src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwLJ_GeomP1P1_avx_256_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_256_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  88     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  97     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  98     __m256           dummy_mask,cutoff_mask;
  99     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 100     __m256           one     = _mm256_set1_ps(1.0);
 101     __m256           two     = _mm256_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_ps(fr->ic->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = _mm256_set1_ps(fr->ic->k_rf);
 116     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 117     crf              = _mm256_set1_ps(fr->ic->c_rf);
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128     j_coord_offsetE = 0;
 129     j_coord_offsetF = 0;
 130     j_coord_offsetG = 0;
 131     j_coord_offsetH = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 157
 158         fix0             = _mm256_setzero_ps();
 159         fiy0             = _mm256_setzero_ps();
 160         fiz0             = _mm256_setzero_ps();
 161
 162         /* Load parameters for i particles */
 163         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 164         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm256_setzero_ps();
 168         vvdwsum          = _mm256_setzero_ps();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             jnrC             = jjnr[jidx+2];
 178             jnrD             = jjnr[jidx+3];
 179             jnrE             = jjnr[jidx+4];
 180             jnrF             = jjnr[jidx+5];
 181             jnrG             = jjnr[jidx+6];
 182             jnrH             = jjnr[jidx+7];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185             j_coord_offsetC  = DIM*jnrC;
 186             j_coord_offsetD  = DIM*jnrD;
 187             j_coord_offsetE  = DIM*jnrE;
 188             j_coord_offsetF  = DIM*jnrF;
 189             j_coord_offsetG  = DIM*jnrG;
 190             j_coord_offsetH  = DIM*jnrH;
 191
 192             /* load j atom coordinates */
 193             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 195                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 196                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 197                                                  &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm256_sub_ps(ix0,jx0);
 201             dy00             = _mm256_sub_ps(iy0,jy0);
 202             dz00             = _mm256_sub_ps(iz0,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 206
 207             rinv00           = avx256_invsqrt_f(rsq00);
 208
 209             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 210
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 213                                                                  charge+jnrC+0,charge+jnrD+0,
 214                                                                  charge+jnrE+0,charge+jnrF+0,
 215                                                                  charge+jnrG+0,charge+jnrH+0);
 216             vdwjidx0A        = 2*vdwtype[jnrA+0];
 217             vdwjidx0B        = 2*vdwtype[jnrB+0];
 218             vdwjidx0C        = 2*vdwtype[jnrC+0];
 219             vdwjidx0D        = 2*vdwtype[jnrD+0];
 220             vdwjidx0E        = 2*vdwtype[jnrE+0];
 221             vdwjidx0F        = 2*vdwtype[jnrF+0];
 222             vdwjidx0G        = 2*vdwtype[jnrG+0];
 223             vdwjidx0H        = 2*vdwtype[jnrH+0];
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq00             = _mm256_mul_ps(iq0,jq0);
 231             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 232                                             vdwioffsetptr0+vdwjidx0B,
 233                                             vdwioffsetptr0+vdwjidx0C,
 234                                             vdwioffsetptr0+vdwjidx0D,
 235                                             vdwioffsetptr0+vdwjidx0E,
 236                                             vdwioffsetptr0+vdwjidx0F,
 237                                             vdwioffsetptr0+vdwjidx0G,
 238                                             vdwioffsetptr0+vdwjidx0H,
 239                                             &c6_00,&c12_00);
 240
 241             /* REACTION-FIELD ELECTROSTATICS */
 242             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 243             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 244
 245             /* LENNARD-JONES DISPERSION/REPULSION */
 246
 247             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 248             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 249             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 250             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 251             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 252
 253             /* Update potential sum for this i atom from the interaction with this j atom. */
 254             velecsum         = _mm256_add_ps(velecsum,velec);
 255             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 256
 257             fscal            = _mm256_add_ps(felec,fvdw);
 258
 259             /* Calculate temporary vectorial force */
 260             tx               = _mm256_mul_ps(fscal,dx00);
 261             ty               = _mm256_mul_ps(fscal,dy00);
 262             tz               = _mm256_mul_ps(fscal,dz00);
 263
 264             /* Update vectorial force */
 265             fix0             = _mm256_add_ps(fix0,tx);
 266             fiy0             = _mm256_add_ps(fiy0,ty);
 267             fiz0             = _mm256_add_ps(fiz0,tz);
 268
 269             fjptrA             = f+j_coord_offsetA;
 270             fjptrB             = f+j_coord_offsetB;
 271             fjptrC             = f+j_coord_offsetC;
 272             fjptrD             = f+j_coord_offsetD;
 273             fjptrE             = f+j_coord_offsetE;
 274             fjptrF             = f+j_coord_offsetF;
 275             fjptrG             = f+j_coord_offsetG;
 276             fjptrH             = f+j_coord_offsetH;
 277             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 278
 279             /* Inner loop uses 44 flops */
 280         }
 281
 282         if(jidx<j_index_end)
 283         {
 284
 285             /* Get j neighbor index, and coordinate index */
 286             jnrlistA         = jjnr[jidx];
 287             jnrlistB         = jjnr[jidx+1];
 288             jnrlistC         = jjnr[jidx+2];
 289             jnrlistD         = jjnr[jidx+3];
 290             jnrlistE         = jjnr[jidx+4];
 291             jnrlistF         = jjnr[jidx+5];
 292             jnrlistG         = jjnr[jidx+6];
 293             jnrlistH         = jjnr[jidx+7];
 294             /* Sign of each element will be negative for non-real atoms.
