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 "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_VdwLJ_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 107     real             *vftab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_elec->data;
 130     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 135     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 136     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm_setzero_ps();
 173         fiy0             = _mm_setzero_ps();
 174         fiz0             = _mm_setzero_ps();
 175         fix1             = _mm_setzero_ps();
 176         fiy1             = _mm_setzero_ps();
 177         fiz1             = _mm_setzero_ps();
 178         fix2             = _mm_setzero_ps();
 179         fiy2             = _mm_setzero_ps();
 180         fiz2             = _mm_setzero_ps();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_ps();
 184         vvdwsum          = _mm_setzero_ps();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199
 200             /* load j atom coordinates */
 201             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                               x+j_coord_offsetC,x+j_coord_offsetD,
 203                                               &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm_sub_ps(ix0,jx0);
 207             dy00             = _mm_sub_ps(iy0,jy0);
 208             dz00             = _mm_sub_ps(iz0,jz0);
 209             dx10             = _mm_sub_ps(ix1,jx0);
 210             dy10             = _mm_sub_ps(iy1,jy0);
 211             dz10             = _mm_sub_ps(iz1,jz0);
 212             dx20             = _mm_sub_ps(ix2,jx0);
 213             dy20             = _mm_sub_ps(iy2,jy0);
 214             dz20             = _mm_sub_ps(iz2,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 220
 221             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 222             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 223             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 224
 225             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 229                                                               charge+jnrC+0,charge+jnrD+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234
 235             fjx0             = _mm_setzero_ps();
 236             fjy0             = _mm_setzero_ps();
 237             fjz0             = _mm_setzero_ps();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             r00              = _mm_mul_ps(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm_mul_ps(iq0,jq0);
 247             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 248                                          vdwparam+vdwioffset0+vdwjidx0B,
 249                                          vdwparam+vdwioffset0+vdwjidx0C,
 250                                          vdwparam+vdwioffset0+vdwjidx0D,
 251                                          &c6_00,&c12_00);
 252
 253             /* Calculate table index by multiplying r with table scale and truncate to integer */
 254             rt               = _mm_mul_ps(r00,vftabscale);
 255             vfitab           = _mm_cvttps_epi32(rt);
 256 #ifdef __XOP__
 257             vfeps            = _mm_frcz_ps(rt);
 258 #else
 259             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 260 #endif
 261             twovfeps         = _mm_add_ps(vfeps,vfeps);
 262             vfitab           = _mm_slli_epi32(vfitab,2);
 263
 264             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 265             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 266             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 267             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 268             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 269             _MM_TRANSPOSE4_PS(Y,F,G,H);
 270             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 271             VV               = _mm_macc_ps(vfeps,Fp,Y);
 272             velec            = _mm_mul_ps(qq00,VV);
 273             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 274             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 275
 276             /* LENNARD-JONES DISPERSION/REPULSION */
 277
 278             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 279             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 280             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 281             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 282             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velecsum         = _mm_add_ps(velecsum,velec);
 286             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 287
 288             fscal            = _mm_add_ps(felec,fvdw);
 289
 290              /* Update vectorial force */
 291             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 292             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 293             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 294
 295             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 296             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 297             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             r10              = _mm_mul_ps(rsq10,rinv10);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_ps(iq1,jq0);
 307
 308             /* Calculate table index by multiplying r with table scale and truncate to integer */
 309             rt               = _mm_mul_ps(r10,vftabscale);
 310             vfitab           = _mm_cvttps_epi32(rt);
 311 #ifdef __XOP__
 312             vfeps            = _mm_frcz_ps(rt);
 313 #else
 314             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 315 #endif
 316             twovfeps         = _mm_add_ps(vfeps,vfeps);
 317             vfitab           = _mm_slli_epi32(vfitab,2);
 318
