src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  89     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  98     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  99     __m256i          ewitab;
 100     __m128i          ewitab_lo,ewitab_hi;
 101     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 102     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 103     real             *ewtab;
 104     __m256           dummy_mask,cutoff_mask;
 105     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 106     __m256           one     = _mm256_set1_ps(1.0);
 107     __m256           two     = _mm256_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 126     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 127     beta2            = _mm256_mul_ps(beta,beta);
 128     beta3            = _mm256_mul_ps(beta,beta2);
 129
 130     ewtab            = fr->ic->tabq_coul_FDV0;
 131     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 132     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138     j_coord_offsetC = 0;
 139     j_coord_offsetD = 0;
 140     j_coord_offsetE = 0;
 141     j_coord_offsetF = 0;
 142     j_coord_offsetG = 0;
 143     j_coord_offsetH = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 169
 170         fix0             = _mm256_setzero_ps();
 171         fiy0             = _mm256_setzero_ps();
 172         fiz0             = _mm256_setzero_ps();
 173
 174         /* Load parameters for i particles */
 175         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 176         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm256_setzero_ps();
 180         vvdwsum          = _mm256_setzero_ps();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             jnrC             = jjnr[jidx+2];
 190             jnrD             = jjnr[jidx+3];
 191             jnrE             = jjnr[jidx+4];
 192             jnrF             = jjnr[jidx+5];
 193             jnrG             = jjnr[jidx+6];
 194             jnrH             = jjnr[jidx+7];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199             j_coord_offsetE  = DIM*jnrE;
 200             j_coord_offsetF  = DIM*jnrF;
 201             j_coord_offsetG  = DIM*jnrG;
 202             j_coord_offsetH  = DIM*jnrH;
 203
 204             /* load j atom coordinates */
 205             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 207                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 208                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 209                                                  &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm256_sub_ps(ix0,jx0);
 213             dy00             = _mm256_sub_ps(iy0,jy0);
 214             dz00             = _mm256_sub_ps(iz0,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 218
 219             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 220
 221             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 222
 223             /* Load parameters for j particles */
 224             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 225                                                                  charge+jnrC+0,charge+jnrD+0,
 226                                                                  charge+jnrE+0,charge+jnrF+0,
 227                                                                  charge+jnrG+0,charge+jnrH+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232             vdwjidx0E        = 2*vdwtype[jnrE+0];
 233             vdwjidx0F        = 2*vdwtype[jnrF+0];
 234             vdwjidx0G        = 2*vdwtype[jnrG+0];
 235             vdwjidx0H        = 2*vdwtype[jnrH+0];
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             r00              = _mm256_mul_ps(rsq00,rinv00);
 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             /* EWALD ELECTROSTATICS */
 256
 257             /* Analytical PME correction */
 258             zeta2            = _mm256_mul_ps(beta2,rsq00);
 259             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 260             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 261             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 262             felec            = _mm256_mul_ps(qq00,felec);
 263             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 264             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 265             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 266             velec            = _mm256_mul_ps(qq00,velec);
 267
 268             /* LENNARD-JONES DISPERSION/REPULSION */
 269
 270             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 271             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 272             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 273             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 274             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velecsum         = _mm256_add_ps(velecsum,velec);
 278             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 279
 280             fscal            = _mm256_add_ps(felec,fvdw);
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm256_mul_ps(fscal,dx00);
 284             ty               = _mm256_mul_ps(fscal,dy00);
 285             tz               = _mm256_mul_ps(fscal,dz00);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm256_add_ps(fix0,tx);
 289             fiy0             = _mm256_add_ps(fiy0,ty);
 290             fiz0             = _mm256_add_ps(fiz0,tz);
 291
 292             fjptrA             = f+j_coord_offsetA;
 293             fjptrB             = f+j_coord_offsetB;
 294             fjptrC             = f+j_coord_offsetC;
 295             fjptrD             = f+j_coord_offsetD;
 296             fjptrE             = f+j_coord_offsetE;
 297             fjptrF             = f+j_coord_offsetF;
 298             fjptrG             = f+j_coord_offsetG;
 299             fjptrH             = f+j_coord_offsetH;
 300             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 301
 302             /* Inner loop uses 96 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             r00              = _mm256_mul_ps(rsq00,rinv00);
