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) 2012,2013,2014, 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_ElecRFCut_VdwLJSh_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_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     __m256           dummy_mask,cutoff_mask;
 100     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 101     __m256           one     = _mm256_set1_ps(1.0);
 102     __m256           two     = _mm256_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm256_set1_ps(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     krf              = _mm256_set1_ps(fr->ic->k_rf);
 117     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 118     crf              = _mm256_set1_ps(fr->ic->c_rf);
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 124     rcutoff_scalar   = fr->rcoulomb;
 125     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 126     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 127
 128     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 129     rvdw             = _mm256_set1_ps(fr->rvdw);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137     j_coord_offsetE = 0;
 138     j_coord_offsetF = 0;
 139     j_coord_offsetG = 0;
 140     j_coord_offsetH = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     for(iidx=0;iidx<4*DIM;iidx++)
 146     {
 147         scratch[iidx] = 0.0;
 148     }
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 166
 167         fix0             = _mm256_setzero_ps();
 168         fiy0             = _mm256_setzero_ps();
 169         fiz0             = _mm256_setzero_ps();
 170
 171         /* Load parameters for i particles */
 172         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 173         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm256_setzero_ps();
 177         vvdwsum          = _mm256_setzero_ps();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             jnrC             = jjnr[jidx+2];
 187             jnrD             = jjnr[jidx+3];
 188             jnrE             = jjnr[jidx+4];
 189             jnrF             = jjnr[jidx+5];
 190             jnrG             = jjnr[jidx+6];
 191             jnrH             = jjnr[jidx+7];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194             j_coord_offsetC  = DIM*jnrC;
 195             j_coord_offsetD  = DIM*jnrD;
 196             j_coord_offsetE  = DIM*jnrE;
 197             j_coord_offsetF  = DIM*jnrF;
 198             j_coord_offsetG  = DIM*jnrG;
 199             j_coord_offsetH  = DIM*jnrH;
 200
 201             /* load j atom coordinates */
 202             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 203                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 204                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 205                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 206                                                  &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm256_sub_ps(ix0,jx0);
 210             dy00             = _mm256_sub_ps(iy0,jy0);
 211             dz00             = _mm256_sub_ps(iz0,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 215
 216             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 217
 218             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 222                                                                  charge+jnrC+0,charge+jnrD+0,
 223                                                                  charge+jnrE+0,charge+jnrF+0,
 224                                                                  charge+jnrG+0,charge+jnrH+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227             vdwjidx0C        = 2*vdwtype[jnrC+0];
 228             vdwjidx0D        = 2*vdwtype[jnrD+0];
 229             vdwjidx0E        = 2*vdwtype[jnrE+0];
 230             vdwjidx0F        = 2*vdwtype[jnrF+0];
 231             vdwjidx0G        = 2*vdwtype[jnrG+0];
 232             vdwjidx0H        = 2*vdwtype[jnrH+0];
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 239             {
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             qq00             = _mm256_mul_ps(iq0,jq0);
 243             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 244                                             vdwioffsetptr0+vdwjidx0B,
 245                                             vdwioffsetptr0+vdwjidx0C,
 246                                             vdwioffsetptr0+vdwjidx0D,
 247                                             vdwioffsetptr0+vdwjidx0E,
 248                                             vdwioffsetptr0+vdwjidx0F,
 249                                             vdwioffsetptr0+vdwjidx0G,
 250                                             vdwioffsetptr0+vdwjidx0H,
 251                                             &c6_00,&c12_00);
 252
 253             /* REACTION-FIELD ELECTROSTATICS */
 254             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 255             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 256
 257             /* LENNARD-JONES DISPERSION/REPULSION */
 258
 259             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 260             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 261             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 262             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) ,
