src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwNone_GeomW3W3_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_ElecEw_VdwNone_GeomW3W3_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Water3
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwNone_GeomW3W3_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     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  81     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  82     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  83     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  84     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  86     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  87     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  88     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  89     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  90     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  92     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m256i          ewitab;
  96     __m128i          ewitab_lo,ewitab_hi;
  97     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  99     real             *ewtab;
 100     __m256           dummy_mask,cutoff_mask;
 101     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 102     __m256           one     = _mm256_set1_ps(1.0);
 103     __m256           two     = _mm256_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm256_set1_ps(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117
 118     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 119     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 120     beta2            = _mm256_mul_ps(beta,beta);
 121     beta3            = _mm256_mul_ps(beta,beta2);
 122
 123     ewtab            = fr->ic->tabq_coul_FDV0;
 124     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 125     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 130     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 131     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 132
 133     jq0              = _mm256_set1_ps(charge[inr+0]);
 134     jq1              = _mm256_set1_ps(charge[inr+1]);
 135     jq2              = _mm256_set1_ps(charge[inr+2]);
 136     qq00             = _mm256_mul_ps(iq0,jq0);
 137     qq01             = _mm256_mul_ps(iq0,jq1);
 138     qq02             = _mm256_mul_ps(iq0,jq2);
 139     qq10             = _mm256_mul_ps(iq1,jq0);
 140     qq11             = _mm256_mul_ps(iq1,jq1);
 141     qq12             = _mm256_mul_ps(iq1,jq2);
 142     qq20             = _mm256_mul_ps(iq2,jq0);
 143     qq21             = _mm256_mul_ps(iq2,jq1);
 144     qq22             = _mm256_mul_ps(iq2,jq2);
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150     j_coord_offsetC = 0;
 151     j_coord_offsetD = 0;
 152     j_coord_offsetE = 0;
 153     j_coord_offsetF = 0;
 154     j_coord_offsetG = 0;
 155     j_coord_offsetH = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 182
 183         fix0             = _mm256_setzero_ps();
 184         fiy0             = _mm256_setzero_ps();
 185         fiz0             = _mm256_setzero_ps();
 186         fix1             = _mm256_setzero_ps();
 187         fiy1             = _mm256_setzero_ps();
 188         fiz1             = _mm256_setzero_ps();
 189         fix2             = _mm256_setzero_ps();
 190         fiy2             = _mm256_setzero_ps();
 191         fiz2             = _mm256_setzero_ps();
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm256_setzero_ps();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             jnrC             = jjnr[jidx+2];
 204             jnrD             = jjnr[jidx+3];
 205             jnrE             = jjnr[jidx+4];
 206             jnrF             = jjnr[jidx+5];
 207             jnrG             = jjnr[jidx+6];
 208             jnrH             = jjnr[jidx+7];
 209             j_coord_offsetA  = DIM*jnrA;
 210             j_coord_offsetB  = DIM*jnrB;
 211             j_coord_offsetC  = DIM*jnrC;
 212             j_coord_offsetD  = DIM*jnrD;
 213             j_coord_offsetE  = DIM*jnrE;
 214             j_coord_offsetF  = DIM*jnrF;
 215             j_coord_offsetG  = DIM*jnrG;
 216             j_coord_offsetH  = DIM*jnrH;
 217
 218             /* load j atom coordinates */
 219             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 220                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 221                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 222                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 223                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 224
 225             /* Calculate displacement vector */
 226             dx00             = _mm256_sub_ps(ix0,jx0);
 227             dy00             = _mm256_sub_ps(iy0,jy0);
 228             dz00             = _mm256_sub_ps(iz0,jz0);
 229             dx01             = _mm256_sub_ps(ix0,jx1);
 230             dy01             = _mm256_sub_ps(iy0,jy1);
 231             dz01             = _mm256_sub_ps(iz0,jz1);
 232             dx02             = _mm256_sub_ps(ix0,jx2);
 233             dy02             = _mm256_sub_ps(iy0,jy2);
 234             dz02             = _mm256_sub_ps(iz0,jz2);
 235             dx10             = _mm256_sub_ps(ix1,jx0);
 236             dy10             = _mm256_sub_ps(iy1,jy0);
 237             dz10             = _mm256_sub_ps(iz1,jz0);
 238             dx11             = _mm256_sub_ps(ix1,jx1);
 239             dy11             = _mm256_sub_ps(iy1,jy1);
 240             dz11             = _mm256_sub_ps(iz1,jz1);
 241             dx12             = _mm256_sub_ps(ix1,jx2);
 242             dy12             = _mm256_sub_ps(iy1,jy2);
 243             dz12             = _mm256_sub_ps(iz1,jz2);
 244             dx20             = _mm256_sub_ps(ix2,jx0);
 245             dy20             = _mm256_sub_ps(iy2,jy0);
 246             dz20             = _mm256_sub_ps(iz2,jz0);
 247             dx21             = _mm256_sub_ps(ix2,jx1);
 248             dy21             = _mm256_sub_ps(iy2,jy1);
 249             dz21             = _mm256_sub_ps(iz2,jz1);
 250             dx22             = _mm256_sub_ps(ix2,jx2);
 251             dy22             = _mm256_sub_ps(iy2,jy2);
 252             dz22             = _mm256_sub_ps(iz2,jz2);
 253
 254             /* Calculate squared distance and things based on it */
 255             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 256             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 257             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 258             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 259             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 260             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 261             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 262             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 263             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 264
 265             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 266             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 267             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 268             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 269             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 270             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 271             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 272             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 273             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 274
 275             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 276             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 277             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 278             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 279             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 280             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 281             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 282             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 283             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 284
 285             fjx0             = _mm256_setzero_ps();
 286             fjy0             = _mm256_setzero_ps();
 287             fjz0             = _mm256_setzero_ps();
 288             fjx1             = _mm256_setzero_ps();
 289             fjy1             = _mm256_setzero_ps();
 290             fjz1             = _mm256_setzero_ps();
 291             fjx2             = _mm256_setzero_ps();
 292             fjy2             = _mm256_setzero_ps();
 293             fjz2             = _mm256_setzero_ps();
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             r00              = _mm256_mul_ps(rsq00,rinv00);
 300
 301             /* EWALD ELECTROSTATICS */
 302
 303             /* Analytical PME correction */
 304             zeta2            = _mm256_mul_ps(beta2,rsq00);
 305             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 306             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 307             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 308             felec            = _mm256_mul_ps(qq00,felec);
 309             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 310             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 311             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 312             velec            = _mm256_mul_ps(qq00,velec);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velecsum         = _mm256_add_ps(velecsum,velec);
 316
 317             fscal            = felec;
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm256_mul_ps(fscal,dx00);
 321             ty               = _mm256_mul_ps(fscal,dy00);
 322             tz               = _mm256_mul_ps(fscal,dz00);
 323
 324             /* Update vectorial force */
 325             fix0             = _mm256_add_ps(fix0,tx);
 326             fiy0             = _mm256_add_ps(fiy0,ty);
 327             fiz0             = _mm256_add_ps(fiz0,tz);
 328
 329             fjx0             = _mm256_add_ps(fjx0,tx);
 330             fjy0             = _mm256_add_ps(fjy0,ty);
 331             fjz0             = _mm256_add_ps(fjz0,tz);
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             r01              = _mm256_mul_ps(rsq01,rinv01);
 338
 339             /* EWALD ELECTROSTATICS */
 340
 341             /* Analytical PME correction */
 342             zeta2            = _mm256_mul_ps(beta2,rsq01);
 343             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 344             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 345             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 346             felec            = _mm256_mul_ps(qq01,felec);
 347             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 348             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 349             velec            = _mm256_sub_ps(rinv01,pmecorrV);
 350             velec            = _mm256_mul_ps(qq01,velec);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velecsum         = _mm256_add_ps(velecsum,velec);
 354
 355             fscal            = felec;
 356
 357             /* Calculate temporary vectorial force */
 358             tx               = _mm256_mul_ps(fscal,dx01);
 359             ty               = _mm256_mul_ps(fscal,dy01);
 360             tz               = _mm256_mul_ps(fscal,dz01);
 361
 362             /* Update vectorial force */
 363             fix0             = _mm256_add_ps(fix0,tx);
 364             fiy0             = _mm256_add_ps(fiy0,ty);
 365             fiz0             = _mm256_add_ps(fiz0,tz);
 366
 367             fjx1             = _mm256_add_ps(fjx1,tx);
 368             fjy1             = _mm256_add_ps(fjy1,ty);
 369             fjz1             = _mm256_add_ps(fjz1,tz);
 370
 371             /**************************
 372              * CALCULATE INTERACTIONS *
 373              **************************/
 374
 375             r02              = _mm256_mul_ps(rsq02,rinv02);
 376
 377             /* EWALD ELECTROSTATICS */
 378
 379             /* Analytical PME correction */
 380             zeta2            = _mm256_mul_ps(beta2,rsq02);
 381             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 382             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 383             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 384             felec            = _mm256_mul_ps(qq02,felec);
