src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwCSTab_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_ElecRFCut_VdwCSTab_GeomP1P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_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     __m256i          vfitab;
  86     __m128i          vfitab_lo,vfitab_hi;
  87     __m128i          ifour       = _mm_set1_epi32(4);
  88     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  89     real             *vftab;
  90     __m256           dummy_mask,cutoff_mask;
  91     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  92     __m256           one     = _mm256_set1_ps(1.0);
  93     __m256           two     = _mm256_set1_ps(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm256_set1_ps(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107     krf              = _mm256_set1_ps(fr->ic->k_rf);
 108     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 109     crf              = _mm256_set1_ps(fr->ic->c_rf);
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     vftab            = kernel_data->table_vdw->data;
 115     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 116
 117     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 118     rcutoff_scalar   = fr->rcoulomb;
 119     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 120     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128     j_coord_offsetE = 0;
 129     j_coord_offsetF = 0;
 130     j_coord_offsetG = 0;
 131     j_coord_offsetH = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 157
 158         fix0             = _mm256_setzero_ps();
 159         fiy0             = _mm256_setzero_ps();
 160         fiz0             = _mm256_setzero_ps();
 161
 162         /* Load parameters for i particles */
 163         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 164         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm256_setzero_ps();
 168         vvdwsum          = _mm256_setzero_ps();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             jnrC             = jjnr[jidx+2];
 178             jnrD             = jjnr[jidx+3];
 179             jnrE             = jjnr[jidx+4];
 180             jnrF             = jjnr[jidx+5];
 181             jnrG             = jjnr[jidx+6];
 182             jnrH             = jjnr[jidx+7];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185             j_coord_offsetC  = DIM*jnrC;
 186             j_coord_offsetD  = DIM*jnrD;
 187             j_coord_offsetE  = DIM*jnrE;
 188             j_coord_offsetF  = DIM*jnrF;
 189             j_coord_offsetG  = DIM*jnrG;
 190             j_coord_offsetH  = DIM*jnrH;
 191
 192             /* load j atom coordinates */
 193             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 195                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 196                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 197                                                  &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm256_sub_ps(ix0,jx0);
 201             dy00             = _mm256_sub_ps(iy0,jy0);
 202             dz00             = _mm256_sub_ps(iz0,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 206
 207             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 208
 209             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 210
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 213                                                                  charge+jnrC+0,charge+jnrD+0,
 214                                                                  charge+jnrE+0,charge+jnrF+0,
 215                                                                  charge+jnrG+0,charge+jnrH+0);
 216             vdwjidx0A        = 2*vdwtype[jnrA+0];
 217             vdwjidx0B        = 2*vdwtype[jnrB+0];
 218             vdwjidx0C        = 2*vdwtype[jnrC+0];
 219             vdwjidx0D        = 2*vdwtype[jnrD+0];
 220             vdwjidx0E        = 2*vdwtype[jnrE+0];
 221             vdwjidx0F        = 2*vdwtype[jnrF+0];
 222             vdwjidx0G        = 2*vdwtype[jnrG+0];
 223             vdwjidx0H        = 2*vdwtype[jnrH+0];
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 230             {
 231
 232             r00              = _mm256_mul_ps(rsq00,rinv00);
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             qq00             = _mm256_mul_ps(iq0,jq0);
 236             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 237                                             vdwioffsetptr0+vdwjidx0B,
 238                                             vdwioffsetptr0+vdwjidx0C,
 239                                             vdwioffsetptr0+vdwjidx0D,
 240                                             vdwioffsetptr0+vdwjidx0E,
 241                                             vdwioffsetptr0+vdwjidx0F,
 242                                             vdwioffsetptr0+vdwjidx0G,
 243                                             vdwioffsetptr0+vdwjidx0H,
 244                                             &c6_00,&c12_00);
 245
 246             /* Calculate table index by multiplying r with table scale and truncate to integer */
 247             rt               = _mm256_mul_ps(r00,vftabscale);
