS *
 935              **************************/
 936
 937             r20              = _mm_mul_ps(rsq20,rinv20);
 938
 939             /* Compute parameters for interactions between i and j atoms */
 940             qq20             = _mm_mul_ps(iq2,jq0);
 941
 942             /* Calculate table index by multiplying r with table scale and truncate to integer */
 943             rt               = _mm_mul_ps(r20,vftabscale);
 944             vfitab           = _mm_cvttps_epi32(rt);
 945 #ifdef __XOP__
 946             vfeps            = _mm_frcz_ps(rt);
 947 #else
 948             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 949 #endif
 950             twovfeps         = _mm_add_ps(vfeps,vfeps);
 951             vfitab           = _mm_slli_epi32(vfitab,2);
 952
 953             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 954             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 955             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 956             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 957             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 958             _MM_TRANSPOSE4_PS(Y,F,G,H);
 959             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 960             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 961             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 962
 963             fscal            = felec;
 964
 965              /* Update vectorial force */
 966             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 967             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 968             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 969
 970             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 971             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 972             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 973
 974             fjptrA             = f+j_coord_offsetA;
 975             fjptrB             = f+j_coord_offsetB;
 976             fjptrC             = f+j_coord_offsetC;
 977             fjptrD             = f+j_coord_offsetD;
 978
 979             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 980
 981             /* Inner loop uses 134 flops */
 982         }
 983
 984         if(jidx<j_index_end)
 985         {
 986
 987             /* Get j neighbor index, and coordinate index */
 988             jnrlistA         = jjnr[jidx];
 989             jnrlistB         = jjnr[jidx+1];
 990             jnrlistC         = jjnr[jidx+2];
 991             jnrlistD         = jjnr[jidx+3];
 992             /* Sign of each element will be negative for non-real atoms.
 993              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 994              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 995              */
 996             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 997             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 998             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 999             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1000             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1001             j_coord_offsetA  = DIM*jnrA;
1002             j_coord_offsetB  = DIM*jnrB;
1003             j_coord_offsetC  = DIM*jnrC;
1004             j_coord_offsetD  = DIM*jnrD;
1005
1006             /* load j atom coordinates */
1007             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1008                                               x+j_coord_offsetC,x+j_coord_offsetD,
1009                                               &jx0,&jy0,&jz0);
1010
1011             /* Calculate displacement vector */
1012             dx00             = _mm_sub_ps(ix0,jx0);
1013             dy00             = _mm_sub_ps(iy0,jy0);
1014             dz00             = _mm_sub_ps(iz0,jz0);
1015             dx10             = _mm_sub_ps(ix1,jx0);
1016             dy10             = _mm_sub_ps(iy1,jy0);
1017             dz10             = _mm_sub_ps(iz1,jz0);
1018             dx20             = _mm_sub_ps(ix2,jx0);
1019             dy20             = _mm_sub_ps(iy2,jy0);
1020             dz20             = _mm_sub_ps(iz2,jz0);
1021
1022             /* Calculate squared distance and things based on it */
1023             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1024             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1025             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1026
1027             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1028             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1029             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1030
1031             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1032
1033             /* Load parameters for j particles */
1034             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1035                                                               charge+jnrC+0,charge+jnrD+0);
1036             vdwjidx0A        = 2*vdwtype[jnrA+0];
1037             vdwjidx0B        = 2*vdwtype[jnrB+0];
1038             vdwjidx0C        = 2*vdwtype[jnrC+0];
1039             vdwjidx0D        = 2*vdwtype[jnrD+0];
1040
1041             fjx0             = _mm_setzero_ps();
1042             fjy0             = _mm_setzero_ps();
1043             fjz0             = _mm_setzero_ps();
1044
1045             /**************************
1046              * CALCULATE INTERACTIONS *
1047              **************************/
1048
1049             r00              = _mm_mul_ps(rsq00,rinv00);
1050             r00              = _mm_andnot_ps(dummy_mask,r00);
1051
1052             /* Compute parameters for interactions between i and j atoms */
1053             qq00             = _mm_mul_ps(iq0,jq0);
1054             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1055                                          vdwparam+vdwioffset0+vdwjidx0B,
1056                                          vdwparam+vdwioffset0+vdwjidx0C,
1057                                          vdwparam+vdwioffset0+vdwjidx0D,
1058                                          &c6_00,&c12_00);
1059
1060             /* Calculate table index by multiplying r with table scale and truncate to integer */
1061             rt               = _mm_mul_ps(r00,vftabscale);
1062             vfitab           = _mm_cvttps_epi32(rt);
1063 #ifdef __XOP__
1064             vfeps            = _mm_frcz_ps(rt);
1065 #else
1066             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1067 #endif
1068             twovfeps         = _mm_add_ps(vfeps,vfeps);
1069             vfitab           = _mm_slli_epi32(vfitab,2);
1070