 263                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 264             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 265
 266             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             velec            = _mm256_and_ps(velec,cutoff_mask);
 270             velecsum         = _mm256_add_ps(velecsum,velec);
 271             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 272             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 273
 274             fscal            = _mm256_add_ps(felec,fvdw);
 275
 276             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm256_mul_ps(fscal,dx00);
 280             ty               = _mm256_mul_ps(fscal,dy00);
 281             tz               = _mm256_mul_ps(fscal,dz00);
 282
 283             /* Update vectorial force */
 284             fix0             = _mm256_add_ps(fix0,tx);
 285             fiy0             = _mm256_add_ps(fiy0,ty);
 286             fiz0             = _mm256_add_ps(fiz0,tz);
 287
 288             fjptrA             = f+j_coord_offsetA;
 289             fjptrB             = f+j_coord_offsetB;
 290             fjptrC             = f+j_coord_offsetC;
 291             fjptrD             = f+j_coord_offsetD;
 292             fjptrE             = f+j_coord_offsetE;
 293             fjptrF             = f+j_coord_offsetF;
 294             fjptrG             = f+j_coord_offsetG;
 295             fjptrH             = f+j_coord_offsetH;
 296             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 297
 298             }
 299
 300             /* Inner loop uses 54 flops */
 301         }
 302
 303         if(jidx<j_index_end)
 304         {
 305
 306             /* Get j neighbor index, and coordinate index */
 307             jnrlistA         = jjnr[jidx];
 308             jnrlistB         = jjnr[jidx+1];
 309             jnrlistC         = jjnr[jidx+2];
 310             jnrlistD         = jjnr[jidx+3];
 311             jnrlistE         = jjnr[jidx+4];
 312             jnrlistF         = jjnr[jidx+5];
 313             jnrlistG         = jjnr[jidx+6];
 314             jnrlistH         = jjnr[jidx+7];
 315             /* Sign of each element will be negative for non-real atoms.
 316              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 317              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 318              */
 319             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 320                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 321
 322             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 323             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 324             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 325             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 326             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 327             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 328             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 329             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 330             j_coord_offsetA  = DIM*jnrA;
 331             j_coord_offsetB  = DIM*jnrB;
 332             j_coord_offsetC  = DIM*jnrC;
 333             j_coord_offsetD  = DIM*jnrD;
 334             j_coord_offsetE  = DIM*jnrE;
 335             j_coord_offsetF  = DIM*jnrF;
 336             j_coord_offsetG  = DIM*jnrG;
 337             j_coord_offsetH  = DIM*jnrH;
 338
 339             /* load j atom coordinates */
 340             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 341                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 342                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 343                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 344                                                  &jx0,&jy0,&jz0);
 345
 346             /* Calculate displacement vector */
 347             dx00             = _mm256_sub_ps(ix0,jx0);
 348             dy00             = _mm256_sub_ps(iy0,jy0);
 349             dz00             = _mm256_sub_ps(iz0,jz0);
 350
 351             /* Calculate squared distance and things based on it */
 352             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 353
 354             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 355
 356             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 357
 358             /* Load parameters for j particles */
 359             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 360                                                                  charge+jnrC+0,charge+jnrD+0,
 361                                                                  charge+jnrE+0,charge+jnrF+0,
 362                                                                  charge+jnrG+0,charge+jnrH+0);
 363             vdwjidx0A        = 2*vdwtype[jnrA+0];
 364             vdwjidx0B        = 2*vdwtype[jnrB+0];
 365             vdwjidx0C        = 2*vdwtype[jnrC+0];
 366             vdwjidx0D        = 2*vdwtype[jnrD+0];
 367             vdwjidx0E        = 2*vdwtype[jnrE+0];
 368             vdwjidx0F        = 2*vdwtype[jnrF+0];
 369             vdwjidx0G        = 2*vdwtype[jnrG+0];
 370             vdwjidx0H        = 2*vdwtype[jnrH+0];
 371
 372             /**************************
 373              * CALCULATE INTERACTIONS *
 374              **************************/
 375
 376             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 377             {
