src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "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_ElecRFCut_VdwLJSh_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_ElecRFCut_VdwLJSh_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 126     rcutoff_scalar   = fr->rcoulomb;
 127     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 128     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 129
 130     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 131     rvdw             = _mm256_set1_ps(fr->rvdw);
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139     j_coord_offsetE = 0;
 140     j_coord_offsetF = 0;
 141     j_coord_offsetG = 0;
 142     j_coord_offsetH = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 168
 169         fix0             = _mm256_setzero_ps();
 170         fiy0             = _mm256_setzero_ps();
 171         fiz0             = _mm256_setzero_ps();
 172
 173         /* Load parameters for i particles */
 174         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 175         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm256_setzero_ps();
 179         vvdwsum          = _mm256_setzero_ps();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             jnrC             = jjnr[jidx+2];
 189             jnrD             = jjnr[jidx+3];
 190             jnrE             = jjnr[jidx+4];
 191             jnrF             = jjnr[jidx+5];
 192             jnrG             = jjnr[jidx+6];
 193             jnrH             = jjnr[jidx+7];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198             j_coord_offsetE  = DIM*jnrE;
 199             j_coord_offsetF  = DIM*jnrF;
 200             j_coord_offsetG  = DIM*jnrG;
 201             j_coord_offsetH  = DIM*jnrH;
 202
 203             /* load j atom coordinates */
 204             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 206                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 207                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 208                                                  &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm256_sub_ps(ix0,jx0);
 212             dy00             = _mm256_sub_ps(iy0,jy0);
 213             dz00             = _mm256_sub_ps(iz0,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 217
 218             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 219
 220             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 224                                                                  charge+jnrC+0,charge+jnrD+0,
 225                                                                  charge+jnrE+0,charge+jnrF+0,
 226                                                                  charge+jnrG+0,charge+jnrH+0);
 227             vdwjidx0A        = 2*vdwtype[jnrA+0];
 228             vdwjidx0B        = 2*vdwtype[jnrB+0];
 229             vdwjidx0C        = 2*vdwtype[jnrC+0];
 230             vdwjidx0D        = 2*vdwtype[jnrD+0];
 231             vdwjidx0E        = 2*vdwtype[jnrE+0];
 232             vdwjidx0F        = 2*vdwtype[jnrF+0];
 233             vdwjidx0G        = 2*vdwtype[jnrG+0];
 234             vdwjidx0H        = 2*vdwtype[jnrH+0];
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 241             {
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             qq00             = _mm256_mul_ps(iq0,jq0);
 245             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 246                                             vdwioffsetptr0+vdwjidx0B,
 247                                             vdwioffsetptr0+vdwjidx0C,
 248                                             vdwioffsetptr0+vdwjidx0D,
 249                                             vdwioffsetptr0+vdwjidx0E,
 250                                             vdwioffsetptr0+vdwjidx0F,
 251                                             vdwioffsetptr0+vdwjidx0G,
 252                                             vdwioffsetptr0+vdwjidx0H,
 253                                             &c6_00,&c12_00);
 254
 255             /* REACTION-FIELD ELECTROSTATICS */
 256             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 257             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 258
 259             /* LENNARD-JONES DISPERSION/REPULSION */
 260
 261             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 262             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 263             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 264             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 265                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 266             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 267
 268             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 269
 270             /* Update potential sum for this i atom from the interaction with this j atom. */
 271             velec            = _mm256_and_ps(velec,cutoff_mask);
 272             velecsum         = _mm256_add_ps(velecsum,velec);
 273             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 274             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 275
 276             fscal            = _mm256_add_ps(felec,fvdw);
 277
