src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwCSTab_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwCSTab_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     krf              = _mm_set1_ps(fr->ic->k_rf);
 124     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 125     crf              = _mm_set1_ps(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 136     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 137     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->rcoulomb;
 142     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 143     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = jnrC = jnrD = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149     j_coord_offsetC = 0;
 150     j_coord_offsetD = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm_setzero_ps();
 179         fiy0             = _mm_setzero_ps();
 180         fiz0             = _mm_setzero_ps();
 181         fix1             = _mm_setzero_ps();
 182         fiy1             = _mm_setzero_ps();
 183         fiz1             = _mm_setzero_ps();
 184         fix2             = _mm_setzero_ps();
 185         fiy2             = _mm_setzero_ps();
 186         fiz2             = _mm_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm_setzero_ps();
 190         vvdwsum          = _mm_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               x+j_coord_offsetC,x+j_coord_offsetD,
 209                                               &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm_sub_ps(ix0,jx0);
 213             dy00             = _mm_sub_ps(iy0,jy0);
 214             dz00             = _mm_sub_ps(iz0,jz0);
 215             dx10             = _mm_sub_ps(ix1,jx0);
 216             dy10             = _mm_sub_ps(iy1,jy0);
 217             dz10             = _mm_sub_ps(iz1,jz0);
 218             dx20             = _mm_sub_ps(ix2,jx0);
 219             dy20             = _mm_sub_ps(iy2,jy0);
 220             dz20             = _mm_sub_ps(iz2,jz0);
 221
 222             /* Calculate squared distance and things based on it */
 223             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 224             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 225             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 226
 227             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 228             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 229             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 230
 231             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 232             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 233             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 234
 235             /* Load parameters for j particles */
 236             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 237                                                               charge+jnrC+0,charge+jnrD+0);
 238             vdwjidx0A        = 2*vdwtype[jnrA+0];
 239             vdwjidx0B        = 2*vdwtype[jnrB+0];
 240             vdwjidx0C        = 2*vdwtype[jnrC+0];
 241             vdwjidx0D        = 2*vdwtype[jnrD+0];
 242
 243             fjx0             = _mm_setzero_ps();
 244             fjy0             = _mm_setzero_ps();
 245             fjz0             = _mm_setzero_ps();
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             if (gmx_mm_any_lt(rsq00,rcutoff2))
 252             {
 253
 254             r00              = _mm_mul_ps(rsq00,rinv00);
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq00             = _mm_mul_ps(iq0,jq0);
 258             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 259                                          vdwparam+vdwioffset0+vdwjidx0B,
 260                                          vdwparam+vdwioffset0+vdwjidx0C,
 261                                          vdwparam+vdwioffset0+vdwjidx0D,
 262                                          &c6_00,&c12_00);
 263
 264             /* Calculate table index by multiplying r with table scale and truncate to integer */
 265             rt               = _mm_mul_ps(r00,vftabscale);
 266             vfitab           = _mm_cvttps_epi32(rt);
 267 #ifdef __XOP__
 268             vfeps            = _mm_frcz_ps(rt);
 269 #else
 270             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 271 #endif
 272             twovfeps         = _mm_add_ps(vfeps,vfeps);
 273             vfitab           = _mm_slli_epi32(vfitab,3);
 274
 275             /* REACTION-FIELD ELECTROSTATICS */
 276             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 277             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 278
 279             /* CUBIC SPLINE TABLE DISPERSION */
 280             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 281             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 282             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 283             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 284             _MM_TRANSPOSE4_PS(Y,F,G,H);
 285             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 286             VV               = _mm_macc_ps(vfeps,Fp,Y);