 378
 379             /* Compute parameters for interactions between i and j atoms */
 380             qq00             = _mm256_mul_ps(iq0,jq0);
 381             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 382                                             vdwioffsetptr0+vdwjidx0B,
 383                                             vdwioffsetptr0+vdwjidx0C,
 384                                             vdwioffsetptr0+vdwjidx0D,
 385                                             vdwioffsetptr0+vdwjidx0E,
 386                                             vdwioffsetptr0+vdwjidx0F,
 387                                             vdwioffsetptr0+vdwjidx0G,
 388                                             vdwioffsetptr0+vdwjidx0H,
 389                                             &c6_00,&c12_00);
 390
 391             /* REACTION-FIELD ELECTROSTATICS */
 392             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 393             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 394
 395             /* LENNARD-JONES DISPERSION/REPULSION */
 396
 397             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 398             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 399             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 400             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) ,
 401                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 402             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 403
 404             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velec            = _mm256_and_ps(velec,cutoff_mask);
 408             velec            = _mm256_andnot_ps(dummy_mask,velec);
 409             velecsum         = _mm256_add_ps(velecsum,velec);
 410             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 411             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 412             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 413
 414             fscal            = _mm256_add_ps(felec,fvdw);
 415
 416             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 417
 418             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 419
 420             /* Calculate temporary vectorial force */
 421             tx               = _mm256_mul_ps(fscal,dx00);
 422             ty               = _mm256_mul_ps(fscal,dy00);
 423             tz               = _mm256_mul_ps(fscal,dz00);
 424
 425             /* Update vectorial force */
 426             fix0             = _mm256_add_ps(fix0,tx);
 427             fiy0             = _mm256_add_ps(fiy0,ty);
 428             fiz0             = _mm256_add_ps(fiz0,tz);
 429
 430             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 431             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 432             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 433             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 434             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 435             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 436             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 437             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 438             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 439
 440             }
 441
 442             /* Inner loop uses 54 flops */
 443         }
 444
 445         /* End of innermost loop */
 446
 447         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 448                                                  f+i_coord_offset,fshift+i_shift_offset);
 449
 450         ggid                        = gid[iidx];
 451         /* Update potential energies */
 452         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 453         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 454
 455         /* Increment number of inner iterations */
 456         inneriter                  += j_index_end - j_index_start;
 457
 458         /* Outer loop uses 9 flops */
 459     }
 460
 461     /* Increment number of outer iterations */
 462     outeriter        += nri;
 463
 464     /* Update outer/inner flops */
 465
 466     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
 467 }
 468 /*
 469  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_256_single
 470  * Electrostatics interaction: ReactionField
 471  * VdW interaction:            LennardJones
 472  * Geometry:                   Particle-Particle
 473  * Calculate force/pot:        Force
 474  */
 475 void
 476 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_256_single
 477                     (t_nblist                    * gmx_restrict       nlist,
 478                      rvec                        * gmx_restrict          xx,
 479                      rvec                        * gmx_restrict          ff,
 480                      t_forcerec                  * gmx_restrict          fr,
 481                      t_mdatoms                   * gmx_restrict     mdatoms,
 482                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 483                      t_nrnb                      * gmx_restrict        nrnb)
 484 {
 485     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 486      * just 0 for non-waters.
 487      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 488      * jnr indices corresponding to data put in the four positions in the SIMD register.
