src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCoul_VdwLJ_GeomP1P1_avx_256_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_avx_256_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  75     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  84     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  85     __m256           dummy_mask,cutoff_mask;
  86     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  87     __m256           one     = _mm256_set1_ps(1.0);
  88     __m256           two     = _mm256_set1_ps(2.0);
  89     x                = xx[0];
  90     f                = ff[0];
  91
  92     nri              = nlist->nri;
  93     iinr             = nlist->iinr;
  94     jindex           = nlist->jindex;
  95     jjnr             = nlist->jjnr;
  96     shiftidx         = nlist->shift;
  97     gid              = nlist->gid;
  98     shiftvec         = fr->shift_vec[0];
  99     fshift           = fr->fshift[0];
 100     facel            = _mm256_set1_ps(fr->epsfac);
 101     charge           = mdatoms->chargeA;
 102     nvdwtype         = fr->ntype;
 103     vdwparam         = fr->nbfp;
 104     vdwtype          = mdatoms->typeA;
 105
 106     /* Avoid stupid compiler warnings */
 107     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 108     j_coord_offsetA = 0;
 109     j_coord_offsetB = 0;
 110     j_coord_offsetC = 0;
 111     j_coord_offsetD = 0;
 112     j_coord_offsetE = 0;
 113     j_coord_offsetF = 0;
 114     j_coord_offsetG = 0;
 115     j_coord_offsetH = 0;
 116
 117     outeriter        = 0;
 118     inneriter        = 0;
 119
 120     for(iidx=0;iidx<4*DIM;iidx++)
 121     {
 122         scratch[iidx] = 0.0;
 123     }
 124
 125     /* Start outer loop over neighborlists */
 126     for(iidx=0; iidx<nri; iidx++)
 127     {
 128         /* Load shift vector for this list */
 129         i_shift_offset   = DIM*shiftidx[iidx];
 130
 131         /* Load limits for loop over neighbors */
 132         j_index_start    = jindex[iidx];
 133         j_index_end      = jindex[iidx+1];
 134
 135         /* Get outer coordinate index */
 136         inr              = iinr[iidx];
 137         i_coord_offset   = DIM*inr;
 138
 139         /* Load i particle coords and add shift vector */
 140         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 141
 142         fix0             = _mm256_setzero_ps();
 143         fiy0             = _mm256_setzero_ps();
 144         fiz0             = _mm256_setzero_ps();
 145
 146         /* Load parameters for i particles */
 147         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 148         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 149
 150         /* Reset potential sums */
 151         velecsum         = _mm256_setzero_ps();
 152         vvdwsum          = _mm256_setzero_ps();
 153
 154         /* Start inner kernel loop */
 155         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 156         {
 157
 158             /* Get j neighbor index, and coordinate index */
 159             jnrA             = jjnr[jidx];
 160             jnrB             = jjnr[jidx+1];
 161             jnrC             = jjnr[jidx+2];
 162             jnrD             = jjnr[jidx+3];
 163             jnrE             = jjnr[jidx+4];
 164             jnrF             = jjnr[jidx+5];
 165             jnrG             = jjnr[jidx+6];
 166             jnrH             = jjnr[jidx+7];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169             j_coord_offsetC  = DIM*jnrC;
 170             j_coord_offsetD  = DIM*jnrD;
 171             j_coord_offsetE  = DIM*jnrE;
 172             j_coord_offsetF  = DIM*jnrF;
 173             j_coord_offsetG  = DIM*jnrG;
 174             j_coord_offsetH  = DIM*jnrH;
 175
 176             /* load j atom coordinates */
 177             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 179                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 180                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 181                                                  &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _mm256_sub_ps(ix0,jx0);
 185             dy00             = _mm256_sub_ps(iy0,jy0);
 186             dz00             = _mm256_sub_ps(iz0,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 190
 191             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 192
