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) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_avx_256_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  88     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  97     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  98     __m256           dummy_mask,cutoff_mask;
  99     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 100     __m256           one     = _mm256_set1_ps(1.0);
 101     __m256           two     = _mm256_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm256_set1_ps(fr->ic->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = _mm256_set1_ps(fr->ic->k_rf);
 116     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 117     crf              = _mm256_set1_ps(fr->ic->c_rf);
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->ic->rcoulomb;
 124     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 125     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 126
 127     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 128     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136     j_coord_offsetE = 0;
 137     j_coord_offsetF = 0;
 138     j_coord_offsetG = 0;
 139     j_coord_offsetH = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 165
 166         fix0             = _mm256_setzero_ps();
 167         fiy0             = _mm256_setzero_ps();
 168         fiz0             = _mm256_setzero_ps();
 169
 170         /* Load parameters for i particles */
 171         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 172         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm256_setzero_ps();
 176         vvdwsum          = _mm256_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             jnrE             = jjnr[jidx+4];
 188             jnrF             = jjnr[jidx+5];
 189             jnrG             = jjnr[jidx+6];
 190             jnrH             = jjnr[jidx+7];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195             j_coord_offsetE  = DIM*jnrE;
 196             j_coord_offsetF  = DIM*jnrF;
 197             j_coord_offsetG  = DIM*jnrG;
 198             j_coord_offsetH  = DIM*jnrH;
 199
 200             /* load j atom coordinates */
 201             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 203                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 204                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 205                                                  &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm256_sub_ps(ix0,jx0);
 209             dy00             = _mm256_sub_ps(iy0,jy0);
 210             dz00             = _mm256_sub_ps(iz0,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 214
 215             rinv00           = avx256_invsqrt_f(rsq00);
 216
 217             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 221                                                                  charge+jnrC+0,charge+jnrD+0,
 222                                                                  charge+jnrE+0,charge+jnrF+0,
 223                                                                  charge+jnrG+0,charge+jnrH+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228             vdwjidx0E        = 2*vdwtype[jnrE+0];
 229             vdwjidx0F        = 2*vdwtype[jnrF+0];
 230             vdwjidx0G        = 2*vdwtype[jnrG+0];
 231             vdwjidx0H        = 2*vdwtype[jnrH+0];
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 238             {
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             qq00             = _mm256_mul_ps(iq0,jq0);
 242             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 243                                             vdwioffsetptr0+vdwjidx0B,
 244                                             vdwioffsetptr0+vdwjidx0C,
 245                                             vdwioffsetptr0+vdwjidx0D,
 246                                             vdwioffsetptr0+vdwjidx0E,
 247                                             vdwioffsetptr0+vdwjidx0F,
 248                                             vdwioffsetptr0+vdwjidx0G,
 249                                             vdwioffsetptr0+vdwjidx0H,
 250                                             &c6_00,&c12_00);
 251
 252             /* REACTION-FIELD ELECTROSTATICS */
 253             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 254             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 255
 256             /* LENNARD-JONES DISPERSION/REPULSION */
 257
 258             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 259             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 260             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 261             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) ,
 262                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 263             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 264
 265             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             velec            = _mm256_and_ps(velec,cutoff_mask);
 269             velecsum         = _mm256_add_ps(velecsum,velec);
 270             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 271             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 272