 489      */
 490     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 491     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 492     int              jnrA,jnrB,jnrC,jnrD;
 493     int              jnrE,jnrF,jnrG,jnrH;
 494     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 495     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 496     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 497     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 498     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 499     real             rcutoff_scalar;
 500     real             *shiftvec,*fshift,*x,*f;
 501     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 502     real             scratch[4*DIM];
 503     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 504     real *           vdwioffsetptr0;
 505     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 506     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 507     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 508     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 509     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 510     real             *charge;
 511     int              nvdwtype;
 512     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 513     int              *vdwtype;
 514     real             *vdwparam;
 515     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 516     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 517     __m256           dummy_mask,cutoff_mask;
 518     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 519     __m256           one     = _mm256_set1_ps(1.0);
 520     __m256           two     = _mm256_set1_ps(2.0);
 521     x                = xx[0];
 522     f                = ff[0];
 523
 524     nri              = nlist->nri;
 525     iinr             = nlist->iinr;
 526     jindex           = nlist->jindex;
 527     jjnr             = nlist->jjnr;
 528     shiftidx         = nlist->shift;
 529     gid              = nlist->gid;
 530     shiftvec         = fr->shift_vec[0];
 531     fshift           = fr->fshift[0];
 532     facel            = _mm256_set1_ps(fr->epsfac);
 533     charge           = mdatoms->chargeA;
 534     krf              = _mm256_set1_ps(fr->ic->k_rf);
 535     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 536     crf              = _mm256_set1_ps(fr->ic->c_rf);
 537     nvdwtype         = fr->ntype;
 538     vdwparam         = fr->nbfp;
 539     vdwtype          = mdatoms->typeA;
 540
 541     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 542     rcutoff_scalar   = fr->rcoulomb;
 543     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 544     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 545
 546     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 547     rvdw             = _mm256_set1_ps(fr->rvdw);
 548
 549     /* Avoid stupid compiler warnings */
 550     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 551     j_coord_offsetA = 0;
 552     j_coord_offsetB = 0;
 553     j_coord_offsetC = 0;
 554     j_coord_offsetD = 0;
 555     j_coord_offsetE = 0;
 556     j_coord_offsetF = 0;
 557     j_coord_offsetG = 0;
 558     j_coord_offsetH = 0;
 559
 560     outeriter        = 0;
 561     inneriter        = 0;
 562
 563     for(iidx=0;iidx<4*DIM;iidx++)
 564     {
 565         scratch[iidx] = 0.0;
 566     }
 567
 568     /* Start outer loop over neighborlists */
 569     for(iidx=0; iidx<nri; iidx++)
 570     {
 571         /* Load shift vector for this list */
 572         i_shift_offset   = DIM*shiftidx[iidx];
 573
 574         /* Load limits for loop over neighbors */
 575         j_index_start    = jindex[iidx];
 576         j_index_end      = jindex[iidx+1];
 577
 578         /* Get outer coordinate index */
 579         inr              = iinr[iidx];
 580         i_coord_offset   = DIM*inr;
 581
 582         /* Load i particle coords and add shift vector */
 583         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 584
 585         fix0             = _mm256_setzero_ps();
 586         fiy0             = _mm256_setzero_ps();
 587         fiz0             = _mm256_setzero_ps();
 588
 589         /* Load parameters for i particles */
 590         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 591         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 592
 593         /* Start inner kernel loop */
 594         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 595         {
 596
 597             /* Get j neighbor index, and coordinate index */
 598             jnrA             = jjnr[jidx];
 599             jnrB             = jjnr[jidx+1];
 600             jnrC             = jjnr[jidx+2];
 601             jnrD             = jjnr[jidx+3];
 602             jnrE             = jjnr[jidx+4];
 603             jnrF             = jjnr[jidx+5];
 604             jnrG             = jjnr[jidx+6];
 605             jnrH             = jjnr[jidx+7];
 606             j_coord_offsetA  = DIM*jnrA;
 607             j_coord_offsetB  = DIM*jnrB;
 608             j_coord_offsetC  = DIM*jnrC;
 609             j_coord_offsetD  = DIM*jnrD;
 610             j_coord_offsetE  = DIM*jnrE;
 611             j_coord_offsetF  = DIM*jnrF;