 319             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 320             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 321             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 322             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 323             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 324             _MM_TRANSPOSE4_PS(Y,F,G,H);
 325             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 326             VV               = _mm_macc_ps(vfeps,Fp,Y);
 327             velec            = _mm_mul_ps(qq10,VV);
 328             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 329             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 330
 331             /* Update potential sum for this i atom from the interaction with this j atom. */
 332             velecsum         = _mm_add_ps(velecsum,velec);
 333
 334             fscal            = felec;
 335
 336              /* Update vectorial force */
 337             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 338             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 339             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 340
 341             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 342             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 343             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 344
 345             /**************************
 346              * CALCULATE INTERACTIONS *
 347              **************************/
 348
 349             r20              = _mm_mul_ps(rsq20,rinv20);
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq20             = _mm_mul_ps(iq2,jq0);
 353
 354             /* Calculate table index by multiplying r with table scale and truncate to integer */
 355             rt               = _mm_mul_ps(r20,vftabscale);
 356             vfitab           = _mm_cvttps_epi32(rt);
 357 #ifdef __XOP__
 358             vfeps            = _mm_frcz_ps(rt);
 359 #else
 360             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 361 #endif
 362             twovfeps         = _mm_add_ps(vfeps,vfeps);
 363             vfitab           = _mm_slli_epi32(vfitab,2);
 364
 365             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 366             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 367             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 368             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 369             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 370             _MM_TRANSPOSE4_PS(Y,F,G,H);
 371             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 372             VV               = _mm_macc_ps(vfeps,Fp,Y);
 373             velec            = _mm_mul_ps(qq20,VV);
 374             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 375             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velecsum         = _mm_add_ps(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382              /* Update vectorial force */
 383             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 384             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 385             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 386
 387             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 388             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 389             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 390
 391             fjptrA             = f+j_coord_offsetA;
 392             fjptrB             = f+j_coord_offsetB;
 393             fjptrC             = f+j_coord_offsetC;
 394             fjptrD             = f+j_coord_offsetD;
 395
 396             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 397
 398             /* Inner loop uses 151 flops */
 399         }
 400
 401         if(jidx<j_index_end)
 402         {
 403
 404             /* Get j neighbor index, and coordinate index */
 405             jnrlistA         = jjnr[jidx];
 406             jnrlistB         = jjnr[jidx+1];
 407             jnrlistC         = jjnr[jidx+2];
 408             jnrlistD         = jjnr[jidx+3];
 409             /* Sign of each element will be negative for non-real atoms.
 410              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 411              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 412              */
 413             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 414             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 415             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 416             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 417             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 418             j_coord_offsetA  = DIM*jnrA;
 419             j_coord_offsetB  = DIM*jnrB;
 420             j_coord_offsetC  = DIM*jnrC;
 421             j_coord_offsetD  = DIM*jnrD;
 422
 423             /* load j atom coordinates */
 424             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 425                                               x+j_coord_offsetC,x+j_coord_offsetD,
 426                                               &jx0,&jy0,&jz0);
 427
 428             /* Calculate displacement vector */
 429             dx00             = _mm_sub_ps(ix0,jx0);
 430             dy00             = _mm_sub_ps(iy0,jy0);
 431             dz00             = _mm_sub_ps(iz0,jz0);
 432             dx10             = _mm_sub_ps(ix1,jx0);
 433             dy10             = _mm_sub_ps(iy1,jy0);
 434             dz10             = _mm_sub_ps(iz1,jz0);
 435             dx20             = _mm_sub_ps(ix2,jx0);
 436             dy20             = _mm_sub_ps(iy2,jy0);
 437             dz20             = _mm_sub_ps(iz2,jz0);
 438
 439             /* Calculate squared distance and things based on it */
 440             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 441             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 442             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 443