 248             vfitab           = _mm256_cvttps_epi32(rt);
 249             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 250             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 251             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 252             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 253             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 254             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 255
 256             /* REACTION-FIELD ELECTROSTATICS */
 257             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 258             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 259
 260             /* CUBIC SPLINE TABLE DISPERSION */
 261             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 262                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 263             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 264                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 265             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 266                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 267             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 268                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 269             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 270             Heps             = _mm256_mul_ps(vfeps,H);
 271             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 272             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 273             vvdw6            = _mm256_mul_ps(c6_00,VV);
 274             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 275             fvdw6            = _mm256_mul_ps(c6_00,FF);
 276
 277             /* CUBIC SPLINE TABLE REPULSION */
 278             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 279             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 280             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 281                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 282             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 283                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 284             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 285                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 286             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 287                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 288             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 289             Heps             = _mm256_mul_ps(vfeps,H);
 290             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 291             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 292             vvdw12           = _mm256_mul_ps(c12_00,VV);
 293             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 294             fvdw12           = _mm256_mul_ps(c12_00,FF);
 295             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 296             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 297
 298             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velec            = _mm256_and_ps(velec,cutoff_mask);
 302             velecsum         = _mm256_add_ps(velecsum,velec);
 303             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 304             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 305
 306             fscal            = _mm256_add_ps(felec,fvdw);
 307
 308             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm256_mul_ps(fscal,dx00);
 312             ty               = _mm256_mul_ps(fscal,dy00);
 313             tz               = _mm256_mul_ps(fscal,dz00);
 314
 315             /* Update vectorial force */
 316             fix0             = _mm256_add_ps(fix0,tx);
 317             fiy0             = _mm256_add_ps(fiy0,ty);
 318             fiz0             = _mm256_add_ps(fiz0,tz);
 319
 320             fjptrA             = f+j_coord_offsetA;
 321             fjptrB             = f+j_coord_offsetB;
 322             fjptrC             = f+j_coord_offsetC;
 323             fjptrD             = f+j_coord_offsetD;
 324             fjptrE             = f+j_coord_offsetE;
 325             fjptrF             = f+j_coord_offsetF;
 326             fjptrG             = f+j_coord_offsetG;
 327             fjptrH             = f+j_coord_offsetH;
 328             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 329
 330             }
 331
 332             /* Inner loop uses 72 flops */
 333         }
 334
 335         if(jidx<j_index_end)
 336         {
 337
 338             /* Get j neighbor index, and coordinate index */
 339             jnrlistA         = jjnr[jidx];
 340             jnrlistB         = jjnr[jidx+1];
 341             jnrlistC         = jjnr[jidx+2];
 342             jnrlistD         = jjnr[jidx+3];
 343             jnrlistE         = jjnr[jidx+4];
 344             jnrlistF         = jjnr[jidx+5];
 345             jnrlistG         = jjnr[jidx+6];
 346             jnrlistH         = jjnr[jidx+7];
 347             /* Sign of each element will be negative for non-real atoms.
