src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRF_VdwLJ_GeomP1P1_avx_256_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_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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  91     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 100     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 101     __m256           dummy_mask,cutoff_mask;
 102     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 103     __m256           one     = _mm256_set1_ps(1.0);
 104     __m256           two     = _mm256_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm256_set1_ps(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     krf              = _mm256_set1_ps(fr->ic->k_rf);
 119     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 120     crf              = _mm256_set1_ps(fr->ic->c_rf);
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129     j_coord_offsetC = 0;
 130     j_coord_offsetD = 0;
 131     j_coord_offsetE = 0;
 132     j_coord_offsetF = 0;
 133     j_coord_offsetG = 0;
 134     j_coord_offsetH = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 160
 161         fix0             = _mm256_setzero_ps();
 162         fiy0             = _mm256_setzero_ps();
 163         fiz0             = _mm256_setzero_ps();
 164
 165         /* Load parameters for i particles */
 166         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 167         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm256_setzero_ps();
 171         vvdwsum          = _mm256_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182             jnrE             = jjnr[jidx+4];
 183             jnrF             = jjnr[jidx+5];
 184             jnrG             = jjnr[jidx+6];
 185             jnrH             = jjnr[jidx+7];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190             j_coord_offsetE  = DIM*jnrE;
 191             j_coord_offsetF  = DIM*jnrF;
 192             j_coord_offsetG  = DIM*jnrG;
 193             j_coord_offsetH  = DIM*jnrH;
 194
 195             /* load j atom coordinates */
 196             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 198                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 199                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 200                                                  &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm256_sub_ps(ix0,jx0);
 204             dy00             = _mm256_sub_ps(iy0,jy0);
 205             dz00             = _mm256_sub_ps(iz0,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 209
 210             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 211
 212             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 216                                                                  charge+jnrC+0,charge+jnrD+0,
 217                                                                  charge+jnrE+0,charge+jnrF+0,
 218                                                                  charge+jnrG+0,charge+jnrH+0);
 219             vdwjidx0A        = 2*vdwtype[jnrA+0];
 220             vdwjidx0B        = 2*vdwtype[jnrB+0];
 221             vdwjidx0C        = 2*vdwtype[jnrC+0];
 222             vdwjidx0D        = 2*vdwtype[jnrD+0];
 223             vdwjidx0E        = 2*vdwtype[jnrE+0];
 224             vdwjidx0F        = 2*vdwtype[jnrF+0];
 225             vdwjidx0G        = 2*vdwtype[jnrG+0];
 226             vdwjidx0H        = 2*vdwtype[jnrH+0];
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             qq00             = _mm256_mul_ps(iq0,jq0);
 234             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 235                                             vdwioffsetptr0+vdwjidx0B,
 236                                             vdwioffsetptr0+vdwjidx0C,
 237                                             vdwioffsetptr0+vdwjidx0D,
 238                                             vdwioffsetptr0+vdwjidx0E,
 239                                             vdwioffsetptr0+vdwjidx0F,
 240                                             vdwioffsetptr0+vdwjidx0G,
 241                                             vdwioffsetptr0+vdwjidx0H,
 242                                             &c6_00,&c12_00);
 243
 244             /* REACTION-FIELD ELECTROSTATICS */
 245             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 246             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 247
 248             /* LENNARD-JONES DISPERSION/REPULSION */
 249
 250             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 251             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 252             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 253             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 254             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 255
 256             /* Update potential sum for this i atom from the interaction with this j atom. */
 257             velecsum         = _mm256_add_ps(velecsum,velec);
 258             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 259
 260             fscal            = _mm256_add_ps(felec,fvdw);
 261
 262             /* Calculate temporary vectorial force */
 263             tx               = _mm256_mul_ps(fscal,dx00);
 264             ty               = _mm256_mul_ps(fscal,dy00);
 265             tz               = _mm256_mul_ps(fscal,dz00);
 266
 267             /* Update vectorial force */
 268             fix0             = _mm256_add_ps(fix0,tx);
 269             fiy0             = _mm256_add_ps(fiy0,ty);
 270             fiz0             = _mm256_add_ps(fiz0,tz);
 271
 272             fjptrA             = f+j_coord_offsetA;
 273             fjptrB             = f+j_coord_offsetB;
 274             fjptrC             = f+j_coord_offsetC;
 275             fjptrD             = f+j_coord_offsetD;
 276             fjptrE             = f+j_coord_offsetE;
 277             fjptrF             = f+j_coord_offsetF;
 278             fjptrG             = f+j_coord_offsetG;
 279             fjptrH             = f+j_coord_offsetH;
 280             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 281
 282             /* Inner loop uses 44 flops */
 283         }
 284
 285         if(jidx<j_index_end)
 286         {
 287
 288             /* Get j neighbor index, and coordinate index */
 289             jnrlistA         = jjnr[jidx];
 290             jnrlistB         = jjnr[jidx+1];
 291             jnrlistC         = jjnr[jidx+2];
 292             jnrlistD         = jjnr[jidx+3];
 293             jnrlistE         = jjnr[jidx+4];
 294             jnrlistF         = jjnr[jidx+5];
 295             jnrlistG         = jjnr[jidx+6];
 296             jnrlistH         = jjnr[jidx+7];
 297             /* Sign of each element will be negative for non-real atoms.
