src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecGB_VdwNone_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_ElecGB_VdwNone_GeomP1P1_VF_avx_256_single
  38  * Electrostatics interaction: GeneralizedBorn
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecGB_VdwNone_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     __m256i          gbitab;
  80     __m128i          gbitab_lo,gbitab_hi;
  81     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  82     __m256           minushalf = _mm256_set1_ps(-0.5);
  83     real             *invsqrta,*dvda,*gbtab;
  84     __m256i          vfitab;
  85     __m128i          vfitab_lo,vfitab_hi;
  86     __m128i          ifour       = _mm_set1_epi32(4);
  87     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  88     real             *vftab;
  89     __m256           dummy_mask,cutoff_mask;
  90     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  91     __m256           one     = _mm256_set1_ps(1.0);
  92     __m256           two     = _mm256_set1_ps(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm256_set1_ps(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 106
 107     invsqrta         = fr->invsqrta;
 108     dvda             = fr->dvda;
 109     gbtabscale       = _mm256_set1_ps(fr->gbtab.scale);
 110     gbtab            = fr->gbtab.data;
 111     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117     j_coord_offsetC = 0;
 118     j_coord_offsetD = 0;
 119     j_coord_offsetE = 0;
 120     j_coord_offsetF = 0;
 121     j_coord_offsetG = 0;
 122     j_coord_offsetH = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     for(iidx=0;iidx<4*DIM;iidx++)
 128     {
 129         scratch[iidx] = 0.0;
 130     }
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm256_setzero_ps();
 150         fiy0             = _mm256_setzero_ps();
 151         fiz0             = _mm256_setzero_ps();
 152
 153         /* Load parameters for i particles */
 154         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 155         isai0            = _mm256_set1_ps(invsqrta[inr+0]);
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm256_setzero_ps();
 159         vgbsum           = _mm256_setzero_ps();
 160         dvdasum          = _mm256_setzero_ps();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             jnrC             = jjnr[jidx+2];
 170             jnrD             = jjnr[jidx+3];
 171             jnrE             = jjnr[jidx+4];
 172             jnrF             = jjnr[jidx+5];
 173             jnrG             = jjnr[jidx+6];
 174             jnrH             = jjnr[jidx+7];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177             j_coord_offsetC  = DIM*jnrC;
 178             j_coord_offsetD  = DIM*jnrD;
 179             j_coord_offsetE  = DIM*jnrE;
 180             j_coord_offsetF  = DIM*jnrF;
 181             j_coord_offsetG  = DIM*jnrG;
 182             j_coord_offsetH  = DIM*jnrH;
 183
 184             /* load j atom coordinates */
 185             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 187                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 188                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 189                                                  &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm256_sub_ps(ix0,jx0);
 193             dy00             = _mm256_sub_ps(iy0,jy0);
 194             dz00             = _mm256_sub_ps(iz0,jz0);
 195
 196             /* Calculate squared distance and things based on it */
 197             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 198
 199             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 200
 201             /* Load parameters for j particles */
 202             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 203                                                                  charge+jnrC+0,charge+jnrD+0,
 204                                                                  charge+jnrE+0,charge+jnrF+0,
 205                                                                  charge+jnrG+0,charge+jnrH+0);
 206             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 207                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 208                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 209                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 210
 211             /**************************
 212              * CALCULATE INTERACTIONS *
 213              **************************/
 214
 215             r00              = _mm256_mul_ps(rsq00,rinv00);
 216
 217             /* Compute parameters for interactions between i and j atoms */
 218             qq00             = _mm256_mul_ps(iq0,jq0);
 219
 220             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 221             isaprod          = _mm256_mul_ps(isai0,isaj0);
 222             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 223             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 224
 225             /* Calculate generalized born table index - this is a separate table from the normal one,
 226              * but we use the same procedure by multiplying r with scale and truncating to integer.
