src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwNone_GeomW4W4_avx_128_fma_double.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_128_fma_double 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_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwNone_GeomW4W4_VF_avx_128_fma_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx1A,vdwjidx1B;
  89     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  90     int              vdwjidx2A,vdwjidx2B;
  91     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  92     int              vdwjidx3A,vdwjidx3B;
  93     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  94     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  95     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  96     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  97     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  98     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  99     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 100     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 101     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 102     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 103     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 104     real             *charge;
 105     __m128i          ewitab;
 106     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     real             *ewtab;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125
 126     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 134     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 135     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 136
 137     jq1              = _mm_set1_pd(charge[inr+1]);
 138     jq2              = _mm_set1_pd(charge[inr+2]);
 139     jq3              = _mm_set1_pd(charge[inr+3]);
 140     qq11             = _mm_mul_pd(iq1,jq1);
 141     qq12             = _mm_mul_pd(iq1,jq2);
 142     qq13             = _mm_mul_pd(iq1,jq3);
 143     qq21             = _mm_mul_pd(iq2,jq1);
 144     qq22             = _mm_mul_pd(iq2,jq2);
 145     qq23             = _mm_mul_pd(iq2,jq3);
 146     qq31             = _mm_mul_pd(iq3,jq1);
 147     qq32             = _mm_mul_pd(iq3,jq2);
 148     qq33             = _mm_mul_pd(iq3,jq3);
 149
 150     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 174                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 175
 176         fix1             = _mm_setzero_pd();
 177         fiy1             = _mm_setzero_pd();
 178         fiz1             = _mm_setzero_pd();
 179         fix2             = _mm_setzero_pd();
 180         fiy2             = _mm_setzero_pd();
 181         fiz2             = _mm_setzero_pd();
 182         fix3             = _mm_setzero_pd();
 183         fiy3             = _mm_setzero_pd();
 184         fiz3             = _mm_setzero_pd();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm_setzero_pd();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             j_coord_offsetA  = DIM*jnrA;
 197             j_coord_offsetB  = DIM*jnrB;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 201                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 202
 203             /* Calculate displacement vector */
 204             dx11             = _mm_sub_pd(ix1,jx1);
 205             dy11             = _mm_sub_pd(iy1,jy1);
 206             dz11             = _mm_sub_pd(iz1,jz1);
 207             dx12             = _mm_sub_pd(ix1,jx2);
 208             dy12             = _mm_sub_pd(iy1,jy2);
 209             dz12             = _mm_sub_pd(iz1,jz2);
 210             dx13             = _mm_sub_pd(ix1,jx3);
 211             dy13             = _mm_sub_pd(iy1,jy3);
 212             dz13             = _mm_sub_pd(iz1,jz3);
 213             dx21             = _mm_sub_pd(ix2,jx1);
 214             dy21             = _mm_sub_pd(iy2,jy1);
 215             dz21             = _mm_sub_pd(iz2,jz1);
 216             dx22             = _mm_sub_pd(ix2,jx2);
 217             dy22             = _mm_sub_pd(iy2,jy2);
 218             dz22             = _mm_sub_pd(iz2,jz2);
 219             dx23             = _mm_sub_pd(ix2,jx3);
 220             dy23             = _mm_sub_pd(iy2,jy3);
 221             dz23             = _mm_sub_pd(iz2,jz3);
 222             dx31             = _mm_sub_pd(ix3,jx1);
 223             dy31             = _mm_sub_pd(iy3,jy1);
 224             dz31             = _mm_sub_pd(iz3,jz1);
 225             dx32             = _mm_sub_pd(ix3,jx2);
 226             dy32             = _mm_sub_pd(iy3,jy2);
 227             dz32             = _mm_sub_pd(iz3,jz2);
 228             dx33             = _mm_sub_pd(ix3,jx3);
 229             dy33             = _mm_sub_pd(iy3,jy3);
 230             dz33             = _mm_sub_pd(iz3,jz3);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 234             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 235             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 236             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 237             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 238             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 239             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 240             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 241             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 242
 243             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 244             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 245             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 246             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 247             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 248             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 249             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 250             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 251             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 252
 253             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 254             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 255             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 256             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 257             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 258             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 259             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 260             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 261             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 262
 263             fjx1             = _mm_setzero_pd();
 264             fjy1             = _mm_setzero_pd();
 265             fjz1             = _mm_setzero_pd();
 266             fjx2             = _mm_setzero_pd();
 267             fjy2             = _mm_setzero_pd();
 268             fjz2             = _mm_setzero_pd();
 269             fjx3             = _mm_setzero_pd();
 270             fjy3             = _mm_setzero_pd();
 271             fjz3             = _mm_setzero_pd();
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             r11              = _mm_mul_pd(rsq11,rinv11);
 278
 279             /* EWALD ELECTROSTATICS */
 280
 281             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 282             ewrt             = _mm_mul_pd(r11,ewtabscale);
 283             ewitab           = _mm_cvttpd_epi32(ewrt);
 284 #ifdef __XOP__
 285             eweps            = _mm_frcz_pd(ewrt);
 286 #else
 287             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 288 #endif
 289             twoeweps         = _mm_add_pd(eweps,eweps);
 290             ewitab           = _mm_slli_epi32(ewitab,2);
 291             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 292             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 293             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 294             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 295             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 296             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 297             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 298             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 299             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 300             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velecsum         = _mm_add_pd(velecsum,velec);
 304
 305             fscal            = felec;
 306
 307             /* Update vectorial force */
 308             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 309             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 310             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 311
 312             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 313             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 314             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             r12              = _mm_mul_pd(rsq12,rinv12);
 321
 322             /* EWALD ELECTROSTATICS */
 323
 324             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 325             ewrt             = _mm_mul_pd(r12,ewtabscale);
 326             ewitab           = _mm_cvttpd_epi32(ewrt);
 327 #ifdef __XOP__
 328             eweps            = _mm_frcz_pd(ewrt);
 329 #else
 330             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 331 #endif
 332             twoeweps         = _mm_add_pd(eweps,eweps);
 333             ewitab           = _mm_slli_epi32(ewitab,2);
 334             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 335             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 336             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 337             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 338             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 339             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 340             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 341             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 342             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 343             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velecsum         = _mm_add_pd(velecsum,velec);
 347
 348             fscal            = felec;
 349
 350             /* Update vectorial force */
 351             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 352             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 353             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 354
 355             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 356             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 357             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             r13              = _mm_mul_pd(rsq13,rinv13);
 364
 365             /* EWALD ELECTROSTATICS */
 366
 367             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 368             ewrt             = _mm_mul_pd(r13,ewtabscale);
 369             ewitab           = _mm_cvttpd_epi32(ewrt);
 370 #ifdef __XOP__
 371             eweps            = _mm_frcz_pd(ewrt);
 372 #else
 373             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 374 #endif