 385             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 386             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 387             velec            = _mm256_sub_ps(rinv02,pmecorrV);
 388             velec            = _mm256_mul_ps(qq02,velec);
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velecsum         = _mm256_add_ps(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm256_mul_ps(fscal,dx02);
 397             ty               = _mm256_mul_ps(fscal,dy02);
 398             tz               = _mm256_mul_ps(fscal,dz02);
 399
 400             /* Update vectorial force */
 401             fix0             = _mm256_add_ps(fix0,tx);
 402             fiy0             = _mm256_add_ps(fiy0,ty);
 403             fiz0             = _mm256_add_ps(fiz0,tz);
 404
 405             fjx2             = _mm256_add_ps(fjx2,tx);
 406             fjy2             = _mm256_add_ps(fjy2,ty);
 407             fjz2             = _mm256_add_ps(fjz2,tz);
 408
 409             /**************************
 410              * CALCULATE INTERACTIONS *
 411              **************************/
 412
 413             r10              = _mm256_mul_ps(rsq10,rinv10);
 414
 415             /* EWALD ELECTROSTATICS */
 416
 417             /* Analytical PME correction */
 418             zeta2            = _mm256_mul_ps(beta2,rsq10);
 419             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 420             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 421             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 422             felec            = _mm256_mul_ps(qq10,felec);
 423             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 424             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 425             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 426             velec            = _mm256_mul_ps(qq10,velec);
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velecsum         = _mm256_add_ps(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             /* Calculate temporary vectorial force */
 434             tx               = _mm256_mul_ps(fscal,dx10);
 435             ty               = _mm256_mul_ps(fscal,dy10);
 436             tz               = _mm256_mul_ps(fscal,dz10);
 437
 438             /* Update vectorial force */
 439             fix1             = _mm256_add_ps(fix1,tx);
 440             fiy1             = _mm256_add_ps(fiy1,ty);
 441             fiz1             = _mm256_add_ps(fiz1,tz);
 442
 443             fjx0             = _mm256_add_ps(fjx0,tx);
 444             fjy0             = _mm256_add_ps(fjy0,ty);
 445             fjz0             = _mm256_add_ps(fjz0,tz);
 446
 447             /**************************
 448              * CALCULATE INTERACTIONS *
 449              **************************/
 450
 451             r11              = _mm256_mul_ps(rsq11,rinv11);
 452
 453             /* EWALD ELECTROSTATICS */
 454
 455             /* Analytical PME correction */
 456             zeta2            = _mm256_mul_ps(beta2,rsq11);
 457             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 458             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 459             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 460             felec            = _mm256_mul_ps(qq11,felec);
 461             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 462             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 463             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 464             velec            = _mm256_mul_ps(qq11,velec);
 465
 466             /* Update potential sum for this i atom from the interaction with this j atom. */
 467             velecsum         = _mm256_add_ps(velecsum,velec);
 468
 469             fscal            = felec;
 470
 471             /* Calculate temporary vectorial force */
 472             tx               = _mm256_mul_ps(fscal,dx11);
 473             ty               = _mm256_mul_ps(fscal,dy11);
 474             tz               = _mm256_mul_ps(fscal,dz11);
 475
 476             /* Update vectorial force */
 477             fix1             = _mm256_add_ps(fix1,tx);
 478             fiy1             = _mm256_add_ps(fiy1,ty);
 479             fiz1             = _mm256_add_ps(fiz1,tz);
 480
 481             fjx1             = _mm256_add_ps(fjx1,tx);
 482             fjy1             = _mm256_add_ps(fjy1,ty);
 483             fjz1             = _mm256_add_ps(fjz1,tz);
 484
 485             /**************************
 486              * CALCULATE INTERACTIONS *
 487              **************************/
 488
 489             r12              = _mm256_mul_ps(rsq12,rinv12);
 490
 491             /* EWALD ELECTROSTATICS */
 492
 493             /* Analytical PME correction */
 494             zeta2            = _mm256_mul_ps(beta2,rsq12);
 495             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 496             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 497             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 498             felec            = _mm256_mul_ps(qq12,felec);
 499             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 500             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 501             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 502             velec            = _mm256_mul_ps(qq12,velec);
 503
 504             /* Update potential sum for this i atom from the interaction with this j atom. */
 505             velecsum         = _mm256_add_ps(velecsum,velec);
 506
 507             fscal            = felec;
 508
 509             /* Calculate temporary vectorial force */
 510             tx               = _mm256_mul_ps(fscal,dx12);
 511             ty               = _mm256_mul_ps(fscal,dy12);
 512             tz               = _mm256_mul_ps(fscal,dz12);
 513
 514             /* Update vectorial force */
 515             fix1             = _mm256_add_ps(fix1,tx);
 516             fiy1             = _mm256_add_ps(fiy1,ty);
 517             fiz1             = _mm256_add_ps(fiz1,tz);
 518
 519             fjx2             = _mm256_add_ps(fjx2,tx);
 520             fjy2             = _mm256_add_ps(fjy2,ty);
 521             fjz2             = _mm256_add_ps(fjz2,tz);
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             r20              = _mm256_mul_ps(rsq20,rinv20);
 528
 529             /* EWALD ELECTROSTATICS */
 530
 531             /* Analytical PME correction */
 532             zeta2            = _mm256_mul_ps(beta2,rsq20);
 533             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 534             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 535             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 536             felec            = _mm256_mul_ps(qq20,felec);
 537             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 538             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 539             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 540             velec            = _mm256_mul_ps(qq20,velec);
 541
 542             /* Update potential sum for this i atom from the interaction with this j atom. */
 543             velecsum         = _mm256_add_ps(velecsum,velec);
 544
 545             fscal            = felec;
 546
 547             /* Calculate temporary vectorial force */
 548             tx               = _mm256_mul_ps(fscal,dx20);
 549             ty               = _mm256_mul_ps(fscal,dy20);
 550             tz               = _mm256_mul_ps(fscal,dz20);
 551
 552             /* Update vectorial force */
 553             fix2             = _mm256_add_ps(fix2,tx);
 554             fiy2             = _mm256_add_ps(fiy2,ty);
 555             fiz2             = _mm256_add_ps(fiz2,tz);
 556
 557             fjx0             = _mm256_add_ps(fjx0,tx);
 558             fjy0             = _mm256_add_ps(fjy0,ty);
 559             fjz0             = _mm256_add_ps(fjz0,tz);
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             r21              = _mm256_mul_ps(rsq21,rinv21);
 566
 567             /* EWALD ELECTROSTATICS */
 568
 569             /* Analytical PME correction */
 570             zeta2            = _mm256_mul_ps(beta2,rsq21);
 571             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
 572             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 573             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 574             felec            = _mm256_mul_ps(qq21,felec);
 575             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 576             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 577             velec            = _mm256_sub_ps(rinv21,pmecorrV);
 578             velec            = _mm256_mul_ps(qq21,velec);
 579
 580             /* Update potential sum for this i atom from the interaction with this j atom. */
 581             velecsum         = _mm256_add_ps(velecsum,velec);
 582
 583             fscal            = felec;
 584
 585             /* Calculate temporary vectorial force */
 586             tx               = _mm256_mul_ps(fscal,dx21);
 587             ty               = _mm256_mul_ps(fscal,dy21);
 588             tz               = _mm256_mul_ps(fscal,dz21);
 589
 590             /* Update vectorial force */
 591             fix2             = _mm256_add_ps(fix2,tx);
 592             fiy2             = _mm256_add_ps(fiy2,ty);
 593             fiz2             = _mm256_add_ps(fiz2,tz);
 594
 595             fjx1             = _mm256_add_ps(fjx1,tx);
 596             fjy1             = _mm256_add_ps(fjy1,ty);
 597             fjz1             = _mm256_add_ps(fjz1,tz);
 598
 599             /**************************
 600              * CALCULATE INTERACTIONS *
 601              **************************/
 602
 603             r22              = _mm256_mul_ps(rsq22,rinv22);
 604
 605             /* EWALD ELECTROSTATICS */
 606
 607             /* Analytical PME correction */
 608             zeta2            = _mm256_mul_ps(beta2,rsq22);
 609             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
 610             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 611             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 612             felec            = _mm256_mul_ps(qq22,felec);
 613             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 614             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 615             velec            = _mm256_sub_ps(rinv22,pmecorrV);
 616             velec            = _mm256_mul_ps(qq22,velec);
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             velecsum         = _mm256_add_ps(velecsum,velec);
 620
 621             fscal            = felec;
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm256_mul_ps(fscal,dx22);
 625             ty               = _mm256_mul_ps(fscal,dy22);
 626             tz               = _mm256_mul_ps(fscal,dz22);
 627
 628             /* Update vectorial force */
 629             fix2             = _mm256_add_ps(fix2,tx);
 630             fiy2             = _mm256_add_ps(fiy2,ty);
 631             fiz2             = _mm256_add_ps(fiz2,tz);
 632
 633             fjx2             = _mm256_add_ps(fjx2,tx);
 634             fjy2             = _mm256_add_ps(fjy2,ty);
 635             fjz2             = _mm256_add_ps(fjz2,tz);
 636
 637             fjptrA             = f+j_coord_offsetA;
 638             fjptrB             = f+j_coord_offsetB;
 639             fjptrC             = f+j_coord_offsetC;
 640             fjptrD             = f+j_coord_offsetD;
 641             fjptrE             = f+j_coord_offsetE;
 642             fjptrF             = f+j_coord_offsetF;
 643             fjptrG             = f+j_coord_offsetG;
 644             fjptrH             = f+j_coord_offsetH;
 645
 646             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 647                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 648
 649             /* Inner loop uses 756 flops */
 650         }
 651
 652         if(jidx<j_index_end)
 653         {
 654
 655             /* Get j neighbor index, and coordinate index */
 656             jnrlistA         = jjnr[jidx];
 657             jnrlistB         = jjnr[jidx+1];
 658             jnrlistC         = jjnr[jidx+2];
 659             jnrlistD         = jjnr[jidx+3];
 660             jnrlistE         = jjnr[jidx+4];
 661             jnrlistF         = jjnr[jidx+5];
 662             jnrlistG         = jjnr[jidx+6];
 663             jnrlistH         = jjnr[jidx+7];
 664             /* Sign of each element will be negative for non-real atoms.