 193             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 194
 195             /* Load parameters for j particles */
 196             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 197                                                                  charge+jnrC+0,charge+jnrD+0,
 198                                                                  charge+jnrE+0,charge+jnrF+0,
 199                                                                  charge+jnrG+0,charge+jnrH+0);
 200             vdwjidx0A        = 2*vdwtype[jnrA+0];
 201             vdwjidx0B        = 2*vdwtype[jnrB+0];
 202             vdwjidx0C        = 2*vdwtype[jnrC+0];
 203             vdwjidx0D        = 2*vdwtype[jnrD+0];
 204             vdwjidx0E        = 2*vdwtype[jnrE+0];
 205             vdwjidx0F        = 2*vdwtype[jnrF+0];
 206             vdwjidx0G        = 2*vdwtype[jnrG+0];
 207             vdwjidx0H        = 2*vdwtype[jnrH+0];
 208
 209             /**************************
 210              * CALCULATE INTERACTIONS *
 211              **************************/
 212
 213             /* Compute parameters for interactions between i and j atoms */
 214             qq00             = _mm256_mul_ps(iq0,jq0);
 215             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 216                                             vdwioffsetptr0+vdwjidx0B,
 217                                             vdwioffsetptr0+vdwjidx0C,
 218                                             vdwioffsetptr0+vdwjidx0D,
 219                                             vdwioffsetptr0+vdwjidx0E,
 220                                             vdwioffsetptr0+vdwjidx0F,
 221                                             vdwioffsetptr0+vdwjidx0G,
 222                                             vdwioffsetptr0+vdwjidx0H,
 223                                             &c6_00,&c12_00);
 224
 225             /* COULOMB ELECTROSTATICS */
 226             velec            = _mm256_mul_ps(qq00,rinv00);
 227             felec            = _mm256_mul_ps(velec,rinvsq00);
 228
 229             /* LENNARD-JONES DISPERSION/REPULSION */
 230
 231             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 232             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 233             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 234             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 235             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 236
 237             /* Update potential sum for this i atom from the interaction with this j atom. */
 238             velecsum         = _mm256_add_ps(velecsum,velec);
 239             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 240
 241             fscal            = _mm256_add_ps(felec,fvdw);
 242
 243             /* Calculate temporary vectorial force */
 244             tx               = _mm256_mul_ps(fscal,dx00);
 245             ty               = _mm256_mul_ps(fscal,dy00);
 246             tz               = _mm256_mul_ps(fscal,dz00);
 247
 248             /* Update vectorial force */
 249             fix0             = _mm256_add_ps(fix0,tx);
 250             fiy0             = _mm256_add_ps(fiy0,ty);
 251             fiz0             = _mm256_add_ps(fiz0,tz);
 252
 253             fjptrA             = f+j_coord_offsetA;
 254             fjptrB             = f+j_coord_offsetB;
 255             fjptrC             = f+j_coord_offsetC;
 256             fjptrD             = f+j_coord_offsetD;
 257             fjptrE             = f+j_coord_offsetE;
 258             fjptrF             = f+j_coord_offsetF;
 259             fjptrG             = f+j_coord_offsetG;
 260             fjptrH             = f+j_coord_offsetH;
 261             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 262
 263             /* Inner loop uses 39 flops */
 264         }
 265
 266         if(jidx<j_index_end)
 267         {
 268
 269             /* Get j neighbor index, and coordinate index */
 270             jnrlistA         = jjnr[jidx];
 271             jnrlistB         = jjnr[jidx+1];
 272             jnrlistC         = jjnr[jidx+2];
 273             jnrlistD         = jjnr[jidx+3];
 274             jnrlistE         = jjnr[jidx+4];
 275             jnrlistF         = jjnr[jidx+5];
 276             jnrlistG         = jjnr[jidx+6];
 277             jnrlistH         = jjnr[jidx+7];
 278             /* Sign of each element will be negative for non-real atoms.