 227              */
 228             rt               = _mm256_mul_ps(r00,gbscale);
 229             gbitab           = _mm256_cvttps_epi32(rt);
 230             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 231             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 232             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 233             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 234             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 235             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 236             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 237                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 238             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 239                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 240             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 241                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 242             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 243                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 244             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 245             Heps             = _mm256_mul_ps(gbeps,H);
 246             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 247             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 248             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 249
 250             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 251             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 252             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 253             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 254             fjptrA           = dvda+jnrA;
 255             fjptrB           = dvda+jnrB;
 256             fjptrC           = dvda+jnrC;
 257             fjptrD           = dvda+jnrD;
 258             fjptrE           = dvda+jnrE;
 259             fjptrF           = dvda+jnrF;
 260             fjptrG           = dvda+jnrG;
 261             fjptrH           = dvda+jnrH;
 262             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 263                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 264             velec            = _mm256_mul_ps(qq00,rinv00);
 265             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             velecsum         = _mm256_add_ps(velecsum,velec);
 269             vgbsum           = _mm256_add_ps(vgbsum,vgb);
 270
 271             fscal            = felec;
 272
 273             /* Calculate temporary vectorial force */
 274             tx               = _mm256_mul_ps(fscal,dx00);
 275             ty               = _mm256_mul_ps(fscal,dy00);
 276             tz               = _mm256_mul_ps(fscal,dz00);
 277
 278             /* Update vectorial force */
 279             fix0             = _mm256_add_ps(fix0,tx);
 280             fiy0             = _mm256_add_ps(fiy0,ty);
 281             fiz0             = _mm256_add_ps(fiz0,tz);
 282
 283             fjptrA             = f+j_coord_offsetA;
 284             fjptrB             = f+j_coord_offsetB;
 285             fjptrC             = f+j_coord_offsetC;
 286             fjptrD             = f+j_coord_offsetD;
 287             fjptrE             = f+j_coord_offsetE;
 288             fjptrF             = f+j_coord_offsetF;
 289             fjptrG             = f+j_coord_offsetG;
 290             fjptrH             = f+j_coord_offsetH;
 291             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 292
 293             /* Inner loop uses 57 flops */
 294         }
 295
 296         if(jidx<j_index_end)
 297         {
 298
 299             /* Get j neighbor index, and coordinate index */
 300             jnrlistA         = jjnr[jidx];
 301             jnrlistB         = jjnr[jidx+1];
 302             jnrlistC         = jjnr[jidx+2];
 303             jnrlistD         = jjnr[jidx+3];
 304             jnrlistE         = jjnr[jidx+4];
 305             jnrlistF         = jjnr[jidx+5];
 306             jnrlistG         = jjnr[jidx+6];
 307             jnrlistH         = jjnr[jidx+7];
 308             /* Sign of each element will be negative for non-real atoms.
 309              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 310              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 311              */
 312             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 313                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 314
 315             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 316             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 317             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 318             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 319             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 320             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 321             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 322             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 323             j_coord_offsetA  = DIM*jnrA;
 324             j_coord_offsetB  = DIM*jnrB;
 325             j_coord_offsetC  = DIM*jnrC;
 326             j_coord_offsetD  = DIM*jnrD;
 327             j_coord_offsetE  = DIM*jnrE;
 328             j_coord_offsetF  = DIM*jnrF;
 329             j_coord_offsetG  = DIM*jnrG;
 330             j_coord_offsetH  = DIM*jnrH;
 331
 332             /* load j atom coordinates */
 333             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 334                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 335                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 336                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 337                                                  &jx0,&jy0,&jz0);
 338
 339             /* Calculate displacement vector */
 340             dx00             = _mm256_sub_ps(ix0,jx0);
 341             dy00             = _mm256_sub_ps(iy0,jy0);
 342             dz00             = _mm256_sub_ps(iz0,jz0);
 343
 344             /* Calculate squared distance and things based on it */
 345             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 346
 347             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 348
 349             /* Load parameters for j particles */
 350             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 351                                                                  charge+jnrC+0,charge+jnrD+0,
 352                                                                  charge+jnrE+0,charge+jnrF+0,
 353                                                                  charge+jnrG+0,charge+jnrH+0);
 354             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 355                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 356                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 357                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             r00              = _mm256_mul_ps(rsq00,rinv00);
 364             r00              = _mm256_andnot_ps(dummy_mask,r00);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq00             = _mm256_mul_ps(iq0,jq0);
 368
 369             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 370             isaprod          = _mm256_mul_ps(isai0,isaj0);
 371             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 372             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 373
 374             /* Calculate generalized born table index - this is a separate table from the normal one,
 375              * but we use the same procedure by multiplying r with scale and truncating to integer.