 444             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 445             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 446             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 447
 448             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 449
 450             /* Load parameters for j particles */
 451             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 452                                                               charge+jnrC+0,charge+jnrD+0);
 453             vdwjidx0A        = 2*vdwtype[jnrA+0];
 454             vdwjidx0B        = 2*vdwtype[jnrB+0];
 455             vdwjidx0C        = 2*vdwtype[jnrC+0];
 456             vdwjidx0D        = 2*vdwtype[jnrD+0];
 457
 458             fjx0             = _mm_setzero_ps();
 459             fjy0             = _mm_setzero_ps();
 460             fjz0             = _mm_setzero_ps();
 461
 462             /**************************
 463              * CALCULATE INTERACTIONS *
 464              **************************/
 465
 466             r00              = _mm_mul_ps(rsq00,rinv00);
 467             r00              = _mm_andnot_ps(dummy_mask,r00);
 468
 469             /* Compute parameters for interactions between i and j atoms */
 470             qq00             = _mm_mul_ps(iq0,jq0);
 471             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 472                                          vdwparam+vdwioffset0+vdwjidx0B,
 473                                          vdwparam+vdwioffset0+vdwjidx0C,
 474                                          vdwparam+vdwioffset0+vdwjidx0D,
 475                                          &c6_00,&c12_00);
 476
 477             /* Calculate table index by multiplying r with table scale and truncate to integer */
 478             rt               = _mm_mul_ps(r00,vftabscale);
 479             vfitab           = _mm_cvttps_epi32(rt);
 480 #ifdef __XOP__
 481             vfeps            = _mm_frcz_ps(rt);
 482 #else
 483             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 484 #endif
 485             twovfeps         = _mm_add_ps(vfeps,vfeps);
 486             vfitab           = _mm_slli_epi32(vfitab,2);
 487
 488             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 489             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 490             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 491             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 492             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 493             _MM_TRANSPOSE4_PS(Y,F,G,H);
 494             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 495             VV               = _mm_macc_ps(vfeps,Fp,Y);
 496             velec            = _mm_mul_ps(qq00,VV);
 497             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 498             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 499
 500             /* LENNARD-JONES DISPERSION/REPULSION */
 501
 502             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 503             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 504             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 505             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 506             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             velec            = _mm_andnot_ps(dummy_mask,velec);
 510             velecsum         = _mm_add_ps(velecsum,velec);
 511             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 512             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 513
 514             fscal            = _mm_add_ps(felec,fvdw);
 515
 516             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 517
 518              /* Update vectorial force */
 519             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 520             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 521             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 522
 523             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 524             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 525             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             r10              = _mm_mul_ps(rsq10,rinv10);
 532             r10              = _mm_andnot_ps(dummy_mask,r10);
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq10             = _mm_mul_ps(iq1,jq0);
 536
 537             /* Calculate table index by multiplying r with table scale and truncate to integer */
 538             rt               = _mm_mul_ps(r10,vftabscale);
 539             vfitab           = _mm_cvttps_epi32(rt);
 540 #ifdef __XOP__
 541             vfeps            = _mm_frcz_ps(rt);
 542 #else
 543             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 544 #endif
 545             twovfeps         = _mm_add_ps(vfeps,vfeps);
 546             vfitab           = _mm_slli_epi32(vfitab,2);
 547
 548             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 549             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 550             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 551             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 552             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 553             _MM_TRANSPOSE4_PS(Y,F,G,H);
 554             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 555             VV               = _mm_macc_ps(vfeps,Fp,Y);
 556             velec            = _mm_mul_ps(qq10,VV);
 557             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 558             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             velec            = _mm_andnot_ps(dummy_mask,velec);
 562             velecsum         = _mm_add_ps(velecsum,velec);
 563
 564             fscal            = felec;
 565
 566             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 567
 568              /* Update vectorial force */
 569             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 570             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 571             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 572