 279              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 280              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 281              */
 282             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 283                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 284
 285             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 286             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 287             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 288             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 289             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 290             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 291             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 292             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 293             j_coord_offsetA  = DIM*jnrA;
 294             j_coord_offsetB  = DIM*jnrB;
 295             j_coord_offsetC  = DIM*jnrC;
 296             j_coord_offsetD  = DIM*jnrD;
 297             j_coord_offsetE  = DIM*jnrE;
 298             j_coord_offsetF  = DIM*jnrF;
 299             j_coord_offsetG  = DIM*jnrG;
 300             j_coord_offsetH  = DIM*jnrH;
 301
 302             /* load j atom coordinates */
 303             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 304                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 305                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 306                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 307                                                  &jx0,&jy0,&jz0);
 308
 309             /* Calculate displacement vector */
 310             dx00             = _mm256_sub_ps(ix0,jx0);
 311             dy00             = _mm256_sub_ps(iy0,jy0);
 312             dz00             = _mm256_sub_ps(iz0,jz0);
 313
 314             /* Calculate squared distance and things based on it */
 315             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 316
 317             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 318
 319             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 320
 321             /* Load parameters for j particles */
 322             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 323                                                                  charge+jnrC+0,charge+jnrD+0,
 324                                                                  charge+jnrE+0,charge+jnrF+0,
 325                                                                  charge+jnrG+0,charge+jnrH+0);
 326             vdwjidx0A        = 2*vdwtype[jnrA+0];
 327             vdwjidx0B        = 2*vdwtype[jnrB+0];
 328             vdwjidx0C        = 2*vdwtype[jnrC+0];
 329             vdwjidx0D        = 2*vdwtype[jnrD+0];
 330             vdwjidx0E        = 2*vdwtype[jnrE+0];
 331             vdwjidx0F        = 2*vdwtype[jnrF+0];
 332             vdwjidx0G        = 2*vdwtype[jnrG+0];
 333             vdwjidx0H        = 2*vdwtype[jnrH+0];
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq00             = _mm256_mul_ps(iq0,jq0);
 341             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 342                                             vdwioffsetptr0+vdwjidx0B,
 343                                             vdwioffsetptr0+vdwjidx0C,
 344                                             vdwioffsetptr0+vdwjidx0D,
 345                                             vdwioffsetptr0+vdwjidx0E,
 346                                             vdwioffsetptr0+vdwjidx0F,
 347                                             vdwioffsetptr0+vdwjidx0G,
 348                                             vdwioffsetptr0+vdwjidx0H,
 349                                             &c6_00,&c12_00);
 350
 351             /* COULOMB ELECTROSTATICS */
 352             velec            = _mm256_mul_ps(qq00,rinv00);
 353             felec            = _mm256_mul_ps(velec,rinvsq00);
 354
 355             /* LENNARD-JONES DISPERSION/REPULSION */
 356
 357             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 358             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 359             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 360             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 361             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velec            = _mm256_andnot_ps(dummy_mask,velec);
 365             velecsum         = _mm256_add_ps(velecsum,velec);
 366             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 367             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 368
 369             fscal            = _mm256_add_ps(felec,fvdw);
 370
 371             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 372
 373             /* Calculate temporary vectorial force */
 374             tx               = _mm256_mul_ps(fscal,dx00);
 375             ty               = _mm256_mul_ps(fscal,dy00);
 376             tz               = _mm256_mul_ps(fscal,dz00);
 377
 378             /* Update vectorial force */
 379             fix0             = _mm256_add_ps(fix0,tx);
 380             fiy0             = _mm256_add_ps(fiy0,ty);
 381             fiz0             = _mm256_add_ps(fiz0,tz);
 382
 383             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 384             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 385             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 386             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 387             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 388             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 389             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 390             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 391             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 392
 393             /* Inner loop uses 39 flops */
 394         }
 395
 396         /* End of innermost loop */
 397
 398         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 399                                                  f+i_coord_offset,fshift+i_shift_offset);
 400
 401         ggid                        = gid[iidx];
 402         /* Update potential energies */
 403         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 404         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 405
 406         /* Increment number of inner iterations */
 407         inneriter                  += j_index_end - j_index_start;
 408
 409         /* Outer loop uses 9 flops */
 410     }
 411
 412     /* Increment number of outer iterations */
 413     outeriter        += nri;
 414
 415     /* Update outer/inner flops */
 416
 417     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*39);
 418 }
 419 /*
 420  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_256_single
 421  * Electrostatics interaction: Coulomb
 422  * VdW interaction:            LennardJones
 423  * Geometry:                   Particle-Particle
 424  * Calculate force/pot:        Force
 425  */
 426 void
 427 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_256_single
 428                     (t_nblist * gmx_restrict                nlist,
 429                      rvec * gmx_restrict                    xx,
 430                      rvec * gmx_restrict                    ff,
 431                      t_forcerec * gmx_restrict              fr,
 432                      t_mdatoms * gmx_restrict               mdatoms,
 433                      nb_kernel_data_t * gmx_restrict        kernel_data,
 434                      t_nrnb * gmx_restrict                  nrnb)
 435 {
 436     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 437      * just 0 for non-waters.