 665              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 666              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 667              */
 668             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 669                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 670
 671             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 672             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 673             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 674             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 675             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 676             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 677             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 678             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 679             j_coord_offsetA  = DIM*jnrA;
 680             j_coord_offsetB  = DIM*jnrB;
 681             j_coord_offsetC  = DIM*jnrC;
 682             j_coord_offsetD  = DIM*jnrD;
 683             j_coord_offsetE  = DIM*jnrE;
 684             j_coord_offsetF  = DIM*jnrF;
 685             j_coord_offsetG  = DIM*jnrG;
 686             j_coord_offsetH  = DIM*jnrH;
 687
 688             /* load j atom coordinates */
 689             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 690                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 691                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 692                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 693                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 694
 695             /* Calculate displacement vector */
 696             dx00             = _mm256_sub_ps(ix0,jx0);
 697             dy00             = _mm256_sub_ps(iy0,jy0);
 698             dz00             = _mm256_sub_ps(iz0,jz0);
 699             dx01             = _mm256_sub_ps(ix0,jx1);
 700             dy01             = _mm256_sub_ps(iy0,jy1);
 701             dz01             = _mm256_sub_ps(iz0,jz1);
 702             dx02             = _mm256_sub_ps(ix0,jx2);
 703             dy02             = _mm256_sub_ps(iy0,jy2);
 704             dz02             = _mm256_sub_ps(iz0,jz2);
 705             dx10             = _mm256_sub_ps(ix1,jx0);
 706             dy10             = _mm256_sub_ps(iy1,jy0);
 707             dz10             = _mm256_sub_ps(iz1,jz0);
 708             dx11             = _mm256_sub_ps(ix1,jx1);
 709             dy11             = _mm256_sub_ps(iy1,jy1);
 710             dz11             = _mm256_sub_ps(iz1,jz1);
 711             dx12             = _mm256_sub_ps(ix1,jx2);
 712             dy12             = _mm256_sub_ps(iy1,jy2);
 713             dz12             = _mm256_sub_ps(iz1,jz2);
 714             dx20             = _mm256_sub_ps(ix2,jx0);
 715             dy20             = _mm256_sub_ps(iy2,jy0);
 716             dz20             = _mm256_sub_ps(iz2,jz0);
 717             dx21             = _mm256_sub_ps(ix2,jx1);
 718             dy21             = _mm256_sub_ps(iy2,jy1);
 719             dz21             = _mm256_sub_ps(iz2,jz1);
 720             dx22             = _mm256_sub_ps(ix2,jx2);
 721             dy22             = _mm256_sub_ps(iy2,jy2);
 722             dz22             = _mm256_sub_ps(iz2,jz2);
 723
 724             /* Calculate squared distance and things based on it */
 725             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 726             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 727             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 728             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 729             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 730             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 731             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 732             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 733             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 734
 735             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 736             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 737             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 738             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 739             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 740             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 741             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 742             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 743             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 744
 745             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 746             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 747             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 748             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 749             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 750             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 751             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 752             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 753             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 754
 755             fjx0             = _mm256_setzero_ps();
 756             fjy0             = _mm256_setzero_ps();
 757             fjz0             = _mm256_setzero_ps();
 758             fjx1             = _mm256_setzero_ps();
 759             fjy1             = _mm256_setzero_ps();
 760             fjz1             = _mm256_setzero_ps();
 761             fjx2             = _mm256_setzero_ps();
 762             fjy2             = _mm256_setzero_ps();
 763             fjz2             = _mm256_setzero_ps();
 764
 765             /**************************
 766              * CALCULATE INTERACTIONS *
 767              **************************/
 768
 769             r00              = _mm256_mul_ps(rsq00,rinv00);
 770             r00              = _mm256_andnot_ps(dummy_mask,r00);
 771
 772             /* EWALD ELECTROSTATICS */
 773
 774             /* Analytical PME correction */
 775             zeta2            = _mm256_mul_ps(beta2,rsq00);
 776             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 777             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 778             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 779             felec            = _mm256_mul_ps(qq00,felec);
 780             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 781             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 782             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 783             velec            = _mm256_mul_ps(qq00,velec);
 784
 785             /* Update potential sum for this i atom from the interaction with this j atom. */
 786             velec            = _mm256_andnot_ps(dummy_mask,velec);
 787             velecsum         = _mm256_add_ps(velecsum,velec);
 788
 789             fscal            = felec;
 790
 791             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 792
 793             /* Calculate temporary vectorial force */
 794             tx               = _mm256_mul_ps(fscal,dx00);
 795             ty               = _mm256_mul_ps(fscal,dy00);
 796             tz               = _mm256_mul_ps(fscal,dz00);
 797
 798             /* Update vectorial force */
 799             fix0             = _mm256_add_ps(fix0,tx);
 800             fiy0             = _mm256_add_ps(fiy0,ty);
 801             fiz0             = _mm256_add_ps(fiz0,tz);
 802
 803             fjx0             = _mm256_add_ps(fjx0,tx);
 804             fjy0             = _mm256_add_ps(fjy0,ty);
 805             fjz0             = _mm256_add_ps(fjz0,tz);
 806
 807             /**************************
 808              * CALCULATE INTERACTIONS *
 809              **************************/
 810
 811             r01              = _mm256_mul_ps(rsq01,rinv01);
 812             r01              = _mm256_andnot_ps(dummy_mask,r01);
 813
 814             /* EWALD ELECTROSTATICS */
 815
 816             /* Analytical PME correction */
 817             zeta2            = _mm256_mul_ps(beta2,rsq01);
 818             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
 819             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 820             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 821             felec            = _mm256_mul_ps(qq01,felec);
 822             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 823             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 824             velec            = _mm256_sub_ps(rinv01,pmecorrV);
 825             velec            = _mm256_mul_ps(qq01,velec);
 826
 827             /* Update potential sum for this i atom from the interaction with this j atom. */
 828             velec            = _mm256_andnot_ps(dummy_mask,velec);
 829             velecsum         = _mm256_add_ps(velecsum,velec);
 830
 831             fscal            = felec;
 832
 833             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 834
 835             /* Calculate temporary vectorial force */
 836             tx               = _mm256_mul_ps(fscal,dx01);
 837             ty               = _mm256_mul_ps(fscal,dy01);
 838             tz               = _mm256_mul_ps(fscal,dz01);
 839
 840             /* Update vectorial force */
 841             fix0             = _mm256_add_ps(fix0,tx);
 842             fiy0             = _mm256_add_ps(fiy0,ty);
 843             fiz0             = _mm256_add_ps(fiz0,tz);
 844
 845             fjx1             = _mm256_add_ps(fjx1,tx);
 846             fjy1             = _mm256_add_ps(fjy1,ty);
 847             fjz1             = _mm256_add_ps(fjz1,tz);
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             r02              = _mm256_mul_ps(rsq02,rinv02);
 854             r02              = _mm256_andnot_ps(dummy_mask,r02);
 855
 856             /* EWALD ELECTROSTATICS */
 857
 858             /* Analytical PME correction */
 859             zeta2            = _mm256_mul_ps(beta2,rsq02);
 860             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
 861             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 862             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 863             felec            = _mm256_mul_ps(qq02,felec);
 864             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 865             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 866             velec            = _mm256_sub_ps(rinv02,pmecorrV);
 867             velec            = _mm256_mul_ps(qq02,velec);
 868
 869             /* Update potential sum for this i atom from the interaction with this j atom. */
 870             velec            = _mm256_andnot_ps(dummy_mask,velec);
 871             velecsum         = _mm256_add_ps(velecsum,velec);
 872
 873             fscal            = felec;
 874
 875             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 876
 877             /* Calculate temporary vectorial force */
 878             tx               = _mm256_mul_ps(fscal,dx02);
 879             ty               = _mm256_mul_ps(fscal,dy02);
 880             tz               = _mm256_mul_ps(fscal,dz02);
 881
 882             /* Update vectorial force */
 883             fix0             = _mm256_add_ps(fix0,tx);
 884             fiy0             = _mm256_add_ps(fiy0,ty);
 885             fiz0             = _mm256_add_ps(fiz0,tz);
 886
 887             fjx2             = _mm256_add_ps(fjx2,tx);
 888             fjy2             = _mm256_add_ps(fjy2,ty);
 889             fjz2             = _mm256_add_ps(fjz2,tz);
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             r10              = _mm256_mul_ps(rsq10,rinv10);
 896             r10              = _mm256_andnot_ps(dummy_mask,r10);
 897
 898             /* EWALD ELECTROSTATICS */
 899
 900             /* Analytical PME correction */
 901             zeta2            = _mm256_mul_ps(beta2,rsq10);
 902             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 903             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 904             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 905             felec            = _mm256_mul_ps(qq10,felec);
 906             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 907             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 908             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 909             velec            = _mm256_mul_ps(qq10,velec);
 910
 911             /* Update potential sum for this i atom from the interaction with this j atom. */
 912             velec            = _mm256_andnot_ps(dummy_mask,velec);
 913             velecsum         = _mm256_add_ps(velecsum,velec);
 914
 915             fscal            = felec;
 916
 917             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 918
 919             /* Calculate temporary vectorial force */
 920             tx               = _mm256_mul_ps(fscal,dx10);
