src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwNone_GeomW3W3_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3W3_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwNone_GeomW3W3_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  97     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  98     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  99     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 100     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 101     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 102     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 103     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 104     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 105     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 106     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 107     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 108     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 109     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 110     real             *charge;
 111     __m256           dummy_mask,cutoff_mask;
 112     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 113     __m256           one     = _mm256_set1_ps(1.0);
 114     __m256           two     = _mm256_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_ps(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     krf              = _mm256_set1_ps(fr->ic->k_rf);
 129     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 130     crf              = _mm256_set1_ps(fr->ic->c_rf);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 135     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 136     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 137
 138     jq0              = _mm256_set1_ps(charge[inr+0]);
 139     jq1              = _mm256_set1_ps(charge[inr+1]);
 140     jq2              = _mm256_set1_ps(charge[inr+2]);
 141     qq00             = _mm256_mul_ps(iq0,jq0);
 142     qq01             = _mm256_mul_ps(iq0,jq1);
 143     qq02             = _mm256_mul_ps(iq0,jq2);
 144     qq10             = _mm256_mul_ps(iq1,jq0);
 145     qq11             = _mm256_mul_ps(iq1,jq1);
 146     qq12             = _mm256_mul_ps(iq1,jq2);
 147     qq20             = _mm256_mul_ps(iq2,jq0);
 148     qq21             = _mm256_mul_ps(iq2,jq1);
 149     qq22             = _mm256_mul_ps(iq2,jq2);
 150
 151     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 152     rcutoff_scalar   = fr->rcoulomb;
 153     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 154     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160     j_coord_offsetC = 0;
 161     j_coord_offsetD = 0;
 162     j_coord_offsetE = 0;
 163     j_coord_offsetF = 0;
 164     j_coord_offsetG = 0;
 165     j_coord_offsetH = 0;
 166
 167     outeriter        = 0;
 168     inneriter        = 0;
 169
 170     for(iidx=0;iidx<4*DIM;iidx++)
 171     {
 172         scratch[iidx] = 0.0;
 173     }
 174
 175     /* Start outer loop over neighborlists */
 176     for(iidx=0; iidx<nri; iidx++)
 177     {
 178         /* Load shift vector for this list */
 179         i_shift_offset   = DIM*shiftidx[iidx];
 180
 181         /* Load limits for loop over neighbors */
 182         j_index_start    = jindex[iidx];
 183         j_index_end      = jindex[iidx+1];
 184
 185         /* Get outer coordinate index */
 186         inr              = iinr[iidx];
 187         i_coord_offset   = DIM*inr;
 188
 189         /* Load i particle coords and add shift vector */
 190         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 191                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 192
 193         fix0             = _mm256_setzero_ps();
 194         fiy0             = _mm256_setzero_ps();
 195         fiz0             = _mm256_setzero_ps();
 196         fix1             = _mm256_setzero_ps();
 197         fiy1             = _mm256_setzero_ps();
 198         fiz1             = _mm256_setzero_ps();
 199         fix2             = _mm256_setzero_ps();
 200         fiy2             = _mm256_setzero_ps();
 201         fiz2             = _mm256_setzero_ps();
 202
 203         /* Reset potential sums */
 204         velecsum         = _mm256_setzero_ps();
 205
 206         /* Start inner kernel loop */
 207         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 208         {
 209
 210             /* Get j neighbor index, and coordinate index */
 211             jnrA             = jjnr[jidx];
 212             jnrB             = jjnr[jidx+1];
 213             jnrC             = jjnr[jidx+2];
 214             jnrD             = jjnr[jidx+3];
 215             jnrE             = jjnr[jidx+4];
 216             jnrF             = jjnr[jidx+5];
 217             jnrG             = jjnr[jidx+6];
 218             jnrH             = jjnr[jidx+7];
 219             j_coord_offsetA  = DIM*jnrA;
 220             j_coord_offsetB  = DIM*jnrB;
 221             j_coord_offsetC  = DIM*jnrC;
 222             j_coord_offsetD  = DIM*jnrD;
 223             j_coord_offsetE  = DIM*jnrE;
 224             j_coord_offsetF  = DIM*jnrF;
 225             j_coord_offsetG  = DIM*jnrG;
 226             j_coord_offsetH  = DIM*jnrH;
 227
 228             /* load j atom coordinates */
 229             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 230                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 231                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 232                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 233                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 234
 235             /* Calculate displacement vector */
 236             dx00             = _mm256_sub_ps(ix0,jx0);
 237             dy00             = _mm256_sub_ps(iy0,jy0);
 238             dz00             = _mm256_sub_ps(iz0,jz0);
 239             dx01             = _mm256_sub_ps(ix0,jx1);
 240             dy01             = _mm256_sub_ps(iy0,jy1);
 241             dz01             = _mm256_sub_ps(iz0,jz1);
 242             dx02             = _mm256_sub_ps(ix0,jx2);
 243             dy02             = _mm256_sub_ps(iy0,jy2);
 244             dz02             = _mm256_sub_ps(iz0,jz2);
 245             dx10             = _mm256_sub_ps(ix1,jx0);
 246             dy10             = _mm256_sub_ps(iy1,jy0);
 247             dz10             = _mm256_sub_ps(iz1,jz0);
 248             dx11             = _mm256_sub_ps(ix1,jx1);
 249             dy11             = _mm256_sub_ps(iy1,jy1);
 250             dz11             = _mm256_sub_ps(iz1,jz1);
 251             dx12             = _mm256_sub_ps(ix1,jx2);
 252             dy12             = _mm256_sub_ps(iy1,jy2);
 253             dz12             = _mm256_sub_ps(iz1,jz2);
 254             dx20             = _mm256_sub_ps(ix2,jx0);
 255             dy20             = _mm256_sub_ps(iy2,jy0);
 256             dz20             = _mm256_sub_ps(iz2,jz0);
 257             dx21             = _mm256_sub_ps(ix2,jx1);
 258             dy21             = _mm256_sub_ps(iy2,jy1);
 259             dz21             = _mm256_sub_ps(iz2,jz1);
 260             dx22             = _mm256_sub_ps(ix2,jx2);
 261             dy22             = _mm256_sub_ps(iy2,jy2);
 262             dz22             = _mm256_sub_ps(iz2,jz2);
 263
 264             /* Calculate squared distance and things based on it */
 265             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 266             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 267             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 268             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 269             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 270             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 271             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 272             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 273             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 274
 275             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 276             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 277             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 278             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 279             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 280             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 281             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 282             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 283             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 284
 285             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 286             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 287             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 288             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 289             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 290             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 291             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 292             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 293             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 294
 295             fjx0             = _mm256_setzero_ps();
 296             fjy0             = _mm256_setzero_ps();
 297             fjz0             = _mm256_setzero_ps();
 298             fjx1             = _mm256_setzero_ps();
 299             fjy1             = _mm256_setzero_ps();
 300             fjz1             = _mm256_setzero_ps();
 301             fjx2             = _mm256_setzero_ps();
 302             fjy2             = _mm256_setzero_ps();
 303             fjz2             = _mm256_setzero_ps();
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 310             {
 311
 312             /* REACTION-FIELD ELECTROSTATICS */
 313             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 314             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 315
 316             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velec            = _mm256_and_ps(velec,cutoff_mask);
 320             velecsum         = _mm256_add_ps(velecsum,velec);
 321
 322             fscal            = felec;
 323
 324             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 325
 326             /* Calculate temporary vectorial force */
 327             tx               = _mm256_mul_ps(fscal,dx00);
 328             ty               = _mm256_mul_ps(fscal,dy00);
 329             tz               = _mm256_mul_ps(fscal,dz00);
 330
 331             /* Update vectorial force */
 332             fix0             = _mm256_add_ps(fix0,tx);
 333             fiy0             = _mm256_add_ps(fiy0,ty);
 334             fiz0             = _mm256_add_ps(fiz0,tz);
 335
 336             fjx0             = _mm256_add_ps(fjx0,tx);
 337             fjy0             = _mm256_add_ps(fjy0,ty);
 338             fjz0             = _mm256_add_ps(fjz0,tz);
 339
 340             }
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 347             {