 612             j_coord_offsetG  = DIM*jnrG;
 613             j_coord_offsetH  = DIM*jnrH;
 614
 615             /* load j atom coordinates */
 616             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 617                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 618                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 619                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 620                                                  &jx0,&jy0,&jz0);
 621
 622             /* Calculate displacement vector */
 623             dx00             = _mm256_sub_ps(ix0,jx0);
 624             dy00             = _mm256_sub_ps(iy0,jy0);
 625             dz00             = _mm256_sub_ps(iz0,jz0);
 626
 627             /* Calculate squared distance and things based on it */
 628             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 629
 630             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 631
 632             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 633
 634             /* Load parameters for j particles */
 635             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 636                                                                  charge+jnrC+0,charge+jnrD+0,
 637                                                                  charge+jnrE+0,charge+jnrF+0,
 638                                                                  charge+jnrG+0,charge+jnrH+0);
 639             vdwjidx0A        = 2*vdwtype[jnrA+0];
 640             vdwjidx0B        = 2*vdwtype[jnrB+0];
 641             vdwjidx0C        = 2*vdwtype[jnrC+0];
 642             vdwjidx0D        = 2*vdwtype[jnrD+0];
 643             vdwjidx0E        = 2*vdwtype[jnrE+0];
 644             vdwjidx0F        = 2*vdwtype[jnrF+0];
 645             vdwjidx0G        = 2*vdwtype[jnrG+0];
 646             vdwjidx0H        = 2*vdwtype[jnrH+0];
 647
 648             /**************************
 649              * CALCULATE INTERACTIONS *
 650              **************************/
 651
 652             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 653             {
 654
 655             /* Compute parameters for interactions between i and j atoms */
 656             qq00             = _mm256_mul_ps(iq0,jq0);
 657             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 658                                             vdwioffsetptr0+vdwjidx0B,
 659                                             vdwioffsetptr0+vdwjidx0C,
 660                                             vdwioffsetptr0+vdwjidx0D,
 661                                             vdwioffsetptr0+vdwjidx0E,
 662                                             vdwioffsetptr0+vdwjidx0F,
 663                                             vdwioffsetptr0+vdwjidx0G,
 664                                             vdwioffsetptr0+vdwjidx0H,
 665                                             &c6_00,&c12_00);
 666
 667             /* REACTION-FIELD ELECTROSTATICS */
 668             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 669
 670             /* LENNARD-JONES DISPERSION/REPULSION */
 671
 672             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 673             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 674
 675             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 676
 677             fscal            = _mm256_add_ps(felec,fvdw);
 678
 679             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 680
 681             /* Calculate temporary vectorial force */
 682             tx               = _mm256_mul_ps(fscal,dx00);
 683             ty               = _mm256_mul_ps(fscal,dy00);
 684             tz               = _mm256_mul_ps(fscal,dz00);
 685
 686             /* Update vectorial force */
 687             fix0             = _mm256_add_ps(fix0,tx);
 688             fiy0             = _mm256_add_ps(fiy0,ty);
 689             fiz0             = _mm256_add_ps(fiz0,tz);
 690
 691             fjptrA             = f+j_coord_offsetA;
 692             fjptrB             = f+j_coord_offsetB;
 693             fjptrC             = f+j_coord_offsetC;
 694             fjptrD             = f+j_coord_offsetD;
 695             fjptrE             = f+j_coord_offsetE;
 696             fjptrF             = f+j_coord_offsetF;
 697             fjptrG             = f+j_coord_offsetG;
 698             fjptrH             = f+j_coord_offsetH;
 699             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 700
 701             }
 702
 703             /* Inner loop uses 37 flops */
 704         }
 705
 706         if(jidx<j_index_end)
 707         {
 708
 709             /* Get j neighbor index, and coordinate index */
 710             jnrlistA         = jjnr[jidx];
 711             jnrlistB         = jjnr[jidx+1];
 712             jnrlistC         = jjnr[jidx+2];
 713             jnrlistD         = jjnr[jidx+3];
 714             jnrlistE         = jjnr[jidx+4];
 715             jnrlistF         = jjnr[jidx+5];
 716             jnrlistG         = jjnr[jidx+6];
 717             jnrlistH         = jjnr[jidx+7];
 718             /* Sign of each element will be negative for non-real atoms.