 278             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 279
 280             /* Calculate temporary vectorial force */
 281             tx               = _mm256_mul_ps(fscal,dx00);
 282             ty               = _mm256_mul_ps(fscal,dy00);
 283             tz               = _mm256_mul_ps(fscal,dz00);
 284
 285             /* Update vectorial force */
 286             fix0             = _mm256_add_ps(fix0,tx);
 287             fiy0             = _mm256_add_ps(fiy0,ty);
 288             fiz0             = _mm256_add_ps(fiz0,tz);
 289
 290             fjptrA             = f+j_coord_offsetA;
 291             fjptrB             = f+j_coord_offsetB;
 292             fjptrC             = f+j_coord_offsetC;
 293             fjptrD             = f+j_coord_offsetD;
 294             fjptrE             = f+j_coord_offsetE;
 295             fjptrF             = f+j_coord_offsetF;
 296             fjptrG             = f+j_coord_offsetG;
 297             fjptrH             = f+j_coord_offsetH;
 298             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 299
 300             }
 301
 302             /* Inner loop uses 54 flops */
 303         }
 304
 305         if(jidx<j_index_end)
 306         {
 307
 308             /* Get j neighbor index, and coordinate index */
 309             jnrlistA         = jjnr[jidx];
 310             jnrlistB         = jjnr[jidx+1];
 311             jnrlistC         = jjnr[jidx+2];
 312             jnrlistD         = jjnr[jidx+3];
 313             jnrlistE         = jjnr[jidx+4];
 314             jnrlistF         = jjnr[jidx+5];
 315             jnrlistG         = jjnr[jidx+6];
 316             jnrlistH         = jjnr[jidx+7];
 317             /* Sign of each element will be negative for non-real atoms.
 318              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 319              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 320              */
 321             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 322                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 323
 324             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 325             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 326             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 327             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 328             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 329             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 330             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 331             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 332             j_coord_offsetA  = DIM*jnrA;
 333             j_coord_offsetB  = DIM*jnrB;
 334             j_coord_offsetC  = DIM*jnrC;
 335             j_coord_offsetD  = DIM*jnrD;
 336             j_coord_offsetE  = DIM*jnrE;
 337             j_coord_offsetF  = DIM*jnrF;
 338             j_coord_offsetG  = DIM*jnrG;
 339             j_coord_offsetH  = DIM*jnrH;
 340
 341             /* load j atom coordinates */
 342             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 343                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 344                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 345                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 346                                                  &jx0,&jy0,&jz0);
 347
 348             /* Calculate displacement vector */
 349             dx00             = _mm256_sub_ps(ix0,jx0);
 350             dy00             = _mm256_sub_ps(iy0,jy0);
 351             dz00             = _mm256_sub_ps(iz0,jz0);
 352
 353             /* Calculate squared distance and things based on it */
 354             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 355
 356             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 357
 358             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 359
 360             /* Load parameters for j particles */
 361             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 362                                                                  charge+jnrC+0,charge+jnrD+0,
 363                                                                  charge+jnrE+0,charge+jnrF+0,
 364                                                                  charge+jnrG+0,charge+jnrH+0);
 365             vdwjidx0A        = 2*vdwtype[jnrA+0];
 366             vdwjidx0B        = 2*vdwtype[jnrB+0];
 367             vdwjidx0C        = 2*vdwtype[jnrC+0];
 368             vdwjidx0D        = 2*vdwtype[jnrD+0];
 369             vdwjidx0E        = 2*vdwtype[jnrE+0];
 370             vdwjidx0F        = 2*vdwtype[jnrF+0];
 371             vdwjidx0G        = 2*vdwtype[jnrG+0];
 372             vdwjidx0H        = 2*vdwtype[jnrH+0];
 373
 374             /**************************
 375              * CALCULATE INTERACTIONS *
 376              **************************/
 377
 378             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 379             {
 380
 381             /* Compute parameters for interactions between i and j atoms */
 382             qq00             = _mm256_mul_ps(iq0,jq0);
 383             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 384                                             vdwioffsetptr0+vdwjidx0B,