 298              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 299              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 300              */
 301             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 302                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 303
 304             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 305             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 306             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 307             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 308             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 309             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 310             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 311             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 312             j_coord_offsetA  = DIM*jnrA;
 313             j_coord_offsetB  = DIM*jnrB;
 314             j_coord_offsetC  = DIM*jnrC;
 315             j_coord_offsetD  = DIM*jnrD;
 316             j_coord_offsetE  = DIM*jnrE;
 317             j_coord_offsetF  = DIM*jnrF;
 318             j_coord_offsetG  = DIM*jnrG;
 319             j_coord_offsetH  = DIM*jnrH;
 320
 321             /* load j atom coordinates */
 322             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 323                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 324                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 325                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 326                                                  &jx0,&jy0,&jz0);
 327
 328             /* Calculate displacement vector */
 329             dx00             = _mm256_sub_ps(ix0,jx0);
 330             dy00             = _mm256_sub_ps(iy0,jy0);
 331             dz00             = _mm256_sub_ps(iz0,jz0);
 332
 333             /* Calculate squared distance and things based on it */
 334             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 335
 336             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 337
 338             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 339
 340             /* Load parameters for j particles */
 341             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 342                                                                  charge+jnrC+0,charge+jnrD+0,
 343                                                                  charge+jnrE+0,charge+jnrF+0,
 344                                                                  charge+jnrG+0,charge+jnrH+0);
 345             vdwjidx0A        = 2*vdwtype[jnrA+0];
 346             vdwjidx0B        = 2*vdwtype[jnrB+0];
 347             vdwjidx0C        = 2*vdwtype[jnrC+0];
 348             vdwjidx0D        = 2*vdwtype[jnrD+0];
 349             vdwjidx0E        = 2*vdwtype[jnrE+0];
 350             vdwjidx0F        = 2*vdwtype[jnrF+0];
 351             vdwjidx0G        = 2*vdwtype[jnrG+0];
 352             vdwjidx0H        = 2*vdwtype[jnrH+0];
 353
 354             /**************************
 355              * CALCULATE INTERACTIONS *
 356              **************************/
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq00             = _mm256_mul_ps(iq0,jq0);
 360             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 361                                             vdwioffsetptr0+vdwjidx0B,
 362                                             vdwioffsetptr0+vdwjidx0C,
 363                                             vdwioffsetptr0+vdwjidx0D,
 364                                             vdwioffsetptr0+vdwjidx0E,
 365                                             vdwioffsetptr0+vdwjidx0F,
 366                                             vdwioffsetptr0+vdwjidx0G,
 367                                             vdwioffsetptr0+vdwjidx0H,
 368                                             &c6_00,&c12_00);
 369
 370             /* REACTION-FIELD ELECTROSTATICS */
 371             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 372             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 373
 374             /* LENNARD-JONES DISPERSION/REPULSION */
 375
 376             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 377             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 378             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 379             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 380             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velec            = _mm256_andnot_ps(dummy_mask,velec);
 384             velecsum         = _mm256_add_ps(velecsum,velec);
 385             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 386             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 387
 388             fscal            = _mm256_add_ps(felec,fvdw);
 389
 390             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 391
 392             /* Calculate temporary vectorial force */
 393             tx               = _mm256_mul_ps(fscal,dx00);
 394             ty               = _mm256_mul_ps(fscal,dy00);
 395             tz               = _mm256_mul_ps(fscal,dz00);
 396
 397             /* Update vectorial force */
 398             fix0             = _mm256_add_ps(fix0,tx);
 399             fiy0             = _mm256_add_ps(fiy0,ty);
 400             fiz0             = _mm256_add_ps(fiz0,tz);
 401
 402             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 403             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 404             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 405             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 406             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 407             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 408             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 409             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 410             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 411
 412             /* Inner loop uses 44 flops */
 413         }
 414
 415         /* End of innermost loop */
 416
 417         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 418                                                  f+i_coord_offset,fshift+i_shift_offset);
 419
 420         ggid                        = gid[iidx];
 421         /* Update potential energies */
 422         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 423         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 424
 425         /* Increment number of inner iterations */
 426         inneriter                  += j_index_end - j_index_start;
 427
 428         /* Outer loop uses 9 flops */
 429     }
 430
 431     /* Increment number of outer iterations */
 432     outeriter        += nri;
 433
 434     /* Update outer/inner flops */
 435
 436     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
 437 }
 438 /*
 439  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_256_single
 440  * Electrostatics interaction: ReactionField
 441  * VdW interaction:            LennardJones
 442  * Geometry:                   Particle-Particle
 443  * Calculate force/pot:        Force
 444  */
 445 void
 446 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_256_single
 447                     (t_nblist                    * gmx_restrict       nlist,
 448                      rvec                        * gmx_restrict          xx,
 449                      rvec                        * gmx_restrict          ff,
 450                      t_forcerec                  * gmx_restrict          fr,
 451                      t_mdatoms                   * gmx_restrict     mdatoms,
 452                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 453                      t_nrnb                      * gmx_restrict        nrnb)
 454 {
 455     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 456      * just 0 for non-waters.