 379             r00              = _mm256_andnot_ps(dummy_mask,r00);
 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             /* EWALD ELECTROSTATICS */
 394
 395             /* Analytical PME correction */
 396             zeta2            = _mm256_mul_ps(beta2,rsq00);
 397             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 398             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 399             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 400             felec            = _mm256_mul_ps(qq00,felec);
 401             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 402             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 403             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 404             velec            = _mm256_mul_ps(qq00,velec);
 405
 406             /* LENNARD-JONES DISPERSION/REPULSION */
 407
 408             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 409             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 410             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 411             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 412             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velec            = _mm256_andnot_ps(dummy_mask,velec);
 416             velecsum         = _mm256_add_ps(velecsum,velec);
 417             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 418             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 419
 420             fscal            = _mm256_add_ps(felec,fvdw);
 421
 422             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 423
 424             /* Calculate temporary vectorial force */
 425             tx               = _mm256_mul_ps(fscal,dx00);
 426             ty               = _mm256_mul_ps(fscal,dy00);
 427             tz               = _mm256_mul_ps(fscal,dz00);
 428
 429             /* Update vectorial force */
 430             fix0             = _mm256_add_ps(fix0,tx);
 431             fiy0             = _mm256_add_ps(fiy0,ty);
 432             fiz0             = _mm256_add_ps(fiz0,tz);
 433
 434             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 435             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 436             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 437             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 438             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 439             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 440             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 441             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 442             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 443
 444             /* Inner loop uses 97 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*97);
 469 }
 470 /*
 471  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_256_single
 472  * Electrostatics interaction: Ewald
 473  * VdW interaction:            LennardJones
 474  * Geometry:                   Particle-Particle
 475  * Calculate force/pot:        Force
 476  */
 477 void
 478 nb_kernel_ElecEw_VdwLJ_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     __m256i          ewitab;
 520     __m128i          ewitab_lo,ewitab_hi;
 521     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 522     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 523     real             *ewtab;
 524     __m256           dummy_mask,cutoff_mask;
 525     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 526     __m256           one     = _mm256_set1_ps(1.0);
 527     __m256           two     = _mm256_set1_ps(2.0);
 528     x                = xx[0];
 529     f                = ff[0];
 530
 531     nri              = nlist->nri;
 532     iinr             = nlist->iinr;
 533     jindex           = nlist->jindex;
 534     jjnr             = nlist->jjnr;
 535     shiftidx         = nlist->shift;
 536     gid              = nlist->gid;
 537     shiftvec         = fr->shift_vec[0];
 538     fshift           = fr->fshift[0];
 539     facel            = _mm256_set1_ps(fr->epsfac);
 540     charge           = mdatoms->chargeA;
 541     nvdwtype         = fr->ntype;
 542     vdwparam         = fr->nbfp;
 543     vdwtype          = mdatoms->typeA;
 544
 545     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 546     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 547     beta2            = _mm256_mul_ps(beta,beta);
 548     beta3            = _mm256_mul_ps(beta,beta2);
 549
 550     ewtab            = fr->ic->tabq_coul_F;
 551     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 552     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 553
 554     /* Avoid stupid compiler warnings */
 555     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 556     j_coord_offsetA = 0;
 557     j_coord_offsetB = 0;
 558     j_coord_offsetC = 0;
 559     j_coord_offsetD = 0;
 560     j_coord_offsetE = 0;
 561     j_coord_offsetF = 0;
 562     j_coord_offsetG = 0;
 563     j_coord_offsetH = 0;
 564
 565     outeriter        = 0;
 566     inneriter        = 0;
 567
 568     for(iidx=0;iidx<4*DIM;iidx++)
 569     {
 570         scratch[iidx] = 0.0;
 571     }
 572
 573     /* Start outer loop over neighborlists */
 574     for(iidx=0; iidx<nri; iidx++)
 575     {
 576         /* Load shift vector for this list */
 577         i_shift_offset   = DIM*shiftidx[iidx];
 578
 579         /* Load limits for loop over neighbors */
 580         j_index_start    = jindex[iidx];
 581         j_index_end      = jindex[iidx+1];
 582
 583         /* Get outer coordinate index */
 584         inr              = iinr[iidx];
 585         i_coord_offset   = DIM*inr;
 586
 587         /* Load i particle coords and add shift vector */
 588         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 589
 590         fix0             = _mm256_setzero_ps();
 591         fiy0             = _mm256_setzero_ps();
 592         fiz0             = _mm256_setzero_ps();
 593
 594         /* Load parameters for i particles */
 595         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 596         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 597