 573             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 574             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 575             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             r20              = _mm_mul_ps(rsq20,rinv20);
 582             r20              = _mm_andnot_ps(dummy_mask,r20);
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq20             = _mm_mul_ps(iq2,jq0);
 586
 587             /* Calculate table index by multiplying r with table scale and truncate to integer */
 588             rt               = _mm_mul_ps(r20,vftabscale);
 589             vfitab           = _mm_cvttps_epi32(rt);
 590 #ifdef __XOP__
 591             vfeps            = _mm_frcz_ps(rt);
 592 #else
 593             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 594 #endif
 595             twovfeps         = _mm_add_ps(vfeps,vfeps);
 596             vfitab           = _mm_slli_epi32(vfitab,2);
 597
 598             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 599             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 600             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 601             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 602             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 603             _MM_TRANSPOSE4_PS(Y,F,G,H);
 604             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 605             VV               = _mm_macc_ps(vfeps,Fp,Y);
 606             velec            = _mm_mul_ps(qq20,VV);
 607             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 608             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 609
 610             /* Update potential sum for this i atom from the interaction with this j atom. */
 611             velec            = _mm_andnot_ps(dummy_mask,velec);
 612             velecsum         = _mm_add_ps(velecsum,velec);
 613
 614             fscal            = felec;
 615
 616             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 617
 618              /* Update vectorial force */
 619             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 620             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 621             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 622
 623             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 624             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 625             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 626
 627             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 628             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 629             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 630             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 631
 632             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 633
 634             /* Inner loop uses 154 flops */
 635         }
 636
 637         /* End of innermost loop */
 638
 639         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 640                                               f+i_coord_offset,fshift+i_shift_offset);
 641
 642         ggid                        = gid[iidx];
 643         /* Update potential energies */
 644         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 645         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 646
 647         /* Increment number of inner iterations */
 648         inneriter                  += j_index_end - j_index_start;
 649
 650         /* Outer loop uses 20 flops */
 651     }
 652
 653     /* Increment number of outer iterations */
 654     outeriter        += nri;
 655
 656     /* Update outer/inner flops */
 657
 658     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 659 }
 660 /*
 661  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_single
 662  * Electrostatics interaction: CubicSplineTable
 663  * VdW interaction:            LennardJones
 664  * Geometry:                   Water3-Particle
 665  * Calculate force/pot:        Force
 666  */
 667 void
 668 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_single
 669                     (t_nblist                    * gmx_restrict       nlist,
 670                      rvec                        * gmx_restrict          xx,
 671                      rvec                        * gmx_restrict          ff,
 672                      t_forcerec                  * gmx_restrict          fr,
 673                      t_mdatoms                   * gmx_restrict     mdatoms,
 674                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 675                      t_nrnb                      * gmx_restrict        nrnb)
 676 {
 677     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 678      * just 0 for non-waters.
 679      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 680      * jnr indices corresponding to data put in the four positions in the SIMD register.
 681      */
 682     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 683     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 684     int              jnrA,jnrB,jnrC,jnrD;
 685     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 686     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 687     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 688     real             rcutoff_scalar;
 689     real             *shiftvec,*fshift,*x,*f;
 690     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 691     real             scratch[4*DIM];
 692     __m128           fscal,rcutoff,rcutoff2,jidxall;
 693     int              vdwioffset0;
 694     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 695     int              vdwioffset1;
 696     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 697     int              vdwioffset2;