 348              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 349              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 350              */
 351             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 352                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 353
 354             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 355             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 356             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 357             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 358             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 359             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 360             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 361             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 362             j_coord_offsetA  = DIM*jnrA;
 363             j_coord_offsetB  = DIM*jnrB;
 364             j_coord_offsetC  = DIM*jnrC;
 365             j_coord_offsetD  = DIM*jnrD;
 366             j_coord_offsetE  = DIM*jnrE;
 367             j_coord_offsetF  = DIM*jnrF;
 368             j_coord_offsetG  = DIM*jnrG;
 369             j_coord_offsetH  = DIM*jnrH;
 370
 371             /* load j atom coordinates */
 372             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 373                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 374                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 375                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 376                                                  &jx0,&jy0,&jz0);
 377
 378             /* Calculate displacement vector */
 379             dx00             = _mm256_sub_ps(ix0,jx0);
 380             dy00             = _mm256_sub_ps(iy0,jy0);
 381             dz00             = _mm256_sub_ps(iz0,jz0);
 382
 383             /* Calculate squared distance and things based on it */
 384             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 385
 386             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 387
 388             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 389
 390             /* Load parameters for j particles */
 391             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 392                                                                  charge+jnrC+0,charge+jnrD+0,
 393                                                                  charge+jnrE+0,charge+jnrF+0,
 394                                                                  charge+jnrG+0,charge+jnrH+0);
 395             vdwjidx0A        = 2*vdwtype[jnrA+0];
 396             vdwjidx0B        = 2*vdwtype[jnrB+0];
 397             vdwjidx0C        = 2*vdwtype[jnrC+0];
 398             vdwjidx0D        = 2*vdwtype[jnrD+0];
 399             vdwjidx0E        = 2*vdwtype[jnrE+0];
 400             vdwjidx0F        = 2*vdwtype[jnrF+0];
 401             vdwjidx0G        = 2*vdwtype[jnrG+0];
 402             vdwjidx0H        = 2*vdwtype[jnrH+0];
 403
 404             /**************************
 405              * CALCULATE INTERACTIONS *
 406              **************************/
 407
 408             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 409             {
 410
 411             r00              = _mm256_mul_ps(rsq00,rinv00);
 412             r00              = _mm256_andnot_ps(dummy_mask,r00);
 413
 414             /* Compute parameters for interactions between i and j atoms */
 415             qq00             = _mm256_mul_ps(iq0,jq0);
 416             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 417                                             vdwioffsetptr0+vdwjidx0B,
 418                                             vdwioffsetptr0+vdwjidx0C,
 419                                             vdwioffsetptr0+vdwjidx0D,
 420                                             vdwioffsetptr0+vdwjidx0E,
 421                                             vdwioffsetptr0+vdwjidx0F,
 422                                             vdwioffsetptr0+vdwjidx0G,
 423                                             vdwioffsetptr0+vdwjidx0H,
 424                                             &c6_00,&c12_00);
 425
 426             /* Calculate table index by multiplying r with table scale and truncate to integer */
 427             rt               = _mm256_mul_ps(r00,vftabscale);
 428             vfitab           = _mm256_cvttps_epi32(rt);
 429             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 430             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 431             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 432             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 433             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 434             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 435
 436             /* REACTION-FIELD ELECTROSTATICS */
 437             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 438             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 439
 440             /* CUBIC SPLINE TABLE DISPERSION */
 441             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 442                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 443             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 444                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 445             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 446                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 447             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 448                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 449             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 450             Heps             = _mm256_mul_ps(vfeps,H);
 451             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 452             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 453             vvdw6            = _mm256_mul_ps(c6_00,VV);
 454             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 455             fvdw6            = _mm256_mul_ps(c6_00,FF);
 456
 457             /* CUBIC SPLINE TABLE REPULSION */