 348
 349             /* REACTION-FIELD ELECTROSTATICS */
 350             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 351             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 352
 353             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velec            = _mm256_and_ps(velec,cutoff_mask);
 357             velecsum         = _mm256_add_ps(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm256_mul_ps(fscal,dx01);
 365             ty               = _mm256_mul_ps(fscal,dy01);
 366             tz               = _mm256_mul_ps(fscal,dz01);
 367
 368             /* Update vectorial force */
 369             fix0             = _mm256_add_ps(fix0,tx);
 370             fiy0             = _mm256_add_ps(fiy0,ty);
 371             fiz0             = _mm256_add_ps(fiz0,tz);
 372
 373             fjx1             = _mm256_add_ps(fjx1,tx);
 374             fjy1             = _mm256_add_ps(fjy1,ty);
 375             fjz1             = _mm256_add_ps(fjz1,tz);
 376
 377             }
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 384             {
 385
 386             /* REACTION-FIELD ELECTROSTATICS */
 387             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 388             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 389
 390             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velec            = _mm256_and_ps(velec,cutoff_mask);
 394             velecsum         = _mm256_add_ps(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 399
 400             /* Calculate temporary vectorial force */
 401             tx               = _mm256_mul_ps(fscal,dx02);
 402             ty               = _mm256_mul_ps(fscal,dy02);
 403             tz               = _mm256_mul_ps(fscal,dz02);
 404
 405             /* Update vectorial force */
 406             fix0             = _mm256_add_ps(fix0,tx);
 407             fiy0             = _mm256_add_ps(fiy0,ty);
 408             fiz0             = _mm256_add_ps(fiz0,tz);
 409
 410             fjx2             = _mm256_add_ps(fjx2,tx);
 411             fjy2             = _mm256_add_ps(fjy2,ty);
 412             fjz2             = _mm256_add_ps(fjz2,tz);
 413
 414             }
 415
 416             /**************************
 417              * CALCULATE INTERACTIONS *
 418              **************************/
 419
 420             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 421             {
 422
 423             /* REACTION-FIELD ELECTROSTATICS */
 424             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 425             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 426
 427             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 428
 429             /* Update potential sum for this i atom from the interaction with this j atom. */
 430             velec            = _mm256_and_ps(velec,cutoff_mask);
 431             velecsum         = _mm256_add_ps(velecsum,velec);
 432
 433             fscal            = felec;
 434
 435             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 436
 437             /* Calculate temporary vectorial force */
 438             tx               = _mm256_mul_ps(fscal,dx10);
 439             ty               = _mm256_mul_ps(fscal,dy10);
 440             tz               = _mm256_mul_ps(fscal,dz10);
 441
 442             /* Update vectorial force */
 443             fix1             = _mm256_add_ps(fix1,tx);
 444             fiy1             = _mm256_add_ps(fiy1,ty);
 445             fiz1             = _mm256_add_ps(fiz1,tz);
 446
 447             fjx0             = _mm256_add_ps(fjx0,tx);
 448             fjy0             = _mm256_add_ps(fjy0,ty);
 449             fjz0             = _mm256_add_ps(fjz0,tz);
 450
 451             }
 452
 453             /**************************
 454              * CALCULATE INTERACTIONS *
 455              **************************/
 456
 457             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 458             {
 459
 460             /* REACTION-FIELD ELECTROSTATICS */
 461             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 462             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 463
 464             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 465
 466             /* Update potential sum for this i atom from the interaction with this j atom. */
 467             velec            = _mm256_and_ps(velec,cutoff_mask);
 468             velecsum         = _mm256_add_ps(velecsum,velec);
 469
 470             fscal            = felec;
 471
 472             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 473
 474             /* Calculate temporary vectorial force */
 475             tx               = _mm256_mul_ps(fscal,dx11);
 476             ty               = _mm256_mul_ps(fscal,dy11);
 477             tz               = _mm256_mul_ps(fscal,dz11);
 478
 479             /* Update vectorial force */
 480             fix1             = _mm256_add_ps(fix1,tx);
 481             fiy1             = _mm256_add_ps(fiy1,ty);
 482             fiz1             = _mm256_add_ps(fiz1,tz);
 483
 484             fjx1             = _mm256_add_ps(fjx1,tx);
 485             fjy1             = _mm256_add_ps(fjy1,ty);
 486             fjz1             = _mm256_add_ps(fjz1,tz);
 487
 488             }
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 495             {
 496
 497             /* REACTION-FIELD ELECTROSTATICS */
 498             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 499             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 500
 501             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velec            = _mm256_and_ps(velec,cutoff_mask);
 505             velecsum         = _mm256_add_ps(velecsum,velec);
 506
 507             fscal            = felec;
 508
 509             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm256_mul_ps(fscal,dx12);
 513             ty               = _mm256_mul_ps(fscal,dy12);
 514             tz               = _mm256_mul_ps(fscal,dz12);
 515
 516             /* Update vectorial force */
 517             fix1             = _mm256_add_ps(fix1,tx);
 518             fiy1             = _mm256_add_ps(fiy1,ty);
 519             fiz1             = _mm256_add_ps(fiz1,tz);
 520
 521             fjx2             = _mm256_add_ps(fjx2,tx);
 522             fjy2             = _mm256_add_ps(fjy2,ty);
 523             fjz2             = _mm256_add_ps(fjz2,tz);
 524
 525             }
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 532             {
 533
 534             /* REACTION-FIELD ELECTROSTATICS */
 535             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 536             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 537
 538             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 539
 540             /* Update potential sum for this i atom from the interaction with this j atom. */
 541             velec            = _mm256_and_ps(velec,cutoff_mask);
 542             velecsum         = _mm256_add_ps(velecsum,velec);
 543
 544             fscal            = felec;
 545
 546             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm256_mul_ps(fscal,dx20);
 550             ty               = _mm256_mul_ps(fscal,dy20);
 551             tz               = _mm256_mul_ps(fscal,dz20);
 552
 553             /* Update vectorial force */
 554             fix2             = _mm256_add_ps(fix2,tx);
 555             fiy2             = _mm256_add_ps(fiy2,ty);
 556             fiz2             = _mm256_add_ps(fiz2,tz);
 557
 558             fjx0             = _mm256_add_ps(fjx0,tx);
 559             fjy0             = _mm256_add_ps(fjy0,ty);
 560             fjz0             = _mm256_add_ps(fjz0,tz);
 561
 562             }
 563
 564             /**************************
 565              * CALCULATE INTERACTIONS *
 566              **************************/
 567
 568             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 569             {
 570
 571             /* REACTION-FIELD ELECTROSTATICS */
 572             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
 573             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
 574
 575             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm256_and_ps(velec,cutoff_mask);
 579             velecsum         = _mm256_add_ps(velecsum,velec);
 580
 581             fscal            = felec;
 582
 583             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 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
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 606             {
 607
 608             /* REACTION-FIELD ELECTROSTATICS */
 609             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
 610             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
 611
 612             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             velec            = _mm256_and_ps(velec,cutoff_mask);
 616             velecsum         = _mm256_add_ps(velecsum,velec);
 617
 618             fscal            = felec;
 619
 620             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 621
 622             /* Calculate temporary vectorial force */
 623             tx               = _mm256_mul_ps(fscal,dx22);
 624             ty               = _mm256_mul_ps(fscal,dy22);
 625             tz               = _mm256_mul_ps(fscal,dz22);
 626
 627             /* Update vectorial force */
 628             fix2             = _mm256_add_ps(fix2,tx);
 629             fiy2             = _mm256_add_ps(fiy2,ty);
 630             fiz2             = _mm256_add_ps(fiz2,tz);
 631
 632             fjx2             = _mm256_add_ps(fjx2,tx);
 633             fjy2             = _mm256_add_ps(fjy2,ty);
 634             fjz2             = _mm256_add_ps(fjz2,tz);
 635
 636             }
 637
 638             fjptrA             = f+j_coord_offsetA;
 639             fjptrB             = f+j_coord_offsetB;
 640             fjptrC             = f+j_coord_offsetC;
 641             fjptrD             = f+j_coord_offsetD;
 642             fjptrE             = f+j_coord_offsetE;
 643             fjptrF             = f+j_coord_offsetF;
 644             fjptrG             = f+j_coord_offsetG;
 645             fjptrH             = f+j_coord_offsetH;
 646
 647             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 648                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 649
 650             /* Inner loop uses 324 flops */
 651         }
 652
 653         if(jidx<j_index_end)
 654         {
 655
 656             /* Get j neighbor index, and coordinate index */
 657             jnrlistA         = jjnr[jidx];
 658             jnrlistB         = jjnr[jidx+1];
 659             jnrlistC         = jjnr[jidx+2];
 660             jnrlistD         = jjnr[jidx+3];
 661             jnrlistE         = jjnr[jidx+4];
 662             jnrlistF         = jjnr[jidx+5];
 663             jnrlistG         = jjnr[jidx+6];
 664             jnrlistH         = jjnr[jidx+7];
 665             /* Sign of each element will be negative for non-real atoms.