 375             twoeweps         = _mm_add_pd(eweps,eweps);
 376             ewitab           = _mm_slli_epi32(ewitab,2);
 377             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 378             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 379             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 380             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 381             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 382             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 383             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 384             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 385             velec            = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
 386             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velecsum         = _mm_add_pd(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             /* Update vectorial force */
 394             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 395             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 396             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 397
 398             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 399             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 400             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 401
 402             /**************************
 403              * CALCULATE INTERACTIONS *
 404              **************************/
 405
 406             r21              = _mm_mul_pd(rsq21,rinv21);
 407
 408             /* EWALD ELECTROSTATICS */
 409
 410             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 411             ewrt             = _mm_mul_pd(r21,ewtabscale);
 412             ewitab           = _mm_cvttpd_epi32(ewrt);
 413 #ifdef __XOP__
 414             eweps            = _mm_frcz_pd(ewrt);
 415 #else
 416             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 417 #endif
 418             twoeweps         = _mm_add_pd(eweps,eweps);
 419             ewitab           = _mm_slli_epi32(ewitab,2);
 420             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 421             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 422             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 423             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 424             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 425             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 426             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 427             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 428             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 429             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 430
 431             /* Update potential sum for this i atom from the interaction with this j atom. */
 432             velecsum         = _mm_add_pd(velecsum,velec);
 433
 434             fscal            = felec;
 435
 436             /* Update vectorial force */
 437             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 438             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 439             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 440
 441             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 442             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 443             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             r22              = _mm_mul_pd(rsq22,rinv22);
 450
 451             /* EWALD ELECTROSTATICS */
 452
 453             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 454             ewrt             = _mm_mul_pd(r22,ewtabscale);
 455             ewitab           = _mm_cvttpd_epi32(ewrt);
 456 #ifdef __XOP__
 457             eweps            = _mm_frcz_pd(ewrt);
 458 #else
 459             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 460 #endif
 461             twoeweps         = _mm_add_pd(eweps,eweps);
 462             ewitab           = _mm_slli_epi32(ewitab,2);
 463             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 464             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 465             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 466             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 467             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 468             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 469             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 470             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 471             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 472             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 473
 474             /* Update potential sum for this i atom from the interaction with this j atom. */
 475             velecsum         = _mm_add_pd(velecsum,velec);
 476
 477             fscal            = felec;
 478
 479             /* Update vectorial force */
 480             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 481             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 482             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 483
 484             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 485             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 486             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 487
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             r23              = _mm_mul_pd(rsq23,rinv23);
 493
 494             /* EWALD ELECTROSTATICS */
 495
 496             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 497             ewrt             = _mm_mul_pd(r23,ewtabscale);
 498             ewitab           = _mm_cvttpd_epi32(ewrt);
 499 #ifdef __XOP__
 500             eweps            = _mm_frcz_pd(ewrt);
 501 #else
 502             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 503 #endif
 504             twoeweps         = _mm_add_pd(eweps,eweps);
 505             ewitab           = _mm_slli_epi32(ewitab,2);
 506             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 507             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 508             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 509             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 510             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 511             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 512             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 513             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 514             velec            = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
 515             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
 516
 517             /* Update potential sum for this i atom from the interaction with this j atom. */
 518             velecsum         = _mm_add_pd(velecsum,velec);
 519
 520             fscal            = felec;
 521
 522             /* Update vectorial force */
 523             fix2             = _mm_macc_pd(dx23,fscal,fix2);
 524             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
 525             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
 526
 527             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
 528             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
 529             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             r31              = _mm_mul_pd(rsq31,rinv31);
 536
 537             /* EWALD ELECTROSTATICS */
 538
 539             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 540             ewrt             = _mm_mul_pd(r31,ewtabscale);
 541             ewitab           = _mm_cvttpd_epi32(ewrt);
 542 #ifdef __XOP__
 543             eweps            = _mm_frcz_pd(ewrt);
 544 #else
 545             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 546 #endif
 547             twoeweps         = _mm_add_pd(eweps,eweps);
 548             ewitab           = _mm_slli_epi32(ewitab,2);
 549             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 550             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 551             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 552             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 553             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 554             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 555             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 556             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 557             velec            = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
 558             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             velecsum         = _mm_add_pd(velecsum,velec);
 562
 563             fscal            = felec;
 564
 565             /* Update vectorial force */
 566             fix3             = _mm_macc_pd(dx31,fscal,fix3);
 567             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
 568             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
 569
 570             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
 571             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
 572             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
 573
 574             /**************************
 575              * CALCULATE INTERACTIONS *
 576              **************************/
 577
 578             r32              = _mm_mul_pd(rsq32,rinv32);
 579
 580             /* EWALD ELECTROSTATICS */
 581
 582             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 583             ewrt             = _mm_mul_pd(r32,ewtabscale);
 584             ewitab           = _mm_cvttpd_epi32(ewrt);
 585 #ifdef __XOP__
 586             eweps            = _mm_frcz_pd(ewrt);
 587 #else
 588             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 589 #endif
 590             twoeweps         = _mm_add_pd(eweps,eweps);
 591             ewitab           = _mm_slli_epi32(ewitab,2);
 592             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 593             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 594             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 595             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 596             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 597             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 598             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 599             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 600             velec            = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
 601             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
 602
 603             /* Update potential sum for this i atom from the interaction with this j atom. */
 604             velecsum         = _mm_add_pd(velecsum,velec);
 605
 606             fscal            = felec;
 607
 608             /* Update vectorial force */
 609             fix3             = _mm_macc_pd(dx32,fscal,fix3);
 610             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
 611             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