 287             vvdw6            = _mm_mul_ps(c6_00,VV);
 288             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 289             fvdw6            = _mm_mul_ps(c6_00,FF);
 290
 291             /* CUBIC SPLINE TABLE REPULSION */
 292             vfitab           = _mm_add_epi32(vfitab,ifour);
 293             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 294             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 295             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 296             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 297             _MM_TRANSPOSE4_PS(Y,F,G,H);
 298             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 299             VV               = _mm_macc_ps(vfeps,Fp,Y);
 300             vvdw12           = _mm_mul_ps(c12_00,VV);
 301             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 302             fvdw12           = _mm_mul_ps(c12_00,FF);
 303             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 304             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 305
 306             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             velec            = _mm_and_ps(velec,cutoff_mask);
 310             velecsum         = _mm_add_ps(velecsum,velec);
 311             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 312             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 313
 314             fscal            = _mm_add_ps(felec,fvdw);
 315
 316             fscal            = _mm_and_ps(fscal,cutoff_mask);
 317
 318              /* Update vectorial force */
 319             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 320             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 321             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 322
 323             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 324             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 325             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 326
 327             }
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             if (gmx_mm_any_lt(rsq10,rcutoff2))
 334             {
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq10             = _mm_mul_ps(iq1,jq0);
 338
 339             /* REACTION-FIELD ELECTROSTATICS */
 340             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 341             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 342
 343             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velec            = _mm_and_ps(velec,cutoff_mask);
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             fscal            = _mm_and_ps(fscal,cutoff_mask);
 352
 353              /* Update vectorial force */
 354             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 355             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 356             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 357
 358             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 359             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 360             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 361
 362             }
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             if (gmx_mm_any_lt(rsq20,rcutoff2))
 369             {
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq20             = _mm_mul_ps(iq2,jq0);
 373
 374             /* REACTION-FIELD ELECTROSTATICS */
 375             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 376             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 377
 378             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velec            = _mm_and_ps(velec,cutoff_mask);
 382             velecsum         = _mm_add_ps(velecsum,velec);
 383
 384             fscal            = felec;
 385
 386             fscal            = _mm_and_ps(fscal,cutoff_mask);
 387
 388              /* Update vectorial force */
 389             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 390             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 391             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 392
 393             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 394             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 395             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 396
 397             }
 398
 399             fjptrA             = f+j_coord_offsetA;
 400             fjptrB             = f+j_coord_offsetB;
 401             fjptrC             = f+j_coord_offsetC;
 402             fjptrD             = f+j_coord_offsetD;
 403
 404             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 405
 406             /* Inner loop uses 153 flops */
 407         }
 408
 409         if(jidx<j_index_end)
 410         {
 411
 412             /* Get j neighbor index, and coordinate index */
 413             jnrlistA         = jjnr[jidx];
 414             jnrlistB         = jjnr[jidx+1];
 415             jnrlistC         = jjnr[jidx+2];
 416             jnrlistD         = jjnr[jidx+3];
 417             /* Sign of each element will be negative for non-real atoms.