 438      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 439      * jnr indices corresponding to data put in the four positions in the SIMD register.
 440      */
 441     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 442     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 443     int              jnrA,jnrB,jnrC,jnrD;
 444     int              jnrE,jnrF,jnrG,jnrH;
 445     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 446     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 447     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 448     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 449     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 450     real             rcutoff_scalar;
 451     real             *shiftvec,*fshift,*x,*f;
 452     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 453     real             scratch[4*DIM];
 454     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 455     real *           vdwioffsetptr0;
 456     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 457     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 458     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 459     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 460     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 461     real             *charge;
 462     int              nvdwtype;
 463     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 464     int              *vdwtype;
 465     real             *vdwparam;
 466     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 467     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 468     __m256           dummy_mask,cutoff_mask;
 469     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 470     __m256           one     = _mm256_set1_ps(1.0);
 471     __m256           two     = _mm256_set1_ps(2.0);
 472     x                = xx[0];
 473     f                = ff[0];
 474
 475     nri              = nlist->nri;
 476     iinr             = nlist->iinr;
 477     jindex           = nlist->jindex;
 478     jjnr             = nlist->jjnr;
 479     shiftidx         = nlist->shift;
 480     gid              = nlist->gid;
 481     shiftvec         = fr->shift_vec[0];
 482     fshift           = fr->fshift[0];
 483     facel            = _mm256_set1_ps(fr->epsfac);
 484     charge           = mdatoms->chargeA;
 485     nvdwtype         = fr->ntype;
 486     vdwparam         = fr->nbfp;
 487     vdwtype          = mdatoms->typeA;
 488
 489     /* Avoid stupid compiler warnings */
 490     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 491     j_coord_offsetA = 0;
 492     j_coord_offsetB = 0;
 493     j_coord_offsetC = 0;
 494     j_coord_offsetD = 0;
 495     j_coord_offsetE = 0;
 496     j_coord_offsetF = 0;
 497     j_coord_offsetG = 0;
 498     j_coord_offsetH = 0;
 499
 500     outeriter        = 0;
 501     inneriter        = 0;
 502
 503     for(iidx=0;iidx<4*DIM;iidx++)
 504     {
 505         scratch[iidx] = 0.0;
 506     }
 507
 508     /* Start outer loop over neighborlists */
 509     for(iidx=0; iidx<nri; iidx++)
 510     {
 511         /* Load shift vector for this list */
 512         i_shift_offset   = DIM*shiftidx[iidx];
 513
 514         /* Load limits for loop over neighbors */
 515         j_index_start    = jindex[iidx];
 516         j_index_end      = jindex[iidx+1];
 517
 518         /* Get outer coordinate index */
 519         inr              = iinr[iidx];
 520         i_coord_offset   = DIM*inr;
 521
 522         /* Load i particle coords and add shift vector */
 523         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 524
 525         fix0             = _mm256_setzero_ps();
 526         fiy0             = _mm256_setzero_ps();
 527         fiz0             = _mm256_setzero_ps();
 528
 529         /* Load parameters for i particles */
 530         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 531         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 532
 533         /* Start inner kernel loop */
 534         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 535         {
 536
 537             /* Get j neighbor index, and coordinate index */
 538             jnrA             = jjnr[jidx];
 539             jnrB             = jjnr[jidx+1];
 540             jnrC             = jjnr[jidx+2];
 541             jnrD             = jjnr[jidx+3];
 542             jnrE             = jjnr[jidx+4];
 543             jnrF             = jjnr[jidx+5];
 544             jnrG             = jjnr[jidx+6];
 545             jnrH             = jjnr[jidx+7];
 546             j_coord_offsetA  = DIM*jnrA;
 547             j_coord_offsetB  = DIM*jnrB;
 548             j_coord_offsetC  = DIM*jnrC;
 549             j_coord_offsetD  = DIM*jnrD;
 550             j_coord_offsetE  = DIM*jnrE;
 551             j_coord_offsetF  = DIM*jnrF;
 552             j_coord_offsetG  = DIM*jnrG;