 273             fscal            = _mm256_add_ps(felec,fvdw);
 274
 275             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 276
 277             /* Calculate temporary vectorial force */
 278             tx               = _mm256_mul_ps(fscal,dx00);
 279             ty               = _mm256_mul_ps(fscal,dy00);
 280             tz               = _mm256_mul_ps(fscal,dz00);
 281
 282             /* Update vectorial force */
 283             fix0             = _mm256_add_ps(fix0,tx);
 284             fiy0             = _mm256_add_ps(fiy0,ty);
 285             fiz0             = _mm256_add_ps(fiz0,tz);
 286
 287             fjptrA             = f+j_coord_offsetA;
 288             fjptrB             = f+j_coord_offsetB;
 289             fjptrC             = f+j_coord_offsetC;
 290             fjptrD             = f+j_coord_offsetD;
 291             fjptrE             = f+j_coord_offsetE;
 292             fjptrF             = f+j_coord_offsetF;
 293             fjptrG             = f+j_coord_offsetG;
 294             fjptrH             = f+j_coord_offsetH;
 295             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 296
 297             }
 298
 299             /* Inner loop uses 54 flops */
 300         }
 301
 302         if(jidx<j_index_end)
 303         {
 304
 305             /* Get j neighbor index, and coordinate index */
 306             jnrlistA         = jjnr[jidx];
 307             jnrlistB         = jjnr[jidx+1];
 308             jnrlistC         = jjnr[jidx+2];
 309             jnrlistD         = jjnr[jidx+3];
 310             jnrlistE         = jjnr[jidx+4];
 311             jnrlistF         = jjnr[jidx+5];
 312             jnrlistG         = jjnr[jidx+6];
 313             jnrlistH         = jjnr[jidx+7];
 314             /* Sign of each element will be negative for non-real atoms.
 315              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 316              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 317              */
 318             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 319                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 320
 321             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 322             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 323             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 324             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 325             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 326             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 327             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 328             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 329             j_coord_offsetA  = DIM*jnrA;
 330             j_coord_offsetB  = DIM*jnrB;
 331             j_coord_offsetC  = DIM*jnrC;
 332             j_coord_offsetD  = DIM*jnrD;
 333             j_coord_offsetE  = DIM*jnrE;
 334             j_coord_offsetF  = DIM*jnrF;
 335             j_coord_offsetG  = DIM*jnrG;
 336             j_coord_offsetH  = DIM*jnrH;
 337
 338             /* load j atom coordinates */
 339             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 340                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 341                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 342                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 343                                                  &jx0,&jy0,&jz0);
 344
 345             /* Calculate displacement vector */
 346             dx00             = _mm256_sub_ps(ix0,jx0);
 347             dy00             = _mm256_sub_ps(iy0,jy0);
 348             dz00             = _mm256_sub_ps(iz0,jz0);
 349
 350             /* Calculate squared distance and things based on it */
 351             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 352
 353             rinv00           = avx256_invsqrt_f(rsq00);
 354
 355             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 356
 357             /* Load parameters for j particles */
 358             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 359                                                                  charge+jnrC+0,charge+jnrD+0,
 360                                                                  charge+jnrE+0,charge+jnrF+0,
 361                                                                  charge+jnrG+0,charge+jnrH+0);
 362             vdwjidx0A        = 2*vdwtype[jnrA+0];
 363             vdwjidx0B        = 2*vdwtype[jnrB+0];
 364             vdwjidx0C        = 2*vdwtype[jnrC+0];
 365             vdwjidx0D        = 2*vdwtype[jnrD+0];
 366             vdwjidx0E        = 2*vdwtype[jnrE+0];
 367             vdwjidx0F        = 2*vdwtype[jnrF+0];
 368             vdwjidx0G        = 2*vdwtype[jnrG+0];
 369             vdwjidx0H        = 2*vdwtype[jnrH+0];
 370
 371             /**************************
 372              * CALCULATE INTERACTIONS *
 373              **************************/
 374
 375             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 376             {
 377
 378             /* Compute parameters for interactions between i and j atoms */
 379             qq00             = _mm256_mul_ps(iq0,jq0);
 380             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 381                                             vdwioffsetptr0+vdwjidx0B,