 666              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 667              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 668              */
 669             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 670                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 671
 672             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 673             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 674             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 675             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 676             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 677             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 678             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 679             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 680             j_coord_offsetA  = DIM*jnrA;
 681             j_coord_offsetB  = DIM*jnrB;
 682             j_coord_offsetC  = DIM*jnrC;
 683             j_coord_offsetD  = DIM*jnrD;
 684             j_coord_offsetE  = DIM*jnrE;
 685             j_coord_offsetF  = DIM*jnrF;
 686             j_coord_offsetG  = DIM*jnrG;
 687             j_coord_offsetH  = DIM*jnrH;
 688
 689             /* load j atom coordinates */
 690             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 691                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 692                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 693                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 694                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 695
 696             /* Calculate displacement vector */
 697             dx00             = _mm256_sub_ps(ix0,jx0);
 698             dy00             = _mm256_sub_ps(iy0,jy0);
 699             dz00             = _mm256_sub_ps(iz0,jz0);
 700             dx01             = _mm256_sub_ps(ix0,jx1);
 701             dy01             = _mm256_sub_ps(iy0,jy1);
 702             dz01             = _mm256_sub_ps(iz0,jz1);
 703             dx02             = _mm256_sub_ps(ix0,jx2);
 704             dy02             = _mm256_sub_ps(iy0,jy2);
 705             dz02             = _mm256_sub_ps(iz0,jz2);
 706             dx10             = _mm256_sub_ps(ix1,jx0);
 707             dy10             = _mm256_sub_ps(iy1,jy0);
 708             dz10             = _mm256_sub_ps(iz1,jz0);
 709             dx11             = _mm256_sub_ps(ix1,jx1);
 710             dy11             = _mm256_sub_ps(iy1,jy1);
 711             dz11             = _mm256_sub_ps(iz1,jz1);
 712             dx12             = _mm256_sub_ps(ix1,jx2);
 713             dy12             = _mm256_sub_ps(iy1,jy2);
 714             dz12             = _mm256_sub_ps(iz1,jz2);
 715             dx20             = _mm256_sub_ps(ix2,jx0);
 716             dy20             = _mm256_sub_ps(iy2,jy0);
 717             dz20             = _mm256_sub_ps(iz2,jz0);
 718             dx21             = _mm256_sub_ps(ix2,jx1);
 719             dy21             = _mm256_sub_ps(iy2,jy1);
 720             dz21             = _mm256_sub_ps(iz2,jz1);
 721             dx22             = _mm256_sub_ps(ix2,jx2);
 722             dy22             = _mm256_sub_ps(iy2,jy2);
 723             dz22             = _mm256_sub_ps(iz2,jz2);
 724
 725             /* Calculate squared distance and things based on it */
 726             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 727             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 728             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 729             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 730             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 731             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 732             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 733             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 734             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 735
 736             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 737             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 738             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 739             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 740             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 741             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 742             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 743             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 744             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 745
 746             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 747             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
 748             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
 749             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 750             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
 751             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
 752             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 753             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
 754             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
 755
 756             fjx0             = _mm256_setzero_ps();
 757             fjy0             = _mm256_setzero_ps();
 758             fjz0             = _mm256_setzero_ps();
 759             fjx1             = _mm256_setzero_ps();
 760             fjy1             = _mm256_setzero_ps();
 761             fjz1             = _mm256_setzero_ps();
 762             fjx2             = _mm256_setzero_ps();
 763             fjy2             = _mm256_setzero_ps();
 764             fjz2             = _mm256_setzero_ps();
 765
 766             /**************************
 767              * CALCULATE INTERACTIONS *
 768              **************************/
 769
 770             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 771             {
 772
 773             /* REACTION-FIELD ELECTROSTATICS */
 774             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 775             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 776
 777             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 778
 779             /* Update potential sum for this i atom from the interaction with this j atom. */
 780             velec            = _mm256_and_ps(velec,cutoff_mask);
 781             velec            = _mm256_andnot_ps(dummy_mask,velec);
 782             velecsum         = _mm256_add_ps(velecsum,velec);
 783
 784             fscal            = felec;
 785
 786             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 787
 788             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 789
 790             /* Calculate temporary vectorial force */
 791             tx               = _mm256_mul_ps(fscal,dx00);
 792             ty               = _mm256_mul_ps(fscal,dy00);
 793             tz               = _mm256_mul_ps(fscal,dz00);
 794
 795             /* Update vectorial force */
 796             fix0             = _mm256_add_ps(fix0,tx);
 797             fiy0             = _mm256_add_ps(fiy0,ty);
 798             fiz0             = _mm256_add_ps(fiz0,tz);
 799
 800             fjx0             = _mm256_add_ps(fjx0,tx);
 801             fjy0             = _mm256_add_ps(fjy0,ty);
 802             fjz0             = _mm256_add_ps(fjz0,tz);
 803
 804             }
 805
 806             /**************************
 807              * CALCULATE INTERACTIONS *
 808              **************************/
 809
 810             if (gmx_mm256_any_lt(rsq01,rcutoff2))
 811             {
 812
 813             /* REACTION-FIELD ELECTROSTATICS */
 814             velec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_add_ps(rinv01,_mm256_mul_ps(krf,rsq01)),crf));
 815             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
 816
 817             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
 818
 819             /* Update potential sum for this i atom from the interaction with this j atom. */
 820             velec            = _mm256_and_ps(velec,cutoff_mask);
 821             velec            = _mm256_andnot_ps(dummy_mask,velec);
 822             velecsum         = _mm256_add_ps(velecsum,velec);
 823
 824             fscal            = felec;
 825
 826             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 827
 828             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 829
 830             /* Calculate temporary vectorial force */
 831             tx               = _mm256_mul_ps(fscal,dx01);
 832             ty               = _mm256_mul_ps(fscal,dy01);
 833             tz               = _mm256_mul_ps(fscal,dz01);
 834
 835             /* Update vectorial force */
 836             fix0             = _mm256_add_ps(fix0,tx);
 837             fiy0             = _mm256_add_ps(fiy0,ty);
 838             fiz0             = _mm256_add_ps(fiz0,tz);
 839
 840             fjx1             = _mm256_add_ps(fjx1,tx);
 841             fjy1             = _mm256_add_ps(fjy1,ty);
 842             fjz1             = _mm256_add_ps(fjz1,tz);
 843
 844             }
 845
 846             /**************************
 847              * CALCULATE INTERACTIONS *
 848              **************************/
 849
 850             if (gmx_mm256_any_lt(rsq02,rcutoff2))
 851             {
 852
 853             /* REACTION-FIELD ELECTROSTATICS */
 854             velec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_add_ps(rinv02,_mm256_mul_ps(krf,rsq02)),crf));
 855             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
 856
 857             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
 858
 859             /* Update potential sum for this i atom from the interaction with this j atom. */
 860             velec            = _mm256_and_ps(velec,cutoff_mask);
 861             velec            = _mm256_andnot_ps(dummy_mask,velec);
 862             velecsum         = _mm256_add_ps(velecsum,velec);
 863
 864             fscal            = felec;
 865
 866             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 867
 868             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 869
 870             /* Calculate temporary vectorial force */
 871             tx               = _mm256_mul_ps(fscal,dx02);
 872             ty               = _mm256_mul_ps(fscal,dy02);
 873             tz               = _mm256_mul_ps(fscal,dz02);
 874
 875             /* Update vectorial force */
 876             fix0             = _mm256_add_ps(fix0,tx);
 877             fiy0             = _mm256_add_ps(fiy0,ty);
 878             fiz0             = _mm256_add_ps(fiz0,tz);
 879
 880             fjx2             = _mm256_add_ps(fjx2,tx);
 881             fjy2             = _mm256_add_ps(fjy2,ty);
 882             fjz2             = _mm256_add_ps(fjz2,tz);
 883
 884             }
 885
 886             /**************************
 887              * CALCULATE INTERACTIONS *
 888              **************************/
 889
 890             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 891             {
 892
 893             /* REACTION-FIELD ELECTROSTATICS */
 894             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 895             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 896
 897             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 898
 899             /* Update potential sum for this i atom from the interaction with this j atom. */
 900             velec            = _mm256_and_ps(velec,cutoff_mask);
 901             velec            = _mm256_andnot_ps(dummy_mask,velec);
 902             velecsum         = _mm256_add_ps(velecsum,velec);
 903
 904             fscal            = felec;