 376              */
 377             rt               = _mm256_mul_ps(r00,gbscale);
 378             gbitab           = _mm256_cvttps_epi32(rt);
 379             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 380             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 381             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 382             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 383             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 384             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 385             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 386                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 387             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 388                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 389             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 390                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 391             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 392                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 393             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 394             Heps             = _mm256_mul_ps(gbeps,H);
 395             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 396             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 397             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 398
 399             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 400             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 401             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 402             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
 403             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 404             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 405             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 406             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 407             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 408             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 409             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
 410             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
 411             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
 412             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
 413             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 414                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 415             velec            = _mm256_mul_ps(qq00,rinv00);
 416             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 417
 418             /* Update potential sum for this i atom from the interaction with this j atom. */
 419             velec            = _mm256_andnot_ps(dummy_mask,velec);
 420             velecsum         = _mm256_add_ps(velecsum,velec);
 421             vgb              = _mm256_andnot_ps(dummy_mask,vgb);
 422             vgbsum           = _mm256_add_ps(vgbsum,vgb);
 423
 424             fscal            = felec;
 425
 426             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 427
 428             /* Calculate temporary vectorial force */
 429             tx               = _mm256_mul_ps(fscal,dx00);
 430             ty               = _mm256_mul_ps(fscal,dy00);
 431             tz               = _mm256_mul_ps(fscal,dz00);
 432
 433             /* Update vectorial force */
 434             fix0             = _mm256_add_ps(fix0,tx);
 435             fiy0             = _mm256_add_ps(fiy0,ty);
 436             fiz0             = _mm256_add_ps(fiz0,tz);
 437
 438             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 439             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 440             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 441             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 442             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 443             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 444             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 445             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 446             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 447
 448             /* Inner loop uses 58 flops */
 449         }
 450
 451         /* End of innermost loop */
 452
 453         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 454                                                  f+i_coord_offset,fshift+i_shift_offset);
 455
 456         ggid                        = gid[iidx];
 457         /* Update potential energies */
 458         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 459         gmx_mm256_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 460         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai0,isai0));
 461         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
 462
 463         /* Increment number of inner iterations */
 464         inneriter                  += j_index_end - j_index_start;
 465
 466         /* Outer loop uses 9 flops */
 467     }
 468
 469     /* Increment number of outer iterations */
 470     outeriter        += nri;
 471
 472     /* Update outer/inner flops */
 473
 474     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*9 + inneriter*58);
 475 }
 476 /*
 477  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_F_avx_256_single
 478  * Electrostatics interaction: GeneralizedBorn
 479  * VdW interaction:            None
 480  * Geometry:                   Particle-Particle
 481  * Calculate force/pot:        Force
 482  */
 483 void
 484 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_avx_256_single
 485                     (t_nblist * gmx_restrict                nlist,
 486                      rvec * gmx_restrict                    xx,
 487                      rvec * gmx_restrict                    ff,
 488                      t_forcerec * gmx_restrict              fr,
 489                      t_mdatoms * gmx_restrict               mdatoms,
 490                      nb_kernel_data_t * gmx_restrict        kernel_data,
 491                      t_nrnb * gmx_restrict                  nrnb)
 492 {
 493     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 494      * just 0 for non-waters.
 495      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 496      * jnr indices corresponding to data put in the four positions in the SIMD register.