 382                                             vdwioffsetptr0+vdwjidx0C,
 383                                             vdwioffsetptr0+vdwjidx0D,
 384                                             vdwioffsetptr0+vdwjidx0E,
 385                                             vdwioffsetptr0+vdwjidx0F,
 386                                             vdwioffsetptr0+vdwjidx0G,
 387                                             vdwioffsetptr0+vdwjidx0H,
 388                                             &c6_00,&c12_00);
 389
 390             /* REACTION-FIELD ELECTROSTATICS */
 391             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 392             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 393
 394             /* LENNARD-JONES DISPERSION/REPULSION */
 395
 396             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 397             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 398             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 399             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) ,
 400                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 401             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 402
 403             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 404
 405             /* Update potential sum for this i atom from the interaction with this j atom. */
 406             velec            = _mm256_and_ps(velec,cutoff_mask);
 407             velec            = _mm256_andnot_ps(dummy_mask,velec);
 408             velecsum         = _mm256_add_ps(velecsum,velec);
 409             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 410             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 411             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 412
 413             fscal            = _mm256_add_ps(felec,fvdw);
 414
 415             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 416
 417             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 418
 419             /* Calculate temporary vectorial force */
 420             tx               = _mm256_mul_ps(fscal,dx00);
 421             ty               = _mm256_mul_ps(fscal,dy00);
 422             tz               = _mm256_mul_ps(fscal,dz00);
 423
 424             /* Update vectorial force */
 425             fix0             = _mm256_add_ps(fix0,tx);
 426             fiy0             = _mm256_add_ps(fiy0,ty);
 427             fiz0             = _mm256_add_ps(fiz0,tz);
 428
 429             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 430             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 431             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 432             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 433             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 434             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 435             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 436             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 437             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 438
 439             }
 440
 441             /* Inner loop uses 54 flops */
 442         }
 443
 444         /* End of innermost loop */
 445
 446         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 447                                                  f+i_coord_offset,fshift+i_shift_offset);
 448
 449         ggid                        = gid[iidx];
 450         /* Update potential energies */
 451         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 452         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 453
 454         /* Increment number of inner iterations */
 455         inneriter                  += j_index_end - j_index_start;
 456
 457         /* Outer loop uses 9 flops */
 458     }
 459
 460     /* Increment number of outer iterations */
 461     outeriter        += nri;
 462
 463     /* Update outer/inner flops */
 464
 465     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
 466 }
 467 /*
 468  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_256_single
 469  * Electrostatics interaction: ReactionField
 470  * VdW interaction:            LennardJones
 471  * Geometry:                   Particle-Particle
 472  * Calculate force/pot:        Force
 473  */
 474 void
 475 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_256_single
 476                     (t_nblist                    * gmx_restrict       nlist,
 477                      rvec                        * gmx_restrict          xx,
 478                      rvec                        * gmx_restrict          ff,
 479                      struct t_forcerec           * gmx_restrict          fr,
 480                      t_mdatoms                   * gmx_restrict     mdatoms,
 481                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 482                      t_nrnb                      * gmx_restrict        nrnb)
 483 {
 484     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 485      * just 0 for non-waters.
 486      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 487      * jnr indices corresponding to data put in the four positions in the SIMD register.