 905
 906             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 907
 908             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 909
 910             /* Calculate temporary vectorial force */
 911             tx               = _mm256_mul_ps(fscal,dx10);
 912             ty               = _mm256_mul_ps(fscal,dy10);
 913             tz               = _mm256_mul_ps(fscal,dz10);
 914
 915             /* Update vectorial force */
 916             fix1             = _mm256_add_ps(fix1,tx);
 917             fiy1             = _mm256_add_ps(fiy1,ty);
 918             fiz1             = _mm256_add_ps(fiz1,tz);
 919
 920             fjx0             = _mm256_add_ps(fjx0,tx);
 921             fjy0             = _mm256_add_ps(fjy0,ty);
 922             fjz0             = _mm256_add_ps(fjz0,tz);
 923
 924             }
 925
 926             /**************************
 927              * CALCULATE INTERACTIONS *
 928              **************************/
 929
 930             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 931             {
 932
 933             /* REACTION-FIELD ELECTROSTATICS */
 934             velec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_add_ps(rinv11,_mm256_mul_ps(krf,rsq11)),crf));
 935             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
 936
 937             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
 938
 939             /* Update potential sum for this i atom from the interaction with this j atom. */
 940             velec            = _mm256_and_ps(velec,cutoff_mask);
 941             velec            = _mm256_andnot_ps(dummy_mask,velec);
 942             velecsum         = _mm256_add_ps(velecsum,velec);
 943
 944             fscal            = felec;
 945
 946             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 947
 948             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 949
 950             /* Calculate temporary vectorial force */
 951             tx               = _mm256_mul_ps(fscal,dx11);
 952             ty               = _mm256_mul_ps(fscal,dy11);
 953             tz               = _mm256_mul_ps(fscal,dz11);
 954
 955             /* Update vectorial force */
 956             fix1             = _mm256_add_ps(fix1,tx);
 957             fiy1             = _mm256_add_ps(fiy1,ty);
 958             fiz1             = _mm256_add_ps(fiz1,tz);
 959
 960             fjx1             = _mm256_add_ps(fjx1,tx);
 961             fjy1             = _mm256_add_ps(fjy1,ty);
 962             fjz1             = _mm256_add_ps(fjz1,tz);
 963
 964             }
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 971             {
 972
 973             /* REACTION-FIELD ELECTROSTATICS */
 974             velec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_add_ps(rinv12,_mm256_mul_ps(krf,rsq12)),crf));
 975             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
 976
 977             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
 978
 979             /* Update potential sum for this i atom from the interaction with this j atom. */
 980             velec            = _mm256_and_ps(velec,cutoff_mask);
 981             velec            = _mm256_andnot_ps(dummy_mask,velec);
 982             velecsum         = _mm256_add_ps(velecsum,velec);
 983
 984             fscal            = felec;
 985
 986             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 987
 988             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 989
 990             /* Calculate temporary vectorial force */
 991             tx               = _mm256_mul_ps(fscal,dx12);
 992             ty               = _mm256_mul_ps(fscal,dy12);
 993             tz               = _mm256_mul_ps(fscal,dz12);
 994
 995             /* Update vectorial force */
 996             fix1             = _mm256_add_ps(fix1,tx);
 997             fiy1             = _mm256_add_ps(fiy1,ty);
 998             fiz1             = _mm256_add_ps(fiz1,tz);
 999
1000             fjx2             = _mm256_add_ps(fjx2,tx);
1001             fjy2             = _mm256_add_ps(fjy2,ty);
1002             fjz2             = _mm256_add_ps(fjz2,tz);
1003
1004             }
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1011             {
1012
1013             /* REACTION-FIELD ELECTROSTATICS */
1014             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
1015             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1016
1017             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1018
1019             /* Update potential sum for this i atom from the interaction with this j atom. */
1020             velec            = _mm256_and_ps(velec,cutoff_mask);
1021             velec            = _mm256_andnot_ps(dummy_mask,velec);
1022             velecsum         = _mm256_add_ps(velecsum,velec);
1023
1024             fscal            = felec;
1025
1026             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1027
1028             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1029
1030             /* Calculate temporary vectorial force */
1031             tx               = _mm256_mul_ps(fscal,dx20);
1032             ty               = _mm256_mul_ps(fscal,dy20);
1033             tz               = _mm256_mul_ps(fscal,dz20);
1034
1035             /* Update vectorial force */
1036             fix2             = _mm256_add_ps(fix2,tx);
1037             fiy2             = _mm256_add_ps(fiy2,ty);
1038             fiz2             = _mm256_add_ps(fiz2,tz);
1039
1040             fjx0             = _mm256_add_ps(fjx0,tx);
1041             fjy0             = _mm256_add_ps(fjy0,ty);
1042             fjz0             = _mm256_add_ps(fjz0,tz);
1043
1044             }
1045
1046             /**************************
1047              * CALCULATE INTERACTIONS *
1048              **************************/
1049
1050             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1051             {
1052
1053             /* REACTION-FIELD ELECTROSTATICS */
1054             velec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_add_ps(rinv21,_mm256_mul_ps(krf,rsq21)),crf));
1055             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1056
1057             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1058
1059             /* Update potential sum for this i atom from the interaction with this j atom. */
1060             velec            = _mm256_and_ps(velec,cutoff_mask);
1061             velec            = _mm256_andnot_ps(dummy_mask,velec);
1062             velecsum         = _mm256_add_ps(velecsum,velec);
1063
1064             fscal            = felec;
1065
1066             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1067
1068             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1069
1070             /* Calculate temporary vectorial force */
1071             tx               = _mm256_mul_ps(fscal,dx21);
1072             ty               = _mm256_mul_ps(fscal,dy21);
1073             tz               = _mm256_mul_ps(fscal,dz21);
1074
1075             /* Update vectorial force */
1076             fix2             = _mm256_add_ps(fix2,tx);
1077             fiy2             = _mm256_add_ps(fiy2,ty);
1078             fiz2             = _mm256_add_ps(fiz2,tz);
1079
1080             fjx1             = _mm256_add_ps(fjx1,tx);
1081             fjy1             = _mm256_add_ps(fjy1,ty);
1082             fjz1             = _mm256_add_ps(fjz1,tz);
1083
1084             }
1085
1086             /**************************
1087              * CALCULATE INTERACTIONS *
1088              **************************/
1089
1090             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1091             {
1092
1093             /* REACTION-FIELD ELECTROSTATICS */
1094             velec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_add_ps(rinv22,_mm256_mul_ps(krf,rsq22)),crf));
1095             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1096
1097             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1098
1099             /* Update potential sum for this i atom from the interaction with this j atom. */
1100             velec            = _mm256_and_ps(velec,cutoff_mask);
1101             velec            = _mm256_andnot_ps(dummy_mask,velec);
1102             velecsum         = _mm256_add_ps(velecsum,velec);
1103
1104             fscal            = felec;
1105
1106             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1107
1108             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1109
1110             /* Calculate temporary vectorial force */
1111             tx               = _mm256_mul_ps(fscal,dx22);
1112             ty               = _mm256_mul_ps(fscal,dy22);
1113             tz               = _mm256_mul_ps(fscal,dz22);
1114
1115             /* Update vectorial force */
1116             fix2             = _mm256_add_ps(fix2,tx);
1117             fiy2             = _mm256_add_ps(fiy2,ty);
1118             fiz2             = _mm256_add_ps(fiz2,tz);
1119
1120             fjx2             = _mm256_add_ps(fjx2,tx);
1121             fjy2             = _mm256_add_ps(fjy2,ty);
1122             fjz2             = _mm256_add_ps(fjz2,tz);
1123
1124             }
1125
1126             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1127             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1128             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1129             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1130             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1131             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1132             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1133             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1134
1135             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1136                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1137
1138             /* Inner loop uses 324 flops */
1139         }
1140
1141         /* End of innermost loop */
1142
1143         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1144                                                  f+i_coord_offset,fshift+i_shift_offset);
1145
1146         ggid                        = gid[iidx];
1147         /* Update potential energies */
1148         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1149
1150         /* Increment number of inner iterations */
1151         inneriter                  += j_index_end - j_index_start;
1152
1153         /* Outer loop uses 19 flops */
1154     }
1155
1156     /* Increment number of outer iterations */
1157     outeriter        += nri;
1158
1159     /* Update outer/inner flops */
1160
1161     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*324);
1162 }
1163 /*
1164  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3W3_F_avx_256_single
1165  * Electrostatics interaction: ReactionField
1166  * VdW interaction:            None
1167  * Geometry:                   Water3-Water3
1168  * Calculate force/pot:        Force
1169  */
1170 void
1171 nb_kernel_ElecRFCut_VdwNone_GeomW3W3_F_avx_256_single
1172                     (t_nblist                    * gmx_restrict       nlist,
1173                      rvec                        * gmx_restrict          xx,
1174                      rvec                        * gmx_restrict          ff,
1175                      t_forcerec                  * gmx_restrict          fr,
1176                      t_mdatoms                   * gmx_restrict     mdatoms,
1177                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1178                      t_nrnb                      * gmx_restrict        nrnb)
1179 {
1180     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1181      * just 0 for non-waters.