 295              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 296              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 297              */
 298             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 299                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 300
 301             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 302             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 303             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 304             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 305             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 306             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 307             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 308             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 309             j_coord_offsetA  = DIM*jnrA;
 310             j_coord_offsetB  = DIM*jnrB;
 311             j_coord_offsetC  = DIM*jnrC;
 312             j_coord_offsetD  = DIM*jnrD;
 313             j_coord_offsetE  = DIM*jnrE;
 314             j_coord_offsetF  = DIM*jnrF;
 315             j_coord_offsetG  = DIM*jnrG;
 316             j_coord_offsetH  = DIM*jnrH;
 317
 318             /* load j atom coordinates */
 319             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 320                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 321                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 322                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 323                                                  &jx0,&jy0,&jz0);
 324
 325             /* Calculate displacement vector */
 326             dx00             = _mm256_sub_ps(ix0,jx0);
 327             dy00             = _mm256_sub_ps(iy0,jy0);
 328             dz00             = _mm256_sub_ps(iz0,jz0);
 329
 330             /* Calculate squared distance and things based on it */
 331             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 332
 333             rinv00           = avx256_invsqrt_f(rsq00);
 334
 335             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 336
 337             /* Load parameters for j particles */
 338             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 339                                                                  charge+jnrC+0,charge+jnrD+0,
 340                                                                  charge+jnrE+0,charge+jnrF+0,
 341                                                                  charge+jnrG+0,charge+jnrH+0);
 342             vdwjidx0A        = 2*vdwtype[jnrA+0];
 343             vdwjidx0B        = 2*vdwtype[jnrB+0];
 344             vdwjidx0C        = 2*vdwtype[jnrC+0];
 345             vdwjidx0D        = 2*vdwtype[jnrD+0];
 346             vdwjidx0E        = 2*vdwtype[jnrE+0];
 347             vdwjidx0F        = 2*vdwtype[jnrF+0];
 348             vdwjidx0G        = 2*vdwtype[jnrG+0];
 349             vdwjidx0H        = 2*vdwtype[jnrH+0];
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             /* Compute parameters for interactions between i and j atoms */
 356             qq00             = _mm256_mul_ps(iq0,jq0);
 357             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 358                                             vdwioffsetptr0+vdwjidx0B,
 359                                             vdwioffsetptr0+vdwjidx0C,
 360                                             vdwioffsetptr0+vdwjidx0D,
 361                                             vdwioffsetptr0+vdwjidx0E,
 362                                             vdwioffsetptr0+vdwjidx0F,
 363                                             vdwioffsetptr0+vdwjidx0G,
 364                                             vdwioffsetptr0+vdwjidx0H,
 365                                             &c6_00,&c12_00);
 366
 367             /* REACTION-FIELD ELECTROSTATICS */
 368             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 369             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 370
 371             /* LENNARD-JONES DISPERSION/REPULSION */
 372
 373             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 374             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 375             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 376             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 377             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velec            = _mm256_andnot_ps(dummy_mask,velec);
 381             velecsum         = _mm256_add_ps(velecsum,velec);
 382             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 383             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 384
 385             fscal            = _mm256_add_ps(felec,fvdw);
 386
 387             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 388
 389             /* Calculate temporary vectorial force */
 390             tx               = _mm256_mul_ps(fscal,dx00);
 391             ty               = _mm256_mul_ps(fscal,dy00);
 392             tz               = _mm256_mul_ps(fscal,dz00);
 393
 394             /* Update vectorial force */
 395             fix0             = _mm256_add_ps(fix0,tx);
 396             fiy0             = _mm256_add_ps(fiy0,ty);
 397             fiz0             = _mm256_add_ps(fiz0,tz);
 398
 399             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 400             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 401             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 402             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 403             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 404             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 405             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 406             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 407             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 408