 612
 613             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
 614             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
 615             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             r33              = _mm_mul_pd(rsq33,rinv33);
 622
 623             /* EWALD ELECTROSTATICS */
 624
 625             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 626             ewrt             = _mm_mul_pd(r33,ewtabscale);
 627             ewitab           = _mm_cvttpd_epi32(ewrt);
 628 #ifdef __XOP__
 629             eweps            = _mm_frcz_pd(ewrt);
 630 #else
 631             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 632 #endif
 633             twoeweps         = _mm_add_pd(eweps,eweps);
 634             ewitab           = _mm_slli_epi32(ewitab,2);
 635             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 636             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 637             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 638             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 639             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 640             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 641             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 642             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 643             velec            = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
 644             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
 645
 646             /* Update potential sum for this i atom from the interaction with this j atom. */
 647             velecsum         = _mm_add_pd(velecsum,velec);
 648
 649             fscal            = felec;
 650
 651             /* Update vectorial force */
 652             fix3             = _mm_macc_pd(dx33,fscal,fix3);
 653             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
 654             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
 655
 656             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
 657             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
 658             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
 659
 660             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA+DIM,f+j_coord_offsetB+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 661
 662             /* Inner loop uses 396 flops */
 663         }
 664
 665         if(jidx<j_index_end)
 666         {
 667
 668             jnrA             = jjnr[jidx];
 669             j_coord_offsetA  = DIM*jnrA;
 670
 671             /* load j atom coordinates */
 672             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
 673                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 674
 675             /* Calculate displacement vector */
 676             dx11             = _mm_sub_pd(ix1,jx1);
 677             dy11             = _mm_sub_pd(iy1,jy1);
 678             dz11             = _mm_sub_pd(iz1,jz1);
 679             dx12             = _mm_sub_pd(ix1,jx2);
 680             dy12             = _mm_sub_pd(iy1,jy2);
 681             dz12             = _mm_sub_pd(iz1,jz2);
 682             dx13             = _mm_sub_pd(ix1,jx3);
 683             dy13             = _mm_sub_pd(iy1,jy3);
 684             dz13             = _mm_sub_pd(iz1,jz3);
 685             dx21             = _mm_sub_pd(ix2,jx1);
 686             dy21             = _mm_sub_pd(iy2,jy1);
 687             dz21             = _mm_sub_pd(iz2,jz1);
 688             dx22             = _mm_sub_pd(ix2,jx2);
 689             dy22             = _mm_sub_pd(iy2,jy2);
 690             dz22             = _mm_sub_pd(iz2,jz2);
 691             dx23             = _mm_sub_pd(ix2,jx3);
 692             dy23             = _mm_sub_pd(iy2,jy3);
 693             dz23             = _mm_sub_pd(iz2,jz3);
 694             dx31             = _mm_sub_pd(ix3,jx1);
 695             dy31             = _mm_sub_pd(iy3,jy1);
 696             dz31             = _mm_sub_pd(iz3,jz1);
 697             dx32             = _mm_sub_pd(ix3,jx2);
 698             dy32             = _mm_sub_pd(iy3,jy2);
 699             dz32             = _mm_sub_pd(iz3,jz2);
 700             dx33             = _mm_sub_pd(ix3,jx3);
 701             dy33             = _mm_sub_pd(iy3,jy3);
 702             dz33             = _mm_sub_pd(iz3,jz3);
 703
 704             /* Calculate squared distance and things based on it */
 705             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 706             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 707             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 708             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 709             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 710             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 711             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 712             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 713             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 714
 715             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 716             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 717             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 718             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 719             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 720             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 721             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 722             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 723             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 724
 725             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 726             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 727             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 728             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 729             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 730             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 731             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 732             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 733             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 734
 735             fjx1             = _mm_setzero_pd();
 736             fjy1             = _mm_setzero_pd();
 737             fjz1             = _mm_setzero_pd();
 738             fjx2             = _mm_setzero_pd();
 739             fjy2             = _mm_setzero_pd();
 740             fjz2             = _mm_setzero_pd();
 741             fjx3             = _mm_setzero_pd();
 742             fjy3             = _mm_setzero_pd();
 743             fjz3             = _mm_setzero_pd();
 744
 745             /**************************
 746              * CALCULATE INTERACTIONS *
 747              **************************/
 748
 749             r11              = _mm_mul_pd(rsq11,rinv11);
 750
 751             /* EWALD ELECTROSTATICS */
 752
 753             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 754             ewrt             = _mm_mul_pd(r11,ewtabscale);
 755             ewitab           = _mm_cvttpd_epi32(ewrt);
 756 #ifdef __XOP__
 757             eweps            = _mm_frcz_pd(ewrt);
 758 #else
 759             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 760 #endif
 761             twoeweps         = _mm_add_pd(eweps,eweps);
 762             ewitab           = _mm_slli_epi32(ewitab,2);
 763             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 764             ewtabD           = _mm_setzero_pd();
 765             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 766             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 767             ewtabFn          = _mm_setzero_pd();
 768             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 769             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 770             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 771             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 772             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 773
 774             /* Update potential sum for this i atom from the interaction with this j atom. */
 775             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 776             velecsum         = _mm_add_pd(velecsum,velec);
 777
 778             fscal            = felec;
 779
 780             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 781
 782             /* Update vectorial force */
 783             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 784             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 785             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 786
 787             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 788             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 789             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             r12              = _mm_mul_pd(rsq12,rinv12);
 796
 797             /* EWALD ELECTROSTATICS */
 798
 799             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 800             ewrt             = _mm_mul_pd(r12,ewtabscale);
 801             ewitab           = _mm_cvttpd_epi32(ewrt);
 802 #ifdef __XOP__
 803             eweps            = _mm_frcz_pd(ewrt);
 804 #else
 805             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 806 #endif
 807             twoeweps         = _mm_add_pd(eweps,eweps);
 808             ewitab           = _mm_slli_epi32(ewitab,2);
 809             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 810             ewtabD           = _mm_setzero_pd();
 811             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 812             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 813             ewtabFn          = _mm_setzero_pd();
 814             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 815             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 816             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 817             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 818             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 819
 820             /* Update potential sum for this i atom from the interaction with this j atom. */
 821             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 822             velecsum         = _mm_add_pd(velecsum,velec);
 823
 824             fscal            = felec;
 825
 826             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 827
 828             /* Update vectorial force */
 829             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 830             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 831             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 832
 833             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 834             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 835             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 836
 837             /**************************
 838              * CALCULATE INTERACTIONS *
 839              **************************/
 840
 841             r13              = _mm_mul_pd(rsq13,rinv13);