1182      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1183      * jnr indices corresponding to data put in the four positions in the SIMD register.
1184      */
1185     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1186     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1187     int              jnrA,jnrB,jnrC,jnrD;
1188     int              jnrE,jnrF,jnrG,jnrH;
1189     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1190     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1191     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1192     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1193     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1194     real             rcutoff_scalar;
1195     real             *shiftvec,*fshift,*x,*f;
1196     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1197     real             scratch[4*DIM];
1198     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1199     real *           vdwioffsetptr0;
1200     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1201     real *           vdwioffsetptr1;
1202     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1203     real *           vdwioffsetptr2;
1204     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1205     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1206     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1207     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1208     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1209     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1210     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1211     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1212     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1213     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1214     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1215     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1216     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1217     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1218     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1219     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1220     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1221     real             *charge;
1222     __m256           dummy_mask,cutoff_mask;
1223     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1224     __m256           one     = _mm256_set1_ps(1.0);
1225     __m256           two     = _mm256_set1_ps(2.0);
1226     x                = xx[0];
1227     f                = ff[0];
1228
1229     nri              = nlist->nri;
1230     iinr             = nlist->iinr;
1231     jindex           = nlist->jindex;
1232     jjnr             = nlist->jjnr;
1233     shiftidx         = nlist->shift;
1234     gid              = nlist->gid;
1235     shiftvec         = fr->shift_vec[0];
1236     fshift           = fr->fshift[0];
1237     facel            = _mm256_set1_ps(fr->epsfac);
1238     charge           = mdatoms->chargeA;
1239     krf              = _mm256_set1_ps(fr->ic->k_rf);
1240     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
1241     crf              = _mm256_set1_ps(fr->ic->c_rf);
1242
1243     /* Setup water-specific parameters */
1244     inr              = nlist->iinr[0];
1245     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1246     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1247     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1248
1249     jq0              = _mm256_set1_ps(charge[inr+0]);
1250     jq1              = _mm256_set1_ps(charge[inr+1]);
1251     jq2              = _mm256_set1_ps(charge[inr+2]);
1252     qq00             = _mm256_mul_ps(iq0,jq0);
1253     qq01             = _mm256_mul_ps(iq0,jq1);
1254     qq02             = _mm256_mul_ps(iq0,jq2);
1255     qq10             = _mm256_mul_ps(iq1,jq0);
1256     qq11             = _mm256_mul_ps(iq1,jq1);
1257     qq12             = _mm256_mul_ps(iq1,jq2);
1258     qq20             = _mm256_mul_ps(iq2,jq0);
1259     qq21             = _mm256_mul_ps(iq2,jq1);
1260     qq22             = _mm256_mul_ps(iq2,jq2);
1261
1262     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1263     rcutoff_scalar   = fr->rcoulomb;
1264     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1265     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1266
1267     /* Avoid stupid compiler warnings */
1268     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1269     j_coord_offsetA = 0;
1270     j_coord_offsetB = 0;
1271     j_coord_offsetC = 0;
1272     j_coord_offsetD = 0;
1273     j_coord_offsetE = 0;
1274     j_coord_offsetF = 0;
1275     j_coord_offsetG = 0;
1276     j_coord_offsetH = 0;
1277
1278     outeriter        = 0;
1279     inneriter        = 0;
1280
1281     for(iidx=0;iidx<4*DIM;iidx++)
1282     {
1283         scratch[iidx] = 0.0;
1284     }
1285
1286     /* Start outer loop over neighborlists */
1287     for(iidx=0; iidx<nri; iidx++)
1288     {
1289         /* Load shift vector for this list */
1290         i_shift_offset   = DIM*shiftidx[iidx];
1291
1292         /* Load limits for loop over neighbors */
1293         j_index_start    = jindex[iidx];
1294         j_index_end      = jindex[iidx+1];
1295
1296         /* Get outer coordinate index */
1297         inr              = iinr[iidx];
1298         i_coord_offset   = DIM*inr;
1299
1300         /* Load i particle coords and add shift vector */
1301         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1302                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1303
1304         fix0             = _mm256_setzero_ps();
1305         fiy0             = _mm256_setzero_ps();
1306         fiz0             = _mm256_setzero_ps();
1307         fix1             = _mm256_setzero_ps();
1308         fiy1             = _mm256_setzero_ps();
1309         fiz1             = _mm256_setzero_ps();
1310         fix2             = _mm256_setzero_ps();
1311         fiy2             = _mm256_setzero_ps();
1312         fiz2             = _mm256_setzero_ps();
1313
1314         /* Start inner kernel loop */
1315         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1316         {
1317
1318             /* Get j neighbor index, and coordinate index */
1319             jnrA             = jjnr[jidx];
1320             jnrB             = jjnr[jidx+1];
1321             jnrC             = jjnr[jidx+2];
1322             jnrD             = jjnr[jidx+3];
1323             jnrE             = jjnr[jidx+4];
1324             jnrF             = jjnr[jidx+5];
1325             jnrG             = jjnr[jidx+6];
1326             jnrH             = jjnr[jidx+7];
1327             j_coord_offsetA  = DIM*jnrA;
1328             j_coord_offsetB  = DIM*jnrB;
1329             j_coord_offsetC  = DIM*jnrC;
1330             j_coord_offsetD  = DIM*jnrD;
1331             j_coord_offsetE  = DIM*jnrE;
1332             j_coord_offsetF  = DIM*jnrF;
1333             j_coord_offsetG  = DIM*jnrG;
1334             j_coord_offsetH  = DIM*jnrH;
1335
1336             /* load j atom coordinates */
1337             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1338                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1339                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1340                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1341                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1342
1343             /* Calculate displacement vector */
1344             dx00             = _mm256_sub_ps(ix0,jx0);
1345             dy00             = _mm256_sub_ps(iy0,jy0);
1346             dz00             = _mm256_sub_ps(iz0,jz0);
1347             dx01             = _mm256_sub_ps(ix0,jx1);
1348             dy01             = _mm256_sub_ps(iy0,jy1);
1349             dz01             = _mm256_sub_ps(iz0,jz1);
1350             dx02             = _mm256_sub_ps(ix0,jx2);
1351             dy02             = _mm256_sub_ps(iy0,jy2);
1352             dz02             = _mm256_sub_ps(iz0,jz2);
1353             dx10             = _mm256_sub_ps(ix1,jx0);
1354             dy10             = _mm256_sub_ps(iy1,jy0);
1355             dz10             = _mm256_sub_ps(iz1,jz0);
1356             dx11             = _mm256_sub_ps(ix1,jx1);
1357             dy11             = _mm256_sub_ps(iy1,jy1);
1358             dz11             = _mm256_sub_ps(iz1,jz1);
1359             dx12             = _mm256_sub_ps(ix1,jx2);
1360             dy12             = _mm256_sub_ps(iy1,jy2);
1361             dz12             = _mm256_sub_ps(iz1,jz2);
1362             dx20             = _mm256_sub_ps(ix2,jx0);
1363             dy20             = _mm256_sub_ps(iy2,jy0);
1364             dz20             = _mm256_sub_ps(iz2,jz0);
1365             dx21             = _mm256_sub_ps(ix2,jx1);
1366             dy21             = _mm256_sub_ps(iy2,jy1);
1367             dz21             = _mm256_sub_ps(iz2,jz1);
1368             dx22             = _mm256_sub_ps(ix2,jx2);
1369             dy22             = _mm256_sub_ps(iy2,jy2);
1370             dz22             = _mm256_sub_ps(iz2,jz2);
1371
1372             /* Calculate squared distance and things based on it */
1373             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1374             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1375             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1376             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1377             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1378             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1379             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1380             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1381             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1382
1383             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1384             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1385             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1386             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1387             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1388             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1389             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1390             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1391             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1392
1393             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1394             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1395             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1396             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1397             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1398             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1399             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1400             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1401             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1402
1403             fjx0             = _mm256_setzero_ps();
1404             fjy0             = _mm256_setzero_ps();
1405             fjz0             = _mm256_setzero_ps();
1406             fjx1             = _mm256_setzero_ps();
1407             fjy1             = _mm256_setzero_ps();
1408             fjz1             = _mm256_setzero_ps();
1409             fjx2             = _mm256_setzero_ps();
1410             fjy2             = _mm256_setzero_ps();
1411             fjz2             = _mm256_setzero_ps();
1412
1413             /**************************
1414              * CALCULATE INTERACTIONS *
1415              **************************/
1416
1417             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1418             {
1419
1420             /* REACTION-FIELD ELECTROSTATICS */
1421             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1422
1423             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1424
1425             fscal            = felec;