 409             /* Inner loop uses 44 flops */
 410         }
 411
 412         /* End of innermost loop */
 413
 414         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 415                                                  f+i_coord_offset,fshift+i_shift_offset);
 416
 417         ggid                        = gid[iidx];
 418         /* Update potential energies */
 419         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 420         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 421
 422         /* Increment number of inner iterations */
 423         inneriter                  += j_index_end - j_index_start;
 424
 425         /* Outer loop uses 9 flops */
 426     }
 427
 428     /* Increment number of outer iterations */
 429     outeriter        += nri;
 430
 431     /* Update outer/inner flops */
 432
 433     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
 434 }
 435 /*
 436  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_256_single
 437  * Electrostatics interaction: ReactionField
 438  * VdW interaction:            LennardJones
 439  * Geometry:                   Particle-Particle
 440  * Calculate force/pot:        Force
 441  */
 442 void
 443 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_256_single
 444                     (t_nblist                    * gmx_restrict       nlist,
 445                      rvec                        * gmx_restrict          xx,
 446                      rvec                        * gmx_restrict          ff,
 447                      struct t_forcerec           * gmx_restrict          fr,
 448                      t_mdatoms                   * gmx_restrict     mdatoms,
 449                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 450                      t_nrnb                      * gmx_restrict        nrnb)
 451 {
 452     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 453      * just 0 for non-waters.
 454      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 455      * jnr indices corresponding to data put in the four positions in the SIMD register.
 456      */
 457     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 458     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 459     int              jnrA,jnrB,jnrC,jnrD;
 460     int              jnrE,jnrF,jnrG,jnrH;
 461     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 462     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 463     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 464     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 465     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 466     real             rcutoff_scalar;
 467     real             *shiftvec,*fshift,*x,*f;
 468     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 469     real             scratch[4*DIM];
 470     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 471     real *           vdwioffsetptr0;
 472     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 473     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 474     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 475     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 476     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 477     real             *charge;
 478     int              nvdwtype;
 479     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 480     int              *vdwtype;
 481     real             *vdwparam;
 482     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 483     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 484     __m256           dummy_mask,cutoff_mask;
 485     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 486     __m256           one     = _mm256_set1_ps(1.0);
 487     __m256           two     = _mm256_set1_ps(2.0);
 488     x                = xx[0];
 489     f                = ff[0];
 490
 491     nri              = nlist->nri;
 492     iinr             = nlist->iinr;
 493     jindex           = nlist->jindex;
 494     jjnr             = nlist->jjnr;
 495     shiftidx         = nlist->shift;
 496     gid              = nlist->gid;
 497     shiftvec         = fr->shift_vec[0];
 498     fshift           = fr->fshift[0];
 499     facel            = _mm256_set1_ps(fr->ic->epsfac);
 500     charge           = mdatoms->chargeA;
 501     krf              = _mm256_set1_ps(fr->ic->k_rf);
 502     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 503     crf              = _mm256_set1_ps(fr->ic->c_rf);
 504     nvdwtype         = fr->ntype;
 505     vdwparam         = fr->nbfp;
 506     vdwtype          = mdatoms->typeA;
 507
 508     /* Avoid stupid compiler warnings */
 509     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 510     j_coord_offsetA = 0;
 511     j_coord_offsetB = 0;
 512     j_coord_offsetC = 0;
 513     j_coord_offsetD = 0;
 514     j_coord_offsetE = 0;
 515     j_coord_offsetF = 0;
 516     j_coord_offsetG = 0;
 517     j_coord_offsetH = 0;
 518
 519     outeriter        = 0;
 520     inneriter        = 0;
 521
 522     for(iidx=0;iidx<4*DIM;iidx++)
 523     {
 524         scratch[iidx] = 0.0;
 525     }
 526
 527     /* Start outer loop over neighborlists */
 528     for(iidx=0; iidx<nri; iidx++)
 529     {
 530         /* Load shift vector for this list */
 531         i_shift_offset   = DIM*shiftidx[iidx];
 532
 533         /* Load limits for loop over neighbors */
 534         j_index_start    = jindex[iidx];
 535         j_index_end      = jindex[iidx+1];
 536
 537         /* Get outer coordinate index */
 538         inr              = iinr[iidx];
 539         i_coord_offset   = DIM*inr;
 540
 541         /* Load i particle coords and add shift vector */
 542         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 543
 544         fix0             = _mm256_setzero_ps();