 921             ty               = _mm256_mul_ps(fscal,dy10);
 922             tz               = _mm256_mul_ps(fscal,dz10);
 923
 924             /* Update vectorial force */
 925             fix1             = _mm256_add_ps(fix1,tx);
 926             fiy1             = _mm256_add_ps(fiy1,ty);
 927             fiz1             = _mm256_add_ps(fiz1,tz);
 928
 929             fjx0             = _mm256_add_ps(fjx0,tx);
 930             fjy0             = _mm256_add_ps(fjy0,ty);
 931             fjz0             = _mm256_add_ps(fjz0,tz);
 932
 933             /**************************
 934              * CALCULATE INTERACTIONS *
 935              **************************/
 936
 937             r11              = _mm256_mul_ps(rsq11,rinv11);
 938             r11              = _mm256_andnot_ps(dummy_mask,r11);
 939
 940             /* EWALD ELECTROSTATICS */
 941
 942             /* Analytical PME correction */
 943             zeta2            = _mm256_mul_ps(beta2,rsq11);
 944             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
 945             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 946             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 947             felec            = _mm256_mul_ps(qq11,felec);
 948             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 949             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 950             velec            = _mm256_sub_ps(rinv11,pmecorrV);
 951             velec            = _mm256_mul_ps(qq11,velec);
 952
 953             /* Update potential sum for this i atom from the interaction with this j atom. */
 954             velec            = _mm256_andnot_ps(dummy_mask,velec);
 955             velecsum         = _mm256_add_ps(velecsum,velec);
 956
 957             fscal            = felec;
 958
 959             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 960
 961             /* Calculate temporary vectorial force */
 962             tx               = _mm256_mul_ps(fscal,dx11);
 963             ty               = _mm256_mul_ps(fscal,dy11);
 964             tz               = _mm256_mul_ps(fscal,dz11);
 965
 966             /* Update vectorial force */
 967             fix1             = _mm256_add_ps(fix1,tx);
 968             fiy1             = _mm256_add_ps(fiy1,ty);
 969             fiz1             = _mm256_add_ps(fiz1,tz);
 970
 971             fjx1             = _mm256_add_ps(fjx1,tx);
 972             fjy1             = _mm256_add_ps(fjy1,ty);
 973             fjz1             = _mm256_add_ps(fjz1,tz);
 974
 975             /**************************
 976              * CALCULATE INTERACTIONS *
 977              **************************/
 978
 979             r12              = _mm256_mul_ps(rsq12,rinv12);
 980             r12              = _mm256_andnot_ps(dummy_mask,r12);
 981
 982             /* EWALD ELECTROSTATICS */
 983
 984             /* Analytical PME correction */
 985             zeta2            = _mm256_mul_ps(beta2,rsq12);
 986             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
 987             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 988             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 989             felec            = _mm256_mul_ps(qq12,felec);
 990             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 991             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 992             velec            = _mm256_sub_ps(rinv12,pmecorrV);
 993             velec            = _mm256_mul_ps(qq12,velec);
 994
 995             /* Update potential sum for this i atom from the interaction with this j atom. */
 996             velec            = _mm256_andnot_ps(dummy_mask,velec);
 997             velecsum         = _mm256_add_ps(velecsum,velec);
 998
 999             fscal            = felec;
1000
1001             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1002
1003             /* Calculate temporary vectorial force */
1004             tx               = _mm256_mul_ps(fscal,dx12);
1005             ty               = _mm256_mul_ps(fscal,dy12);
1006             tz               = _mm256_mul_ps(fscal,dz12);
1007
1008             /* Update vectorial force */
1009             fix1             = _mm256_add_ps(fix1,tx);
1010             fiy1             = _mm256_add_ps(fiy1,ty);
1011             fiz1             = _mm256_add_ps(fiz1,tz);
1012
1013             fjx2             = _mm256_add_ps(fjx2,tx);
1014             fjy2             = _mm256_add_ps(fjy2,ty);
1015             fjz2             = _mm256_add_ps(fjz2,tz);
1016
1017             /**************************
1018              * CALCULATE INTERACTIONS *
1019              **************************/
1020
1021             r20              = _mm256_mul_ps(rsq20,rinv20);
1022             r20              = _mm256_andnot_ps(dummy_mask,r20);
1023
1024             /* EWALD ELECTROSTATICS */
1025
1026             /* Analytical PME correction */
1027             zeta2            = _mm256_mul_ps(beta2,rsq20);
1028             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1029             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1030             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1031             felec            = _mm256_mul_ps(qq20,felec);
1032             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1033             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1034             velec            = _mm256_sub_ps(rinv20,pmecorrV);
1035             velec            = _mm256_mul_ps(qq20,velec);
1036
1037             /* Update potential sum for this i atom from the interaction with this j atom. */
1038             velec            = _mm256_andnot_ps(dummy_mask,velec);
1039             velecsum         = _mm256_add_ps(velecsum,velec);
1040
1041             fscal            = felec;
1042
1043             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1044
1045             /* Calculate temporary vectorial force */
1046             tx               = _mm256_mul_ps(fscal,dx20);
1047             ty               = _mm256_mul_ps(fscal,dy20);
1048             tz               = _mm256_mul_ps(fscal,dz20);
1049
1050             /* Update vectorial force */
1051             fix2             = _mm256_add_ps(fix2,tx);
1052             fiy2             = _mm256_add_ps(fiy2,ty);
1053             fiz2             = _mm256_add_ps(fiz2,tz);
1054
1055             fjx0             = _mm256_add_ps(fjx0,tx);
1056             fjy0             = _mm256_add_ps(fjy0,ty);
1057             fjz0             = _mm256_add_ps(fjz0,tz);
1058
1059             /**************************
1060              * CALCULATE INTERACTIONS *
1061              **************************/
1062
1063             r21              = _mm256_mul_ps(rsq21,rinv21);
1064             r21              = _mm256_andnot_ps(dummy_mask,r21);
1065
1066             /* EWALD ELECTROSTATICS */
1067
1068             /* Analytical PME correction */
1069             zeta2            = _mm256_mul_ps(beta2,rsq21);
1070             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1071             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1072             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1073             felec            = _mm256_mul_ps(qq21,felec);
1074             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1075             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1076             velec            = _mm256_sub_ps(rinv21,pmecorrV);
1077             velec            = _mm256_mul_ps(qq21,velec);
1078
1079             /* Update potential sum for this i atom from the interaction with this j atom. */
1080             velec            = _mm256_andnot_ps(dummy_mask,velec);
1081             velecsum         = _mm256_add_ps(velecsum,velec);
1082
1083             fscal            = felec;
1084
1085             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1086
1087             /* Calculate temporary vectorial force */
1088             tx               = _mm256_mul_ps(fscal,dx21);
1089             ty               = _mm256_mul_ps(fscal,dy21);
1090             tz               = _mm256_mul_ps(fscal,dz21);
1091
1092             /* Update vectorial force */
1093             fix2             = _mm256_add_ps(fix2,tx);
1094             fiy2             = _mm256_add_ps(fiy2,ty);
1095             fiz2             = _mm256_add_ps(fiz2,tz);
1096
1097             fjx1             = _mm256_add_ps(fjx1,tx);
1098             fjy1             = _mm256_add_ps(fjy1,ty);
1099             fjz1             = _mm256_add_ps(fjz1,tz);
1100
1101             /**************************
1102              * CALCULATE INTERACTIONS *
1103              **************************/
1104
1105             r22              = _mm256_mul_ps(rsq22,rinv22);
1106             r22              = _mm256_andnot_ps(dummy_mask,r22);
1107
1108             /* EWALD ELECTROSTATICS */
1109
1110             /* Analytical PME correction */
1111             zeta2            = _mm256_mul_ps(beta2,rsq22);
1112             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1113             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1114             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1115             felec            = _mm256_mul_ps(qq22,felec);
1116             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
1117             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
1118             velec            = _mm256_sub_ps(rinv22,pmecorrV);
1119             velec            = _mm256_mul_ps(qq22,velec);
1120
1121             /* Update potential sum for this i atom from the interaction with this j atom. */
1122             velec            = _mm256_andnot_ps(dummy_mask,velec);
1123             velecsum         = _mm256_add_ps(velecsum,velec);
1124
1125             fscal            = felec;
1126
1127             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1128
1129             /* Calculate temporary vectorial force */
1130             tx               = _mm256_mul_ps(fscal,dx22);
1131             ty               = _mm256_mul_ps(fscal,dy22);
1132             tz               = _mm256_mul_ps(fscal,dz22);
1133
1134             /* Update vectorial force */
1135             fix2             = _mm256_add_ps(fix2,tx);
1136             fiy2             = _mm256_add_ps(fiy2,ty);
1137             fiz2             = _mm256_add_ps(fiz2,tz);
1138
1139             fjx2             = _mm256_add_ps(fjx2,tx);
1140             fjy2             = _mm256_add_ps(fjy2,ty);
1141             fjz2             = _mm256_add_ps(fjz2,tz);
1142
1143             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1144             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1145             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1146             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1147             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1148             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1149             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1150             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1151
1152             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1153                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1154
1155             /* Inner loop uses 765 flops */
1156         }
1157
1158         /* End of innermost loop */
1159
1160         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1161                                                  f+i_coord_offset,fshift+i_shift_offset);
1162
1163         ggid                        = gid[iidx];
1164         /* Update potential energies */
1165         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1166
1167         /* Increment number of inner iterations */
1168         inneriter                  += j_index_end - j_index_start;
1169
1170         /* Outer loop uses 19 flops */
1171     }
1172
1173     /* Increment number of outer iterations */
1174     outeriter        += nri;
1175
1176     /* Update outer/inner flops */
1177
1178     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*765);
1179 }
1180 /*
1181  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3W3_F_avx_256_single
1182  * Electrostatics interaction: Ewald
1183  * VdW interaction:            None
1184  * Geometry:                   Water3-Water3
1185  * Calculate force/pot:        Force
1186  */
1187 void
1188 nb_kernel_ElecEw_VdwNone_GeomW3W3_F_avx_256_single
1189                     (t_nblist * gmx_restrict                nlist,
1190                      rvec * gmx_restrict                    xx,
1191                      rvec * gmx_restrict                    ff,
1192                      t_forcerec * gmx_restrict              fr,
1193                      t_mdatoms * gmx_restrict               mdatoms,
1194                      nb_kernel_data_t * gmx_restrict        kernel_data,
1195                      t_nrnb * gmx_restrict                  nrnb)
1196 {
1197     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1198      * just 0 for non-waters.