 842
 843             /* EWALD ELECTROSTATICS */
 844
 845             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 846             ewrt             = _mm_mul_pd(r13,ewtabscale);
 847             ewitab           = _mm_cvttpd_epi32(ewrt);
 848 #ifdef __XOP__
 849             eweps            = _mm_frcz_pd(ewrt);
 850 #else
 851             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 852 #endif
 853             twoeweps         = _mm_add_pd(eweps,eweps);
 854             ewitab           = _mm_slli_epi32(ewitab,2);
 855             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 856             ewtabD           = _mm_setzero_pd();
 857             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 858             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 859             ewtabFn          = _mm_setzero_pd();
 860             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 861             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 862             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 863             velec            = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
 864             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 865
 866             /* Update potential sum for this i atom from the interaction with this j atom. */
 867             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 868             velecsum         = _mm_add_pd(velecsum,velec);
 869
 870             fscal            = felec;
 871
 872             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 873
 874             /* Update vectorial force */
 875             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 876             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 877             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 878
 879             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 880             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 881             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 882
 883             /**************************
 884              * CALCULATE INTERACTIONS *
 885              **************************/
 886
 887             r21              = _mm_mul_pd(rsq21,rinv21);
 888
 889             /* EWALD ELECTROSTATICS */
 890
 891             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 892             ewrt             = _mm_mul_pd(r21,ewtabscale);
 893             ewitab           = _mm_cvttpd_epi32(ewrt);
 894 #ifdef __XOP__
 895             eweps            = _mm_frcz_pd(ewrt);
 896 #else
 897             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 898 #endif
 899             twoeweps         = _mm_add_pd(eweps,eweps);
 900             ewitab           = _mm_slli_epi32(ewitab,2);
 901             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 902             ewtabD           = _mm_setzero_pd();
 903             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 904             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 905             ewtabFn          = _mm_setzero_pd();
 906             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 907             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 908             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 909             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 910             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 911
 912             /* Update potential sum for this i atom from the interaction with this j atom. */
 913             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 914             velecsum         = _mm_add_pd(velecsum,velec);
 915
 916             fscal            = felec;
 917
 918             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 919
 920             /* Update vectorial force */
 921             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 922             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 923             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 924
 925             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 926             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 927             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 928
 929             /**************************
 930              * CALCULATE INTERACTIONS *
 931              **************************/
 932
 933             r22              = _mm_mul_pd(rsq22,rinv22);
 934
 935             /* EWALD ELECTROSTATICS */
 936
 937             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 938             ewrt             = _mm_mul_pd(r22,ewtabscale);
 939             ewitab           = _mm_cvttpd_epi32(ewrt);
 940 #ifdef __XOP__
 941             eweps            = _mm_frcz_pd(ewrt);
 942 #else
 943             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 944 #endif
 945             twoeweps         = _mm_add_pd(eweps,eweps);
 946             ewitab           = _mm_slli_epi32(ewitab,2);
 947             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 948             ewtabD           = _mm_setzero_pd();
 949             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 950             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 951             ewtabFn          = _mm_setzero_pd();
 952             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 953             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 954             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 955             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 956             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 957
 958             /* Update potential sum for this i atom from the interaction with this j atom. */
 959             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 960             velecsum         = _mm_add_pd(velecsum,velec);
 961
 962             fscal            = felec;
 963
 964             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 965
 966             /* Update vectorial force */
 967             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 968             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 969             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 970
 971             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 972             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 973             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 974
 975             /**************************
 976              * CALCULATE INTERACTIONS *
 977              **************************/
 978
 979             r23              = _mm_mul_pd(rsq23,rinv23);
 980
 981             /* EWALD ELECTROSTATICS */
 982
 983             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 984             ewrt             = _mm_mul_pd(r23,ewtabscale);
 985             ewitab           = _mm_cvttpd_epi32(ewrt);
 986 #ifdef __XOP__
 987             eweps            = _mm_frcz_pd(ewrt);
 988 #else
 989             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 990 #endif
 991             twoeweps         = _mm_add_pd(eweps,eweps);
 992             ewitab           = _mm_slli_epi32(ewitab,2);
 993             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 994             ewtabD           = _mm_setzero_pd();
 995             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 996             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 997             ewtabFn          = _mm_setzero_pd();
 998             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 999             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1000             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1001             velec            = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
1002             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1003
1004             /* Update potential sum for this i atom from the interaction with this j atom. */
1005             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1006             velecsum         = _mm_add_pd(velecsum,velec);
1007
1008             fscal            = felec;
1009
1010             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1011
1012             /* Update vectorial force */
1013             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1014             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1015             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1016
1017             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1018             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1019             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1020
1021             /**************************
1022              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             r31              = _mm_mul_pd(rsq31,rinv31);
1026
1027             /* EWALD ELECTROSTATICS */
1028
1029             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1030             ewrt             = _mm_mul_pd(r31,ewtabscale);
1031             ewitab           = _mm_cvttpd_epi32(ewrt);
1032 #ifdef __XOP__
1033             eweps            = _mm_frcz_pd(ewrt);
1034 #else
1035             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1036 #endif
1037             twoeweps         = _mm_add_pd(eweps,eweps);
1038             ewitab           = _mm_slli_epi32(ewitab,2);
1039             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1040             ewtabD           = _mm_setzero_pd();
1041             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1042             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1043             ewtabFn          = _mm_setzero_pd();
1044             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1045             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1046             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1047             velec            = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
1048             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1049
1050             /* Update potential sum for this i atom from the interaction with this j atom. */
1051             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1052             velecsum         = _mm_add_pd(velecsum,velec);
1053
1054             fscal            = felec;
1055
1056             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1057
1058             /* Update vectorial force */
1059             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1060             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1061             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1062
1063             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1064             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1065             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             r32              = _mm_mul_pd(rsq32,rinv32);
1072
1073             /* EWALD ELECTROSTATICS */
1074
1075             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1076             ewrt             = _mm_mul_pd(r32,ewtabscale);
1077             ewitab           = _mm_cvttpd_epi32(ewrt);
1078 #ifdef __XOP__