 457      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 458      * jnr indices corresponding to data put in the four positions in the SIMD register.
 459      */
 460     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 461     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 462     int              jnrA,jnrB,jnrC,jnrD;
 463     int              jnrE,jnrF,jnrG,jnrH;
 464     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 465     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 466     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 467     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 468     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 469     real             rcutoff_scalar;
 470     real             *shiftvec,*fshift,*x,*f;
 471     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 472     real             scratch[4*DIM];
 473     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 474     real *           vdwioffsetptr0;
 475     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 476     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 477     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 478     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 479     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 480     real             *charge;
 481     int              nvdwtype;
 482     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 483     int              *vdwtype;
 484     real             *vdwparam;
 485     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 486     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 487     __m256           dummy_mask,cutoff_mask;
 488     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 489     __m256           one     = _mm256_set1_ps(1.0);
 490     __m256           two     = _mm256_set1_ps(2.0);
 491     x                = xx[0];
 492     f                = ff[0];
 493
 494     nri              = nlist->nri;
 495     iinr             = nlist->iinr;
 496     jindex           = nlist->jindex;
 497     jjnr             = nlist->jjnr;
 498     shiftidx         = nlist->shift;
 499     gid              = nlist->gid;
 500     shiftvec         = fr->shift_vec[0];
 501     fshift           = fr->fshift[0];
 502     facel            = _mm256_set1_ps(fr->epsfac);
 503     charge           = mdatoms->chargeA;
 504     krf              = _mm256_set1_ps(fr->ic->k_rf);
 505     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 506     crf              = _mm256_set1_ps(fr->ic->c_rf);
 507     nvdwtype         = fr->ntype;
 508     vdwparam         = fr->nbfp;
 509     vdwtype          = mdatoms->typeA;
 510
 511     /* Avoid stupid compiler warnings */
 512     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 513     j_coord_offsetA = 0;
 514     j_coord_offsetB = 0;
 515     j_coord_offsetC = 0;
 516     j_coord_offsetD = 0;
 517     j_coord_offsetE = 0;
 518     j_coord_offsetF = 0;
 519     j_coord_offsetG = 0;
 520     j_coord_offsetH = 0;
 521
 522     outeriter        = 0;
 523     inneriter        = 0;
 524
 525     for(iidx=0;iidx<4*DIM;iidx++)
 526     {
 527         scratch[iidx] = 0.0;
 528     }
 529
 530     /* Start outer loop over neighborlists */
 531     for(iidx=0; iidx<nri; iidx++)
 532     {
 533         /* Load shift vector for this list */
 534         i_shift_offset   = DIM*shiftidx[iidx];
 535
 536         /* Load limits for loop over neighbors */
 537         j_index_start    = jindex[iidx];
 538         j_index_end      = jindex[iidx+1];
 539
 540         /* Get outer coordinate index */
 541         inr              = iinr[iidx];
 542         i_coord_offset   = DIM*inr;
 543
 544         /* Load i particle coords and add shift vector */
 545         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 546
 547         fix0             = _mm256_setzero_ps();
 548         fiy0             = _mm256_setzero_ps();
 549         fiz0             = _mm256_setzero_ps();
 550
 551         /* Load parameters for i particles */
 552         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 553         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 554
 555         /* Start inner kernel loop */
 556         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 557         {
 558
 559             /* Get j neighbor index, and coordinate index */
 560             jnrA             = jjnr[jidx];
 561             jnrB             = jjnr[jidx+1];
 562             jnrC             = jjnr[jidx+2];
 563             jnrD             = jjnr[jidx+3];
 564             jnrE             = jjnr[jidx+4];
 565             jnrF             = jjnr[jidx+5];
 566             jnrG             = jjnr[jidx+6];
 567             jnrH             = jjnr[jidx+7];
 568             j_coord_offsetA  = DIM*jnrA;
 569             j_coord_offsetB  = DIM*jnrB;
 570             j_coord_offsetC  = DIM*jnrC;
 571             j_coord_offsetD  = DIM*jnrD;