1426
1427             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1428
1429             /* Calculate temporary vectorial force */
1430             tx               = _mm256_mul_ps(fscal,dx00);
1431             ty               = _mm256_mul_ps(fscal,dy00);
1432             tz               = _mm256_mul_ps(fscal,dz00);
1433
1434             /* Update vectorial force */
1435             fix0             = _mm256_add_ps(fix0,tx);
1436             fiy0             = _mm256_add_ps(fiy0,ty);
1437             fiz0             = _mm256_add_ps(fiz0,tz);
1438
1439             fjx0             = _mm256_add_ps(fjx0,tx);
1440             fjy0             = _mm256_add_ps(fjy0,ty);
1441             fjz0             = _mm256_add_ps(fjz0,tz);
1442
1443             }
1444
1445             /**************************
1446              * CALCULATE INTERACTIONS *
1447              **************************/
1448
1449             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1450             {
1451
1452             /* REACTION-FIELD ELECTROSTATICS */
1453             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1454
1455             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1456
1457             fscal            = felec;
1458
1459             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1460
1461             /* Calculate temporary vectorial force */
1462             tx               = _mm256_mul_ps(fscal,dx01);
1463             ty               = _mm256_mul_ps(fscal,dy01);
1464             tz               = _mm256_mul_ps(fscal,dz01);
1465
1466             /* Update vectorial force */
1467             fix0             = _mm256_add_ps(fix0,tx);
1468             fiy0             = _mm256_add_ps(fiy0,ty);
1469             fiz0             = _mm256_add_ps(fiz0,tz);
1470
1471             fjx1             = _mm256_add_ps(fjx1,tx);
1472             fjy1             = _mm256_add_ps(fjy1,ty);
1473             fjz1             = _mm256_add_ps(fjz1,tz);
1474
1475             }
1476
1477             /**************************
1478              * CALCULATE INTERACTIONS *
1479              **************************/
1480
1481             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1482             {
1483
1484             /* REACTION-FIELD ELECTROSTATICS */
1485             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1486
1487             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1488
1489             fscal            = felec;
1490
1491             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1492
1493             /* Calculate temporary vectorial force */
1494             tx               = _mm256_mul_ps(fscal,dx02);
1495             ty               = _mm256_mul_ps(fscal,dy02);
1496             tz               = _mm256_mul_ps(fscal,dz02);
1497
1498             /* Update vectorial force */
1499             fix0             = _mm256_add_ps(fix0,tx);
1500             fiy0             = _mm256_add_ps(fiy0,ty);
1501             fiz0             = _mm256_add_ps(fiz0,tz);
1502
1503             fjx2             = _mm256_add_ps(fjx2,tx);
1504             fjy2             = _mm256_add_ps(fjy2,ty);
1505             fjz2             = _mm256_add_ps(fjz2,tz);
1506
1507             }
1508
1509             /**************************
1510              * CALCULATE INTERACTIONS *
1511              **************************/
1512
1513             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1514             {
1515
1516             /* REACTION-FIELD ELECTROSTATICS */
1517             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1518
1519             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1520
1521             fscal            = felec;
1522
1523             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1524
1525             /* Calculate temporary vectorial force */
1526             tx               = _mm256_mul_ps(fscal,dx10);
1527             ty               = _mm256_mul_ps(fscal,dy10);
1528             tz               = _mm256_mul_ps(fscal,dz10);
1529
1530             /* Update vectorial force */
1531             fix1             = _mm256_add_ps(fix1,tx);
1532             fiy1             = _mm256_add_ps(fiy1,ty);
1533             fiz1             = _mm256_add_ps(fiz1,tz);
1534
1535             fjx0             = _mm256_add_ps(fjx0,tx);
1536             fjy0             = _mm256_add_ps(fjy0,ty);
1537             fjz0             = _mm256_add_ps(fjz0,tz);
1538
1539             }
1540
1541             /**************************
1542              * CALCULATE INTERACTIONS *
1543              **************************/
1544
1545             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1546             {
1547
1548             /* REACTION-FIELD ELECTROSTATICS */
1549             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1550
1551             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1552
1553             fscal            = felec;
1554
1555             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1556
1557             /* Calculate temporary vectorial force */
1558             tx               = _mm256_mul_ps(fscal,dx11);
1559             ty               = _mm256_mul_ps(fscal,dy11);
1560             tz               = _mm256_mul_ps(fscal,dz11);
1561
1562             /* Update vectorial force */
1563             fix1             = _mm256_add_ps(fix1,tx);
1564             fiy1             = _mm256_add_ps(fiy1,ty);
1565             fiz1             = _mm256_add_ps(fiz1,tz);
1566
1567             fjx1             = _mm256_add_ps(fjx1,tx);
1568             fjy1             = _mm256_add_ps(fjy1,ty);
1569             fjz1             = _mm256_add_ps(fjz1,tz);
1570
1571             }
1572
1573             /**************************
1574              * CALCULATE INTERACTIONS *
1575              **************************/
1576
1577             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1578             {
1579
1580             /* REACTION-FIELD ELECTROSTATICS */
1581             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
1582
1583             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
1584
1585             fscal            = felec;
1586
1587             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1588
1589             /* Calculate temporary vectorial force */
1590             tx               = _mm256_mul_ps(fscal,dx12);
1591             ty               = _mm256_mul_ps(fscal,dy12);
1592             tz               = _mm256_mul_ps(fscal,dz12);
1593
1594             /* Update vectorial force */
1595             fix1             = _mm256_add_ps(fix1,tx);
1596             fiy1             = _mm256_add_ps(fiy1,ty);
1597             fiz1             = _mm256_add_ps(fiz1,tz);
1598
1599             fjx2             = _mm256_add_ps(fjx2,tx);
1600             fjy2             = _mm256_add_ps(fjy2,ty);
1601             fjz2             = _mm256_add_ps(fjz2,tz);
1602
1603             }
1604
1605             /**************************
1606              * CALCULATE INTERACTIONS *
1607              **************************/
1608
1609             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1610             {
1611
1612             /* REACTION-FIELD ELECTROSTATICS */
1613             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1614
1615             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1616
1617             fscal            = felec;
1618
1619             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1620
1621             /* Calculate temporary vectorial force */
1622             tx               = _mm256_mul_ps(fscal,dx20);
1623             ty               = _mm256_mul_ps(fscal,dy20);
1624             tz               = _mm256_mul_ps(fscal,dz20);
1625
1626             /* Update vectorial force */
1627             fix2             = _mm256_add_ps(fix2,tx);
1628             fiy2             = _mm256_add_ps(fiy2,ty);
1629             fiz2             = _mm256_add_ps(fiz2,tz);
1630
1631             fjx0             = _mm256_add_ps(fjx0,tx);
1632             fjy0             = _mm256_add_ps(fjy0,ty);
1633             fjz0             = _mm256_add_ps(fjz0,tz);
1634
1635             }
1636
1637             /**************************
1638              * CALCULATE INTERACTIONS *
1639              **************************/
1640
1641             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1642             {
1643
1644             /* REACTION-FIELD ELECTROSTATICS */
1645             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
1646
1647             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
1648
1649             fscal            = felec;
1650
1651             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1652
1653             /* Calculate temporary vectorial force */
1654             tx               = _mm256_mul_ps(fscal,dx21);
1655             ty               = _mm256_mul_ps(fscal,dy21);
1656             tz               = _mm256_mul_ps(fscal,dz21);
1657
1658             /* Update vectorial force */
1659             fix2             = _mm256_add_ps(fix2,tx);
1660             fiy2             = _mm256_add_ps(fiy2,ty);
1661             fiz2             = _mm256_add_ps(fiz2,tz);
1662
1663             fjx1             = _mm256_add_ps(fjx1,tx);
1664             fjy1             = _mm256_add_ps(fjy1,ty);
1665             fjz1             = _mm256_add_ps(fjz1,tz);
1666
1667             }
1668
1669             /**************************
1670              * CALCULATE INTERACTIONS *
1671              **************************/
1672
1673             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1674             {
1675
1676             /* REACTION-FIELD ELECTROSTATICS */
1677             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
1678
1679             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
1680
1681             fscal            = felec;
1682
1683             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1684
1685             /* Calculate temporary vectorial force */
1686             tx               = _mm256_mul_ps(fscal,dx22);
1687             ty               = _mm256_mul_ps(fscal,dy22);
1688             tz               = _mm256_mul_ps(fscal,dz22);
1689
1690             /* Update vectorial force */
1691             fix2             = _mm256_add_ps(fix2,tx);
1692             fiy2             = _mm256_add_ps(fiy2,ty);
1693             fiz2             = _mm256_add_ps(fiz2,tz);
1694
1695             fjx2             = _mm256_add_ps(fjx2,tx);
1696             fjy2             = _mm256_add_ps(fjy2,ty);
1697             fjz2             = _mm256_add_ps(fjz2,tz);
1698
1699             }
1700
1701             fjptrA             = f+j_coord_offsetA;
1702             fjptrB             = f+j_coord_offsetB;
1703             fjptrC             = f+j_coord_offsetC;
1704             fjptrD             = f+j_coord_offsetD;
1705             fjptrE             = f+j_coord_offsetE;
1706             fjptrF             = f+j_coord_offsetF;
1707             fjptrG             = f+j_coord_offsetG;
1708             fjptrH             = f+j_coord_offsetH;
1709
1710             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1711                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1712
1713             /* Inner loop uses 270 flops */
1714         }
1715
1716         if(jidx<j_index_end)
1717         {
1718
1719             /* Get j neighbor index, and coordinate index */
1720             jnrlistA         = jjnr[jidx];
1721             jnrlistB         = jjnr[jidx+1];
1722             jnrlistC         = jjnr[jidx+2];
1723             jnrlistD         = jjnr[jidx+3];
1724             jnrlistE         = jjnr[jidx+4];
1725             jnrlistF         = jjnr[jidx+5];
1726             jnrlistG         = jjnr[jidx+6];
1727             jnrlistH         = jjnr[jidx+7];
1728             /* Sign of each element will be negative for non-real atoms.