 418              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 419              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 420              */
 421             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 422             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 423             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 424             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 425             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 426             j_coord_offsetA  = DIM*jnrA;
 427             j_coord_offsetB  = DIM*jnrB;
 428             j_coord_offsetC  = DIM*jnrC;
 429             j_coord_offsetD  = DIM*jnrD;
 430
 431             /* load j atom coordinates */
 432             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 433                                               x+j_coord_offsetC,x+j_coord_offsetD,
 434                                               &jx0,&jy0,&jz0);
 435
 436             /* Calculate displacement vector */
 437             dx00             = _mm_sub_ps(ix0,jx0);
 438             dy00             = _mm_sub_ps(iy0,jy0);
 439             dz00             = _mm_sub_ps(iz0,jz0);
 440             dx10             = _mm_sub_ps(ix1,jx0);
 441             dy10             = _mm_sub_ps(iy1,jy0);
 442             dz10             = _mm_sub_ps(iz1,jz0);
 443             dx20             = _mm_sub_ps(ix2,jx0);
 444             dy20             = _mm_sub_ps(iy2,jy0);
 445             dz20             = _mm_sub_ps(iz2,jz0);
 446
 447             /* Calculate squared distance and things based on it */
 448             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 449             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 450             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 451
 452             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 453             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 454             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 455
 456             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 457             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 458             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 459
 460             /* Load parameters for j particles */
 461             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 462                                                               charge+jnrC+0,charge+jnrD+0);
 463             vdwjidx0A        = 2*vdwtype[jnrA+0];
 464             vdwjidx0B        = 2*vdwtype[jnrB+0];
 465             vdwjidx0C        = 2*vdwtype[jnrC+0];
 466             vdwjidx0D        = 2*vdwtype[jnrD+0];
 467
 468             fjx0             = _mm_setzero_ps();
 469             fjy0             = _mm_setzero_ps();
 470             fjz0             = _mm_setzero_ps();
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             if (gmx_mm_any_lt(rsq00,rcutoff2))
 477             {
 478
 479             r00              = _mm_mul_ps(rsq00,rinv00);
 480             r00              = _mm_andnot_ps(dummy_mask,r00);
 481
 482             /* Compute parameters for interactions between i and j atoms */
 483             qq00             = _mm_mul_ps(iq0,jq0);
 484             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 485                                          vdwparam+vdwioffset0+vdwjidx0B,
 486                                          vdwparam+vdwioffset0+vdwjidx0C,
 487                                          vdwparam+vdwioffset0+vdwjidx0D,
 488                                          &c6_00,&c12_00);
 489
 490             /* Calculate table index by multiplying r with table scale and truncate to integer */
 491             rt               = _mm_mul_ps(r00,vftabscale);
 492             vfitab           = _mm_cvttps_epi32(rt);
 493 #ifdef __XOP__
 494             vfeps            = _mm_frcz_ps(rt);
 495 #else
 496             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 497 #endif
 498             twovfeps         = _mm_add_ps(vfeps,vfeps);
 499             vfitab           = _mm_slli_epi32(vfitab,3);
 500
 501             /* REACTION-FIELD ELECTROSTATICS */
 502             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 503             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 504
 505             /* CUBIC SPLINE TABLE DISPERSION */
 506             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 507             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 508             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 509             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 510             _MM_TRANSPOSE4_PS(Y,F,G,H);
 511             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 512             VV               = _mm_macc_ps(vfeps,Fp,Y);
 513             vvdw6            = _mm_mul_ps(c6_00,VV);
 514             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 515             fvdw6            = _mm_mul_ps(c6_00,FF);
 516
 517             /* CUBIC SPLINE TABLE REPULSION */
 518             vfitab           = _mm_add_epi32(vfitab,ifour);
 519             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 520             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 521             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 522             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 523             _MM_TRANSPOSE4_PS(Y,F,G,H);
 524             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 525             VV               = _mm_macc_ps(vfeps,Fp,Y);
 526             vvdw12           = _mm_mul_ps(c12_00,VV);
 527             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 528             fvdw12           = _mm_mul_ps(c12_00,FF);
 529             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 530             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 531
 532             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velec            = _mm_and_ps(velec,cutoff_mask);
 536             velec            = _mm_andnot_ps(dummy_mask,velec);
 537             velecsum         = _mm_add_ps(velecsum,velec);
 538             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 539             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 540             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 541