 698     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 699     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 700     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 701     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 702     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 703     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 704     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 705     real             *charge;
 706     int              nvdwtype;
 707     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 708     int              *vdwtype;
 709     real             *vdwparam;
 710     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 711     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 712     __m128i          vfitab;
 713     __m128i          ifour       = _mm_set1_epi32(4);
 714     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 715     real             *vftab;
 716     __m128           dummy_mask,cutoff_mask;
 717     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 718     __m128           one     = _mm_set1_ps(1.0);
 719     __m128           two     = _mm_set1_ps(2.0);
 720     x                = xx[0];
 721     f                = ff[0];
 722
 723     nri              = nlist->nri;
 724     iinr             = nlist->iinr;
 725     jindex           = nlist->jindex;
 726     jjnr             = nlist->jjnr;
 727     shiftidx         = nlist->shift;
 728     gid              = nlist->gid;
 729     shiftvec         = fr->shift_vec[0];
 730     fshift           = fr->fshift[0];
 731     facel            = _mm_set1_ps(fr->epsfac);
 732     charge           = mdatoms->chargeA;
 733     nvdwtype         = fr->ntype;
 734     vdwparam         = fr->nbfp;
 735     vdwtype          = mdatoms->typeA;
 736
 737     vftab            = kernel_data->table_elec->data;
 738     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 739
 740     /* Setup water-specific parameters */
 741     inr              = nlist->iinr[0];
 742     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 743     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 744     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 745     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 746
 747     /* Avoid stupid compiler warnings */
 748     jnrA = jnrB = jnrC = jnrD = 0;
 749     j_coord_offsetA = 0;
 750     j_coord_offsetB = 0;
 751     j_coord_offsetC = 0;
 752     j_coord_offsetD = 0;
 753
 754     outeriter        = 0;
 755     inneriter        = 0;
 756
 757     for(iidx=0;iidx<4*DIM;iidx++)
 758     {
 759         scratch[iidx] = 0.0;
 760     }
 761
 762     /* Start outer loop over neighborlists */
 763     for(iidx=0; iidx<nri; iidx++)
 764     {
 765         /* Load shift vector for this list */
 766         i_shift_offset   = DIM*shiftidx[iidx];
 767
 768         /* Load limits for loop over neighbors */
 769         j_index_start    = jindex[iidx];
 770         j_index_end      = jindex[iidx+1];
 771
 772         /* Get outer coordinate index */
 773         inr              = iinr[iidx];
 774         i_coord_offset   = DIM*inr;
 775
 776         /* Load i particle coords and add shift vector */
 777         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 778                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 779
 780         fix0             = _mm_setzero_ps();
 781         fiy0             = _mm_setzero_ps();
 782         fiz0             = _mm_setzero_ps();
 783         fix1             = _mm_setzero_ps();
 784         fiy1             = _mm_setzero_ps();
 785         fiz1             = _mm_setzero_ps();
 786         fix2             = _mm_setzero_ps();
 787         fiy2             = _mm_setzero_ps();
 788         fiz2             = _mm_setzero_ps();
 789
 790         /* Start inner kernel loop */
 791         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 792         {
 793
 794             /* Get j neighbor index, and coordinate index */
 795             jnrA             = jjnr[jidx];
 796             jnrB             = jjnr[jidx+1];
 797             jnrC             = jjnr[jidx+2];
 798             jnrD             = jjnr[jidx+3];
 799             j_coord_offsetA  = DIM*jnrA;
 800             j_coord_offsetB  = DIM*jnrB;
 801             j_coord_offsetC  = DIM*jnrC;
 802             j_coord_offsetD  = DIM*jnrD;
 803
 804             /* load j atom coordinates */
 805             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 806                                               x+j_coord_offsetC,x+j_coord_offsetD,
 807                                               &jx0,&jy0,&jz0);
 808
 809             /* Calculate displacement vector */
 810             dx00             = _mm_sub_ps(ix0,jx0);
 811             dy00             = _mm_sub_ps(iy0,jy0);
 812             dz00             = _mm_sub_ps(iz0,jz0);
 813             dx10             = _mm_sub_ps(ix1,jx0);
 814             dy10             = _mm_sub_ps(iy1,jy0);
 815             dz10             = _mm_sub_ps(iz1,jz0);
 816             dx20             = _mm_sub_ps(ix2,jx0);
 817             dy20             = _mm_sub_ps(iy2,jy0);
 818             dz20             = _mm_sub_ps(iz2,jz0);
 819
 820             /* Calculate squared distance and things based on it */
 821             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 822             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 823             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 824
 825             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 826             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 827             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 828
 829             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 830
 831             /* Load parameters for j particles */
 832             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 833                                                               charge+jnrC+0,charge+jnrD+0);
 834             vdwjidx0A        = 2*vdwtype[jnrA+0];
 835             vdwjidx0B        = 2*vdwtype[jnrB+0];
 836             vdwjidx0C        = 2*vdwtype[jnrC+0];
 837             vdwjidx0D        = 2*vdwtype[jnrD+0];