 719              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 720              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 721              */
 722             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 723                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 724
 725             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 726             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 727             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 728             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 729             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 730             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 731             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 732             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 733             j_coord_offsetA  = DIM*jnrA;
 734             j_coord_offsetB  = DIM*jnrB;
 735             j_coord_offsetC  = DIM*jnrC;
 736             j_coord_offsetD  = DIM*jnrD;
 737             j_coord_offsetE  = DIM*jnrE;
 738             j_coord_offsetF  = DIM*jnrF;
 739             j_coord_offsetG  = DIM*jnrG;
 740             j_coord_offsetH  = DIM*jnrH;
 741
 742             /* load j atom coordinates */
 743             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 744                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 745                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 746                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 747                                                  &jx0,&jy0,&jz0);
 748
 749             /* Calculate displacement vector */
 750             dx00             = _mm256_sub_ps(ix0,jx0);
 751             dy00             = _mm256_sub_ps(iy0,jy0);
 752             dz00             = _mm256_sub_ps(iz0,jz0);
 753
 754             /* Calculate squared distance and things based on it */
 755             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 756
 757             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 758
 759             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 760
 761             /* Load parameters for j particles */
 762             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 763                                                                  charge+jnrC+0,charge+jnrD+0,
 764                                                                  charge+jnrE+0,charge+jnrF+0,
 765                                                                  charge+jnrG+0,charge+jnrH+0);
 766             vdwjidx0A        = 2*vdwtype[jnrA+0];
 767             vdwjidx0B        = 2*vdwtype[jnrB+0];
 768             vdwjidx0C        = 2*vdwtype[jnrC+0];
 769             vdwjidx0D        = 2*vdwtype[jnrD+0];
 770             vdwjidx0E        = 2*vdwtype[jnrE+0];
 771             vdwjidx0F        = 2*vdwtype[jnrF+0];
 772             vdwjidx0G        = 2*vdwtype[jnrG+0];
 773             vdwjidx0H        = 2*vdwtype[jnrH+0];
 774
 775             /**************************
 776              * CALCULATE INTERACTIONS *
 777              **************************/
 778
 779             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 780             {
 781
 782             /* Compute parameters for interactions between i and j atoms */
 783             qq00             = _mm256_mul_ps(iq0,jq0);
 784             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 785                                             vdwioffsetptr0+vdwjidx0B,
 786                                             vdwioffsetptr0+vdwjidx0C,
 787                                             vdwioffsetptr0+vdwjidx0D,
 788                                             vdwioffsetptr0+vdwjidx0E,
 789                                             vdwioffsetptr0+vdwjidx0F,
 790                                             vdwioffsetptr0+vdwjidx0G,
 791                                             vdwioffsetptr0+vdwjidx0H,
 792                                             &c6_00,&c12_00);
 793
 794             /* REACTION-FIELD ELECTROSTATICS */
 795             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 796
 797             /* LENNARD-JONES DISPERSION/REPULSION */
 798
 799             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 800             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 801
 802             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 803
 804             fscal            = _mm256_add_ps(felec,fvdw);
 805
 806             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 807
 808             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 809
 810             /* Calculate temporary vectorial force */
 811             tx               = _mm256_mul_ps(fscal,dx00);
 812             ty               = _mm256_mul_ps(fscal,dy00);
 813             tz               = _mm256_mul_ps(fscal,dz00);
 814
 815             /* Update vectorial force */
 816             fix0             = _mm256_add_ps(fix0,tx);
 817             fiy0             = _mm256_add_ps(fiy0,ty);
 818             fiz0             = _mm256_add_ps(fiz0,tz);
 819
 820             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 821             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 822             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 823             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 824             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 825             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 826             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 827             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 828             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 829
 830             }
 831
 832             /* Inner loop uses 37 flops */
 833         }
 834
 835         /* End of innermost loop */
 836
 837         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 838                                                  f+i_coord_offset,fshift+i_shift_offset);
 839
 840         /* Increment number of inner iterations */
 841         inneriter                  += j_index_end - j_index_start;
 842
 843         /* Outer loop uses 7 flops */
 844     }
 845
 846     /* Increment number of outer iterations */
 847     outeriter        += nri;
 848
 849     /* Update outer/inner flops */
 850
 851     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 852 }