 545         fiy0             = _mm256_setzero_ps();
 546         fiz0             = _mm256_setzero_ps();
 547
 548         /* Load parameters for i particles */
 549         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 550         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 551
 552         /* Start inner kernel loop */
 553         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 554         {
 555
 556             /* Get j neighbor index, and coordinate index */
 557             jnrA             = jjnr[jidx];
 558             jnrB             = jjnr[jidx+1];
 559             jnrC             = jjnr[jidx+2];
 560             jnrD             = jjnr[jidx+3];
 561             jnrE             = jjnr[jidx+4];
 562             jnrF             = jjnr[jidx+5];
 563             jnrG             = jjnr[jidx+6];
 564             jnrH             = jjnr[jidx+7];
 565             j_coord_offsetA  = DIM*jnrA;
 566             j_coord_offsetB  = DIM*jnrB;
 567             j_coord_offsetC  = DIM*jnrC;
 568             j_coord_offsetD  = DIM*jnrD;
 569             j_coord_offsetE  = DIM*jnrE;
 570             j_coord_offsetF  = DIM*jnrF;
 571             j_coord_offsetG  = DIM*jnrG;
 572             j_coord_offsetH  = DIM*jnrH;
 573
 574             /* load j atom coordinates */
 575             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 576                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 577                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 578                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 579                                                  &jx0,&jy0,&jz0);
 580
 581             /* Calculate displacement vector */
 582             dx00             = _mm256_sub_ps(ix0,jx0);
 583             dy00             = _mm256_sub_ps(iy0,jy0);
 584             dz00             = _mm256_sub_ps(iz0,jz0);
 585
 586             /* Calculate squared distance and things based on it */
 587             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 588
 589             rinv00           = avx256_invsqrt_f(rsq00);
 590
 591             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 592
 593             /* Load parameters for j particles */
 594             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 595                                                                  charge+jnrC+0,charge+jnrD+0,
 596                                                                  charge+jnrE+0,charge+jnrF+0,
 597                                                                  charge+jnrG+0,charge+jnrH+0);
 598             vdwjidx0A        = 2*vdwtype[jnrA+0];
 599             vdwjidx0B        = 2*vdwtype[jnrB+0];
 600             vdwjidx0C        = 2*vdwtype[jnrC+0];
 601             vdwjidx0D        = 2*vdwtype[jnrD+0];
 602             vdwjidx0E        = 2*vdwtype[jnrE+0];
 603             vdwjidx0F        = 2*vdwtype[jnrF+0];
 604             vdwjidx0G        = 2*vdwtype[jnrG+0];
 605             vdwjidx0H        = 2*vdwtype[jnrH+0];
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq00             = _mm256_mul_ps(iq0,jq0);
 613             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 614                                             vdwioffsetptr0+vdwjidx0B,
 615                                             vdwioffsetptr0+vdwjidx0C,
 616                                             vdwioffsetptr0+vdwjidx0D,
 617                                             vdwioffsetptr0+vdwjidx0E,
 618                                             vdwioffsetptr0+vdwjidx0F,
 619                                             vdwioffsetptr0+vdwjidx0G,
 620                                             vdwioffsetptr0+vdwjidx0H,
 621                                             &c6_00,&c12_00);
 622
 623             /* REACTION-FIELD ELECTROSTATICS */
 624             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 625
 626             /* LENNARD-JONES DISPERSION/REPULSION */
 627
 628             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 629             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 630
 631             fscal            = _mm256_add_ps(felec,fvdw);
 632
 633             /* Calculate temporary vectorial force */
 634             tx               = _mm256_mul_ps(fscal,dx00);
 635             ty               = _mm256_mul_ps(fscal,dy00);
 636             tz               = _mm256_mul_ps(fscal,dz00);
 637
 638             /* Update vectorial force */
 639             fix0             = _mm256_add_ps(fix0,tx);
 640             fiy0             = _mm256_add_ps(fiy0,ty);
 641             fiz0             = _mm256_add_ps(fiz0,tz);
 642
 643             fjptrA             = f+j_coord_offsetA;
 644             fjptrB             = f+j_coord_offsetB;
 645             fjptrC             = f+j_coord_offsetC;
 646             fjptrD             = f+j_coord_offsetD;
 647             fjptrE             = f+j_coord_offsetE;
 648             fjptrF             = f+j_coord_offsetF;
 649             fjptrG             = f+j_coord_offsetG;
 650             fjptrH             = f+j_coord_offsetH;
 651             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 652
 653             /* Inner loop uses 34 flops */
 654         }
 655
 656         if(jidx<j_index_end)
 657         {
 658
 659             /* Get j neighbor index, and coordinate index */
 660             jnrlistA         = jjnr[jidx];
 661             jnrlistB         = jjnr[jidx+1];
 662             jnrlistC         = jjnr[jidx+2];
 663             jnrlistD         = jjnr[jidx+3];
 664             jnrlistE         = jjnr[jidx+4];
 665             jnrlistF         = jjnr[jidx+5];
 666             jnrlistG         = jjnr[jidx+6];
 667             jnrlistH         = jjnr[jidx+7];
 668             /* Sign of each element will be negative for non-real atoms.