 385                                             vdwioffsetptr0+vdwjidx0C,
 386                                             vdwioffsetptr0+vdwjidx0D,
 387                                             vdwioffsetptr0+vdwjidx0E,
 388                                             vdwioffsetptr0+vdwjidx0F,
 389                                             vdwioffsetptr0+vdwjidx0G,
 390                                             vdwioffsetptr0+vdwjidx0H,
 391                                             &c6_00,&c12_00);
 392
 393             /* REACTION-FIELD ELECTROSTATICS */
 394             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 395             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 396
 397             /* LENNARD-JONES DISPERSION/REPULSION */
 398
 399             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 400             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 401             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 402             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 403                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 404             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 405
 406             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 407
 408             /* Update potential sum for this i atom from the interaction with this j atom. */
 409             velec            = _mm256_and_ps(velec,cutoff_mask);
 410             velec            = _mm256_andnot_ps(dummy_mask,velec);
 411             velecsum         = _mm256_add_ps(velecsum,velec);
 412             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 413             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 414             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 415
 416             fscal            = _mm256_add_ps(felec,fvdw);
 417
 418             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 419
 420             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 421
 422             /* Calculate temporary vectorial force */
 423             tx               = _mm256_mul_ps(fscal,dx00);
 424             ty               = _mm256_mul_ps(fscal,dy00);
 425             tz               = _mm256_mul_ps(fscal,dz00);
 426
 427             /* Update vectorial force */
 428             fix0             = _mm256_add_ps(fix0,tx);
 429             fiy0             = _mm256_add_ps(fiy0,ty);
 430             fiz0             = _mm256_add_ps(fiz0,tz);
 431
 432             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 433             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 434             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 435             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 436             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 437             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 438             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 439             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 440             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 441
 442             }
 443
 444             /* Inner loop uses 54 flops */
 445         }
 446
 447         /* End of innermost loop */
 448
 449         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 450                                                  f+i_coord_offset,fshift+i_shift_offset);
 451
 452         ggid                        = gid[iidx];
 453         /* Update potential energies */
 454         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 455         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 456
 457         /* Increment number of inner iterations */
 458         inneriter                  += j_index_end - j_index_start;
 459
 460         /* Outer loop uses 9 flops */
 461     }
 462
 463     /* Increment number of outer iterations */
 464     outeriter        += nri;
 465
 466     /* Update outer/inner flops */
 467
 468     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
 469 }
 470 /*
 471  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_256_single
 472  * Electrostatics interaction: ReactionField
 473  * VdW interaction:            LennardJones
 474  * Geometry:                   Particle-Particle
 475  * Calculate force/pot:        Force
 476  */
 477 void
 478 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_256_single
 479                     (t_nblist                    * gmx_restrict       nlist,
 480                      rvec                        * gmx_restrict          xx,
 481                      rvec                        * gmx_restrict          ff,
 482                      t_forcerec                  * gmx_restrict          fr,
 483                      t_mdatoms                   * gmx_restrict     mdatoms,
 484                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 485                      t_nrnb                      * gmx_restrict        nrnb)
 486 {
 487     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 488      * just 0 for non-waters.
 489      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 490      * jnr indices corresponding to data put in the four positions in the SIMD register.