 838
 839             fjx0             = _mm_setzero_ps();
 840             fjy0             = _mm_setzero_ps();
 841             fjz0             = _mm_setzero_ps();
 842
 843             /**************************
 844              * CALCULATE INTERACTIONS *
 845              **************************/
 846
 847             r00              = _mm_mul_ps(rsq00,rinv00);
 848
 849             /* Compute parameters for interactions between i and j atoms */
 850             qq00             = _mm_mul_ps(iq0,jq0);
 851             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 852                                          vdwparam+vdwioffset0+vdwjidx0B,
 853                                          vdwparam+vdwioffset0+vdwjidx0C,
 854                                          vdwparam+vdwioffset0+vdwjidx0D,
 855                                          &c6_00,&c12_00);
 856
 857             /* Calculate table index by multiplying r with table scale and truncate to integer */
 858             rt               = _mm_mul_ps(r00,vftabscale);
 859             vfitab           = _mm_cvttps_epi32(rt);
 860 #ifdef __XOP__
 861             vfeps            = _mm_frcz_ps(rt);
 862 #else
 863             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 864 #endif
 865             twovfeps         = _mm_add_ps(vfeps,vfeps);
 866             vfitab           = _mm_slli_epi32(vfitab,2);
 867
 868             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 869             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 870             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 871             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 872             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 873             _MM_TRANSPOSE4_PS(Y,F,G,H);
 874             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 875             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 876             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 877
 878             /* LENNARD-JONES DISPERSION/REPULSION */
 879
 880             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 881             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 882
 883             fscal            = _mm_add_ps(felec,fvdw);
 884
 885              /* Update vectorial force */
 886             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 887             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 888             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 889
 890             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 891             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 892             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             r10              = _mm_mul_ps(rsq10,rinv10);
 899
 900             /* Compute parameters for interactions between i and j atoms */
 901             qq10             = _mm_mul_ps(iq1,jq0);
 902
 903             /* Calculate table index by multiplying r with table scale and truncate to integer */
 904             rt               = _mm_mul_ps(r10,vftabscale);
 905             vfitab           = _mm_cvttps_epi32(rt);
 906 #ifdef __XOP__
 907             vfeps            = _mm_frcz_ps(rt);
 908 #else
 909             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 910 #endif
 911             twovfeps         = _mm_add_ps(vfeps,vfeps);
 912             vfitab           = _mm_slli_epi32(vfitab,2);
 913
 914             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 915             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 916             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 917             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 918             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 919             _MM_TRANSPOSE4_PS(Y,F,G,H);
 920             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 921             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 922             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 923
 924             fscal            = felec;
 925
 926              /* Update vectorial force */
 927             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 928             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 929             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 930
 931             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 932             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 933             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 934
 935             /**************************
 936              * CALCULATE INTERACTIONS *
 937              **************************/
 938
 939             r20              = _mm_mul_ps(rsq20,rinv20);
 940
 941             /* Compute parameters for interactions between i and j atoms */
 942             qq20             = _mm_mul_ps(iq2,jq0);
 943
 944             /* Calculate table index by multiplying r with table scale and truncate to integer */
 945             rt               = _mm_mul_ps(r20,vftabscale);
 946             vfitab           = _mm_cvttps_epi32(rt);
 947 #ifdef __XOP__
 948             vfeps            = _mm_frcz_ps(rt);
 949 #else
 950             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 951 #endif
 952             twovfeps         = _mm_add_ps(vfeps,vfeps);
 953             vfitab           = _mm_slli_epi32(vfitab,2);
 954
 955             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 956             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 957             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 958             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 959             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 960             _MM_TRANSPOSE4_PS(Y,F,G,H);
 961             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 962             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 963             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 964