 488      */
 489     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 490     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 491     int              jnrA,jnrB,jnrC,jnrD;
 492     int              jnrE,jnrF,jnrG,jnrH;
 493     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 494     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 495     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 496     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 497     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 498     real             rcutoff_scalar;
 499     real             *shiftvec,*fshift,*x,*f;
 500     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 501     real             scratch[4*DIM];
 502     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 503     real *           vdwioffsetptr0;
 504     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 505     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 506     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 507     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 508     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 509     real             *charge;
 510     int              nvdwtype;
 511     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 512     int              *vdwtype;
 513     real             *vdwparam;
 514     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 515     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 516     __m256           dummy_mask,cutoff_mask;
 517     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 518     __m256           one     = _mm256_set1_ps(1.0);
 519     __m256           two     = _mm256_set1_ps(2.0);
 520     x                = xx[0];
 521     f                = ff[0];
 522
 523     nri              = nlist->nri;
 524     iinr             = nlist->iinr;
 525     jindex           = nlist->jindex;
 526     jjnr             = nlist->jjnr;
 527     shiftidx         = nlist->shift;
 528     gid              = nlist->gid;
 529     shiftvec         = fr->shift_vec[0];
 530     fshift           = fr->fshift[0];
 531     facel            = _mm256_set1_ps(fr->ic->epsfac);
 532     charge           = mdatoms->chargeA;
 533     krf              = _mm256_set1_ps(fr->ic->k_rf);
 534     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 535     crf              = _mm256_set1_ps(fr->ic->c_rf);
 536     nvdwtype         = fr->ntype;
 537     vdwparam         = fr->nbfp;
 538     vdwtype          = mdatoms->typeA;
 539
 540     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 541     rcutoff_scalar   = fr->ic->rcoulomb;
 542     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 543     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 544
 545     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 546     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 547
 548     /* Avoid stupid compiler warnings */
 549     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 550     j_coord_offsetA = 0;
 551     j_coord_offsetB = 0;
 552     j_coord_offsetC = 0;
 553     j_coord_offsetD = 0;
 554     j_coord_offsetE = 0;
 555     j_coord_offsetF = 0;
 556     j_coord_offsetG = 0;
 557     j_coord_offsetH = 0;
 558
 559     outeriter        = 0;
 560     inneriter        = 0;
 561
 562     for(iidx=0;iidx<4*DIM;iidx++)
 563     {
 564         scratch[iidx] = 0.0;
 565     }
 566
 567     /* Start outer loop over neighborlists */
 568     for(iidx=0; iidx<nri; iidx++)
 569     {
 570         /* Load shift vector for this list */
 571         i_shift_offset   = DIM*shiftidx[iidx];
 572
 573         /* Load limits for loop over neighbors */
 574         j_index_start    = jindex[iidx];
 575         j_index_end      = jindex[iidx+1];
 576
 577         /* Get outer coordinate index */
 578         inr              = iinr[iidx];
 579         i_coord_offset   = DIM*inr;
 580
 581         /* Load i particle coords and add shift vector */
 582         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 583
 584         fix0             = _mm256_setzero_ps();
 585         fiy0             = _mm256_setzero_ps();
 586         fiz0             = _mm256_setzero_ps();
 587
 588         /* Load parameters for i particles */
 589         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 590         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 591
 592         /* Start inner kernel loop */
 593         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 594         {
 595
 596             /* Get j neighbor index, and coordinate index */
 597             jnrA             = jjnr[jidx];
 598             jnrB             = jjnr[jidx+1];
 599             jnrC             = jjnr[jidx+2];
 600             jnrD             = jjnr[jidx+3];
 601             jnrE             = jjnr[jidx+4];
 602             jnrF             = jjnr[jidx+5];
 603             jnrG             = jjnr[jidx+6];
 604             jnrH             = jjnr[jidx+7];
 605             j_coord_offsetA  = DIM*jnrA;
 606             j_coord_offsetB  = DIM*jnrB;
 607             j_coord_offsetC  = DIM*jnrC;
 608             j_coord_offsetD  = DIM*jnrD;