 553             j_coord_offsetH  = DIM*jnrH;
 554
 555             /* load j atom coordinates */
 556             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 557                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 558                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 559                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 560                                                  &jx0,&jy0,&jz0);
 561
 562             /* Calculate displacement vector */
 563             dx00             = _mm256_sub_ps(ix0,jx0);
 564             dy00             = _mm256_sub_ps(iy0,jy0);
 565             dz00             = _mm256_sub_ps(iz0,jz0);
 566
 567             /* Calculate squared distance and things based on it */
 568             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 569
 570             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 571
 572             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 573
 574             /* Load parameters for j particles */
 575             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 576                                                                  charge+jnrC+0,charge+jnrD+0,
 577                                                                  charge+jnrE+0,charge+jnrF+0,
 578                                                                  charge+jnrG+0,charge+jnrH+0);
 579             vdwjidx0A        = 2*vdwtype[jnrA+0];
 580             vdwjidx0B        = 2*vdwtype[jnrB+0];
 581             vdwjidx0C        = 2*vdwtype[jnrC+0];
 582             vdwjidx0D        = 2*vdwtype[jnrD+0];
 583             vdwjidx0E        = 2*vdwtype[jnrE+0];
 584             vdwjidx0F        = 2*vdwtype[jnrF+0];
 585             vdwjidx0G        = 2*vdwtype[jnrG+0];
 586             vdwjidx0H        = 2*vdwtype[jnrH+0];
 587
 588             /**************************
 589              * CALCULATE INTERACTIONS *
 590              **************************/
 591
 592             /* Compute parameters for interactions between i and j atoms */
 593             qq00             = _mm256_mul_ps(iq0,jq0);
 594             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 595                                             vdwioffsetptr0+vdwjidx0B,
 596                                             vdwioffsetptr0+vdwjidx0C,
 597                                             vdwioffsetptr0+vdwjidx0D,
 598                                             vdwioffsetptr0+vdwjidx0E,
 599                                             vdwioffsetptr0+vdwjidx0F,
 600                                             vdwioffsetptr0+vdwjidx0G,
 601                                             vdwioffsetptr0+vdwjidx0H,
 602                                             &c6_00,&c12_00);
 603
 604             /* COULOMB ELECTROSTATICS */
 605             velec            = _mm256_mul_ps(qq00,rinv00);
 606             felec            = _mm256_mul_ps(velec,rinvsq00);
 607
 608             /* LENNARD-JONES DISPERSION/REPULSION */
 609
 610             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 611             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 612
 613             fscal            = _mm256_add_ps(felec,fvdw);
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm256_mul_ps(fscal,dx00);
 617             ty               = _mm256_mul_ps(fscal,dy00);
 618             tz               = _mm256_mul_ps(fscal,dz00);
 619
 620             /* Update vectorial force */
 621             fix0             = _mm256_add_ps(fix0,tx);
 622             fiy0             = _mm256_add_ps(fiy0,ty);
 623             fiz0             = _mm256_add_ps(fiz0,tz);
 624
 625             fjptrA             = f+j_coord_offsetA;
 626             fjptrB             = f+j_coord_offsetB;
 627             fjptrC             = f+j_coord_offsetC;
 628             fjptrD             = f+j_coord_offsetD;
 629             fjptrE             = f+j_coord_offsetE;
 630             fjptrF             = f+j_coord_offsetF;
 631             fjptrG             = f+j_coord_offsetG;
 632             fjptrH             = f+j_coord_offsetH;
 633             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 634
 635             /* Inner loop uses 33 flops */
 636         }
 637
 638         if(jidx<j_index_end)
 639         {
 640
 641             /* Get j neighbor index, and coordinate index */
 642             jnrlistA         = jjnr[jidx];
 643             jnrlistB         = jjnr[jidx+1];
 644             jnrlistC         = jjnr[jidx+2];
 645             jnrlistD         = jjnr[jidx+3];
 646             jnrlistE         = jjnr[jidx+4];
 647             jnrlistF         = jjnr[jidx+5];
 648             jnrlistG         = jjnr[jidx+6];
 649             jnrlistH         = jjnr[jidx+7];
 650             /* Sign of each element will be negative for non-real atoms.