1079             eweps            = _mm_frcz_pd(ewrt);
1080 #else
1081             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1082 #endif
1083             twoeweps         = _mm_add_pd(eweps,eweps);
1084             ewitab           = _mm_slli_epi32(ewitab,2);
1085             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1086             ewtabD           = _mm_setzero_pd();
1087             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1088             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1089             ewtabFn          = _mm_setzero_pd();
1090             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1091             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1092             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1093             velec            = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
1094             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1095
1096             /* Update potential sum for this i atom from the interaction with this j atom. */
1097             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1098             velecsum         = _mm_add_pd(velecsum,velec);
1099
1100             fscal            = felec;
1101
1102             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1103
1104             /* Update vectorial force */
1105             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1106             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1107             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1108
1109             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1110             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1111             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1112
1113             /**************************
1114              * CALCULATE INTERACTIONS *
1115              **************************/
1116
1117             r33              = _mm_mul_pd(rsq33,rinv33);
1118
1119             /* EWALD ELECTROSTATICS */
1120
1121             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1122             ewrt             = _mm_mul_pd(r33,ewtabscale);
1123             ewitab           = _mm_cvttpd_epi32(ewrt);
1124 #ifdef __XOP__
1125             eweps            = _mm_frcz_pd(ewrt);
1126 #else
1127             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1128 #endif
1129             twoeweps         = _mm_add_pd(eweps,eweps);
1130             ewitab           = _mm_slli_epi32(ewitab,2);
1131             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1132             ewtabD           = _mm_setzero_pd();
1133             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1134             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1135             ewtabFn          = _mm_setzero_pd();
1136             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1137             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1138             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1139             velec            = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
1140             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1141
1142             /* Update potential sum for this i atom from the interaction with this j atom. */
1143             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1144             velecsum         = _mm_add_pd(velecsum,velec);
1145
1146             fscal            = felec;
1147
1148             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1149
1150             /* Update vectorial force */
1151             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1152             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1153             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1154
1155             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1156             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1157             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1158
1159             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1160
1161             /* Inner loop uses 396 flops */
1162         }
1163
1164         /* End of innermost loop */
1165
1166         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1167                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1168
1169         ggid                        = gid[iidx];
1170         /* Update potential energies */
1171         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1172
1173         /* Increment number of inner iterations */
1174         inneriter                  += j_index_end - j_index_start;
1175
1176         /* Outer loop uses 19 flops */
1177     }
1178
1179     /* Increment number of outer iterations */
1180     outeriter        += nri;
1181
1182     /* Update outer/inner flops */
1183
1184     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*396);
1185 }
1186 /*
1187  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_128_fma_double
1188  * Electrostatics interaction: Ewald
1189  * VdW interaction:            None
1190  * Geometry:                   Water4-Water4
1191  * Calculate force/pot:        Force
1192  */
1193 void
1194 nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_128_fma_double
1195                     (t_nblist                    * gmx_restrict       nlist,
1196                      rvec                        * gmx_restrict          xx,
1197                      rvec                        * gmx_restrict          ff,
1198                      t_forcerec                  * gmx_restrict          fr,
1199                      t_mdatoms                   * gmx_restrict     mdatoms,
1200                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1201                      t_nrnb                      * gmx_restrict        nrnb)
1202 {
1203     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1204      * just 0 for non-waters.
1205      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1206      * jnr indices corresponding to data put in the four positions in the SIMD register.
1207      */
1208     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1209     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1210     int              jnrA,jnrB;
1211     int              j_coord_offsetA,j_coord_offsetB;
1212     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1213     real             rcutoff_scalar;
1214     real             *shiftvec,*fshift,*x,*f;
1215     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1216     int              vdwioffset1;
1217     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1218     int              vdwioffset2;
1219     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1220     int              vdwioffset3;
1221     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1222     int              vdwjidx1A,vdwjidx1B;
1223     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1224     int              vdwjidx2A,vdwjidx2B;
1225     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1226     int              vdwjidx3A,vdwjidx3B;
1227     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1228     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1229     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1230     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1231     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1232     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1233     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1234     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1235     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1236     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1237     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1238     real             *charge;
1239     __m128i          ewitab;
1240     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1241     real             *ewtab;
1242     __m128d          dummy_mask,cutoff_mask;
1243     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1244     __m128d          one     = _mm_set1_pd(1.0);
1245     __m128d          two     = _mm_set1_pd(2.0);
1246     x                = xx[0];
1247     f                = ff[0];
1248
1249     nri              = nlist->nri;
1250     iinr             = nlist->iinr;
1251     jindex           = nlist->jindex;
1252     jjnr             = nlist->jjnr;
1253     shiftidx         = nlist->shift;
1254     gid              = nlist->gid;
1255     shiftvec         = fr->shift_vec[0];
1256     fshift           = fr->fshift[0];
1257     facel            = _mm_set1_pd(fr->epsfac);
1258     charge           = mdatoms->chargeA;
1259
1260     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1261     ewtab            = fr->ic->tabq_coul_F;
1262     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1263     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1264
1265     /* Setup water-specific parameters */
1266     inr              = nlist->iinr[0];
1267     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1268     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1269     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1270
1271     jq1              = _mm_set1_pd(charge[inr+1]);
1272     jq2              = _mm_set1_pd(charge[inr+2]);
1273     jq3              = _mm_set1_pd(charge[inr+3]);
1274     qq11             = _mm_mul_pd(iq1,jq1);
1275     qq12             = _mm_mul_pd(iq1,jq2);
1276     qq13             = _mm_mul_pd(iq1,jq3);
1277     qq21             = _mm_mul_pd(iq2,jq1);
1278     qq22             = _mm_mul_pd(iq2,jq2);
1279     qq23             = _mm_mul_pd(iq2,jq3);
1280     qq31             = _mm_mul_pd(iq3,jq1);
1281     qq32             = _mm_mul_pd(iq3,jq2);
1282     qq33             = _mm_mul_pd(iq3,jq3);
1283
1284     /* Avoid stupid compiler warnings */
1285     jnrA = jnrB = 0;
1286     j_coord_offsetA = 0;
1287     j_coord_offsetB = 0;
1288
1289     outeriter        = 0;
1290     inneriter        = 0;
1291
1292     /* Start outer loop over neighborlists */
1293     for(iidx=0; iidx<nri; iidx++)
1294     {
1295         /* Load shift vector for this list */
1296         i_shift_offset   = DIM*shiftidx[iidx];
1297
1298         /* Load limits for loop over neighbors */
1299         j_index_start    = jindex[iidx];
1300         j_index_end      = jindex[iidx+1];
1301
1302         /* Get outer coordinate index */
1303         inr              = iinr[iidx];
1304         i_coord_offset   = DIM*inr;
1305
1306         /* Load i particle coords and add shift vector */
1307         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1308                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1309
1310         fix1             = _mm_setzero_pd();
1311         fiy1             = _mm_setzero_pd();
1312         fiz1             = _mm_setzero_pd();
1313         fix2             = _mm_setzero_pd();
1314         fiy2             = _mm_setzero_pd();
1315         fiz2             = _mm_setzero_pd();
1316         fix3             = _mm_setzero_pd();
1317         fiy3             = _mm_setzero_pd();
1318         fiz3             = _mm_setzero_pd();
1319
1320         /* Start inner kernel loop */
1321         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1322         {
1323
1324             /* Get j neighbor index, and coordinate index */