 542             fscal            = _mm_add_ps(felec,fvdw);
 543
 544             fscal            = _mm_and_ps(fscal,cutoff_mask);
 545
 546             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 547
 548              /* Update vectorial force */
 549             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 550             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 551             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 552
 553             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 554             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 555             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 556
 557             }
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             if (gmx_mm_any_lt(rsq10,rcutoff2))
 564             {
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq10             = _mm_mul_ps(iq1,jq0);
 568
 569             /* REACTION-FIELD ELECTROSTATICS */
 570             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 571             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 572
 573             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 574
 575             /* Update potential sum for this i atom from the interaction with this j atom. */
 576             velec            = _mm_and_ps(velec,cutoff_mask);
 577             velec            = _mm_andnot_ps(dummy_mask,velec);
 578             velecsum         = _mm_add_ps(velecsum,velec);
 579
 580             fscal            = felec;
 581
 582             fscal            = _mm_and_ps(fscal,cutoff_mask);
 583
 584             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 585
 586              /* Update vectorial force */
 587             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 588             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 589             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 590
 591             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 592             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 593             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 594
 595             }
 596
 597             /**************************
 598              * CALCULATE INTERACTIONS *
 599              **************************/
 600
 601             if (gmx_mm_any_lt(rsq20,rcutoff2))
 602             {
 603
 604             /* Compute parameters for interactions between i and j atoms */
 605             qq20             = _mm_mul_ps(iq2,jq0);
 606
 607             /* REACTION-FIELD ELECTROSTATICS */
 608             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 609             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 610
 611             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 612
 613             /* Update potential sum for this i atom from the interaction with this j atom. */
 614             velec            = _mm_and_ps(velec,cutoff_mask);
 615             velec            = _mm_andnot_ps(dummy_mask,velec);
 616             velecsum         = _mm_add_ps(velecsum,velec);
 617
 618             fscal            = felec;
 619
 620             fscal            = _mm_and_ps(fscal,cutoff_mask);
 621
 622             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 623
 624              /* Update vectorial force */
 625             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 626             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 627             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 628
 629             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 630             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 631             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 632
 633             }
 634
 635             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 636             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 637             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 638             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 639
 640             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 641
 642             /* Inner loop uses 154 flops */
 643         }
 644
 645         /* End of innermost loop */
 646
 647         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 648                                               f+i_coord_offset,fshift+i_shift_offset);
 649
 650         ggid                        = gid[iidx];
 651         /* Update potential energies */
 652         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 653         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 654
 655         /* Increment number of inner iterations */
 656         inneriter                  += j_index_end - j_index_start;
 657
 658         /* Outer loop uses 20 flops */
 659     }
 660
 661     /* Increment number of outer iterations */
 662     outeriter        += nri;
 663
 664     /* Update outer/inner flops */
 665
 666     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 667 }
 668 /*
 669  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 670  * Electrostatics interaction: ReactionField
 671  * VdW interaction:            CubicSplineTable
 672  * Geometry:                   Water3-Particle
 673  * Calculate force/pot:        Force
 674  */
 675 void
 676 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 677                     (t_nblist                    * gmx_restrict       nlist,
 678                      rvec                        * gmx_restrict          xx,
 679                      rvec                        * gmx_restrict          ff,
 680                      t_forcerec                  * gmx_restrict          fr,
 681                      t_mdatoms                   * gmx_restrict     mdatoms,
 682                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 683                      t_nrnb                      * gmx_restrict        nrnb)
 684 {
 685     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 686      * just 0 for non-waters.
 687      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 688      * jnr indices corresponding to data put in the four positions in the SIMD register.