 458             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 459             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 460             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 461                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 462             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 463                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 464             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 465                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 466             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 467                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 468             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 469             Heps             = _mm256_mul_ps(vfeps,H);
 470             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 471             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 472             vvdw12           = _mm256_mul_ps(c12_00,VV);
 473             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 474             fvdw12           = _mm256_mul_ps(c12_00,FF);
 475             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 476             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 477
 478             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 479
 480             /* Update potential sum for this i atom from the interaction with this j atom. */
 481             velec            = _mm256_and_ps(velec,cutoff_mask);
 482             velec            = _mm256_andnot_ps(dummy_mask,velec);
 483             velecsum         = _mm256_add_ps(velecsum,velec);
 484             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 485             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 486             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 487
 488             fscal            = _mm256_add_ps(felec,fvdw);
 489
 490             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 491
 492             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 493
 494             /* Calculate temporary vectorial force */
 495             tx               = _mm256_mul_ps(fscal,dx00);
 496             ty               = _mm256_mul_ps(fscal,dy00);
 497             tz               = _mm256_mul_ps(fscal,dz00);
 498
 499             /* Update vectorial force */
 500             fix0             = _mm256_add_ps(fix0,tx);
 501             fiy0             = _mm256_add_ps(fiy0,ty);
 502             fiz0             = _mm256_add_ps(fiz0,tz);
 503
 504             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 505             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 506             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 507             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 508             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 509             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 510             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 511             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 512             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 513
 514             }
 515
 516             /* Inner loop uses 73 flops */
 517         }
 518
 519         /* End of innermost loop */
 520
 521         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 522                                                  f+i_coord_offset,fshift+i_shift_offset);
 523
 524         ggid                        = gid[iidx];
 525         /* Update potential energies */
 526         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 527         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 528
 529         /* Increment number of inner iterations */
 530         inneriter                  += j_index_end - j_index_start;
 531
 532         /* Outer loop uses 9 flops */
 533     }
 534
 535     /* Increment number of outer iterations */
 536     outeriter        += nri;
 537
 538     /* Update outer/inner flops */
 539
 540     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 541 }
 542 /*
 543  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_avx_256_single
 544  * Electrostatics interaction: ReactionField
 545  * VdW interaction:            CubicSplineTable
 546  * Geometry:                   Particle-Particle
 547  * Calculate force/pot:        Force
 548  */
 549 void
 550 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_avx_256_single
 551                     (t_nblist * gmx_restrict                nlist,
 552                      rvec * gmx_restrict                    xx,
 553                      rvec * gmx_restrict                    ff,
 554                      t_forcerec * gmx_restrict              fr,
 555                      t_mdatoms * gmx_restrict               mdatoms,
 556                      nb_kernel_data_t * gmx_restrict        kernel_data,
 557                      t_nrnb * gmx_restrict                  nrnb)
 558 {
 559     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 560      * just 0 for non-waters.
 561      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 562      * jnr indices corresponding to data put in the four positions in the SIMD register.
 563      */
 564     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 565     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 566     int              jnrA,jnrB,jnrC,jnrD;
 567     int              jnrE,jnrF,jnrG,jnrH;
 568     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 569     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 570     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 571     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 572     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 573     real             rcutoff_scalar;
 574     real             *shiftvec,*fshift,*x,*f;
 575     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 576     real             scratch[4*DIM];
 577     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 578     real *           vdwioffsetptr0;