1325             jnrA             = jjnr[jidx];
1326             jnrB             = jjnr[jidx+1];
1327             j_coord_offsetA  = DIM*jnrA;
1328             j_coord_offsetB  = DIM*jnrB;
1329
1330             /* load j atom coordinates */
1331             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1332                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1333
1334             /* Calculate displacement vector */
1335             dx11             = _mm_sub_pd(ix1,jx1);
1336             dy11             = _mm_sub_pd(iy1,jy1);
1337             dz11             = _mm_sub_pd(iz1,jz1);
1338             dx12             = _mm_sub_pd(ix1,jx2);
1339             dy12             = _mm_sub_pd(iy1,jy2);
1340             dz12             = _mm_sub_pd(iz1,jz2);
1341             dx13             = _mm_sub_pd(ix1,jx3);
1342             dy13             = _mm_sub_pd(iy1,jy3);
1343             dz13             = _mm_sub_pd(iz1,jz3);
1344             dx21             = _mm_sub_pd(ix2,jx1);
1345             dy21             = _mm_sub_pd(iy2,jy1);
1346             dz21             = _mm_sub_pd(iz2,jz1);
1347             dx22             = _mm_sub_pd(ix2,jx2);
1348             dy22             = _mm_sub_pd(iy2,jy2);
1349             dz22             = _mm_sub_pd(iz2,jz2);
1350             dx23             = _mm_sub_pd(ix2,jx3);
1351             dy23             = _mm_sub_pd(iy2,jy3);
1352             dz23             = _mm_sub_pd(iz2,jz3);
1353             dx31             = _mm_sub_pd(ix3,jx1);
1354             dy31             = _mm_sub_pd(iy3,jy1);
1355             dz31             = _mm_sub_pd(iz3,jz1);
1356             dx32             = _mm_sub_pd(ix3,jx2);
1357             dy32             = _mm_sub_pd(iy3,jy2);
1358             dz32             = _mm_sub_pd(iz3,jz2);
1359             dx33             = _mm_sub_pd(ix3,jx3);
1360             dy33             = _mm_sub_pd(iy3,jy3);
1361             dz33             = _mm_sub_pd(iz3,jz3);
1362
1363             /* Calculate squared distance and things based on it */
1364             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1365             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1366             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1367             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1368             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1369             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1370             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1371             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1372             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1373
1374             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1375             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1376             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1377             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1378             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1379             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1380             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1381             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1382             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1383
1384             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1385             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1386             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1387             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1388             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1389             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1390             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1391             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1392             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1393
1394             fjx1             = _mm_setzero_pd();
1395             fjy1             = _mm_setzero_pd();
1396             fjz1             = _mm_setzero_pd();
1397             fjx2             = _mm_setzero_pd();
1398             fjy2             = _mm_setzero_pd();
1399             fjz2             = _mm_setzero_pd();
1400             fjx3             = _mm_setzero_pd();
1401             fjy3             = _mm_setzero_pd();
1402             fjz3             = _mm_setzero_pd();
1403
1404             /**************************
1405              * CALCULATE INTERACTIONS *
1406              **************************/
1407
1408             r11              = _mm_mul_pd(rsq11,rinv11);
1409
1410             /* EWALD ELECTROSTATICS */
1411
1412             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1413             ewrt             = _mm_mul_pd(r11,ewtabscale);
1414             ewitab           = _mm_cvttpd_epi32(ewrt);
1415 #ifdef __XOP__
1416             eweps            = _mm_frcz_pd(ewrt);
1417 #else
1418             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1419 #endif
1420             twoeweps         = _mm_add_pd(eweps,eweps);
1421             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1422                                          &ewtabF,&ewtabFn);
1423             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1424             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1425
1426             fscal            = felec;
1427
1428             /* Update vectorial force */
1429             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1430             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1431             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1432
1433             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1434             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1435             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1436
1437             /**************************
1438              * CALCULATE INTERACTIONS *
1439              **************************/
1440
1441             r12              = _mm_mul_pd(rsq12,rinv12);
1442
1443             /* EWALD ELECTROSTATICS */
1444
1445             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1446             ewrt             = _mm_mul_pd(r12,ewtabscale);
1447             ewitab           = _mm_cvttpd_epi32(ewrt);
1448 #ifdef __XOP__
1449             eweps            = _mm_frcz_pd(ewrt);
1450 #else
1451             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1452 #endif
1453             twoeweps         = _mm_add_pd(eweps,eweps);
1454             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1455                                          &ewtabF,&ewtabFn);
1456             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1457             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1458
1459             fscal            = felec;
1460
1461             /* Update vectorial force */
1462             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1463             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1464             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1465
1466             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1467             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1468             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1469
1470             /**************************
1471              * CALCULATE INTERACTIONS *
1472              **************************/
1473
1474             r13              = _mm_mul_pd(rsq13,rinv13);
1475
1476             /* EWALD ELECTROSTATICS */
1477
1478             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1479             ewrt             = _mm_mul_pd(r13,ewtabscale);
1480             ewitab           = _mm_cvttpd_epi32(ewrt);
1481 #ifdef __XOP__
1482             eweps            = _mm_frcz_pd(ewrt);
1483 #else
1484             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1485 #endif
1486             twoeweps         = _mm_add_pd(eweps,eweps);
1487             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1488                                          &ewtabF,&ewtabFn);
1489             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1490             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1491
1492             fscal            = felec;
1493
1494             /* Update vectorial force */
1495             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1496             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1497             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1498
1499             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1500             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1501             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1502
1503             /**************************
1504              * CALCULATE INTERACTIONS *
1505              **************************/
1506
1507             r21              = _mm_mul_pd(rsq21,rinv21);
1508
1509             /* EWALD ELECTROSTATICS */
1510
1511             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1512             ewrt             = _mm_mul_pd(r21,ewtabscale);
1513             ewitab           = _mm_cvttpd_epi32(ewrt);
1514 #ifdef __XOP__
1515             eweps            = _mm_frcz_pd(ewrt);
1516 #else
1517             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1518 #endif
1519             twoeweps         = _mm_add_pd(eweps,eweps);
1520             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1521                                          &ewtabF,&ewtabFn);
1522             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1523             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1524
1525             fscal            = felec;
1526
1527             /* Update vectorial force */
1528             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1529             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1530             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1531
1532             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1533             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1534             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1535
1536             /**************************
1537              * CALCULATE INTERACTIONS *
1538              **************************/
1539
1540             r22              = _mm_mul_pd(rsq22,rinv22);
1541
1542             /* EWALD ELECTROSTATICS */
1543
1544             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1545             ewrt             = _mm_mul_pd(r22,ewtabscale);
1546             ewitab           = _mm_cvttpd_epi32(ewrt);
1547 #ifdef __XOP__
1548             eweps            = _mm_frcz_pd(ewrt);
1549 #else
1550             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1551 #endif
1552             twoeweps         = _mm_add_pd(eweps,eweps);
1553             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1554                                          &ewtabF,&ewtabFn);
1555             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1556             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1557
1558             fscal            = felec;
1559