 689      */
 690     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 691     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 692     int              jnrA,jnrB,jnrC,jnrD;
 693     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 694     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 695     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 696     real             rcutoff_scalar;
 697     real             *shiftvec,*fshift,*x,*f;
 698     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 699     real             scratch[4*DIM];
 700     __m128           fscal,rcutoff,rcutoff2,jidxall;
 701     int              vdwioffset0;
 702     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 703     int              vdwioffset1;
 704     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 705     int              vdwioffset2;
 706     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 707     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 708     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 709     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 710     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 711     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 712     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 713     real             *charge;
 714     int              nvdwtype;
 715     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 716     int              *vdwtype;
 717     real             *vdwparam;
 718     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 719     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 720     __m128i          vfitab;
 721     __m128i          ifour       = _mm_set1_epi32(4);
 722     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 723     real             *vftab;
 724     __m128           dummy_mask,cutoff_mask;
 725     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 726     __m128           one     = _mm_set1_ps(1.0);
 727     __m128           two     = _mm_set1_ps(2.0);
 728     x                = xx[0];
 729     f                = ff[0];
 730
 731     nri              = nlist->nri;
 732     iinr             = nlist->iinr;
 733     jindex           = nlist->jindex;
 734     jjnr             = nlist->jjnr;
 735     shiftidx         = nlist->shift;
 736     gid              = nlist->gid;
 737     shiftvec         = fr->shift_vec[0];
 738     fshift           = fr->fshift[0];
 739     facel            = _mm_set1_ps(fr->epsfac);
 740     charge           = mdatoms->chargeA;
 741     krf              = _mm_set1_ps(fr->ic->k_rf);
 742     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 743     crf              = _mm_set1_ps(fr->ic->c_rf);
 744     nvdwtype         = fr->ntype;
 745     vdwparam         = fr->nbfp;
 746     vdwtype          = mdatoms->typeA;
 747
 748     vftab            = kernel_data->table_vdw->data;
 749     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 750
 751     /* Setup water-specific parameters */
 752     inr              = nlist->iinr[0];
 753     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 754     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 755     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 756     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 757
 758     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 759     rcutoff_scalar   = fr->rcoulomb;
 760     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 761     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 762
 763     /* Avoid stupid compiler warnings */
 764     jnrA = jnrB = jnrC = jnrD = 0;
 765     j_coord_offsetA = 0;
 766     j_coord_offsetB = 0;
 767     j_coord_offsetC = 0;
 768     j_coord_offsetD = 0;
 769
 770     outeriter        = 0;
 771     inneriter        = 0;
 772
 773     for(iidx=0;iidx<4*DIM;iidx++)
 774     {
 775         scratch[iidx] = 0.0;
 776     }
 777
 778     /* Start outer loop over neighborlists */
 779     for(iidx=0; iidx<nri; iidx++)
 780     {
 781         /* Load shift vector for this list */
 782         i_shift_offset   = DIM*shiftidx[iidx];
 783
 784         /* Load limits for loop over neighbors */
 785         j_index_start    = jindex[iidx];
 786         j_index_end      = jindex[iidx+1];
 787
 788         /* Get outer coordinate index */
 789         inr              = iinr[iidx];
 790         i_coord_offset   = DIM*inr;
 791
 792         /* Load i particle coords and add shift vector */
 793         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 794                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 795
 796         fix0             = _mm_setzero_ps();
 797         fiy0             = _mm_setzero_ps();
 798         fiz0             = _mm_setzero_ps();
 799         fix1             = _mm_setzero_ps();
 800         fiy1             = _mm_setzero_ps();
 801         fiz1             = _mm_setzero_ps();
 802         fix2             = _mm_setzero_ps();
 803         fiy2             = _mm_setzero_ps();
 804         fiz2             = _mm_setzero_ps();
 805
 806         /* Start inner kernel loop */
 807         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 808         {
 809
 810             /* Get j neighbor index, and coordinate index */
 811             jnrA             = jjnr[jidx];
 812             jnrB             = jjnr[jidx+1];
 813             jnrC             = jjnr[jidx+2];
 814             jnrD             = jjnr[jidx+3];
 815             j_coord_offsetA  = DIM*jnrA;
 816             j_coord_offsetB  = DIM*jnrB;
 817             j_coord_offsetC  = DIM*jnrC;
 818             j_coord_offsetD  = DIM*jnrD;
 819
 820             /* load j atom coordinates */
 821             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 822                                               x+j_coord_offsetC,x+j_coord_offsetD,
 823                                               &jx0,&jy0,&jz0);
 824
 825             /* Calculate displacement vector */
 826             dx00             = _mm_sub_ps(ix0,jx0);