 491      */
 492     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 493     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 494     int              jnrA,jnrB,jnrC,jnrD;
 495     int              jnrE,jnrF,jnrG,jnrH;
 496     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 497     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 498     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 499     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 500     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 501     real             rcutoff_scalar;
 502     real             *shiftvec,*fshift,*x,*f;
 503     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 504     real             scratch[4*DIM];
 505     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 506     real *           vdwioffsetptr0;
 507     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 508     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 509     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 510     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 511     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 512     real             *charge;
 513     int              nvdwtype;
 514     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 515     int              *vdwtype;
 516     real             *vdwparam;
 517     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 518     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 519     __m256           dummy_mask,cutoff_mask;
 520     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 521     __m256           one     = _mm256_set1_ps(1.0);
 522     __m256           two     = _mm256_set1_ps(2.0);
 523     x                = xx[0];
 524     f                = ff[0];
 525
 526     nri              = nlist->nri;
 527     iinr             = nlist->iinr;
 528     jindex           = nlist->jindex;
 529     jjnr             = nlist->jjnr;
 530     shiftidx         = nlist->shift;
 531     gid              = nlist->gid;
 532     shiftvec         = fr->shift_vec[0];
 533     fshift           = fr->fshift[0];
 534     facel            = _mm256_set1_ps(fr->epsfac);
 535     charge           = mdatoms->chargeA;
 536     krf              = _mm256_set1_ps(fr->ic->k_rf);
 537     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 538     crf              = _mm256_set1_ps(fr->ic->c_rf);
 539     nvdwtype         = fr->ntype;
 540     vdwparam         = fr->nbfp;
 541     vdwtype          = mdatoms->typeA;
 542
 543     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 544     rcutoff_scalar   = fr->rcoulomb;
 545     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 546     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 547
 548     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 549     rvdw             = _mm256_set1_ps(fr->rvdw);
 550
 551     /* Avoid stupid compiler warnings */
 552     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 553     j_coord_offsetA = 0;
 554     j_coord_offsetB = 0;
 555     j_coord_offsetC = 0;
 556     j_coord_offsetD = 0;
 557     j_coord_offsetE = 0;
 558     j_coord_offsetF = 0;
 559     j_coord_offsetG = 0;
 560     j_coord_offsetH = 0;
 561
 562     outeriter        = 0;
 563     inneriter        = 0;
 564
 565     for(iidx=0;iidx<4*DIM;iidx++)
 566     {
 567         scratch[iidx] = 0.0;
 568     }
 569
 570     /* Start outer loop over neighborlists */
 571     for(iidx=0; iidx<nri; iidx++)
 572     {
 573         /* Load shift vector for this list */
 574         i_shift_offset   = DIM*shiftidx[iidx];
 575
 576         /* Load limits for loop over neighbors */
 577         j_index_start    = jindex[iidx];
 578         j_index_end      = jindex[iidx+1];
 579
 580         /* Get outer coordinate index */
 581         inr              = iinr[iidx];
 582         i_coord_offset   = DIM*inr;
 583
 584         /* Load i particle coords and add shift vector */
 585         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 586
 587         fix0             = _mm256_setzero_ps();
 588         fiy0             = _mm256_setzero_ps();
 589         fiz0             = _mm256_setzero_ps();
 590
 591         /* Load parameters for i particles */
 592         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 593         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 594
 595         /* Start inner kernel loop */
 596         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 597         {
 598
 599             /* Get j neighbor index, and coordinate index */
 600             jnrA             = jjnr[jidx];
 601             jnrB             = jjnr[jidx+1];
 602             jnrC             = jjnr[jidx+2];
 603             jnrD             = jjnr[jidx+3];
 604             jnrE             = jjnr[jidx+4];
 605             jnrF             = jjnr[jidx+5];
 606             jnrG             = jjnr[jidx+6];
 607             jnrH             = jjnr[jidx+7];
 608             j_coord_offsetA  = DIM*jnrA;
 609             j_coord_offsetB  = DIM*jnrB;
 610             j_coord_offsetC  = DIM*jnrC;
 611             j_coord_offsetD  = DIM*jnrD;
 612             j_coord_offsetE  = DIM*jnrE;
 613             j_coord_offsetF  = DIM*jnrF;