1560             /* Update vectorial force */
1561             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1562             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1563             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1564
1565             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1566             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1567             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1568
1569             /**************************
1570              * CALCULATE INTERACTIONS *
1571              **************************/
1572
1573             r23              = _mm_mul_pd(rsq23,rinv23);
1574
1575             /* EWALD ELECTROSTATICS */
1576
1577             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1578             ewrt             = _mm_mul_pd(r23,ewtabscale);
1579             ewitab           = _mm_cvttpd_epi32(ewrt);
1580 #ifdef __XOP__
1581             eweps            = _mm_frcz_pd(ewrt);
1582 #else
1583             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1584 #endif
1585             twoeweps         = _mm_add_pd(eweps,eweps);
1586             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1587                                          &ewtabF,&ewtabFn);
1588             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1589             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1590
1591             fscal            = felec;
1592
1593             /* Update vectorial force */
1594             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1595             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1596             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1597
1598             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1599             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1600             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1601
1602             /**************************
1603              * CALCULATE INTERACTIONS *
1604              **************************/
1605
1606             r31              = _mm_mul_pd(rsq31,rinv31);
1607
1608             /* EWALD ELECTROSTATICS */
1609
1610             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1611             ewrt             = _mm_mul_pd(r31,ewtabscale);
1612             ewitab           = _mm_cvttpd_epi32(ewrt);
1613 #ifdef __XOP__
1614             eweps            = _mm_frcz_pd(ewrt);
1615 #else
1616             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1617 #endif
1618             twoeweps         = _mm_add_pd(eweps,eweps);
1619             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1620                                          &ewtabF,&ewtabFn);
1621             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1622             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1623
1624             fscal            = felec;
1625
1626             /* Update vectorial force */
1627             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1628             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1629             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1630
1631             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1632             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1633             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1634
1635             /**************************
1636              * CALCULATE INTERACTIONS *
1637              **************************/
1638
1639             r32              = _mm_mul_pd(rsq32,rinv32);
1640
1641             /* EWALD ELECTROSTATICS */
1642
1643             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1644             ewrt             = _mm_mul_pd(r32,ewtabscale);
1645             ewitab           = _mm_cvttpd_epi32(ewrt);
1646 #ifdef __XOP__
1647             eweps            = _mm_frcz_pd(ewrt);
1648 #else
1649             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1650 #endif
1651             twoeweps         = _mm_add_pd(eweps,eweps);
1652             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1653                                          &ewtabF,&ewtabFn);
1654             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1655             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1656
1657             fscal            = felec;
1658
1659             /* Update vectorial force */
1660             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1661             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1662             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1663
1664             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1665             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1666             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1667
1668             /**************************
1669              * CALCULATE INTERACTIONS *
1670              **************************/
1671
1672             r33              = _mm_mul_pd(rsq33,rinv33);
1673
1674             /* EWALD ELECTROSTATICS */
1675
1676             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1677             ewrt             = _mm_mul_pd(r33,ewtabscale);
1678             ewitab           = _mm_cvttpd_epi32(ewrt);
1679 #ifdef __XOP__
1680             eweps            = _mm_frcz_pd(ewrt);
1681 #else
1682             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1683 #endif
1684             twoeweps         = _mm_add_pd(eweps,eweps);
1685             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1686                                          &ewtabF,&ewtabFn);
1687             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1688             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1689
1690             fscal            = felec;
1691
1692             /* Update vectorial force */
1693             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1694             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1695             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1696
1697             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1698             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1699             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1700
1701             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA+DIM,f+j_coord_offsetB+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1702
1703             /* Inner loop uses 351 flops */
1704         }
1705
1706         if(jidx<j_index_end)
1707         {
1708
1709             jnrA             = jjnr[jidx];
1710             j_coord_offsetA  = DIM*jnrA;
1711
1712             /* load j atom coordinates */
1713             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
1714                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1715
1716             /* Calculate displacement vector */
1717             dx11             = _mm_sub_pd(ix1,jx1);
1718             dy11             = _mm_sub_pd(iy1,jy1);
1719             dz11             = _mm_sub_pd(iz1,jz1);
1720             dx12             = _mm_sub_pd(ix1,jx2);
1721             dy12             = _mm_sub_pd(iy1,jy2);
1722             dz12             = _mm_sub_pd(iz1,jz2);
1723             dx13             = _mm_sub_pd(ix1,jx3);
1724             dy13             = _mm_sub_pd(iy1,jy3);
1725             dz13             = _mm_sub_pd(iz1,jz3);
1726             dx21             = _mm_sub_pd(ix2,jx1);
1727             dy21             = _mm_sub_pd(iy2,jy1);
1728             dz21             = _mm_sub_pd(iz2,jz1);
1729             dx22             = _mm_sub_pd(ix2,jx2);
1730             dy22             = _mm_sub_pd(iy2,jy2);
1731             dz22             = _mm_sub_pd(iz2,jz2);
1732             dx23             = _mm_sub_pd(ix2,jx3);
1733             dy23             = _mm_sub_pd(iy2,jy3);
1734             dz23             = _mm_sub_pd(iz2,jz3);
1735             dx31             = _mm_sub_pd(ix3,jx1);
1736             dy31             = _mm_sub_pd(iy3,jy1);
1737             dz31             = _mm_sub_pd(iz3,jz1);
1738             dx32             = _mm_sub_pd(ix3,jx2);
1739             dy32             = _mm_sub_pd(iy3,jy2);
1740             dz32             = _mm_sub_pd(iz3,jz2);
1741             dx33             = _mm_sub_pd(ix3,jx3);
1742             dy33             = _mm_sub_pd(iy3,jy3);
1743             dz33             = _mm_sub_pd(iz3,jz3);
1744
1745             /* Calculate squared distance and things based on it */
1746             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1747             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1748             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1749             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1750             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1751             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1752             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1753             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1754             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1755
1756             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1757             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1758             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1759             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1760             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1761             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1762             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1763             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1764             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1765
1766             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1767             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1768             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1769             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1770             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1771             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1772             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1773             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1774             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1775
1776             fjx1             = _mm_setzero_pd();
1777             fjy1             = _mm_setzero_pd();
1778             fjz1             = _mm_setzero_pd();
1779             fjx2             = _mm_setzero_pd();
1780             fjy2             = _mm_setzero_pd();
1781             fjz2             = _mm_setzero_pd();
1782             fjx3             = _mm_setzero_pd();
1783             fjy3             = _mm_setzero_pd();
1784             fjz3             = _mm_setzero_pd();
1785
1786             /**************************
1787              * CALCULATE INTERACTIONS *
1788              **************************/
1789
1790             r11              = _mm_mul_pd(rsq11,rinv11);
1791
1792             /* EWALD ELECTROSTATICS */
1793
1794             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1795             ewrt             = _mm_mul_pd(r11,ewtabscale);