1199      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1200      * jnr indices corresponding to data put in the four positions in the SIMD register.
1201      */
1202     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1203     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1204     int              jnrA,jnrB,jnrC,jnrD;
1205     int              jnrE,jnrF,jnrG,jnrH;
1206     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1207     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1208     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1209     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1210     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1211     real             rcutoff_scalar;
1212     real             *shiftvec,*fshift,*x,*f;
1213     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1214     real             scratch[4*DIM];
1215     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1216     real *           vdwioffsetptr0;
1217     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1218     real *           vdwioffsetptr1;
1219     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1220     real *           vdwioffsetptr2;
1221     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1222     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1223     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1224     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1225     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1226     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1227     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1228     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1229     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1230     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1231     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1232     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1233     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1234     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1235     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1236     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1237     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1238     real             *charge;
1239     __m256i          ewitab;
1240     __m128i          ewitab_lo,ewitab_hi;
1241     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1242     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1243     real             *ewtab;
1244     __m256           dummy_mask,cutoff_mask;
1245     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1246     __m256           one     = _mm256_set1_ps(1.0);
1247     __m256           two     = _mm256_set1_ps(2.0);
1248     x                = xx[0];
1249     f                = ff[0];
1250
1251     nri              = nlist->nri;
1252     iinr             = nlist->iinr;
1253     jindex           = nlist->jindex;
1254     jjnr             = nlist->jjnr;
1255     shiftidx         = nlist->shift;
1256     gid              = nlist->gid;
1257     shiftvec         = fr->shift_vec[0];
1258     fshift           = fr->fshift[0];
1259     facel            = _mm256_set1_ps(fr->epsfac);
1260     charge           = mdatoms->chargeA;
1261
1262     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1263     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
1264     beta2            = _mm256_mul_ps(beta,beta);
1265     beta3            = _mm256_mul_ps(beta,beta2);
1266
1267     ewtab            = fr->ic->tabq_coul_F;
1268     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1269     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1270
1271     /* Setup water-specific parameters */
1272     inr              = nlist->iinr[0];
1273     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1274     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1275     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1276
1277     jq0              = _mm256_set1_ps(charge[inr+0]);
1278     jq1              = _mm256_set1_ps(charge[inr+1]);
1279     jq2              = _mm256_set1_ps(charge[inr+2]);
1280     qq00             = _mm256_mul_ps(iq0,jq0);
1281     qq01             = _mm256_mul_ps(iq0,jq1);
1282     qq02             = _mm256_mul_ps(iq0,jq2);
1283     qq10             = _mm256_mul_ps(iq1,jq0);
1284     qq11             = _mm256_mul_ps(iq1,jq1);
1285     qq12             = _mm256_mul_ps(iq1,jq2);
1286     qq20             = _mm256_mul_ps(iq2,jq0);
1287     qq21             = _mm256_mul_ps(iq2,jq1);
1288     qq22             = _mm256_mul_ps(iq2,jq2);
1289
1290     /* Avoid stupid compiler warnings */
1291     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1292     j_coord_offsetA = 0;
1293     j_coord_offsetB = 0;
1294     j_coord_offsetC = 0;
1295     j_coord_offsetD = 0;
1296     j_coord_offsetE = 0;
1297     j_coord_offsetF = 0;
1298     j_coord_offsetG = 0;
1299     j_coord_offsetH = 0;
1300
1301     outeriter        = 0;
1302     inneriter        = 0;
1303
1304     for(iidx=0;iidx<4*DIM;iidx++)
1305     {
1306         scratch[iidx] = 0.0;
1307     }
1308
1309     /* Start outer loop over neighborlists */
1310     for(iidx=0; iidx<nri; iidx++)
1311     {
1312         /* Load shift vector for this list */
1313         i_shift_offset   = DIM*shiftidx[iidx];
1314
1315         /* Load limits for loop over neighbors */
1316         j_index_start    = jindex[iidx];
1317         j_index_end      = jindex[iidx+1];
1318
1319         /* Get outer coordinate index */
1320         inr              = iinr[iidx];
1321         i_coord_offset   = DIM*inr;
1322
1323         /* Load i particle coords and add shift vector */
1324         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1325                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1326
1327         fix0             = _mm256_setzero_ps();
1328         fiy0             = _mm256_setzero_ps();
1329         fiz0             = _mm256_setzero_ps();
1330         fix1             = _mm256_setzero_ps();
1331         fiy1             = _mm256_setzero_ps();
1332         fiz1             = _mm256_setzero_ps();
1333         fix2             = _mm256_setzero_ps();
1334         fiy2             = _mm256_setzero_ps();
1335         fiz2             = _mm256_setzero_ps();
1336
1337         /* Start inner kernel loop */
1338         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1339         {
1340
1341             /* Get j neighbor index, and coordinate index */
1342             jnrA             = jjnr[jidx];
1343             jnrB             = jjnr[jidx+1];
1344             jnrC             = jjnr[jidx+2];
1345             jnrD             = jjnr[jidx+3];
1346             jnrE             = jjnr[jidx+4];
1347             jnrF             = jjnr[jidx+5];
1348             jnrG             = jjnr[jidx+6];
1349             jnrH             = jjnr[jidx+7];
1350             j_coord_offsetA  = DIM*jnrA;
1351             j_coord_offsetB  = DIM*jnrB;
1352             j_coord_offsetC  = DIM*jnrC;
1353             j_coord_offsetD  = DIM*jnrD;
1354             j_coord_offsetE  = DIM*jnrE;
1355             j_coord_offsetF  = DIM*jnrF;
1356             j_coord_offsetG  = DIM*jnrG;
1357             j_coord_offsetH  = DIM*jnrH;
1358
1359             /* load j atom coordinates */
1360             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1361                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1362                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1363                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1364                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1365
1366             /* Calculate displacement vector */
1367             dx00             = _mm256_sub_ps(ix0,jx0);
1368             dy00             = _mm256_sub_ps(iy0,jy0);
1369             dz00             = _mm256_sub_ps(iz0,jz0);
1370             dx01             = _mm256_sub_ps(ix0,jx1);
1371             dy01             = _mm256_sub_ps(iy0,jy1);
1372             dz01             = _mm256_sub_ps(iz0,jz1);
1373             dx02             = _mm256_sub_ps(ix0,jx2);
1374             dy02             = _mm256_sub_ps(iy0,jy2);
1375             dz02             = _mm256_sub_ps(iz0,jz2);
1376             dx10             = _mm256_sub_ps(ix1,jx0);
1377             dy10             = _mm256_sub_ps(iy1,jy0);
1378             dz10             = _mm256_sub_ps(iz1,jz0);
1379             dx11             = _mm256_sub_ps(ix1,jx1);
1380             dy11             = _mm256_sub_ps(iy1,jy1);
1381             dz11             = _mm256_sub_ps(iz1,jz1);
1382             dx12             = _mm256_sub_ps(ix1,jx2);
1383             dy12             = _mm256_sub_ps(iy1,jy2);
1384             dz12             = _mm256_sub_ps(iz1,jz2);
1385             dx20             = _mm256_sub_ps(ix2,jx0);
1386             dy20             = _mm256_sub_ps(iy2,jy0);
1387             dz20             = _mm256_sub_ps(iz2,jz0);
1388             dx21             = _mm256_sub_ps(ix2,jx1);
1389             dy21             = _mm256_sub_ps(iy2,jy1);
1390             dz21             = _mm256_sub_ps(iz2,jz1);
1391             dx22             = _mm256_sub_ps(ix2,jx2);
1392             dy22             = _mm256_sub_ps(iy2,jy2);
1393             dz22             = _mm256_sub_ps(iz2,jz2);
1394
1395             /* Calculate squared distance and things based on it */
1396             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1397             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1398             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1399             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1400             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1401             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1402             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1403             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1404             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1405
1406             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1407             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1408             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1409             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1410             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1411             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1412             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1413             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1414             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1415
1416             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1417             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1418             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1419             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1420             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1421             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1422             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1423             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1424             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1425
1426             fjx0             = _mm256_setzero_ps();
1427             fjy0             = _mm256_setzero_ps();
1428             fjz0             = _mm256_setzero_ps();
1429             fjx1             = _mm256_setzero_ps();
1430             fjy1             = _mm256_setzero_ps();
1431             fjz1             = _mm256_setzero_ps();
1432             fjx2             = _mm256_setzero_ps();
1433             fjy2             = _mm256_setzero_ps();
1434             fjz2             = _mm256_setzero_ps();
1435
1436             /**************************
1437              * CALCULATE INTERACTIONS *
1438              **************************/
1439
1440             r00              = _mm256_mul_ps(rsq00,rinv00);
1441
1442             /* EWALD ELECTROSTATICS */
1443
1444             /* Analytical PME correction */
1445             zeta2            = _mm256_mul_ps(beta2,rsq00);
1446             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1447             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1448             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1449             felec            = _mm256_mul_ps(qq00,felec);
1450
1451             fscal            = felec;
1452
1453             /* Calculate temporary vectorial force */
1454             tx               = _mm256_mul_ps(fscal,dx00);