 827             dy00             = _mm_sub_ps(iy0,jy0);
 828             dz00             = _mm_sub_ps(iz0,jz0);
 829             dx10             = _mm_sub_ps(ix1,jx0);
 830             dy10             = _mm_sub_ps(iy1,jy0);
 831             dz10             = _mm_sub_ps(iz1,jz0);
 832             dx20             = _mm_sub_ps(ix2,jx0);
 833             dy20             = _mm_sub_ps(iy2,jy0);
 834             dz20             = _mm_sub_ps(iz2,jz0);
 835
 836             /* Calculate squared distance and things based on it */
 837             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 838             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 839             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 840
 841             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 842             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 843             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 844
 845             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 846             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 847             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 848
 849             /* Load parameters for j particles */
 850             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 851                                                               charge+jnrC+0,charge+jnrD+0);
 852             vdwjidx0A        = 2*vdwtype[jnrA+0];
 853             vdwjidx0B        = 2*vdwtype[jnrB+0];
 854             vdwjidx0C        = 2*vdwtype[jnrC+0];
 855             vdwjidx0D        = 2*vdwtype[jnrD+0];
 856
 857             fjx0             = _mm_setzero_ps();
 858             fjy0             = _mm_setzero_ps();
 859             fjz0             = _mm_setzero_ps();
 860
 861             /**************************
 862              * CALCULATE INTERACTIONS *
 863              **************************/
 864
 865             if (gmx_mm_any_lt(rsq00,rcutoff2))
 866             {
 867
 868             r00              = _mm_mul_ps(rsq00,rinv00);
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq00             = _mm_mul_ps(iq0,jq0);
 872             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 873                                          vdwparam+vdwioffset0+vdwjidx0B,
 874                                          vdwparam+vdwioffset0+vdwjidx0C,
 875                                          vdwparam+vdwioffset0+vdwjidx0D,
 876                                          &c6_00,&c12_00);
 877
 878             /* Calculate table index by multiplying r with table scale and truncate to integer */
 879             rt               = _mm_mul_ps(r00,vftabscale);
 880             vfitab           = _mm_cvttps_epi32(rt);
 881 #ifdef __XOP__
 882             vfeps            = _mm_frcz_ps(rt);
 883 #else
 884             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 885 #endif
 886             twovfeps         = _mm_add_ps(vfeps,vfeps);
 887             vfitab           = _mm_slli_epi32(vfitab,3);
 888
 889             /* REACTION-FIELD ELECTROSTATICS */
 890             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 891
 892             /* CUBIC SPLINE TABLE DISPERSION */
 893             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 894             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 895             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 896             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 897             _MM_TRANSPOSE4_PS(Y,F,G,H);
 898             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 899             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 900             fvdw6            = _mm_mul_ps(c6_00,FF);
 901
 902             /* CUBIC SPLINE TABLE REPULSION */
 903             vfitab           = _mm_add_epi32(vfitab,ifour);
 904             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 905             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 906             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 907             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 908             _MM_TRANSPOSE4_PS(Y,F,G,H);
 909             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 910             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 911             fvdw12           = _mm_mul_ps(c12_00,FF);
 912             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 913
 914             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 915
 916             fscal            = _mm_add_ps(felec,fvdw);
 917
 918             fscal            = _mm_and_ps(fscal,cutoff_mask);
 919
 920              /* Update vectorial force */
 921             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 922             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 923             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 924
 925             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 926             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 927             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 928
 929             }
 930
 931             /**************************
 932              * CALCULATE INTERACTIONS *
 933              **************************/
 934
 935             if (gmx_mm_any_lt(rsq10,rcutoff2))
 936             {
 937
 938             /* Compute parameters for interactions between i and j atoms */
 939             qq10             = _mm_mul_ps(iq1,jq0);
 940
 941             /* REACTION-FIELD ELECTROSTATICS */
 942             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 943
 944             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 945
 946             fscal            = felec;
 947
 948             fscal            = _mm_and_ps(fscal,cutoff_mask);
 949
 950              /* Update vectorial force */
 951             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 952             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 953             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 954