1729              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1730              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1731              */
1732             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1733                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1734
1735             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1736             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1737             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1738             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1739             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1740             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1741             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1742             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1743             j_coord_offsetA  = DIM*jnrA;
1744             j_coord_offsetB  = DIM*jnrB;
1745             j_coord_offsetC  = DIM*jnrC;
1746             j_coord_offsetD  = DIM*jnrD;
1747             j_coord_offsetE  = DIM*jnrE;
1748             j_coord_offsetF  = DIM*jnrF;
1749             j_coord_offsetG  = DIM*jnrG;
1750             j_coord_offsetH  = DIM*jnrH;
1751
1752             /* load j atom coordinates */
1753             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1754                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1755                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1756                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1757                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1758
1759             /* Calculate displacement vector */
1760             dx00             = _mm256_sub_ps(ix0,jx0);
1761             dy00             = _mm256_sub_ps(iy0,jy0);
1762             dz00             = _mm256_sub_ps(iz0,jz0);
1763             dx01             = _mm256_sub_ps(ix0,jx1);
1764             dy01             = _mm256_sub_ps(iy0,jy1);
1765             dz01             = _mm256_sub_ps(iz0,jz1);
1766             dx02             = _mm256_sub_ps(ix0,jx2);
1767             dy02             = _mm256_sub_ps(iy0,jy2);
1768             dz02             = _mm256_sub_ps(iz0,jz2);
1769             dx10             = _mm256_sub_ps(ix1,jx0);
1770             dy10             = _mm256_sub_ps(iy1,jy0);
1771             dz10             = _mm256_sub_ps(iz1,jz0);
1772             dx11             = _mm256_sub_ps(ix1,jx1);
1773             dy11             = _mm256_sub_ps(iy1,jy1);
1774             dz11             = _mm256_sub_ps(iz1,jz1);
1775             dx12             = _mm256_sub_ps(ix1,jx2);
1776             dy12             = _mm256_sub_ps(iy1,jy2);
1777             dz12             = _mm256_sub_ps(iz1,jz2);
1778             dx20             = _mm256_sub_ps(ix2,jx0);
1779             dy20             = _mm256_sub_ps(iy2,jy0);
1780             dz20             = _mm256_sub_ps(iz2,jz0);
1781             dx21             = _mm256_sub_ps(ix2,jx1);
1782             dy21             = _mm256_sub_ps(iy2,jy1);
1783             dz21             = _mm256_sub_ps(iz2,jz1);
1784             dx22             = _mm256_sub_ps(ix2,jx2);
1785             dy22             = _mm256_sub_ps(iy2,jy2);
1786             dz22             = _mm256_sub_ps(iz2,jz2);
1787
1788             /* Calculate squared distance and things based on it */
1789             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1790             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1791             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1792             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1793             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1794             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1795             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1796             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1797             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1798
1799             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1800             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1801             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1802             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1803             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1804             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1805             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1806             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1807             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1808
1809             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1810             rinvsq01         = _mm256_mul_ps(rinv01,rinv01);
1811             rinvsq02         = _mm256_mul_ps(rinv02,rinv02);
1812             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1813             rinvsq11         = _mm256_mul_ps(rinv11,rinv11);
1814             rinvsq12         = _mm256_mul_ps(rinv12,rinv12);
1815             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1816             rinvsq21         = _mm256_mul_ps(rinv21,rinv21);
1817             rinvsq22         = _mm256_mul_ps(rinv22,rinv22);
1818
1819             fjx0             = _mm256_setzero_ps();
1820             fjy0             = _mm256_setzero_ps();
1821             fjz0             = _mm256_setzero_ps();
1822             fjx1             = _mm256_setzero_ps();
1823             fjy1             = _mm256_setzero_ps();
1824             fjz1             = _mm256_setzero_ps();
1825             fjx2             = _mm256_setzero_ps();
1826             fjy2             = _mm256_setzero_ps();
1827             fjz2             = _mm256_setzero_ps();
1828
1829             /**************************
1830              * CALCULATE INTERACTIONS *
1831              **************************/
1832
1833             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1834             {
1835
1836             /* REACTION-FIELD ELECTROSTATICS */
1837             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1838
1839             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1840
1841             fscal            = felec;
1842
1843             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1844
1845             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1846
1847             /* Calculate temporary vectorial force */
1848             tx               = _mm256_mul_ps(fscal,dx00);
1849             ty               = _mm256_mul_ps(fscal,dy00);
1850             tz               = _mm256_mul_ps(fscal,dz00);
1851
1852             /* Update vectorial force */
1853             fix0             = _mm256_add_ps(fix0,tx);
1854             fiy0             = _mm256_add_ps(fiy0,ty);
1855             fiz0             = _mm256_add_ps(fiz0,tz);
1856
1857             fjx0             = _mm256_add_ps(fjx0,tx);
1858             fjy0             = _mm256_add_ps(fjy0,ty);
1859             fjz0             = _mm256_add_ps(fjz0,tz);
1860
1861             }
1862
1863             /**************************
1864              * CALCULATE INTERACTIONS *
1865              **************************/
1866
1867             if (gmx_mm256_any_lt(rsq01,rcutoff2))
1868             {
1869
1870             /* REACTION-FIELD ELECTROSTATICS */
1871             felec            = _mm256_mul_ps(qq01,_mm256_sub_ps(_mm256_mul_ps(rinv01,rinvsq01),krf2));
1872
1873             cutoff_mask      = _mm256_cmp_ps(rsq01,rcutoff2,_CMP_LT_OQ);
1874
1875             fscal            = felec;
1876
1877             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1878
1879             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1880
1881             /* Calculate temporary vectorial force */
1882             tx               = _mm256_mul_ps(fscal,dx01);
1883             ty               = _mm256_mul_ps(fscal,dy01);
1884             tz               = _mm256_mul_ps(fscal,dz01);
1885
1886             /* Update vectorial force */
1887             fix0             = _mm256_add_ps(fix0,tx);
1888             fiy0             = _mm256_add_ps(fiy0,ty);
1889             fiz0             = _mm256_add_ps(fiz0,tz);
1890
1891             fjx1             = _mm256_add_ps(fjx1,tx);
1892             fjy1             = _mm256_add_ps(fjy1,ty);
1893             fjz1             = _mm256_add_ps(fjz1,tz);
1894
1895             }
1896
1897             /**************************
1898              * CALCULATE INTERACTIONS *
1899              **************************/
1900
1901             if (gmx_mm256_any_lt(rsq02,rcutoff2))
1902             {
1903
1904             /* REACTION-FIELD ELECTROSTATICS */
1905             felec            = _mm256_mul_ps(qq02,_mm256_sub_ps(_mm256_mul_ps(rinv02,rinvsq02),krf2));
1906
1907             cutoff_mask      = _mm256_cmp_ps(rsq02,rcutoff2,_CMP_LT_OQ);
1908
1909             fscal            = felec;
1910
1911             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1912
1913             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1914
1915             /* Calculate temporary vectorial force */
1916             tx               = _mm256_mul_ps(fscal,dx02);
1917             ty               = _mm256_mul_ps(fscal,dy02);
1918             tz               = _mm256_mul_ps(fscal,dz02);
1919
1920             /* Update vectorial force */
1921             fix0             = _mm256_add_ps(fix0,tx);
1922             fiy0             = _mm256_add_ps(fiy0,ty);
1923             fiz0             = _mm256_add_ps(fiz0,tz);
1924
1925             fjx2             = _mm256_add_ps(fjx2,tx);