1455             ty               = _mm256_mul_ps(fscal,dy00);
1456             tz               = _mm256_mul_ps(fscal,dz00);
1457
1458             /* Update vectorial force */
1459             fix0             = _mm256_add_ps(fix0,tx);
1460             fiy0             = _mm256_add_ps(fiy0,ty);
1461             fiz0             = _mm256_add_ps(fiz0,tz);
1462
1463             fjx0             = _mm256_add_ps(fjx0,tx);
1464             fjy0             = _mm256_add_ps(fjy0,ty);
1465             fjz0             = _mm256_add_ps(fjz0,tz);
1466
1467             /**************************
1468              * CALCULATE INTERACTIONS *
1469              **************************/
1470
1471             r01              = _mm256_mul_ps(rsq01,rinv01);
1472
1473             /* EWALD ELECTROSTATICS */
1474
1475             /* Analytical PME correction */
1476             zeta2            = _mm256_mul_ps(beta2,rsq01);
1477             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1478             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1479             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1480             felec            = _mm256_mul_ps(qq01,felec);
1481
1482             fscal            = felec;
1483
1484             /* Calculate temporary vectorial force */
1485             tx               = _mm256_mul_ps(fscal,dx01);
1486             ty               = _mm256_mul_ps(fscal,dy01);
1487             tz               = _mm256_mul_ps(fscal,dz01);
1488
1489             /* Update vectorial force */
1490             fix0             = _mm256_add_ps(fix0,tx);
1491             fiy0             = _mm256_add_ps(fiy0,ty);
1492             fiz0             = _mm256_add_ps(fiz0,tz);
1493
1494             fjx1             = _mm256_add_ps(fjx1,tx);
1495             fjy1             = _mm256_add_ps(fjy1,ty);
1496             fjz1             = _mm256_add_ps(fjz1,tz);
1497
1498             /**************************
1499              * CALCULATE INTERACTIONS *
1500              **************************/
1501
1502             r02              = _mm256_mul_ps(rsq02,rinv02);
1503
1504             /* EWALD ELECTROSTATICS */
1505
1506             /* Analytical PME correction */
1507             zeta2            = _mm256_mul_ps(beta2,rsq02);
1508             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1509             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1510             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1511             felec            = _mm256_mul_ps(qq02,felec);
1512
1513             fscal            = felec;
1514
1515             /* Calculate temporary vectorial force */
1516             tx               = _mm256_mul_ps(fscal,dx02);
1517             ty               = _mm256_mul_ps(fscal,dy02);
1518             tz               = _mm256_mul_ps(fscal,dz02);
1519
1520             /* Update vectorial force */
1521             fix0             = _mm256_add_ps(fix0,tx);
1522             fiy0             = _mm256_add_ps(fiy0,ty);
1523             fiz0             = _mm256_add_ps(fiz0,tz);
1524
1525             fjx2             = _mm256_add_ps(fjx2,tx);
1526             fjy2             = _mm256_add_ps(fjy2,ty);
1527             fjz2             = _mm256_add_ps(fjz2,tz);
1528
1529             /**************************
1530              * CALCULATE INTERACTIONS *
1531              **************************/
1532
1533             r10              = _mm256_mul_ps(rsq10,rinv10);
1534
1535             /* EWALD ELECTROSTATICS */
1536
1537             /* Analytical PME correction */
1538             zeta2            = _mm256_mul_ps(beta2,rsq10);
1539             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1540             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1541             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1542             felec            = _mm256_mul_ps(qq10,felec);
1543
1544             fscal            = felec;
1545
1546             /* Calculate temporary vectorial force */
1547             tx               = _mm256_mul_ps(fscal,dx10);
1548             ty               = _mm256_mul_ps(fscal,dy10);
1549             tz               = _mm256_mul_ps(fscal,dz10);
1550
1551             /* Update vectorial force */
1552             fix1             = _mm256_add_ps(fix1,tx);
1553             fiy1             = _mm256_add_ps(fiy1,ty);
1554             fiz1             = _mm256_add_ps(fiz1,tz);
1555
1556             fjx0             = _mm256_add_ps(fjx0,tx);
1557             fjy0             = _mm256_add_ps(fjy0,ty);
1558             fjz0             = _mm256_add_ps(fjz0,tz);
1559
1560             /**************************
1561              * CALCULATE INTERACTIONS *
1562              **************************/
1563
1564             r11              = _mm256_mul_ps(rsq11,rinv11);
1565
1566             /* EWALD ELECTROSTATICS */
1567
1568             /* Analytical PME correction */
1569             zeta2            = _mm256_mul_ps(beta2,rsq11);
1570             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1571             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1572             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1573             felec            = _mm256_mul_ps(qq11,felec);
1574
1575             fscal            = felec;
1576
1577             /* Calculate temporary vectorial force */
1578             tx               = _mm256_mul_ps(fscal,dx11);
1579             ty               = _mm256_mul_ps(fscal,dy11);
1580             tz               = _mm256_mul_ps(fscal,dz11);
1581
1582             /* Update vectorial force */
1583             fix1             = _mm256_add_ps(fix1,tx);
1584             fiy1             = _mm256_add_ps(fiy1,ty);
1585             fiz1             = _mm256_add_ps(fiz1,tz);
1586
1587             fjx1             = _mm256_add_ps(fjx1,tx);
1588             fjy1             = _mm256_add_ps(fjy1,ty);
1589             fjz1             = _mm256_add_ps(fjz1,tz);
1590
1591             /**************************
1592              * CALCULATE INTERACTIONS *
1593              **************************/
1594
1595             r12              = _mm256_mul_ps(rsq12,rinv12);
1596
1597             /* EWALD ELECTROSTATICS */
1598
1599             /* Analytical PME correction */
1600             zeta2            = _mm256_mul_ps(beta2,rsq12);
1601             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
1602             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1603             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1604             felec            = _mm256_mul_ps(qq12,felec);
1605
1606             fscal            = felec;
1607
1608             /* Calculate temporary vectorial force */
1609             tx               = _mm256_mul_ps(fscal,dx12);
1610             ty               = _mm256_mul_ps(fscal,dy12);
1611             tz               = _mm256_mul_ps(fscal,dz12);
1612
1613             /* Update vectorial force */
1614             fix1             = _mm256_add_ps(fix1,tx);
1615             fiy1             = _mm256_add_ps(fiy1,ty);
1616             fiz1             = _mm256_add_ps(fiz1,tz);
1617
1618             fjx2             = _mm256_add_ps(fjx2,tx);
1619             fjy2             = _mm256_add_ps(fjy2,ty);
1620             fjz2             = _mm256_add_ps(fjz2,tz);
1621
1622             /**************************
1623              * CALCULATE INTERACTIONS *
1624              **************************/
1625
1626             r20              = _mm256_mul_ps(rsq20,rinv20);
1627
1628             /* EWALD ELECTROSTATICS */
1629
1630             /* Analytical PME correction */
1631             zeta2            = _mm256_mul_ps(beta2,rsq20);
1632             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1633             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1634             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1635             felec            = _mm256_mul_ps(qq20,felec);
1636
1637             fscal            = felec;
1638
1639             /* Calculate temporary vectorial force */
1640             tx               = _mm256_mul_ps(fscal,dx20);
1641             ty               = _mm256_mul_ps(fscal,dy20);
1642             tz               = _mm256_mul_ps(fscal,dz20);
1643
1644             /* Update vectorial force */
1645             fix2             = _mm256_add_ps(fix2,tx);
1646             fiy2             = _mm256_add_ps(fiy2,ty);
1647             fiz2             = _mm256_add_ps(fiz2,tz);
1648
1649             fjx0             = _mm256_add_ps(fjx0,tx);
1650             fjy0             = _mm256_add_ps(fjy0,ty);
1651             fjz0             = _mm256_add_ps(fjz0,tz);
1652
1653             /**************************
1654              * CALCULATE INTERACTIONS *
1655              **************************/
1656
1657             r21              = _mm256_mul_ps(rsq21,rinv21);
1658
1659             /* EWALD ELECTROSTATICS */
1660
1661             /* Analytical PME correction */
1662             zeta2            = _mm256_mul_ps(beta2,rsq21);
1663             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
1664             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1665             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1666             felec            = _mm256_mul_ps(qq21,felec);
1667
1668             fscal            = felec;
1669
1670             /* Calculate temporary vectorial force */
1671             tx               = _mm256_mul_ps(fscal,dx21);
1672             ty               = _mm256_mul_ps(fscal,dy21);
1673             tz               = _mm256_mul_ps(fscal,dz21);
1674
1675             /* Update vectorial force */
1676             fix2             = _mm256_add_ps(fix2,tx);
1677             fiy2             = _mm256_add_ps(fiy2,ty);
1678             fiz2             = _mm256_add_ps(fiz2,tz);
1679
1680             fjx1             = _mm256_add_ps(fjx1,tx);
1681             fjy1             = _mm256_add_ps(fjy1,ty);
1682             fjz1             = _mm256_add_ps(fjz1,tz);
1683
1684             /**************************
1685              * CALCULATE INTERACTIONS *
1686              **************************/
1687
1688             r22              = _mm256_mul_ps(rsq22,rinv22);
1689
1690             /* EWALD ELECTROSTATICS */
1691
1692             /* Analytical PME correction */
1693             zeta2            = _mm256_mul_ps(beta2,rsq22);
1694             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
1695             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1696             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1697             felec            = _mm256_mul_ps(qq22,felec);
1698
1699             fscal            = felec;
1700
1701             /* Calculate temporary vectorial force */
1702             tx               = _mm256_mul_ps(fscal,dx22);
1703             ty               = _mm256_mul_ps(fscal,dy22);
1704             tz               = _mm256_mul_ps(fscal,dz22);
1705
1706             /* Update vectorial force */
1707             fix2             = _mm256_add_ps(fix2,tx);
1708             fiy2             = _mm256_add_ps(fiy2,ty);
1709             fiz2             = _mm256_add_ps(fiz2,tz);
1710
1711             fjx2             = _mm256_add_ps(fjx2,tx);
1712             fjy2             = _mm256_add_ps(fjy2,ty);
1713             fjz2             = _mm256_add_ps(fjz2,tz);
1714
1715             fjptrA             = f+j_coord_offsetA;
1716             fjptrB             = f+j_coord_offsetB;
1717             fjptrC             = f+j_coord_offsetC;
1718             fjptrD             = f+j_coord_offsetD;
1719             fjptrE             = f+j_coord_offsetE;
1720             fjptrF             = f+j_coord_offsetF;
1721             fjptrG             = f+j_coord_offsetG;
1722             fjptrH             = f+j_coord_offsetH;
1723
1724             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1725                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1726
1727             /* Inner loop uses 504 flops */
1728         }
1729
1730         if(jidx<j_index_end)
1731         {
1732
1733             /* Get j neighbor index, and coordinate index */
1734             jnrlistA         = jjnr[jidx];
1735             jnrlistB         = jjnr[jidx+1];
1736             jnrlistC         = jjnr[jidx+2];
1737             jnrlistD         = jjnr[jidx+3];
1738             jnrlistE         = jjnr[jidx+4];
1739             jnrlistF         = jjnr[jidx+5];
1740             jnrlistG         = jjnr[jidx+6];
1741             jnrlistH         = jjnr[jidx+7];
1742             /* Sign of each element will be negative for non-real atoms.