 955             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 956             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 957             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 958
 959             }
 960
 961             /**************************
 962              * CALCULATE INTERACTIONS *
 963              **************************/
 964
 965             if (gmx_mm_any_lt(rsq20,rcutoff2))
 966             {
 967
 968             /* Compute parameters for interactions between i and j atoms */
 969             qq20             = _mm_mul_ps(iq2,jq0);
 970
 971             /* REACTION-FIELD ELECTROSTATICS */
 972             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 973
 974             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 975
 976             fscal            = felec;
 977
 978             fscal            = _mm_and_ps(fscal,cutoff_mask);
 979
 980              /* Update vectorial force */
 981             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 982             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 983             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 984
 985             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 986             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 987             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 988
 989             }
 990
 991             fjptrA             = f+j_coord_offsetA;
 992             fjptrB             = f+j_coord_offsetB;
 993             fjptrC             = f+j_coord_offsetC;
 994             fjptrD             = f+j_coord_offsetD;
 995
 996             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 997
 998             /* Inner loop uses 126 flops */
 999         }
1000
1001         if(jidx<j_index_end)
1002         {
1003
1004             /* Get j neighbor index, and coordinate index */
1005             jnrlistA         = jjnr[jidx];
1006             jnrlistB         = jjnr[jidx+1];
1007             jnrlistC         = jjnr[jidx+2];
1008             jnrlistD         = jjnr[jidx+3];
1009             /* Sign of each element will be negative for non-real atoms.
1010              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1011              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1012              */
1013             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1014             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1015             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1016             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1017             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1018             j_coord_offsetA  = DIM*jnrA;
1019             j_coord_offsetB  = DIM*jnrB;
1020             j_coord_offsetC  = DIM*jnrC;
1021             j_coord_offsetD  = DIM*jnrD;
1022
1023             /* load j atom coordinates */
1024             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1025                                               x+j_coord_offsetC,x+j_coord_offsetD,
1026                                               &jx0,&jy0,&jz0);
1027
1028             /* Calculate displacement vector */
1029             dx00             = _mm_sub_ps(ix0,jx0);
1030             dy00             = _mm_sub_ps(iy0,jy0);
1031             dz00             = _mm_sub_ps(iz0,jz0);
1032             dx10             = _mm_sub_ps(ix1,jx0);
1033             dy10             = _mm_sub_ps(iy1,jy0);
1034             dz10             = _mm_sub_ps(iz1,jz0);
1035             dx20             = _mm_sub_ps(ix2,jx0);
1036             dy20             = _mm_sub_ps(iy2,jy0);
1037             dz20             = _mm_sub_ps(iz2,jz0);
1038
1039             /* Calculate squared distance and things based on it */
1040             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1041             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1042             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1043
1044             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1045             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1046             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1047
1048             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1049             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1050             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1051
1052             /* Load parameters for j particles */
1053             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1054                                                               charge+jnrC+0,charge+jnrD+0);
1055             vdwjidx0A        = 2*vdwtype[jnrA+0];
1056             vdwjidx0B        = 2*vdwtype[jnrB+0];
1057             vdwjidx0C        = 2*vdwtype[jnrC+0];
1058             vdwjidx0D        = 2*vdwtype[jnrD+0];
1059
1060             fjx0             = _mm_setzero_ps();
1061             fjy0             = _mm_setzero_ps();
1062             fjz0             = _mm_setzero_ps();
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             if (gmx_mm_any_lt(rsq00,rcutoff2))
1069             {
1070
1071             r00              = _mm_mul_ps(rsq00,rinv00);
1072             r00              = _mm_andnot_ps(dummy_mask,r00);
1073
1074             /* Compute parameters for interactions between i and j atoms */
1075             qq00             = _mm_mul_ps(iq0,jq0);
1076             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1077                                          vdwparam+vdwioffset0+vdwjidx0B,
1078                                          vdwparam+vdwioffset0+vdwjidx0C,
1079                                          vdwparam+vdwioffset0+vdwjidx0D,
1080                                          &c6_00,&c12_00);
1081
1082             /* Calculate table index by multiplying r with table scale and truncate to integer */
1083             rt               = _mm_mul_ps(r00,vftabscale);
1084             vfitab           = _mm_cvttps_epi32(rt);