 614             j_coord_offsetG  = DIM*jnrG;
 615             j_coord_offsetH  = DIM*jnrH;
 616
 617             /* load j atom coordinates */
 618             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 619                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 620                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 621                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 622                                                  &jx0,&jy0,&jz0);
 623
 624             /* Calculate displacement vector */
 625             dx00             = _mm256_sub_ps(ix0,jx0);
 626             dy00             = _mm256_sub_ps(iy0,jy0);
 627             dz00             = _mm256_sub_ps(iz0,jz0);
 628
 629             /* Calculate squared distance and things based on it */
 630             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 631
 632             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 633
 634             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 635
 636             /* Load parameters for j particles */
 637             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 638                                                                  charge+jnrC+0,charge+jnrD+0,
 639                                                                  charge+jnrE+0,charge+jnrF+0,
 640                                                                  charge+jnrG+0,charge+jnrH+0);
 641             vdwjidx0A        = 2*vdwtype[jnrA+0];
 642             vdwjidx0B        = 2*vdwtype[jnrB+0];
 643             vdwjidx0C        = 2*vdwtype[jnrC+0];
 644             vdwjidx0D        = 2*vdwtype[jnrD+0];
 645             vdwjidx0E        = 2*vdwtype[jnrE+0];
 646             vdwjidx0F        = 2*vdwtype[jnrF+0];
 647             vdwjidx0G        = 2*vdwtype[jnrG+0];
 648             vdwjidx0H        = 2*vdwtype[jnrH+0];
 649
 650             /**************************
 651              * CALCULATE INTERACTIONS *
 652              **************************/
 653
 654             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 655             {
 656
 657             /* Compute parameters for interactions between i and j atoms */
 658             qq00             = _mm256_mul_ps(iq0,jq0);
 659             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 660                                             vdwioffsetptr0+vdwjidx0B,
 661                                             vdwioffsetptr0+vdwjidx0C,
 662                                             vdwioffsetptr0+vdwjidx0D,
 663                                             vdwioffsetptr0+vdwjidx0E,
 664                                             vdwioffsetptr0+vdwjidx0F,
 665                                             vdwioffsetptr0+vdwjidx0G,
 666                                             vdwioffsetptr0+vdwjidx0H,
 667                                             &c6_00,&c12_00);
 668
 669             /* REACTION-FIELD ELECTROSTATICS */
 670             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 671
 672             /* LENNARD-JONES DISPERSION/REPULSION */
 673
 674             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 675             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 676
 677             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 678
 679             fscal            = _mm256_add_ps(felec,fvdw);
 680
 681             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 682
 683             /* Calculate temporary vectorial force */
 684             tx               = _mm256_mul_ps(fscal,dx00);
 685             ty               = _mm256_mul_ps(fscal,dy00);
 686             tz               = _mm256_mul_ps(fscal,dz00);
 687
 688             /* Update vectorial force */
 689             fix0             = _mm256_add_ps(fix0,tx);
 690             fiy0             = _mm256_add_ps(fiy0,ty);
 691             fiz0             = _mm256_add_ps(fiz0,tz);
 692
 693             fjptrA             = f+j_coord_offsetA;
 694             fjptrB             = f+j_coord_offsetB;
 695             fjptrC             = f+j_coord_offsetC;
 696             fjptrD             = f+j_coord_offsetD;
 697             fjptrE             = f+j_coord_offsetE;
 698             fjptrF             = f+j_coord_offsetF;
 699             fjptrG             = f+j_coord_offsetG;
 700             fjptrH             = f+j_coord_offsetH;
 701             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 702
 703             }
 704
 705             /* Inner loop uses 37 flops */
 706         }
 707
 708         if(jidx<j_index_end)
 709         {
 710
 711             /* Get j neighbor index, and coordinate index */
 712             jnrlistA         = jjnr[jidx];
 713             jnrlistB         = jjnr[jidx+1];
 714             jnrlistC         = jjnr[jidx+2];
 715             jnrlistD         = jjnr[jidx+3];
 716             jnrlistE         = jjnr[jidx+4];
 717             jnrlistF         = jjnr[jidx+5];
 718             jnrlistG         = jjnr[jidx+6];
 719             jnrlistH         = jjnr[jidx+7];
 720             /* Sign of each element will be negative for non-real atoms.