1926             fjy2             = _mm256_add_ps(fjy2,ty);
1927             fjz2             = _mm256_add_ps(fjz2,tz);
1928
1929             }
1930
1931             /**************************
1932              * CALCULATE INTERACTIONS *
1933              **************************/
1934
1935             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1936             {
1937
1938             /* REACTION-FIELD ELECTROSTATICS */
1939             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1940
1941             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1942
1943             fscal            = felec;
1944
1945             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1946
1947             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1948
1949             /* Calculate temporary vectorial force */
1950             tx               = _mm256_mul_ps(fscal,dx10);
1951             ty               = _mm256_mul_ps(fscal,dy10);
1952             tz               = _mm256_mul_ps(fscal,dz10);
1953
1954             /* Update vectorial force */
1955             fix1             = _mm256_add_ps(fix1,tx);
1956             fiy1             = _mm256_add_ps(fiy1,ty);
1957             fiz1             = _mm256_add_ps(fiz1,tz);
1958
1959             fjx0             = _mm256_add_ps(fjx0,tx);
1960             fjy0             = _mm256_add_ps(fjy0,ty);
1961             fjz0             = _mm256_add_ps(fjz0,tz);
1962
1963             }
1964
1965             /**************************
1966              * CALCULATE INTERACTIONS *
1967              **************************/
1968
1969             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1970             {
1971
1972             /* REACTION-FIELD ELECTROSTATICS */
1973             felec            = _mm256_mul_ps(qq11,_mm256_sub_ps(_mm256_mul_ps(rinv11,rinvsq11),krf2));
1974
1975             cutoff_mask      = _mm256_cmp_ps(rsq11,rcutoff2,_CMP_LT_OQ);
1976
1977             fscal            = felec;
1978
1979             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1980
1981             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1982
1983             /* Calculate temporary vectorial force */
1984             tx               = _mm256_mul_ps(fscal,dx11);
1985             ty               = _mm256_mul_ps(fscal,dy11);
1986             tz               = _mm256_mul_ps(fscal,dz11);
1987
1988             /* Update vectorial force */
1989             fix1             = _mm256_add_ps(fix1,tx);
1990             fiy1             = _mm256_add_ps(fiy1,ty);
1991             fiz1             = _mm256_add_ps(fiz1,tz);
1992
1993             fjx1             = _mm256_add_ps(fjx1,tx);
1994             fjy1             = _mm256_add_ps(fjy1,ty);
1995             fjz1             = _mm256_add_ps(fjz1,tz);
1996
1997             }
1998
1999             /**************************
2000              * CALCULATE INTERACTIONS *
2001              **************************/
2002
2003             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2004             {
2005
2006             /* REACTION-FIELD ELECTROSTATICS */
2007             felec            = _mm256_mul_ps(qq12,_mm256_sub_ps(_mm256_mul_ps(rinv12,rinvsq12),krf2));
2008
2009             cutoff_mask      = _mm256_cmp_ps(rsq12,rcutoff2,_CMP_LT_OQ);
2010
2011             fscal            = felec;
2012
2013             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2014
2015             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2016
2017             /* Calculate temporary vectorial force */
2018             tx               = _mm256_mul_ps(fscal,dx12);
2019             ty               = _mm256_mul_ps(fscal,dy12);
2020             tz               = _mm256_mul_ps(fscal,dz12);
2021
2022             /* Update vectorial force */
2023             fix1             = _mm256_add_ps(fix1,tx);
2024             fiy1             = _mm256_add_ps(fiy1,ty);
2025             fiz1             = _mm256_add_ps(fiz1,tz);
2026
2027             fjx2             = _mm256_add_ps(fjx2,tx);
2028             fjy2             = _mm256_add_ps(fjy2,ty);
2029             fjz2             = _mm256_add_ps(fjz2,tz);
2030
2031             }
2032
2033             /**************************
2034              * CALCULATE INTERACTIONS *
2035              **************************/
2036
2037             if (gmx_mm256_any_lt(rsq20,rcutoff2))
2038             {
2039
2040             /* REACTION-FIELD ELECTROSTATICS */
2041             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
2042
2043             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
2044
2045             fscal            = felec;
2046
2047             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2048
2049             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2050
2051             /* Calculate temporary vectorial force */
2052             tx               = _mm256_mul_ps(fscal,dx20);
2053             ty               = _mm256_mul_ps(fscal,dy20);
2054             tz               = _mm256_mul_ps(fscal,dz20);
2055
2056             /* Update vectorial force */
2057             fix2             = _mm256_add_ps(fix2,tx);
2058             fiy2             = _mm256_add_ps(fiy2,ty);
2059             fiz2             = _mm256_add_ps(fiz2,tz);
2060
2061             fjx0             = _mm256_add_ps(fjx0,tx);
2062             fjy0             = _mm256_add_ps(fjy0,ty);
2063             fjz0             = _mm256_add_ps(fjz0,tz);
2064
2065             }
2066
2067             /**************************
2068              * CALCULATE INTERACTIONS *
2069              **************************/
2070
2071             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2072             {
2073
2074             /* REACTION-FIELD ELECTROSTATICS */
2075             felec            = _mm256_mul_ps(qq21,_mm256_sub_ps(_mm256_mul_ps(rinv21,rinvsq21),krf2));
2076
2077             cutoff_mask      = _mm256_cmp_ps(rsq21,rcutoff2,_CMP_LT_OQ);
2078
2079             fscal            = felec;
2080
2081             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2082
2083             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2084
2085             /* Calculate temporary vectorial force */
2086             tx               = _mm256_mul_ps(fscal,dx21);
2087             ty               = _mm256_mul_ps(fscal,dy21);
2088             tz               = _mm256_mul_ps(fscal,dz21);
2089
2090             /* Update vectorial force */
2091             fix2             = _mm256_add_ps(fix2,tx);
2092             fiy2             = _mm256_add_ps(fiy2,ty);
2093             fiz2             = _mm256_add_ps(fiz2,tz);
2094
2095             fjx1             = _mm256_add_ps(fjx1,tx);
2096             fjy1             = _mm256_add_ps(fjy1,ty);
2097             fjz1             = _mm256_add_ps(fjz1,tz);
2098
2099             }
2100
2101             /**************************
2102              * CALCULATE INTERACTIONS *
2103              **************************/
2104
2105             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2106             {
2107
2108             /* REACTION-FIELD ELECTROSTATICS */
2109             felec            = _mm256_mul_ps(qq22,_mm256_sub_ps(_mm256_mul_ps(rinv22,rinvsq22),krf2));
2110
2111             cutoff_mask      = _mm256_cmp_ps(rsq22,rcutoff2,_CMP_LT_OQ);
2112
2113             fscal            = felec;
2114
2115             fscal            = _mm256_and_ps(fscal,cutoff_mask);
2116
2117             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2118
2119             /* Calculate temporary vectorial force */
2120             tx               = _mm256_mul_ps(fscal,dx22);
2121             ty               = _mm256_mul_ps(fscal,dy22);
2122             tz               = _mm256_mul_ps(fscal,dz22);
2123
2124             /* Update vectorial force */
2125             fix2             = _mm256_add_ps(fix2,tx);
2126             fiy2             = _mm256_add_ps(fiy2,ty);
2127             fiz2             = _mm256_add_ps(fiz2,tz);
2128
2129             fjx2             = _mm256_add_ps(fjx2,tx);
2130             fjy2             = _mm256_add_ps(fjy2,ty);
2131             fjz2             = _mm256_add_ps(fjz2,tz);
2132
2133             }
2134
2135             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2136             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2137             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2138             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2139             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2140             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2141             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2142             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2143
2144             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2145                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2146
2147             /* Inner loop uses 270 flops */
2148         }
2149
2150         /* End of innermost loop */
2151
2152         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2153                                                  f+i_coord_offset,fshift+i_shift_offset);
2154
2155         /* Increment number of inner iterations */
2156         inneriter                  += j_index_end - j_index_start;
2157
2158         /* Outer loop uses 18 flops */
2159     }
2160
2161     /* Increment number of outer iterations */
2162     outeriter        += nri;
2163
2164     /* Update outer/inner flops */
2165
2166     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*270);
2167 }