1743              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1744              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1745              */
1746             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1747                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1748
1749             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1750             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1751             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1752             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1753             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1754             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1755             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1756             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1757             j_coord_offsetA  = DIM*jnrA;
1758             j_coord_offsetB  = DIM*jnrB;
1759             j_coord_offsetC  = DIM*jnrC;
1760             j_coord_offsetD  = DIM*jnrD;
1761             j_coord_offsetE  = DIM*jnrE;
1762             j_coord_offsetF  = DIM*jnrF;
1763             j_coord_offsetG  = DIM*jnrG;
1764             j_coord_offsetH  = DIM*jnrH;
1765
1766             /* load j atom coordinates */
1767             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1768                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1769                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1770                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1771                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1772
1773             /* Calculate displacement vector */
1774             dx00             = _mm256_sub_ps(ix0,jx0);
1775             dy00             = _mm256_sub_ps(iy0,jy0);
1776             dz00             = _mm256_sub_ps(iz0,jz0);
1777             dx01             = _mm256_sub_ps(ix0,jx1);
1778             dy01             = _mm256_sub_ps(iy0,jy1);
1779             dz01             = _mm256_sub_ps(iz0,jz1);
1780             dx02             = _mm256_sub_ps(ix0,jx2);
1781             dy02             = _mm256_sub_ps(iy0,jy2);
1782             dz02             = _mm256_sub_ps(iz0,jz2);
1783             dx10             = _mm256_sub_ps(ix1,jx0);
1784             dy10             = _mm256_sub_ps(iy1,jy0);
1785             dz10             = _mm256_sub_ps(iz1,jz0);
1786             dx11             = _mm256_sub_ps(ix1,jx1);
1787             dy11             = _mm256_sub_ps(iy1,jy1);
1788             dz11             = _mm256_sub_ps(iz1,jz1);
1789             dx12             = _mm256_sub_ps(ix1,jx2);
1790             dy12             = _mm256_sub_ps(iy1,jy2);
1791             dz12             = _mm256_sub_ps(iz1,jz2);
1792             dx20             = _mm256_sub_ps(ix2,jx0);
1793             dy20             = _mm256_sub_ps(iy2,jy0);
1794             dz20             = _mm256_sub_ps(iz2,jz0);
1795             dx21             = _mm256_sub_ps(ix2,jx1);
1796             dy21             = _mm256_sub_ps(iy2,jy1);
1797             dz21             = _mm256_sub_ps(iz2,jz1);
1798             dx22             = _mm256_sub_ps(ix2,jx2);
1799             dy22             = _mm256_sub_ps(iy2,jy2);
1800             dz22             = _mm256_sub_ps(iz2,jz2);
1801
1802             /* Calculate squared distance and things based on it */
1803             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1804             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1805             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1806             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1807             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1808             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1809             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1810             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1811             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1812
1813             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1814             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1815             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1816             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1817             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1818             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1819             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1820             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1821             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1822
1823             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1824             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1825             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1826             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1827             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1828             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1829             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1830             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1831             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1832
1833             fjx0             = _mm256_setzero_ps();
1834             fjy0             = _mm256_setzero_ps();
1835             fjz0             = _mm256_setzero_ps();
1836             fjx1             = _mm256_setzero_ps();
1837             fjy1             = _mm256_setzero_ps();
1838             fjz1             = _mm256_setzero_ps();
1839             fjx2             = _mm256_setzero_ps();
1840             fjy2             = _mm256_setzero_ps();
1841             fjz2             = _mm256_setzero_ps();
1842
1843             /**************************
1844              * CALCULATE INTERACTIONS *
1845              **************************/
1846
1847             r00              = _mm256_mul_ps(rsq00,rinv00);
1848             r00              = _mm256_andnot_ps(dummy_mask,r00);
1849
1850             /* EWALD ELECTROSTATICS */
1851
1852             /* Analytical PME correction */
1853             zeta2            = _mm256_mul_ps(beta2,rsq00);
1854             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1855             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1856             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1857             felec            = _mm256_mul_ps(qq00,felec);
1858
1859             fscal            = felec;
1860
1861             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1862
1863             /* Calculate temporary vectorial force */
1864             tx               = _mm256_mul_ps(fscal,dx00);
1865             ty               = _mm256_mul_ps(fscal,dy00);
1866             tz               = _mm256_mul_ps(fscal,dz00);
1867
1868             /* Update vectorial force */
1869             fix0             = _mm256_add_ps(fix0,tx);
1870             fiy0             = _mm256_add_ps(fiy0,ty);
1871             fiz0             = _mm256_add_ps(fiz0,tz);
1872
1873             fjx0             = _mm256_add_ps(fjx0,tx);
1874             fjy0             = _mm256_add_ps(fjy0,ty);
1875             fjz0             = _mm256_add_ps(fjz0,tz);
1876
1877             /**************************
1878              * CALCULATE INTERACTIONS *
1879              **************************/
1880
1881             r01              = _mm256_mul_ps(rsq01,rinv01);
1882             r01              = _mm256_andnot_ps(dummy_mask,r01);
1883
1884             /* EWALD ELECTROSTATICS */
1885
1886             /* Analytical PME correction */
1887             zeta2            = _mm256_mul_ps(beta2,rsq01);
1888             rinv3            = _mm256_mul_ps(rinvsq01,rinv01);
1889             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1890             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1891             felec            = _mm256_mul_ps(qq01,felec);
1892
1893             fscal            = felec;
1894
1895             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1896
1897             /* Calculate temporary vectorial force */
1898             tx               = _mm256_mul_ps(fscal,dx01);
1899             ty               = _mm256_mul_ps(fscal,dy01);
1900             tz               = _mm256_mul_ps(fscal,dz01);
1901
1902             /* Update vectorial force */
1903             fix0             = _mm256_add_ps(fix0,tx);
1904             fiy0             = _mm256_add_ps(fiy0,ty);
1905             fiz0             = _mm256_add_ps(fiz0,tz);
1906
1907             fjx1             = _mm256_add_ps(fjx1,tx);
1908             fjy1             = _mm256_add_ps(fjy1,ty);
1909             fjz1             = _mm256_add_ps(fjz1,tz);
1910
1911             /**************************
1912              * CALCULATE INTERACTIONS *
1913              **************************/
1914
1915             r02              = _mm256_mul_ps(rsq02,rinv02);
1916             r02              = _mm256_andnot_ps(dummy_mask,r02);
1917
1918             /* EWALD ELECTROSTATICS */
1919
1920             /* Analytical PME correction */
1921             zeta2            = _mm256_mul_ps(beta2,rsq02);
1922             rinv3            = _mm256_mul_ps(rinvsq02,rinv02);
1923             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1924             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1925             felec            = _mm256_mul_ps(qq02,felec);
1926
1927             fscal            = felec;
1928
1929             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1930
1931             /* Calculate temporary vectorial force */
1932             tx               = _mm256_mul_ps(fscal,dx02);
1933             ty               = _mm256_mul_ps(fscal,dy02);
1934             tz               = _mm256_mul_ps(fscal,dz02);
1935
1936             /* Update vectorial force */
1937             fix0             = _mm256_add_ps(fix0,tx);
1938             fiy0             = _mm256_add_ps(fiy0,ty);
1939             fiz0             = _mm256_add_ps(fiz0,tz);
1940
1941             fjx2             = _mm256_add_ps(fjx2,tx);
1942             fjy2             = _mm256_add_ps(fjy2,ty);
1943             fjz2             = _mm256_add_ps(fjz2,tz);
1944
1945             /**************************