 598         /* Start inner kernel loop */
 599         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 600         {
 601
 602             /* Get j neighbor index, and coordinate index */
 603             jnrA             = jjnr[jidx];
 604             jnrB             = jjnr[jidx+1];
 605             jnrC             = jjnr[jidx+2];
 606             jnrD             = jjnr[jidx+3];
 607             jnrE             = jjnr[jidx+4];
 608             jnrF             = jjnr[jidx+5];
 609             jnrG             = jjnr[jidx+6];
 610             jnrH             = jjnr[jidx+7];
 611             j_coord_offsetA  = DIM*jnrA;
 612             j_coord_offsetB  = DIM*jnrB;
 613             j_coord_offsetC  = DIM*jnrC;
 614             j_coord_offsetD  = DIM*jnrD;
 615             j_coord_offsetE  = DIM*jnrE;
 616             j_coord_offsetF  = DIM*jnrF;
 617             j_coord_offsetG  = DIM*jnrG;
 618             j_coord_offsetH  = DIM*jnrH;
 619
 620             /* load j atom coordinates */
 621             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 622                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 623                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 624                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 625                                                  &jx0,&jy0,&jz0);
 626
 627             /* Calculate displacement vector */
 628             dx00             = _mm256_sub_ps(ix0,jx0);
 629             dy00             = _mm256_sub_ps(iy0,jy0);
 630             dz00             = _mm256_sub_ps(iz0,jz0);
 631
 632             /* Calculate squared distance and things based on it */
 633             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 634
 635             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 636
 637             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 638
 639             /* Load parameters for j particles */
 640             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 641                                                                  charge+jnrC+0,charge+jnrD+0,
 642                                                                  charge+jnrE+0,charge+jnrF+0,
 643                                                                  charge+jnrG+0,charge+jnrH+0);
 644             vdwjidx0A        = 2*vdwtype[jnrA+0];
 645             vdwjidx0B        = 2*vdwtype[jnrB+0];
 646             vdwjidx0C        = 2*vdwtype[jnrC+0];
 647             vdwjidx0D        = 2*vdwtype[jnrD+0];
 648             vdwjidx0E        = 2*vdwtype[jnrE+0];
 649             vdwjidx0F        = 2*vdwtype[jnrF+0];
 650             vdwjidx0G        = 2*vdwtype[jnrG+0];
 651             vdwjidx0H        = 2*vdwtype[jnrH+0];
 652
 653             /**************************
 654              * CALCULATE INTERACTIONS *
 655              **************************/
 656
 657             r00              = _mm256_mul_ps(rsq00,rinv00);
 658
 659             /* Compute parameters for interactions between i and j atoms */
 660             qq00             = _mm256_mul_ps(iq0,jq0);
 661             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 662                                             vdwioffsetptr0+vdwjidx0B,
 663                                             vdwioffsetptr0+vdwjidx0C,
 664                                             vdwioffsetptr0+vdwjidx0D,
 665                                             vdwioffsetptr0+vdwjidx0E,
 666                                             vdwioffsetptr0+vdwjidx0F,
 667                                             vdwioffsetptr0+vdwjidx0G,
 668                                             vdwioffsetptr0+vdwjidx0H,
 669                                             &c6_00,&c12_00);
 670
 671             /* EWALD ELECTROSTATICS */
 672
 673             /* Analytical PME correction */
 674             zeta2            = _mm256_mul_ps(beta2,rsq00);
 675             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 676             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 677             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 678             felec            = _mm256_mul_ps(qq00,felec);
 679
 680             /* LENNARD-JONES DISPERSION/REPULSION */
 681
 682             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 683             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 684
 685             fscal            = _mm256_add_ps(felec,fvdw);
 686
 687             /* Calculate temporary vectorial force */
 688             tx               = _mm256_mul_ps(fscal,dx00);
 689             ty               = _mm256_mul_ps(fscal,dy00);
 690             tz               = _mm256_mul_ps(fscal,dz00);
 691
 692             /* Update vectorial force */
 693             fix0             = _mm256_add_ps(fix0,tx);
 694             fiy0             = _mm256_add_ps(fiy0,ty);
 695             fiz0             = _mm256_add_ps(fiz0,tz);
 696
 697             fjptrA             = f+j_coord_offsetA;
 698             fjptrB             = f+j_coord_offsetB;
 699             fjptrC             = f+j_coord_offsetC;
 700             fjptrD             = f+j_coord_offsetD;
 701             fjptrE             = f+j_coord_offsetE;
 702             fjptrF             = f+j_coord_offsetF;
 703             fjptrG             = f+j_coord_offsetG;
 704             fjptrH             = f+j_coord_offsetH;
 705             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 706
 707             /* Inner loop uses 63 flops */
 708         }
 709
 710         if(jidx<j_index_end)
 711         {
 712
 713             /* Get j neighbor index, and coordinate index */
 714             jnrlistA         = jjnr[jidx];
 715             jnrlistB         = jjnr[jidx+1];
 716             jnrlistC         = jjnr[jidx+2];
 717             jnrlistD         = jjnr[jidx+3];
 718             jnrlistE         = jjnr[jidx+4];
 719             jnrlistF         = jjnr[jidx+5];
 720             jnrlistG         = jjnr[jidx+6];
 721             jnrlistH         = jjnr[jidx+7];
 722             /* Sign of each element will be negative for non-real atoms.