 497      */
 498     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 499     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 500     int              jnrA,jnrB,jnrC,jnrD;
 501     int              jnrE,jnrF,jnrG,jnrH;
 502     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 503     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 504     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 505     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 506     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 507     real             rcutoff_scalar;
 508     real             *shiftvec,*fshift,*x,*f;
 509     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 510     real             scratch[4*DIM];
 511     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 512     real *           vdwioffsetptr0;
 513     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 514     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 515     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 516     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 517     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 518     real             *charge;
 519     __m256i          gbitab;
 520     __m128i          gbitab_lo,gbitab_hi;
 521     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 522     __m256           minushalf = _mm256_set1_ps(-0.5);
 523     real             *invsqrta,*dvda,*gbtab;
 524     __m256i          vfitab;
 525     __m128i          vfitab_lo,vfitab_hi;
 526     __m128i          ifour       = _mm_set1_epi32(4);
 527     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 528     real             *vftab;
 529     __m256           dummy_mask,cutoff_mask;
 530     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 531     __m256           one     = _mm256_set1_ps(1.0);
 532     __m256           two     = _mm256_set1_ps(2.0);
 533     x                = xx[0];
 534     f                = ff[0];
 535
 536     nri              = nlist->nri;
 537     iinr             = nlist->iinr;
 538     jindex           = nlist->jindex;
 539     jjnr             = nlist->jjnr;
 540     shiftidx         = nlist->shift;
 541     gid              = nlist->gid;
 542     shiftvec         = fr->shift_vec[0];
 543     fshift           = fr->fshift[0];
 544     facel            = _mm256_set1_ps(fr->epsfac);
 545     charge           = mdatoms->chargeA;
 546
 547     invsqrta         = fr->invsqrta;
 548     dvda             = fr->dvda;
 549     gbtabscale       = _mm256_set1_ps(fr->gbtab.scale);
 550     gbtab            = fr->gbtab.data;
 551     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 552
 553     /* Avoid stupid compiler warnings */
 554     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 555     j_coord_offsetA = 0;
 556     j_coord_offsetB = 0;
 557     j_coord_offsetC = 0;
 558     j_coord_offsetD = 0;
 559     j_coord_offsetE = 0;
 560     j_coord_offsetF = 0;
 561     j_coord_offsetG = 0;
 562     j_coord_offsetH = 0;
 563
 564     outeriter        = 0;
 565     inneriter        = 0;
 566
 567     for(iidx=0;iidx<4*DIM;iidx++)
 568     {
 569         scratch[iidx] = 0.0;
 570     }
 571
 572     /* Start outer loop over neighborlists */
 573     for(iidx=0; iidx<nri; iidx++)
 574     {
 575         /* Load shift vector for this list */
 576         i_shift_offset   = DIM*shiftidx[iidx];
 577
 578         /* Load limits for loop over neighbors */
 579         j_index_start    = jindex[iidx];
 580         j_index_end      = jindex[iidx+1];
 581
 582         /* Get outer coordinate index */
 583         inr              = iinr[iidx];
 584         i_coord_offset   = DIM*inr;
 585
 586         /* Load i particle coords and add shift vector */
 587         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 588
 589         fix0             = _mm256_setzero_ps();
 590         fiy0             = _mm256_setzero_ps();
 591         fiz0             = _mm256_setzero_ps();
 592
 593         /* Load parameters for i particles */
 594         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 595         isai0            = _mm256_set1_ps(invsqrta[inr+0]);
 596
 597         dvdasum          = _mm256_setzero_ps();
 598
 599         /* Start inner kernel loop */
 600         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 601         {
 602
 603             /* Get j neighbor index, and coordinate index */
 604             jnrA             = jjnr[jidx];
 605             jnrB             = jjnr[jidx+1];
 606             jnrC             = jjnr[jidx+2];
 607             jnrD             = jjnr[jidx+3];
 608             jnrE             = jjnr[jidx+4];
 609             jnrF             = jjnr[jidx+5];
 610             jnrG             = jjnr[jidx+6];
 611             jnrH             = jjnr[jidx+7];
 612             j_coord_offsetA  = DIM*jnrA;
 613             j_coord_offsetB  = DIM*jnrB;
 614             j_coord_offsetC  = DIM*jnrC;
 615             j_coord_offsetD  = DIM*jnrD;
 616             j_coord_offsetE  = DIM*jnrE;
 617             j_coord_offsetF  = DIM*jnrF;
 618             j_coord_offsetG  = DIM*jnrG;
 619             j_coord_offsetH  = DIM*jnrH;
 620
 621             /* load j atom coordinates */
 622             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 623                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 624                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 625                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 626                                                  &jx0,&jy0,&jz0);
 627
 628             /* Calculate displacement vector */
 629             dx00             = _mm256_sub_ps(ix0,jx0);
 630             dy00             = _mm256_sub_ps(iy0,jy0);
 631             dz00             = _mm256_sub_ps(iz0,jz0);
 632
 633             /* Calculate squared distance and things based on it */