 579     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 580     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 581     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 582     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 583     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 584     real             *charge;
 585     int              nvdwtype;
 586     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 587     int              *vdwtype;
 588     real             *vdwparam;
 589     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 590     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 591     __m256i          vfitab;
 592     __m128i          vfitab_lo,vfitab_hi;
 593     __m128i          ifour       = _mm_set1_epi32(4);
 594     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 595     real             *vftab;
 596     __m256           dummy_mask,cutoff_mask;
 597     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 598     __m256           one     = _mm256_set1_ps(1.0);
 599     __m256           two     = _mm256_set1_ps(2.0);
 600     x                = xx[0];
 601     f                = ff[0];
 602
 603     nri              = nlist->nri;
 604     iinr             = nlist->iinr;
 605     jindex           = nlist->jindex;
 606     jjnr             = nlist->jjnr;
 607     shiftidx         = nlist->shift;
 608     gid              = nlist->gid;
 609     shiftvec         = fr->shift_vec[0];
 610     fshift           = fr->fshift[0];
 611     facel            = _mm256_set1_ps(fr->epsfac);
 612     charge           = mdatoms->chargeA;
 613     krf              = _mm256_set1_ps(fr->ic->k_rf);
 614     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 615     crf              = _mm256_set1_ps(fr->ic->c_rf);
 616     nvdwtype         = fr->ntype;
 617     vdwparam         = fr->nbfp;
 618     vdwtype          = mdatoms->typeA;
 619
 620     vftab            = kernel_data->table_vdw->data;
 621     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 622
 623     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 624     rcutoff_scalar   = fr->rcoulomb;
 625     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 626     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 627
 628     /* Avoid stupid compiler warnings */
 629     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 630     j_coord_offsetA = 0;
 631     j_coord_offsetB = 0;
 632     j_coord_offsetC = 0;
 633     j_coord_offsetD = 0;
 634     j_coord_offsetE = 0;
 635     j_coord_offsetF = 0;
 636     j_coord_offsetG = 0;
 637     j_coord_offsetH = 0;
 638
 639     outeriter        = 0;
 640     inneriter        = 0;
 641
 642     for(iidx=0;iidx<4*DIM;iidx++)
 643     {
 644         scratch[iidx] = 0.0;
 645     }
 646
 647     /* Start outer loop over neighborlists */
 648     for(iidx=0; iidx<nri; iidx++)
 649     {
 650         /* Load shift vector for this list */
 651         i_shift_offset   = DIM*shiftidx[iidx];
 652
 653         /* Load limits for loop over neighbors */
 654         j_index_start    = jindex[iidx];
 655         j_index_end      = jindex[iidx+1];
 656
 657         /* Get outer coordinate index */
 658         inr              = iinr[iidx];
 659         i_coord_offset   = DIM*inr;
 660
 661         /* Load i particle coords and add shift vector */
 662         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 663
 664         fix0             = _mm256_setzero_ps();
 665         fiy0             = _mm256_setzero_ps();
 666         fiz0             = _mm256_setzero_ps();
 667
 668         /* Load parameters for i particles */
 669         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 670         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 671
 672         /* Start inner kernel loop */
 673         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 674         {
 675
 676             /* Get j neighbor index, and coordinate index */
 677             jnrA             = jjnr[jidx];
 678             jnrB             = jjnr[jidx+1];
 679             jnrC             = jjnr[jidx+2];
 680             jnrD             = jjnr[jidx+3];
 681             jnrE             = jjnr[jidx+4];
 682             jnrF             = jjnr[jidx+5];
 683             jnrG             = jjnr[jidx+6];
 684             jnrH             = jjnr[jidx+7];
 685             j_coord_offsetA  = DIM*jnrA;
 686             j_coord_offsetB  = DIM*jnrB;
 687             j_coord_offsetC  = DIM*jnrC;
 688             j_coord_offsetD  = DIM*jnrD;
 689             j_coord_offsetE  = DIM*jnrE;
 690             j_coord_offsetF  = DIM*jnrF;
 691             j_coord_offsetG  = DIM*jnrG;
 692             j_coord_offsetH  = DIM*jnrH;
 693
 694             /* load j atom coordinates */
 695             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 696                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 697                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 698                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 699                                                  &jx0,&jy0,&jz0);
 700
 701             /* Calculate displacement vector */
 702             dx00             = _mm256_sub_ps(ix0,jx0);
 703             dy00             = _mm256_sub_ps(iy0,jy0);
 704             dz00             = _mm256_sub_ps(iz0,jz0);
 705
 706             /* Calculate squared distance and things based on it */
 707             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 708
 709             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 710
 711             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 712
 713             /* Load parameters for j particles */
 714             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 715                                                                  charge+jnrC+0,charge+jnrD+0,
 716                                                                  charge+jnrE+0,charge+jnrF+0,
 717                                                                  charge+jnrG+0,charge+jnrH+0);
 718             vdwjidx0A        = 2*vdwtype[jnrA+0];
 719             vdwjidx0B        = 2*vdwtype[jnrB+0];
 720             vdwjidx0C        = 2*vdwtype[jnrC+0];