 651              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 652              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 653              */
 654             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 655                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 656
 657             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 658             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 659             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 660             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 661             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 662             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 663             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 664             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 665             j_coord_offsetA  = DIM*jnrA;
 666             j_coord_offsetB  = DIM*jnrB;
 667             j_coord_offsetC  = DIM*jnrC;
 668             j_coord_offsetD  = DIM*jnrD;
 669             j_coord_offsetE  = DIM*jnrE;
 670             j_coord_offsetF  = DIM*jnrF;
 671             j_coord_offsetG  = DIM*jnrG;
 672             j_coord_offsetH  = DIM*jnrH;
 673
 674             /* load j atom coordinates */
 675             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 676                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 677                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 678                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 679                                                  &jx0,&jy0,&jz0);
 680
 681             /* Calculate displacement vector */
 682             dx00             = _mm256_sub_ps(ix0,jx0);
 683             dy00             = _mm256_sub_ps(iy0,jy0);
 684             dz00             = _mm256_sub_ps(iz0,jz0);
 685
 686             /* Calculate squared distance and things based on it */
 687             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 688
 689             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 690
 691             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 692
 693             /* Load parameters for j particles */
 694             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 695                                                                  charge+jnrC+0,charge+jnrD+0,
 696                                                                  charge+jnrE+0,charge+jnrF+0,
 697                                                                  charge+jnrG+0,charge+jnrH+0);
 698             vdwjidx0A        = 2*vdwtype[jnrA+0];
 699             vdwjidx0B        = 2*vdwtype[jnrB+0];
 700             vdwjidx0C        = 2*vdwtype[jnrC+0];
 701             vdwjidx0D        = 2*vdwtype[jnrD+0];
 702             vdwjidx0E        = 2*vdwtype[jnrE+0];
 703             vdwjidx0F        = 2*vdwtype[jnrF+0];
 704             vdwjidx0G        = 2*vdwtype[jnrG+0];
 705             vdwjidx0H        = 2*vdwtype[jnrH+0];
 706
 707             /**************************
 708              * CALCULATE INTERACTIONS *
 709              **************************/
 710
 711             /* Compute parameters for interactions between i and j atoms */
 712             qq00             = _mm256_mul_ps(iq0,jq0);
 713             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 714                                             vdwioffsetptr0+vdwjidx0B,
 715                                             vdwioffsetptr0+vdwjidx0C,
 716                                             vdwioffsetptr0+vdwjidx0D,
 717                                             vdwioffsetptr0+vdwjidx0E,
 718                                             vdwioffsetptr0+vdwjidx0F,
 719                                             vdwioffsetptr0+vdwjidx0G,
 720                                             vdwioffsetptr0+vdwjidx0H,
 721                                             &c6_00,&c12_00);
 722
 723             /* COULOMB ELECTROSTATICS */
 724             velec            = _mm256_mul_ps(qq00,rinv00);
 725             felec            = _mm256_mul_ps(velec,rinvsq00);
 726
 727             /* LENNARD-JONES DISPERSION/REPULSION */
 728
 729             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 730             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 731
 732             fscal            = _mm256_add_ps(felec,fvdw);
 733
 734             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 735
 736             /* Calculate temporary vectorial force */
 737             tx               = _mm256_mul_ps(fscal,dx00);
 738             ty               = _mm256_mul_ps(fscal,dy00);
 739             tz               = _mm256_mul_ps(fscal,dz00);
 740
 741             /* Update vectorial force */
 742             fix0             = _mm256_add_ps(fix0,tx);
 743             fiy0             = _mm256_add_ps(fiy0,ty);
 744             fiz0             = _mm256_add_ps(fiz0,tz);
 745
 746             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 747             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 748             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 749             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 750             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 751             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 752             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 753             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 754             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 755
 756             /* Inner loop uses 33 flops */
 757         }
 758
 759         /* End of innermost loop */
 760
 761         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 762                                                  f+i_coord_offset,fshift+i_shift_offset);
 763
 764         /* Increment number of inner iterations */
 765         inneriter                  += j_index_end - j_index_start;
 766
 767         /* Outer loop uses 7 flops */
 768     }
 769
 770     /* Increment number of outer iterations */
 771     outeriter        += nri;
 772
 773     /* Update outer/inner flops */
 774
 775     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*33);
 776 }