 634             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 635
 636             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 637
 638             /* Load parameters for j particles */
 639             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 640                                                                  charge+jnrC+0,charge+jnrD+0,
 641                                                                  charge+jnrE+0,charge+jnrF+0,
 642                                                                  charge+jnrG+0,charge+jnrH+0);
 643             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 644                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 645                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 646                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 647
 648             /**************************
 649              * CALCULATE INTERACTIONS *
 650              **************************/
 651
 652             r00              = _mm256_mul_ps(rsq00,rinv00);
 653
 654             /* Compute parameters for interactions between i and j atoms */
 655             qq00             = _mm256_mul_ps(iq0,jq0);
 656
 657             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 658             isaprod          = _mm256_mul_ps(isai0,isaj0);
 659             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 660             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 661
 662             /* Calculate generalized born table index - this is a separate table from the normal one,
 663              * but we use the same procedure by multiplying r with scale and truncating to integer.
 664              */
 665             rt               = _mm256_mul_ps(r00,gbscale);
 666             gbitab           = _mm256_cvttps_epi32(rt);
 667             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 668             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 669             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 670             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 671             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 672             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 673             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 674                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 675             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 676                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 677             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 678                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 679             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 680                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 681             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 682             Heps             = _mm256_mul_ps(gbeps,H);
 683             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 684             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 685             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 686
 687             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 688             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 689             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 690             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 691             fjptrA           = dvda+jnrA;
 692             fjptrB           = dvda+jnrB;
 693             fjptrC           = dvda+jnrC;
 694             fjptrD           = dvda+jnrD;
 695             fjptrE           = dvda+jnrE;
 696             fjptrF           = dvda+jnrF;
 697             fjptrG           = dvda+jnrG;
 698             fjptrH           = dvda+jnrH;
 699             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 700                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 701             velec            = _mm256_mul_ps(qq00,rinv00);
 702             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 703
 704             fscal            = felec;
 705
 706             /* Calculate temporary vectorial force */
 707             tx               = _mm256_mul_ps(fscal,dx00);
 708             ty               = _mm256_mul_ps(fscal,dy00);
 709             tz               = _mm256_mul_ps(fscal,dz00);
 710
 711             /* Update vectorial force */
 712             fix0             = _mm256_add_ps(fix0,tx);
 713             fiy0             = _mm256_add_ps(fiy0,ty);
 714             fiz0             = _mm256_add_ps(fiz0,tz);
 715
 716             fjptrA             = f+j_coord_offsetA;
 717             fjptrB             = f+j_coord_offsetB;
 718             fjptrC             = f+j_coord_offsetC;
 719             fjptrD             = f+j_coord_offsetD;
 720             fjptrE             = f+j_coord_offsetE;
 721             fjptrF             = f+j_coord_offsetF;
 722             fjptrG             = f+j_coord_offsetG;
 723             fjptrH             = f+j_coord_offsetH;
 724             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 725
 726             /* Inner loop uses 55 flops */
 727         }
 728
 729         if(jidx<j_index_end)
 730         {
 731
 732             /* Get j neighbor index, and coordinate index */
 733             jnrlistA         = jjnr[jidx];
 734             jnrlistB         = jjnr[jidx+1];
 735             jnrlistC         = jjnr[jidx+2];
 736             jnrlistD         = jjnr[jidx+3];
 737             jnrlistE         = jjnr[jidx+4];
 738             jnrlistF         = jjnr[jidx+5];
 739             jnrlistG         = jjnr[jidx+6];
 740             jnrlistH         = jjnr[jidx+7];
 741             /* Sign of each element will be negative for non-real atoms.