 721             vdwjidx0D        = 2*vdwtype[jnrD+0];
 722             vdwjidx0E        = 2*vdwtype[jnrE+0];
 723             vdwjidx0F        = 2*vdwtype[jnrF+0];
 724             vdwjidx0G        = 2*vdwtype[jnrG+0];
 725             vdwjidx0H        = 2*vdwtype[jnrH+0];
 726
 727             /**************************
 728              * CALCULATE INTERACTIONS *
 729              **************************/
 730
 731             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 732             {
 733
 734             r00              = _mm256_mul_ps(rsq00,rinv00);
 735
 736             /* Compute parameters for interactions between i and j atoms */
 737             qq00             = _mm256_mul_ps(iq0,jq0);
 738             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 739                                             vdwioffsetptr0+vdwjidx0B,
 740                                             vdwioffsetptr0+vdwjidx0C,
 741                                             vdwioffsetptr0+vdwjidx0D,
 742                                             vdwioffsetptr0+vdwjidx0E,
 743                                             vdwioffsetptr0+vdwjidx0F,
 744                                             vdwioffsetptr0+vdwjidx0G,
 745                                             vdwioffsetptr0+vdwjidx0H,
 746                                             &c6_00,&c12_00);
 747
 748             /* Calculate table index by multiplying r with table scale and truncate to integer */
 749             rt               = _mm256_mul_ps(r00,vftabscale);
 750             vfitab           = _mm256_cvttps_epi32(rt);
 751             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 752             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 753             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 754             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 755             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 756             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 757
 758             /* REACTION-FIELD ELECTROSTATICS */
 759             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 760
 761             /* CUBIC SPLINE TABLE DISPERSION */
 762             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 763                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 764             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 765                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 766             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 767                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 768             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 769                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 770             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 771             Heps             = _mm256_mul_ps(vfeps,H);
 772             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 773             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 774             fvdw6            = _mm256_mul_ps(c6_00,FF);
 775
 776             /* CUBIC SPLINE TABLE REPULSION */
 777             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 778             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 779             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 780                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 781             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 782                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 783             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 784                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 785             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 786                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 787             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 788             Heps             = _mm256_mul_ps(vfeps,H);
 789             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 790             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 791             fvdw12           = _mm256_mul_ps(c12_00,FF);
 792             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 793
 794             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 795
 796             fscal            = _mm256_add_ps(felec,fvdw);
 797
 798             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 799
 800             /* Calculate temporary vectorial force */
 801             tx               = _mm256_mul_ps(fscal,dx00);
 802             ty               = _mm256_mul_ps(fscal,dy00);
 803             tz               = _mm256_mul_ps(fscal,dz00);
 804
 805             /* Update vectorial force */
 806             fix0             = _mm256_add_ps(fix0,tx);
 807             fiy0             = _mm256_add_ps(fiy0,ty);
 808             fiz0             = _mm256_add_ps(fiz0,tz);
 809
 810             fjptrA             = f+j_coord_offsetA;
 811             fjptrB             = f+j_coord_offsetB;
 812             fjptrC             = f+j_coord_offsetC;
 813             fjptrD             = f+j_coord_offsetD;
 814             fjptrE             = f+j_coord_offsetE;
 815             fjptrF             = f+j_coord_offsetF;
 816             fjptrG             = f+j_coord_offsetG;
 817             fjptrH             = f+j_coord_offsetH;
 818             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 819
 820             }
 821
 822             /* Inner loop uses 57 flops */
 823         }
 824
 825         if(jidx<j_index_end)
 826         {
 827
 828             /* Get j neighbor index, and coordinate index */
 829             jnrlistA         = jjnr[jidx];
 830             jnrlistB         = jjnr[jidx+1];
 831             jnrlistC         = jjnr[jidx+2];
 832             jnrlistD         = jjnr[jidx+3];
 833             jnrlistE         = jjnr[jidx+4];
 834             jnrlistF         = jjnr[jidx+5];
 835             jnrlistG         = jjnr[jidx+6];
 836             jnrlistH         = jjnr[jidx+7];
 837             /* Sign of each element will be negative for non-real atoms.