 609             j_coord_offsetE  = DIM*jnrE;
 610             j_coord_offsetF  = DIM*jnrF;
 611             j_coord_offsetG  = DIM*jnrG;
 612             j_coord_offsetH  = DIM*jnrH;
 613
 614             /* load j atom coordinates */
 615             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 616                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 617                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 618                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 619                                                  &jx0,&jy0,&jz0);
 620
 621             /* Calculate displacement vector */
 622             dx00             = _mm256_sub_ps(ix0,jx0);
 623             dy00             = _mm256_sub_ps(iy0,jy0);
 624             dz00             = _mm256_sub_ps(iz0,jz0);
 625
 626             /* Calculate squared distance and things based on it */
 627             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 628
 629             rinv00           = avx256_invsqrt_f(rsq00);
 630
 631             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 632
 633             /* Load parameters for j particles */
 634             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 635                                                                  charge+jnrC+0,charge+jnrD+0,
 636                                                                  charge+jnrE+0,charge+jnrF+0,
 637                                                                  charge+jnrG+0,charge+jnrH+0);
 638             vdwjidx0A        = 2*vdwtype[jnrA+0];
 639             vdwjidx0B        = 2*vdwtype[jnrB+0];
 640             vdwjidx0C        = 2*vdwtype[jnrC+0];
 641             vdwjidx0D        = 2*vdwtype[jnrD+0];
 642             vdwjidx0E        = 2*vdwtype[jnrE+0];
 643             vdwjidx0F        = 2*vdwtype[jnrF+0];
 644             vdwjidx0G        = 2*vdwtype[jnrG+0];
 645             vdwjidx0H        = 2*vdwtype[jnrH+0];
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 652             {
 653
 654             /* Compute parameters for interactions between i and j atoms */
 655             qq00             = _mm256_mul_ps(iq0,jq0);
 656             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 657                                             vdwioffsetptr0+vdwjidx0B,
 658                                             vdwioffsetptr0+vdwjidx0C,
 659                                             vdwioffsetptr0+vdwjidx0D,
 660                                             vdwioffsetptr0+vdwjidx0E,
 661                                             vdwioffsetptr0+vdwjidx0F,
 662                                             vdwioffsetptr0+vdwjidx0G,
 663                                             vdwioffsetptr0+vdwjidx0H,
 664                                             &c6_00,&c12_00);
 665
 666             /* REACTION-FIELD ELECTROSTATICS */
 667             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 668
 669             /* LENNARD-JONES DISPERSION/REPULSION */
 670
 671             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 672             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 673
 674             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 675
 676             fscal            = _mm256_add_ps(felec,fvdw);
 677
 678             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 679
 680             /* Calculate temporary vectorial force */
 681             tx               = _mm256_mul_ps(fscal,dx00);
 682             ty               = _mm256_mul_ps(fscal,dy00);
 683             tz               = _mm256_mul_ps(fscal,dz00);
 684
 685             /* Update vectorial force */
 686             fix0             = _mm256_add_ps(fix0,tx);
 687             fiy0             = _mm256_add_ps(fiy0,ty);
 688             fiz0             = _mm256_add_ps(fiz0,tz);
 689
 690             fjptrA             = f+j_coord_offsetA;
 691             fjptrB             = f+j_coord_offsetB;
 692             fjptrC             = f+j_coord_offsetC;
 693             fjptrD             = f+j_coord_offsetD;
 694             fjptrE             = f+j_coord_offsetE;
 695             fjptrF             = f+j_coord_offsetF;
 696             fjptrG             = f+j_coord_offsetG;
 697             fjptrH             = f+j_coord_offsetH;
 698             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 699
 700             }
 701
 702             /* Inner loop uses 37 flops */
 703         }
 704
 705         if(jidx<j_index_end)
 706         {
 707
 708             /* Get j neighbor index, and coordinate index */
 709             jnrlistA         = jjnr[jidx];
 710             jnrlistB         = jjnr[jidx+1];
 711             jnrlistC         = jjnr[jidx+2];
 712             jnrlistD         = jjnr[jidx+3];
 713             jnrlistE         = jjnr[jidx+4];
 714             jnrlistF         = jjnr[jidx+5];
 715             jnrlistG         = jjnr[jidx+6];
 716             jnrlistH         = jjnr[jidx+7];
 717             /* Sign of each element will be negative for non-real atoms.