1946              * CALCULATE INTERACTIONS *
1947              **************************/
1948
1949             r10              = _mm256_mul_ps(rsq10,rinv10);
1950             r10              = _mm256_andnot_ps(dummy_mask,r10);
1951
1952             /* EWALD ELECTROSTATICS */
1953
1954             /* Analytical PME correction */
1955             zeta2            = _mm256_mul_ps(beta2,rsq10);
1956             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1957             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1958             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1959             felec            = _mm256_mul_ps(qq10,felec);
1960
1961             fscal            = felec;
1962
1963             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1964
1965             /* Calculate temporary vectorial force */
1966             tx               = _mm256_mul_ps(fscal,dx10);
1967             ty               = _mm256_mul_ps(fscal,dy10);
1968             tz               = _mm256_mul_ps(fscal,dz10);
1969
1970             /* Update vectorial force */
1971             fix1             = _mm256_add_ps(fix1,tx);
1972             fiy1             = _mm256_add_ps(fiy1,ty);
1973             fiz1             = _mm256_add_ps(fiz1,tz);
1974
1975             fjx0             = _mm256_add_ps(fjx0,tx);
1976             fjy0             = _mm256_add_ps(fjy0,ty);
1977             fjz0             = _mm256_add_ps(fjz0,tz);
1978
1979             /**************************
1980              * CALCULATE INTERACTIONS *
1981              **************************/
1982
1983             r11              = _mm256_mul_ps(rsq11,rinv11);
1984             r11              = _mm256_andnot_ps(dummy_mask,r11);
1985
1986             /* EWALD ELECTROSTATICS */
1987
1988             /* Analytical PME correction */
1989             zeta2            = _mm256_mul_ps(beta2,rsq11);
1990             rinv3            = _mm256_mul_ps(rinvsq11,rinv11);
1991             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1992             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1993             felec            = _mm256_mul_ps(qq11,felec);
1994
1995             fscal            = felec;
1996
1997             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1998
1999             /* Calculate temporary vectorial force */
2000             tx               = _mm256_mul_ps(fscal,dx11);
2001             ty               = _mm256_mul_ps(fscal,dy11);
2002             tz               = _mm256_mul_ps(fscal,dz11);
2003
2004             /* Update vectorial force */
2005             fix1             = _mm256_add_ps(fix1,tx);
2006             fiy1             = _mm256_add_ps(fiy1,ty);
2007             fiz1             = _mm256_add_ps(fiz1,tz);
2008
2009             fjx1             = _mm256_add_ps(fjx1,tx);
2010             fjy1             = _mm256_add_ps(fjy1,ty);
2011             fjz1             = _mm256_add_ps(fjz1,tz);
2012
2013             /**************************
2014              * CALCULATE INTERACTIONS *
2015              **************************/
2016
2017             r12              = _mm256_mul_ps(rsq12,rinv12);
2018             r12              = _mm256_andnot_ps(dummy_mask,r12);
2019
2020             /* EWALD ELECTROSTATICS */
2021
2022             /* Analytical PME correction */
2023             zeta2            = _mm256_mul_ps(beta2,rsq12);
2024             rinv3            = _mm256_mul_ps(rinvsq12,rinv12);
2025             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2026             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2027             felec            = _mm256_mul_ps(qq12,felec);
2028
2029             fscal            = felec;
2030
2031             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2032
2033             /* Calculate temporary vectorial force */
2034             tx               = _mm256_mul_ps(fscal,dx12);
2035             ty               = _mm256_mul_ps(fscal,dy12);
2036             tz               = _mm256_mul_ps(fscal,dz12);
2037
2038             /* Update vectorial force */
2039             fix1             = _mm256_add_ps(fix1,tx);
2040             fiy1             = _mm256_add_ps(fiy1,ty);
2041             fiz1             = _mm256_add_ps(fiz1,tz);
2042
2043             fjx2             = _mm256_add_ps(fjx2,tx);
2044             fjy2             = _mm256_add_ps(fjy2,ty);
2045             fjz2             = _mm256_add_ps(fjz2,tz);
2046
2047             /**************************
2048              * CALCULATE INTERACTIONS *
2049              **************************/
2050
2051             r20              = _mm256_mul_ps(rsq20,rinv20);
2052             r20              = _mm256_andnot_ps(dummy_mask,r20);
2053
2054             /* EWALD ELECTROSTATICS */
2055
2056             /* Analytical PME correction */
2057             zeta2            = _mm256_mul_ps(beta2,rsq20);
2058             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
2059             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2060             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2061             felec            = _mm256_mul_ps(qq20,felec);
2062
2063             fscal            = felec;
2064
2065             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2066
2067             /* Calculate temporary vectorial force */
2068             tx               = _mm256_mul_ps(fscal,dx20);
2069             ty               = _mm256_mul_ps(fscal,dy20);
2070             tz               = _mm256_mul_ps(fscal,dz20);
2071
2072             /* Update vectorial force */
2073             fix2             = _mm256_add_ps(fix2,tx);
2074             fiy2             = _mm256_add_ps(fiy2,ty);
2075             fiz2             = _mm256_add_ps(fiz2,tz);
2076
2077             fjx0             = _mm256_add_ps(fjx0,tx);
2078             fjy0             = _mm256_add_ps(fjy0,ty);
2079             fjz0             = _mm256_add_ps(fjz0,tz);
2080
2081             /**************************
2082              * CALCULATE INTERACTIONS *
2083              **************************/
2084
2085             r21              = _mm256_mul_ps(rsq21,rinv21);
2086             r21              = _mm256_andnot_ps(dummy_mask,r21);
2087
2088             /* EWALD ELECTROSTATICS */
2089
2090             /* Analytical PME correction */
2091             zeta2            = _mm256_mul_ps(beta2,rsq21);
2092             rinv3            = _mm256_mul_ps(rinvsq21,rinv21);
2093             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2094             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2095             felec            = _mm256_mul_ps(qq21,felec);
2096
2097             fscal            = felec;
2098
2099             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2100
2101             /* Calculate temporary vectorial force */
2102             tx               = _mm256_mul_ps(fscal,dx21);
2103             ty               = _mm256_mul_ps(fscal,dy21);
2104             tz               = _mm256_mul_ps(fscal,dz21);
2105
2106             /* Update vectorial force */
2107             fix2             = _mm256_add_ps(fix2,tx);
2108             fiy2             = _mm256_add_ps(fiy2,ty);
2109             fiz2             = _mm256_add_ps(fiz2,tz);
2110
2111             fjx1             = _mm256_add_ps(fjx1,tx);
2112             fjy1             = _mm256_add_ps(fjy1,ty);
2113             fjz1             = _mm256_add_ps(fjz1,tz);
2114
2115             /**************************
2116              * CALCULATE INTERACTIONS *
2117              **************************/
2118
2119             r22              = _mm256_mul_ps(rsq22,rinv22);
2120             r22              = _mm256_andnot_ps(dummy_mask,r22);
2121
2122             /* EWALD ELECTROSTATICS */
2123
2124             /* Analytical PME correction */
2125             zeta2            = _mm256_mul_ps(beta2,rsq22);
2126             rinv3            = _mm256_mul_ps(rinvsq22,rinv22);
2127             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
2128             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
2129             felec            = _mm256_mul_ps(qq22,felec);
2130
2131             fscal            = felec;
2132
2133             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2134
2135             /* Calculate temporary vectorial force */
2136             tx               = _mm256_mul_ps(fscal,dx22);
2137             ty               = _mm256_mul_ps(fscal,dy22);
2138             tz               = _mm256_mul_ps(fscal,dz22);
2139
2140             /* Update vectorial force */
2141             fix2             = _mm256_add_ps(fix2,tx);
2142             fiy2             = _mm256_add_ps(fiy2,ty);
2143             fiz2             = _mm256_add_ps(fiz2,tz);
2144
2145             fjx2             = _mm256_add_ps(fjx2,tx);
2146             fjy2             = _mm256_add_ps(fjy2,ty);
2147             fjz2             = _mm256_add_ps(fjz2,tz);
2148
2149             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2150             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2151             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2152             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2153             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2154             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2155             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2156             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2157
2158             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2159                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2160
2161             /* Inner loop uses 513 flops */
2162         }
2163
2164         /* End of innermost loop */
2165
2166         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2167                                                  f+i_coord_offset,fshift+i_shift_offset);
2168
2169         /* Increment number of inner iterations */
2170         inneriter                  += j_index_end - j_index_start;
2171
2172         /* Outer loop uses 18 flops */
2173     }
2174
2175     /* Increment number of outer iterations */
2176     outeriter        += nri;
2177
2178     /* Update outer/inner flops */
2179
2180     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*513);
2181 }