 669              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 670              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 671              */
 672             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 673                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 674
 675             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 676             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 677             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 678             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 679             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 680             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 681             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 682             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 683             j_coord_offsetA  = DIM*jnrA;
 684             j_coord_offsetB  = DIM*jnrB;
 685             j_coord_offsetC  = DIM*jnrC;
 686             j_coord_offsetD  = DIM*jnrD;
 687             j_coord_offsetE  = DIM*jnrE;
 688             j_coord_offsetF  = DIM*jnrF;
 689             j_coord_offsetG  = DIM*jnrG;
 690             j_coord_offsetH  = DIM*jnrH;
 691
 692             /* load j atom coordinates */
 693             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 694                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 695                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 696                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 697                                                  &jx0,&jy0,&jz0);
 698
 699             /* Calculate displacement vector */
 700             dx00             = _mm256_sub_ps(ix0,jx0);
 701             dy00             = _mm256_sub_ps(iy0,jy0);
 702             dz00             = _mm256_sub_ps(iz0,jz0);
 703
 704             /* Calculate squared distance and things based on it */
 705             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 706
 707             rinv00           = avx256_invsqrt_f(rsq00);
 708
 709             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 710
 711             /* Load parameters for j particles */
 712             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 713                                                                  charge+jnrC+0,charge+jnrD+0,
 714                                                                  charge+jnrE+0,charge+jnrF+0,
 715                                                                  charge+jnrG+0,charge+jnrH+0);
 716             vdwjidx0A        = 2*vdwtype[jnrA+0];
 717             vdwjidx0B        = 2*vdwtype[jnrB+0];
 718             vdwjidx0C        = 2*vdwtype[jnrC+0];
 719             vdwjidx0D        = 2*vdwtype[jnrD+0];
 720             vdwjidx0E        = 2*vdwtype[jnrE+0];
 721             vdwjidx0F        = 2*vdwtype[jnrF+0];
 722             vdwjidx0G        = 2*vdwtype[jnrG+0];
 723             vdwjidx0H        = 2*vdwtype[jnrH+0];
 724
 725             /**************************
 726              * CALCULATE INTERACTIONS *
 727              **************************/
 728
 729             /* Compute parameters for interactions between i and j atoms */
 730             qq00             = _mm256_mul_ps(iq0,jq0);
 731             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 732                                             vdwioffsetptr0+vdwjidx0B,
 733                                             vdwioffsetptr0+vdwjidx0C,
 734                                             vdwioffsetptr0+vdwjidx0D,
 735                                             vdwioffsetptr0+vdwjidx0E,
 736                                             vdwioffsetptr0+vdwjidx0F,
 737                                             vdwioffsetptr0+vdwjidx0G,
 738                                             vdwioffsetptr0+vdwjidx0H,
 739                                             &c6_00,&c12_00);
 740
 741             /* REACTION-FIELD ELECTROSTATICS */
 742             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 743
 744             /* LENNARD-JONES DISPERSION/REPULSION */
 745
 746             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 747             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 748
 749             fscal            = _mm256_add_ps(felec,fvdw);
 750
 751             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 752
 753             /* Calculate temporary vectorial force */
 754             tx               = _mm256_mul_ps(fscal,dx00);
 755             ty               = _mm256_mul_ps(fscal,dy00);
 756             tz               = _mm256_mul_ps(fscal,dz00);
 757
 758             /* Update vectorial force */
 759             fix0             = _mm256_add_ps(fix0,tx);
 760             fiy0             = _mm256_add_ps(fiy0,ty);
 761             fiz0             = _mm256_add_ps(fiz0,tz);
 762
 763             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 764             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 765             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 766             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 767             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 768             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 769             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 770             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 771             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 772
 773             /* Inner loop uses 34 flops */
 774         }
 775
 776         /* End of innermost loop */
 777
 778         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 779                                                  f+i_coord_offset,fshift+i_shift_offset);
 780
 781         /* Increment number of inner iterations */
 782         inneriter                  += j_index_end - j_index_start;
 783
 784         /* Outer loop uses 7 flops */
 785     }
 786
 787     /* Increment number of outer iterations */
 788     outeriter        += nri;
 789
 790     /* Update outer/inner flops */
 791
 792     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 793 }