 965             fscal            = felec;
 966
 967              /* Update vectorial force */
 968             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 969             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 970             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 971
 972             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 973             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 974             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 975
 976             fjptrA             = f+j_coord_offsetA;
 977             fjptrB             = f+j_coord_offsetB;
 978             fjptrC             = f+j_coord_offsetC;
 979             fjptrD             = f+j_coord_offsetD;
 980
 981             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 982
 983             /* Inner loop uses 134 flops */
 984         }
 985
 986         if(jidx<j_index_end)
 987         {
 988
 989             /* Get j neighbor index, and coordinate index */
 990             jnrlistA         = jjnr[jidx];
 991             jnrlistB         = jjnr[jidx+1];
 992             jnrlistC         = jjnr[jidx+2];
 993             jnrlistD         = jjnr[jidx+3];
 994             /* Sign of each element will be negative for non-real atoms.
 995              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 996              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 997              */
 998             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 999             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1000             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1001             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1002             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1003             j_coord_offsetA  = DIM*jnrA;
1004             j_coord_offsetB  = DIM*jnrB;
1005             j_coord_offsetC  = DIM*jnrC;
1006             j_coord_offsetD  = DIM*jnrD;
1007
1008             /* load j atom coordinates */
1009             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1010                                               x+j_coord_offsetC,x+j_coord_offsetD,
1011                                               &jx0,&jy0,&jz0);
1012
1013             /* Calculate displacement vector */
1014             dx00             = _mm_sub_ps(ix0,jx0);
1015             dy00             = _mm_sub_ps(iy0,jy0);
1016             dz00             = _mm_sub_ps(iz0,jz0);
1017             dx10             = _mm_sub_ps(ix1,jx0);
1018             dy10             = _mm_sub_ps(iy1,jy0);
1019             dz10             = _mm_sub_ps(iz1,jz0);
1020             dx20             = _mm_sub_ps(ix2,jx0);
1021             dy20             = _mm_sub_ps(iy2,jy0);
1022             dz20             = _mm_sub_ps(iz2,jz0);
1023
1024             /* Calculate squared distance and things based on it */
1025             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1026             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1027             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1028
1029             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1030             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1031             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1032
1033             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1034
1035             /* Load parameters for j particles */
1036             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1037                                                               charge+jnrC+0,charge+jnrD+0);
1038             vdwjidx0A        = 2*vdwtype[jnrA+0];
1039             vdwjidx0B        = 2*vdwtype[jnrB+0];
1040             vdwjidx0C        = 2*vdwtype[jnrC+0];
1041             vdwjidx0D        = 2*vdwtype[jnrD+0];
1042
1043             fjx0             = _mm_setzero_ps();
1044             fjy0             = _mm_setzero_ps();
1045             fjz0             = _mm_setzero_ps();
1046
1047             /**************************
1048              * CALCULATE INTERACTIONS *
1049              **************************/
1050
1051             r00              = _mm_mul_ps(rsq00,rinv00);
1052             r00              = _mm_andnot_ps(dummy_mask,r00);
1053
1054             /* Compute parameters for interactions between i and j atoms */
1055             qq00             = _mm_mul_ps(iq0,jq0);
1056             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1057                                          vdwparam+vdwioffset0+vdwjidx0B,
1058                                          vdwparam+vdwioffset0+vdwjidx0C,
1059                                          vdwparam+vdwioffset0+vdwjidx0D,
1060                                          &c6_00,&c12_00);
1061
1062             /* Calculate table index by multiplying r with table scale and truncate to integer */
1063             rt               = _mm_mul_ps(r00,vftabscale);
1064             vfitab           = _mm_cvttps_epi32(rt);
1065 #ifdef __XOP__
1066             vfeps            = _mm_frcz_ps(rt);
1067 #else
1068             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1069 #endif
1070             twovfeps         = _mm_add_ps(vfeps,vfeps);
1071             vfitab           = _mm_slli_epi32(vfitab,2);
1072
1073             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1074             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1075             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1076             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1077             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1078             _MM_TRANSPOSE4_PS(Y,F,G,H);
1079             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1080             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1081             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1082
1083             /* LENNARD-JONES DISPERSION/REPULSION */
1084