 572             j_coord_offsetE  = DIM*jnrE;
 573             j_coord_offsetF  = DIM*jnrF;
 574             j_coord_offsetG  = DIM*jnrG;
 575             j_coord_offsetH  = DIM*jnrH;
 576
 577             /* load j atom coordinates */
 578             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 579                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 580                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 581                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 582                                                  &jx0,&jy0,&jz0);
 583
 584             /* Calculate displacement vector */
 585             dx00             = _mm256_sub_ps(ix0,jx0);
 586             dy00             = _mm256_sub_ps(iy0,jy0);
 587             dz00             = _mm256_sub_ps(iz0,jz0);
 588
 589             /* Calculate squared distance and things based on it */
 590             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 591
 592             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 593
 594             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 595
 596             /* Load parameters for j particles */
 597             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 598                                                                  charge+jnrC+0,charge+jnrD+0,
 599                                                                  charge+jnrE+0,charge+jnrF+0,
 600                                                                  charge+jnrG+0,charge+jnrH+0);
 601             vdwjidx0A        = 2*vdwtype[jnrA+0];
 602             vdwjidx0B        = 2*vdwtype[jnrB+0];
 603             vdwjidx0C        = 2*vdwtype[jnrC+0];
 604             vdwjidx0D        = 2*vdwtype[jnrD+0];
 605             vdwjidx0E        = 2*vdwtype[jnrE+0];
 606             vdwjidx0F        = 2*vdwtype[jnrF+0];
 607             vdwjidx0G        = 2*vdwtype[jnrG+0];
 608             vdwjidx0H        = 2*vdwtype[jnrH+0];
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             /* Compute parameters for interactions between i and j atoms */
 615             qq00             = _mm256_mul_ps(iq0,jq0);
 616             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 617                                             vdwioffsetptr0+vdwjidx0B,
 618                                             vdwioffsetptr0+vdwjidx0C,
 619                                             vdwioffsetptr0+vdwjidx0D,
 620                                             vdwioffsetptr0+vdwjidx0E,
 621                                             vdwioffsetptr0+vdwjidx0F,
 622                                             vdwioffsetptr0+vdwjidx0G,
 623                                             vdwioffsetptr0+vdwjidx0H,
 624                                             &c6_00,&c12_00);
 625
 626             /* REACTION-FIELD ELECTROSTATICS */
 627             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 628
 629             /* LENNARD-JONES DISPERSION/REPULSION */
 630
 631             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 632             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 633
 634             fscal            = _mm256_add_ps(felec,fvdw);
 635
 636             /* Calculate temporary vectorial force */
 637             tx               = _mm256_mul_ps(fscal,dx00);
 638             ty               = _mm256_mul_ps(fscal,dy00);
 639             tz               = _mm256_mul_ps(fscal,dz00);
 640
 641             /* Update vectorial force */
 642             fix0             = _mm256_add_ps(fix0,tx);
 643             fiy0             = _mm256_add_ps(fiy0,ty);
 644             fiz0             = _mm256_add_ps(fiz0,tz);
 645
 646             fjptrA             = f+j_coord_offsetA;
 647             fjptrB             = f+j_coord_offsetB;
 648             fjptrC             = f+j_coord_offsetC;
 649             fjptrD             = f+j_coord_offsetD;
 650             fjptrE             = f+j_coord_offsetE;
 651             fjptrF             = f+j_coord_offsetF;
 652             fjptrG             = f+j_coord_offsetG;
 653             fjptrH             = f+j_coord_offsetH;
 654             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 655
 656             /* Inner loop uses 34 flops */
 657         }
 658
 659         if(jidx<j_index_end)
 660         {
 661
 662             /* Get j neighbor index, and coordinate index */
 663             jnrlistA         = jjnr[jidx];
 664             jnrlistB         = jjnr[jidx+1];
 665             jnrlistC         = jjnr[jidx+2];
 666             jnrlistD         = jjnr[jidx+3];
 667             jnrlistE         = jjnr[jidx+4];
 668             jnrlistF         = jjnr[jidx+5];
 669             jnrlistG         = jjnr[jidx+6];
 670             jnrlistH         = jjnr[jidx+7];
 671             /* Sign of each element will be negative for non-real atoms.