 723              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 724              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 725              */
 726             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 727                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 728
 729             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 730             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 731             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 732             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 733             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 734             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 735             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 736             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 737             j_coord_offsetA  = DIM*jnrA;
 738             j_coord_offsetB  = DIM*jnrB;
 739             j_coord_offsetC  = DIM*jnrC;
 740             j_coord_offsetD  = DIM*jnrD;
 741             j_coord_offsetE  = DIM*jnrE;
 742             j_coord_offsetF  = DIM*jnrF;
 743             j_coord_offsetG  = DIM*jnrG;
 744             j_coord_offsetH  = DIM*jnrH;
 745
 746             /* load j atom coordinates */
 747             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 748                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 749                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 750                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 751                                                  &jx0,&jy0,&jz0);
 752
 753             /* Calculate displacement vector */
 754             dx00             = _mm256_sub_ps(ix0,jx0);
 755             dy00             = _mm256_sub_ps(iy0,jy0);
 756             dz00             = _mm256_sub_ps(iz0,jz0);
 757
 758             /* Calculate squared distance and things based on it */
 759             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 760
 761             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 762
 763             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 764
 765             /* Load parameters for j particles */
 766             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 767                                                                  charge+jnrC+0,charge+jnrD+0,
 768                                                                  charge+jnrE+0,charge+jnrF+0,
 769                                                                  charge+jnrG+0,charge+jnrH+0);
 770             vdwjidx0A        = 2*vdwtype[jnrA+0];
 771             vdwjidx0B        = 2*vdwtype[jnrB+0];
 772             vdwjidx0C        = 2*vdwtype[jnrC+0];
 773             vdwjidx0D        = 2*vdwtype[jnrD+0];
 774             vdwjidx0E        = 2*vdwtype[jnrE+0];
 775             vdwjidx0F        = 2*vdwtype[jnrF+0];
 776             vdwjidx0G        = 2*vdwtype[jnrG+0];
 777             vdwjidx0H        = 2*vdwtype[jnrH+0];
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             r00              = _mm256_mul_ps(rsq00,rinv00);
 784             r00              = _mm256_andnot_ps(dummy_mask,r00);
 785
 786             /* Compute parameters for interactions between i and j atoms */
 787             qq00             = _mm256_mul_ps(iq0,jq0);
 788             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 789                                             vdwioffsetptr0+vdwjidx0B,
 790                                             vdwioffsetptr0+vdwjidx0C,
 791                                             vdwioffsetptr0+vdwjidx0D,
 792                                             vdwioffsetptr0+vdwjidx0E,
 793                                             vdwioffsetptr0+vdwjidx0F,
 794                                             vdwioffsetptr0+vdwjidx0G,
 795                                             vdwioffsetptr0+vdwjidx0H,
 796                                             &c6_00,&c12_00);
 797
 798             /* EWALD ELECTROSTATICS */
 799
 800             /* Analytical PME correction */
 801             zeta2            = _mm256_mul_ps(beta2,rsq00);
 802             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 803             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 804             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 805             felec            = _mm256_mul_ps(qq00,felec);
 806
 807             /* LENNARD-JONES DISPERSION/REPULSION */
 808
 809             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 810             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 811
 812             fscal            = _mm256_add_ps(felec,fvdw);
 813
 814             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 815
 816             /* Calculate temporary vectorial force */
 817             tx               = _mm256_mul_ps(fscal,dx00);
 818             ty               = _mm256_mul_ps(fscal,dy00);
 819             tz               = _mm256_mul_ps(fscal,dz00);
 820
 821             /* Update vectorial force */
 822             fix0             = _mm256_add_ps(fix0,tx);
 823             fiy0             = _mm256_add_ps(fiy0,ty);
 824             fiz0             = _mm256_add_ps(fiz0,tz);
 825
 826             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 827             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 828             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 829             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 830             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 831             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 832             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 833             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 834             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 835
 836             /* Inner loop uses 64 flops */
 837         }
 838
 839         /* End of innermost loop */
 840
 841         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 842                                                  f+i_coord_offset,fshift+i_shift_offset);
 843
 844         /* Increment number of inner iterations */
 845         inneriter                  += j_index_end - j_index_start;
 846
 847         /* Outer loop uses 7 flops */
 848     }
 849
 850     /* Increment number of outer iterations */
 851     outeriter        += nri;
 852
 853     /* Update outer/inner flops */
 854
 855     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 856 }