1071             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1072             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1073             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1074             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1075             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1076             _MM_TRANSPOSE4_PS(Y,F,G,H);
1077             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1078             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1079             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1080
1081             /* LENNARD-JONES DISPERSION/REPULSION */
1082
1083             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1084             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1085
1086             fscal            = _mm_add_ps(felec,fvdw);
1087
1088             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1089
1090              /* Update vectorial force */
1091             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1092             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1093             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1094
1095             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1096             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1097             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1098
1099             /**************************
1100              * CALCULATE INTERACTIONS *
1101              **************************/
1102
1103             r10              = _mm_mul_ps(rsq10,rinv10);
1104             r10              = _mm_andnot_ps(dummy_mask,r10);
1105
1106             /* Compute parameters for interactions between i and j atoms */
1107             qq10             = _mm_mul_ps(iq1,jq0);
1108
1109             /* Calculate table index by multiplying r with table scale and truncate to integer */
1110             rt               = _mm_mul_ps(r10,vftabscale);
1111             vfitab           = _mm_cvttps_epi32(rt);
1112 #ifdef __XOP__
1113             vfeps            = _mm_frcz_ps(rt);
1114 #else
1115             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1116 #endif
1117             twovfeps         = _mm_add_ps(vfeps,vfeps);
1118             vfitab           = _mm_slli_epi32(vfitab,2);
1119
1120             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1121             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1122             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1123             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1124             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1125             _MM_TRANSPOSE4_PS(Y,F,G,H);
1126             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1127             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1128             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1129
1130             fscal            = felec;
1131
1132             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1133
1134              /* Update vectorial force */
1135             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1136             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1137             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1138
1139             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1140             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1141             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1142
1143             /**************************
1144              * CALCULATE INTERACTIONS *
1145              **************************/
1146
1147             r20              = _mm_mul_ps(rsq20,rinv20);
1148             r20              = _mm_andnot_ps(dummy_mask,r20);
1149
1150             /* Compute parameters for interactions between i and j atoms */
1151             qq20             = _mm_mul_ps(iq2,jq0);
1152
1153             /* Calculate table index by multiplying r with table scale and truncate to integer */
1154             rt               = _mm_mul_ps(r20,vftabscale);
1155             vfitab           = _mm_cvttps_epi32(rt);
1156 #ifdef __XOP__
1157             vfeps            = _mm_frcz_ps(rt);
1158 #else
1159             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1160 #endif
1161             twovfeps         = _mm_add_ps(vfeps,vfeps);
1162             vfitab           = _mm_slli_epi32(vfitab,2);
1163
1164             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1165             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1166             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1167             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1168             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1169             _MM_TRANSPOSE4_PS(Y,F,G,H);
1170             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1171             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1172             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1173
1174             fscal            = felec;
1175
1176             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1177
1178              /* Update vectorial force */
1179             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1180             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1181             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1182
1183             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1184             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1185             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1186
1187             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1188             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1189             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1190             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1191
1192             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1193
1194             /* Inner loop uses 137 flops */
1195         }
1196
1197         /* End of innermost loop */
1198
1199         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1200                                               f+i_coord_offset,fshift+i_shift_offset);
1201
1202         /* Increment number of inner iterations */
1203         inneriter                  += j_index_end - j_index_start;
1204
1205         /* Outer loop uses 18 flops */
1206     }
1207
1208     /* Increment number of outer iterations */
1209     outeriter        += nri;
1210
1211     /* Update outer/inner flops */
1212
1213     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1214 }