 672              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 673              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 674              */
 675             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 676                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 677
 678             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 679             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 680             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 681             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 682             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 683             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 684             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 685             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 686             j_coord_offsetA  = DIM*jnrA;
 687             j_coord_offsetB  = DIM*jnrB;
 688             j_coord_offsetC  = DIM*jnrC;
 689             j_coord_offsetD  = DIM*jnrD;
 690             j_coord_offsetE  = DIM*jnrE;
 691             j_coord_offsetF  = DIM*jnrF;
 692             j_coord_offsetG  = DIM*jnrG;
 693             j_coord_offsetH  = DIM*jnrH;
 694
 695             /* load j atom coordinates */
 696             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 697                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 698                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 699                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 700                                                  &jx0,&jy0,&jz0);
 701
 702             /* Calculate displacement vector */
 703             dx00             = _mm256_sub_ps(ix0,jx0);
 704             dy00             = _mm256_sub_ps(iy0,jy0);
 705             dz00             = _mm256_sub_ps(iz0,jz0);
 706
 707             /* Calculate squared distance and things based on it */
 708             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 709
 710             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 711
 712             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 713
 714             /* Load parameters for j particles */
 715             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 716                                                                  charge+jnrC+0,charge+jnrD+0,
 717                                                                  charge+jnrE+0,charge+jnrF+0,
 718                                                                  charge+jnrG+0,charge+jnrH+0);
 719             vdwjidx0A        = 2*vdwtype[jnrA+0];
 720             vdwjidx0B        = 2*vdwtype[jnrB+0];
 721             vdwjidx0C        = 2*vdwtype[jnrC+0];
 722             vdwjidx0D        = 2*vdwtype[jnrD+0];
 723             vdwjidx0E        = 2*vdwtype[jnrE+0];
 724             vdwjidx0F        = 2*vdwtype[jnrF+0];
 725             vdwjidx0G        = 2*vdwtype[jnrG+0];
 726             vdwjidx0H        = 2*vdwtype[jnrH+0];
 727
 728             /**************************
 729              * CALCULATE INTERACTIONS *
 730              **************************/
 731
 732             /* Compute parameters for interactions between i and j atoms */
 733             qq00             = _mm256_mul_ps(iq0,jq0);
 734             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 735                                             vdwioffsetptr0+vdwjidx0B,
 736                                             vdwioffsetptr0+vdwjidx0C,
 737                                             vdwioffsetptr0+vdwjidx0D,
 738                                             vdwioffsetptr0+vdwjidx0E,
 739                                             vdwioffsetptr0+vdwjidx0F,
 740                                             vdwioffsetptr0+vdwjidx0G,
 741                                             vdwioffsetptr0+vdwjidx0H,
 742                                             &c6_00,&c12_00);
 743
 744             /* REACTION-FIELD ELECTROSTATICS */
 745             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 746
 747             /* LENNARD-JONES DISPERSION/REPULSION */
 748
 749             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 750             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 751
 752             fscal            = _mm256_add_ps(felec,fvdw);
 753
 754             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 755
 756             /* Calculate temporary vectorial force */
 757             tx               = _mm256_mul_ps(fscal,dx00);
 758             ty               = _mm256_mul_ps(fscal,dy00);
 759             tz               = _mm256_mul_ps(fscal,dz00);
 760
 761             /* Update vectorial force */
 762             fix0             = _mm256_add_ps(fix0,tx);
 763             fiy0             = _mm256_add_ps(fiy0,ty);
 764             fiz0             = _mm256_add_ps(fiz0,tz);
 765
 766             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 767             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 768             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 769             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 770             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 771             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 772             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 773             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 774             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 775
 776             /* Inner loop uses 34 flops */
 777         }
 778
 779         /* End of innermost loop */
 780
 781         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 782                                                  f+i_coord_offset,fshift+i_shift_offset);
 783
 784         /* Increment number of inner iterations */
 785         inneriter                  += j_index_end - j_index_start;
 786
 787         /* Outer loop uses 7 flops */
 788     }
 789
 790     /* Increment number of outer iterations */
 791     outeriter        += nri;
 792
 793     /* Update outer/inner flops */
 794
 795     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 796 }