1085             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1086             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1087
1088             fscal            = _mm_add_ps(felec,fvdw);
1089
1090             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1091
1092              /* Update vectorial force */
1093             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1094             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1095             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1096
1097             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1098             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1099             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1100
1101             /**************************
1102              * CALCULATE INTERACTIONS *
1103              **************************/
1104
1105             r10              = _mm_mul_ps(rsq10,rinv10);
1106             r10              = _mm_andnot_ps(dummy_mask,r10);
1107
1108             /* Compute parameters for interactions between i and j atoms */
1109             qq10             = _mm_mul_ps(iq1,jq0);
1110
1111             /* Calculate table index by multiplying r with table scale and truncate to integer */
1112             rt               = _mm_mul_ps(r10,vftabscale);
1113             vfitab           = _mm_cvttps_epi32(rt);
1114 #ifdef __XOP__
1115             vfeps            = _mm_frcz_ps(rt);
1116 #else
1117             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1118 #endif
1119             twovfeps         = _mm_add_ps(vfeps,vfeps);
1120             vfitab           = _mm_slli_epi32(vfitab,2);
1121
1122             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1123             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1124             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1125             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1126             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1127             _MM_TRANSPOSE4_PS(Y,F,G,H);
1128             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1129             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1130             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1131
1132             fscal            = felec;
1133
1134             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1135
1136              /* Update vectorial force */
1137             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1138             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1139             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1140
1141             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1142             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1143             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1144
1145             /**************************
1146              * CALCULATE INTERACTIONS *
1147              **************************/
1148
1149             r20              = _mm_mul_ps(rsq20,rinv20);
1150             r20              = _mm_andnot_ps(dummy_mask,r20);
1151
1152             /* Compute parameters for interactions between i and j atoms */
1153             qq20             = _mm_mul_ps(iq2,jq0);
1154
1155             /* Calculate table index by multiplying r with table scale and truncate to integer */
1156             rt               = _mm_mul_ps(r20,vftabscale);
1157             vfitab           = _mm_cvttps_epi32(rt);
1158 #ifdef __XOP__
1159             vfeps            = _mm_frcz_ps(rt);
1160 #else
1161             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1162 #endif
1163             twovfeps         = _mm_add_ps(vfeps,vfeps);
1164             vfitab           = _mm_slli_epi32(vfitab,2);
1165
1166             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1167             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1168             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1169             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1170             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1171             _MM_TRANSPOSE4_PS(Y,F,G,H);
1172             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1173             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1174             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1175
1176             fscal            = felec;
1177
1178             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1179
1180              /* Update vectorial force */
1181             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1182             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1183             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1184
1185             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1186             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1187             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1188
1189             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1190             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1191             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1192             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1193
1194             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1195
1196             /* Inner loop uses 137 flops */
1197         }
1198
1199         /* End of innermost loop */
1200
1201         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1202                                               f+i_coord_offset,fshift+i_shift_offset);
1203
1204         /* Increment number of inner iterations */
1205         inneriter                  += j_index_end - j_index_start;
1206
1207         /* Outer loop uses 18 flops */
1208     }
1209
1210     /* Increment number of outer iterations */
1211     outeriter        += nri;
1212
1213     /* Update outer/inner flops */
1214
1215     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1216 }