1085 #ifdef __XOP__
1086             vfeps            = _mm_frcz_ps(rt);
1087 #else
1088             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1089 #endif
1090             twovfeps         = _mm_add_ps(vfeps,vfeps);
1091             vfitab           = _mm_slli_epi32(vfitab,3);
1092
1093             /* REACTION-FIELD ELECTROSTATICS */
1094             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1095
1096             /* CUBIC SPLINE TABLE DISPERSION */
1097             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1098             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1099             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1100             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1101             _MM_TRANSPOSE4_PS(Y,F,G,H);
1102             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1103             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1104             fvdw6            = _mm_mul_ps(c6_00,FF);
1105
1106             /* CUBIC SPLINE TABLE REPULSION */
1107             vfitab           = _mm_add_epi32(vfitab,ifour);
1108             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1109             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1110             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1111             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1112             _MM_TRANSPOSE4_PS(Y,F,G,H);
1113             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1114             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1115             fvdw12           = _mm_mul_ps(c12_00,FF);
1116             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1117
1118             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1119
1120             fscal            = _mm_add_ps(felec,fvdw);
1121
1122             fscal            = _mm_and_ps(fscal,cutoff_mask);
1123
1124             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1125
1126              /* Update vectorial force */
1127             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1128             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1129             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1130
1131             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1132             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1133             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1134
1135             }
1136
1137             /**************************
1138              * CALCULATE INTERACTIONS *
1139              **************************/
1140
1141             if (gmx_mm_any_lt(rsq10,rcutoff2))
1142             {
1143
1144             /* Compute parameters for interactions between i and j atoms */
1145             qq10             = _mm_mul_ps(iq1,jq0);
1146
1147             /* REACTION-FIELD ELECTROSTATICS */
1148             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1149
1150             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1151
1152             fscal            = felec;
1153
1154             fscal            = _mm_and_ps(fscal,cutoff_mask);
1155
1156             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1157
1158              /* Update vectorial force */
1159             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1160             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1161             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1162
1163             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1164             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1165             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1166
1167             }
1168
1169             /**************************
1170              * CALCULATE INTERACTIONS *
1171              **************************/
1172
1173             if (gmx_mm_any_lt(rsq20,rcutoff2))
1174             {
1175
1176             /* Compute parameters for interactions between i and j atoms */
1177             qq20             = _mm_mul_ps(iq2,jq0);
1178
1179             /* REACTION-FIELD ELECTROSTATICS */
1180             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1181
1182             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1183
1184             fscal            = felec;
1185
1186             fscal            = _mm_and_ps(fscal,cutoff_mask);
1187
1188             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1189
1190              /* Update vectorial force */
1191             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1192             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1193             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1194
1195             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1196             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1197             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1198
1199             }
1200
1201             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1202             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1203             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1204             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1205
1206             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1207
1208             /* Inner loop uses 127 flops */
1209         }
1210
1211         /* End of innermost loop */
1212
1213         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1214                                               f+i_coord_offset,fshift+i_shift_offset);
1215
1216         /* Increment number of inner iterations */
1217         inneriter                  += j_index_end - j_index_start;
1218
1219         /* Outer loop uses 18 flops */
1220     }
1221
1222     /* Increment number of outer iterations */
1223     outeriter        += nri;
1224
1225     /* Update outer/inner flops */
1226
1227     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);
1228 }