 742              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 743              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 744              */
 745             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 746                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 747
 748             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 749             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 750             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 751             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 752             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 753             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 754             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 755             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 756             j_coord_offsetA  = DIM*jnrA;
 757             j_coord_offsetB  = DIM*jnrB;
 758             j_coord_offsetC  = DIM*jnrC;
 759             j_coord_offsetD  = DIM*jnrD;
 760             j_coord_offsetE  = DIM*jnrE;
 761             j_coord_offsetF  = DIM*jnrF;
 762             j_coord_offsetG  = DIM*jnrG;
 763             j_coord_offsetH  = DIM*jnrH;
 764
 765             /* load j atom coordinates */
 766             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 767                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 768                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 769                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 770                                                  &jx0,&jy0,&jz0);
 771
 772             /* Calculate displacement vector */
 773             dx00             = _mm256_sub_ps(ix0,jx0);
 774             dy00             = _mm256_sub_ps(iy0,jy0);
 775             dz00             = _mm256_sub_ps(iz0,jz0);
 776
 777             /* Calculate squared distance and things based on it */
 778             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 779
 780             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 781
 782             /* Load parameters for j particles */
 783             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 784                                                                  charge+jnrC+0,charge+jnrD+0,
 785                                                                  charge+jnrE+0,charge+jnrF+0,
 786                                                                  charge+jnrG+0,charge+jnrH+0);
 787             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 788                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 789                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 790                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 791
 792             /**************************
 793              * CALCULATE INTERACTIONS *
 794              **************************/
 795
 796             r00              = _mm256_mul_ps(rsq00,rinv00);
 797             r00              = _mm256_andnot_ps(dummy_mask,r00);
 798
 799             /* Compute parameters for interactions between i and j atoms */
 800             qq00             = _mm256_mul_ps(iq0,jq0);
 801
 802             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 803             isaprod          = _mm256_mul_ps(isai0,isaj0);
 804             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 805             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 806
 807             /* Calculate generalized born table index - this is a separate table from the normal one,
 808              * but we use the same procedure by multiplying r with scale and truncating to integer.
 809              */
 810             rt               = _mm256_mul_ps(r00,gbscale);
 811             gbitab           = _mm256_cvttps_epi32(rt);
 812             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 813             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 814             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 815             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 816             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 817             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 818             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 819                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 820             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 821                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 822             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 823                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 824             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 825                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 826             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 827             Heps             = _mm256_mul_ps(gbeps,H);
 828             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 829             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 830             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 831
 832             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 833             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 834             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 835             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
 836             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 837             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 838             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 839             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 840             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 841             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 842             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
 843             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
 844             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
 845             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
 846             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 847                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 848             velec            = _mm256_mul_ps(qq00,rinv00);
 849             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 850
 851             fscal            = felec;
 852
 853             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 854
 855             /* Calculate temporary vectorial force */
 856             tx               = _mm256_mul_ps(fscal,dx00);
 857             ty               = _mm256_mul_ps(fscal,dy00);
 858             tz               = _mm256_mul_ps(fscal,dz00);
 859
 860             /* Update vectorial force */
 861             fix0             = _mm256_add_ps(fix0,tx);
 862             fiy0             = _mm256_add_ps(fiy0,ty);
 863             fiz0             = _mm256_add_ps(fiz0,tz);
 864
 865             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 866             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 867             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 868             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 869             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 870             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 871             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 872             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 873             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 874
 875             /* Inner loop uses 56 flops */
 876         }
 877
 878         /* End of innermost loop */
 879
 880         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 881                                                  f+i_coord_offset,fshift+i_shift_offset);
 882
 883         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai0,isai0));
 884         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
 885
 886         /* Increment number of inner iterations */
 887         inneriter                  += j_index_end - j_index_start;
 888
 889         /* Outer loop uses 7 flops */
 890     }
 891
 892     /* Increment number of outer iterations */
 893     outeriter        += nri;
 894
 895     /* Update outer/inner flops */
 896
 897     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*56);
 898 }