 838              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 839              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 840              */
 841             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 842                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 843
 844             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 845             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 846             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 847             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 848             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 849             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 850             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 851             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 852             j_coord_offsetA  = DIM*jnrA;
 853             j_coord_offsetB  = DIM*jnrB;
 854             j_coord_offsetC  = DIM*jnrC;
 855             j_coord_offsetD  = DIM*jnrD;
 856             j_coord_offsetE  = DIM*jnrE;
 857             j_coord_offsetF  = DIM*jnrF;
 858             j_coord_offsetG  = DIM*jnrG;
 859             j_coord_offsetH  = DIM*jnrH;
 860
 861             /* load j atom coordinates */
 862             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 863                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 864                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 865                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 866                                                  &jx0,&jy0,&jz0);
 867
 868             /* Calculate displacement vector */
 869             dx00             = _mm256_sub_ps(ix0,jx0);
 870             dy00             = _mm256_sub_ps(iy0,jy0);
 871             dz00             = _mm256_sub_ps(iz0,jz0);
 872
 873             /* Calculate squared distance and things based on it */
 874             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 875
 876             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 877
 878             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 879
 880             /* Load parameters for j particles */
 881             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 882                                                                  charge+jnrC+0,charge+jnrD+0,
 883                                                                  charge+jnrE+0,charge+jnrF+0,
 884                                                                  charge+jnrG+0,charge+jnrH+0);
 885             vdwjidx0A        = 2*vdwtype[jnrA+0];
 886             vdwjidx0B        = 2*vdwtype[jnrB+0];
 887             vdwjidx0C        = 2*vdwtype[jnrC+0];
 888             vdwjidx0D        = 2*vdwtype[jnrD+0];
 889             vdwjidx0E        = 2*vdwtype[jnrE+0];
 890             vdwjidx0F        = 2*vdwtype[jnrF+0];
 891             vdwjidx0G        = 2*vdwtype[jnrG+0];
 892             vdwjidx0H        = 2*vdwtype[jnrH+0];
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 899             {
 900
 901             r00              = _mm256_mul_ps(rsq00,rinv00);
 902             r00              = _mm256_andnot_ps(dummy_mask,r00);
 903
 904             /* Compute parameters for interactions between i and j atoms */
 905             qq00             = _mm256_mul_ps(iq0,jq0);
 906             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 907                                             vdwioffsetptr0+vdwjidx0B,
 908                                             vdwioffsetptr0+vdwjidx0C,
 909                                             vdwioffsetptr0+vdwjidx0D,
 910                                             vdwioffsetptr0+vdwjidx0E,
 911                                             vdwioffsetptr0+vdwjidx0F,
 912                                             vdwioffsetptr0+vdwjidx0G,
 913                                             vdwioffsetptr0+vdwjidx0H,
 914                                             &c6_00,&c12_00);
 915
 916             /* Calculate table index by multiplying r with table scale and truncate to integer */
 917             rt               = _mm256_mul_ps(r00,vftabscale);
 918             vfitab           = _mm256_cvttps_epi32(rt);
 919             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 920             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 921             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 922             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 923             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 924             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 925
 926             /* REACTION-FIELD ELECTROSTATICS */
 927             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 928
 929             /* CUBIC SPLINE TABLE DISPERSION */
 930             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 931                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 932             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 933                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 934             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 935                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 936             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 937                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 938             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 939             Heps             = _mm256_mul_ps(vfeps,H);
 940             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 941             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 942             fvdw6            = _mm256_mul_ps(c6_00,FF);
 943
 944             /* CUBIC SPLINE TABLE REPULSION */
 945             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 946             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 947             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 948                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 949             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 950                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 951             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 952                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 953             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 954                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 955             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 956             Heps             = _mm256_mul_ps(vfeps,H);
 957             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 958             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 959             fvdw12           = _mm256_mul_ps(c12_00,FF);
 960             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 961
 962             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 963
 964             fscal            = _mm256_add_ps(felec,fvdw);
 965
 966             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 967
 968             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 969
 970             /* Calculate temporary vectorial force */
 971             tx               = _mm256_mul_ps(fscal,dx00);
 972             ty               = _mm256_mul_ps(fscal,dy00);
 973             tz               = _mm256_mul_ps(fscal,dz00);
 974
 975             /* Update vectorial force */
 976             fix0             = _mm256_add_ps(fix0,tx);
 977             fiy0             = _mm256_add_ps(fiy0,ty);
 978             fiz0             = _mm256_add_ps(fiz0,tz);
 979
 980             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 981             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 982             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 983             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 984             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 985             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 986             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 987             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 988             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 989
 990             }
 991
 992             /* Inner loop uses 58 flops */
 993         }
 994
 995         /* End of innermost loop */
 996
 997         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 998                                                  f+i_coord_offset,fshift+i_shift_offset);
 999
1000         /* Increment number of inner iterations */
1001         inneriter                  += j_index_end - j_index_start;
1002
1003         /* Outer loop uses 7 flops */
1004     }
1005
1006     /* Increment number of outer iterations */
1007     outeriter        += nri;
1008
1009     /* Update outer/inner flops */
1010
1011     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*58);
1012 }