 721              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 722              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 723              */
 724             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 725                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 726
 727             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 728             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 729             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 730             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 731             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 732             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 733             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 734             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 735             j_coord_offsetA  = DIM*jnrA;
 736             j_coord_offsetB  = DIM*jnrB;
 737             j_coord_offsetC  = DIM*jnrC;
 738             j_coord_offsetD  = DIM*jnrD;
 739             j_coord_offsetE  = DIM*jnrE;
 740             j_coord_offsetF  = DIM*jnrF;
 741             j_coord_offsetG  = DIM*jnrG;
 742             j_coord_offsetH  = DIM*jnrH;
 743
 744             /* load j atom coordinates */
 745             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 746                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 747                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 748                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 749                                                  &jx0,&jy0,&jz0);
 750
 751             /* Calculate displacement vector */
 752             dx00             = _mm256_sub_ps(ix0,jx0);
 753             dy00             = _mm256_sub_ps(iy0,jy0);
 754             dz00             = _mm256_sub_ps(iz0,jz0);
 755
 756             /* Calculate squared distance and things based on it */
 757             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 758
 759             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 760
 761             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 762
 763             /* Load parameters for j particles */
 764             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 765                                                                  charge+jnrC+0,charge+jnrD+0,
 766                                                                  charge+jnrE+0,charge+jnrF+0,
 767                                                                  charge+jnrG+0,charge+jnrH+0);
 768             vdwjidx0A        = 2*vdwtype[jnrA+0];
 769             vdwjidx0B        = 2*vdwtype[jnrB+0];
 770             vdwjidx0C        = 2*vdwtype[jnrC+0];
 771             vdwjidx0D        = 2*vdwtype[jnrD+0];
 772             vdwjidx0E        = 2*vdwtype[jnrE+0];
 773             vdwjidx0F        = 2*vdwtype[jnrF+0];
 774             vdwjidx0G        = 2*vdwtype[jnrG+0];
 775             vdwjidx0H        = 2*vdwtype[jnrH+0];
 776
 777             /**************************
 778              * CALCULATE INTERACTIONS *
 779              **************************/
 780
 781             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 782             {
 783
 784             /* Compute parameters for interactions between i and j atoms */
 785             qq00             = _mm256_mul_ps(iq0,jq0);
 786             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 787                                             vdwioffsetptr0+vdwjidx0B,
 788                                             vdwioffsetptr0+vdwjidx0C,
 789                                             vdwioffsetptr0+vdwjidx0D,
 790                                             vdwioffsetptr0+vdwjidx0E,
 791                                             vdwioffsetptr0+vdwjidx0F,
 792                                             vdwioffsetptr0+vdwjidx0G,
 793                                             vdwioffsetptr0+vdwjidx0H,
 794                                             &c6_00,&c12_00);
 795
 796             /* REACTION-FIELD ELECTROSTATICS */
 797             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 798
 799             /* LENNARD-JONES DISPERSION/REPULSION */
 800
 801             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 802             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 803
 804             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 805
 806             fscal            = _mm256_add_ps(felec,fvdw);
 807
 808             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 809
 810             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 811
 812             /* Calculate temporary vectorial force */
 813             tx               = _mm256_mul_ps(fscal,dx00);
 814             ty               = _mm256_mul_ps(fscal,dy00);
 815             tz               = _mm256_mul_ps(fscal,dz00);
 816
 817             /* Update vectorial force */
 818             fix0             = _mm256_add_ps(fix0,tx);
 819             fiy0             = _mm256_add_ps(fiy0,ty);
 820             fiz0             = _mm256_add_ps(fiz0,tz);
 821
 822             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 823             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 824             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 825             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 826             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 827             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 828             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 829             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 830             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 831
 832             }
 833
 834             /* Inner loop uses 37 flops */
 835         }
 836
 837         /* End of innermost loop */
 838
 839         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 840                                                  f+i_coord_offset,fshift+i_shift_offset);
 841
 842         /* Increment number of inner iterations */
 843         inneriter                  += j_index_end - j_index_start;
 844
 845         /* Outer loop uses 7 flops */
 846     }
 847
 848     /* Increment number of outer iterations */
 849     outeriter        += nri;
 850
 851     /* Update outer/inner flops */
 852
 853     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 854 }