 718              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 719              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 720              */
 721             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 722                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 723
 724             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 725             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 726             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 727             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 728             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 729             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 730             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 731             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 732             j_coord_offsetA  = DIM*jnrA;
 733             j_coord_offsetB  = DIM*jnrB;
 734             j_coord_offsetC  = DIM*jnrC;
 735             j_coord_offsetD  = DIM*jnrD;
 736             j_coord_offsetE  = DIM*jnrE;
 737             j_coord_offsetF  = DIM*jnrF;
 738             j_coord_offsetG  = DIM*jnrG;
 739             j_coord_offsetH  = DIM*jnrH;
 740
 741             /* load j atom coordinates */
 742             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 743                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 744                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 745                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 746                                                  &jx0,&jy0,&jz0);
 747
 748             /* Calculate displacement vector */
 749             dx00             = _mm256_sub_ps(ix0,jx0);
 750             dy00             = _mm256_sub_ps(iy0,jy0);
 751             dz00             = _mm256_sub_ps(iz0,jz0);
 752
 753             /* Calculate squared distance and things based on it */
 754             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 755
 756             rinv00           = avx256_invsqrt_f(rsq00);
 757
 758             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 759
 760             /* Load parameters for j particles */
 761             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 762                                                                  charge+jnrC+0,charge+jnrD+0,
 763                                                                  charge+jnrE+0,charge+jnrF+0,
 764                                                                  charge+jnrG+0,charge+jnrH+0);
 765             vdwjidx0A        = 2*vdwtype[jnrA+0];
 766             vdwjidx0B        = 2*vdwtype[jnrB+0];
 767             vdwjidx0C        = 2*vdwtype[jnrC+0];
 768             vdwjidx0D        = 2*vdwtype[jnrD+0];
 769             vdwjidx0E        = 2*vdwtype[jnrE+0];
 770             vdwjidx0F        = 2*vdwtype[jnrF+0];
 771             vdwjidx0G        = 2*vdwtype[jnrG+0];
 772             vdwjidx0H        = 2*vdwtype[jnrH+0];
 773
 774             /**************************
 775              * CALCULATE INTERACTIONS *
 776              **************************/
 777
 778             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 779             {
 780
 781             /* Compute parameters for interactions between i and j atoms */
 782             qq00             = _mm256_mul_ps(iq0,jq0);
 783             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 784                                             vdwioffsetptr0+vdwjidx0B,
 785                                             vdwioffsetptr0+vdwjidx0C,
 786                                             vdwioffsetptr0+vdwjidx0D,
 787                                             vdwioffsetptr0+vdwjidx0E,
 788                                             vdwioffsetptr0+vdwjidx0F,
 789                                             vdwioffsetptr0+vdwjidx0G,
 790                                             vdwioffsetptr0+vdwjidx0H,
 791                                             &c6_00,&c12_00);
 792
 793             /* REACTION-FIELD ELECTROSTATICS */
 794             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 795
 796             /* LENNARD-JONES DISPERSION/REPULSION */
 797
 798             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 799             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 800
 801             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 802
 803             fscal            = _mm256_add_ps(felec,fvdw);
 804
 805             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 806
 807             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 808
 809             /* Calculate temporary vectorial force */
 810             tx               = _mm256_mul_ps(fscal,dx00);
 811             ty               = _mm256_mul_ps(fscal,dy00);
 812             tz               = _mm256_mul_ps(fscal,dz00);
 813
 814             /* Update vectorial force */
 815             fix0             = _mm256_add_ps(fix0,tx);
 816             fiy0             = _mm256_add_ps(fiy0,ty);
 817             fiz0             = _mm256_add_ps(fiz0,tz);
 818
 819             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 820             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 821             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 822             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 823             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 824             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 825             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 826             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 827             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 828
 829             }
 830
 831             /* Inner loop uses 37 flops */
 832         }
 833
 834         /* End of innermost loop */
 835
 836         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 837                                                  f+i_coord_offset,fshift+i_shift_offset);
 838
 839         /* Increment number of inner iterations */
 840         inneriter                  += j_index_end - j_index_start;
 841
 842         /* Outer loop uses 7 flops */
 843     }
 844
 845     /* Increment number of outer iterations */
 846     outeriter        += nri;
 847
 848     /* Update outer/inner flops */
 849
 850     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 851 }