1796             ewitab           = _mm_cvttpd_epi32(ewrt);
1797 #ifdef __XOP__
1798             eweps            = _mm_frcz_pd(ewrt);
1799 #else
1800             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1801 #endif
1802             twoeweps         = _mm_add_pd(eweps,eweps);
1803             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1804             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1805             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1806
1807             fscal            = felec;
1808
1809             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1810
1811             /* Update vectorial force */
1812             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1813             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1814             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1815
1816             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1817             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1818             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1819
1820             /**************************
1821              * CALCULATE INTERACTIONS *
1822              **************************/
1823
1824             r12              = _mm_mul_pd(rsq12,rinv12);
1825
1826             /* EWALD ELECTROSTATICS */
1827
1828             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1829             ewrt             = _mm_mul_pd(r12,ewtabscale);
1830             ewitab           = _mm_cvttpd_epi32(ewrt);
1831 #ifdef __XOP__
1832             eweps            = _mm_frcz_pd(ewrt);
1833 #else
1834             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1835 #endif
1836             twoeweps         = _mm_add_pd(eweps,eweps);
1837             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1838             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1839             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1840
1841             fscal            = felec;
1842
1843             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1844
1845             /* Update vectorial force */
1846             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1847             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1848             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1849
1850             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1851             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1852             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1853
1854             /**************************
1855              * CALCULATE INTERACTIONS *
1856              **************************/
1857
1858             r13              = _mm_mul_pd(rsq13,rinv13);
1859
1860             /* EWALD ELECTROSTATICS */
1861
1862             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1863             ewrt             = _mm_mul_pd(r13,ewtabscale);
1864             ewitab           = _mm_cvttpd_epi32(ewrt);
1865 #ifdef __XOP__
1866             eweps            = _mm_frcz_pd(ewrt);
1867 #else
1868             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1869 #endif
1870             twoeweps         = _mm_add_pd(eweps,eweps);
1871             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1872             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1873             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1874
1875             fscal            = felec;
1876
1877             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1878
1879             /* Update vectorial force */
1880             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1881             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1882             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1883
1884             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1885             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1886             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1887
1888             /**************************
1889              * CALCULATE INTERACTIONS *
1890              **************************/
1891
1892             r21              = _mm_mul_pd(rsq21,rinv21);
1893
1894             /* EWALD ELECTROSTATICS */
1895
1896             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1897             ewrt             = _mm_mul_pd(r21,ewtabscale);
1898             ewitab           = _mm_cvttpd_epi32(ewrt);
1899 #ifdef __XOP__
1900             eweps            = _mm_frcz_pd(ewrt);
1901 #else
1902             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1903 #endif
1904             twoeweps         = _mm_add_pd(eweps,eweps);
1905             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1906             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1907             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1908
1909             fscal            = felec;
1910
1911             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1912
1913             /* Update vectorial force */
1914             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1915             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1916             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1917
1918             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1919             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1920             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1921
1922             /**************************
1923              * CALCULATE INTERACTIONS *
1924              **************************/
1925
1926             r22              = _mm_mul_pd(rsq22,rinv22);
1927
1928             /* EWALD ELECTROSTATICS */
1929
1930             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1931             ewrt             = _mm_mul_pd(r22,ewtabscale);
1932             ewitab           = _mm_cvttpd_epi32(ewrt);
1933 #ifdef __XOP__
1934             eweps            = _mm_frcz_pd(ewrt);
1935 #else
1936             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1937 #endif
1938             twoeweps         = _mm_add_pd(eweps,eweps);
1939             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1940             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1941             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1942
1943             fscal            = felec;
1944
1945             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1946
1947             /* Update vectorial force */
1948             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1949             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1950             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1951
1952             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1953             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1954             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1955
1956             /**************************
1957              * CALCULATE INTERACTIONS *
1958              **************************/
1959
1960             r23              = _mm_mul_pd(rsq23,rinv23);
1961
1962             /* EWALD ELECTROSTATICS */
1963
1964             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1965             ewrt             = _mm_mul_pd(r23,ewtabscale);
1966             ewitab           = _mm_cvttpd_epi32(ewrt);
1967 #ifdef __XOP__
1968             eweps            = _mm_frcz_pd(ewrt);
1969 #else
1970             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1971 #endif
1972             twoeweps         = _mm_add_pd(eweps,eweps);
1973             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1974             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1975             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1976
1977             fscal            = felec;
1978
1979             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1980
1981             /* Update vectorial force */
1982             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1983             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1984             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1985
1986             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1987             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1988             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1989
1990             /**************************
1991              * CALCULATE INTERACTIONS *
1992              **************************/
1993
1994             r31              = _mm_mul_pd(rsq31,rinv31);
1995
1996             /* EWALD ELECTROSTATICS */
1997
1998             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1999             ewrt             = _mm_mul_pd(r31,ewtabscale);
2000             ewitab           = _mm_cvttpd_epi32(ewrt);
2001 #ifdef __XOP__
2002             eweps            = _mm_frcz_pd(ewrt);
2003 #else
2004             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2005 #endif
2006             twoeweps         = _mm_add_pd(eweps,eweps);
2007             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2008             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2009             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
2010
2011             fscal            = felec;
2012
2013             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2014
2015             /* Update vectorial force */
2016             fix3             = _mm_macc_pd(dx31,fscal,fix3);
2017             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
2018             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
2019
2020             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
2021             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
2022             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
2023
2024             /**************************
2025              * CALCULATE INTERACTIONS *
2026              **************************/
2027
2028             r32              = _mm_mul_pd(rsq32,rinv32);
2029
2030             /* EWALD ELECTROSTATICS */
2031
2032             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2033             ewrt             = _mm_mul_pd(r32,ewtabscale);
2034             ewitab           = _mm_cvttpd_epi32(ewrt);
2035 #ifdef __XOP__
2036             eweps            = _mm_frcz_pd(ewrt);
2037 #else
2038             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2039 #endif
2040             twoeweps         = _mm_add_pd(eweps,eweps);
2041             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2042             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2043             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2044
2045             fscal            = felec;
2046
2047             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2048
2049             /* Update vectorial force */
2050             fix3             = _mm_macc_pd(dx32,fscal,fix3);
2051             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
2052             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
2053
2054             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
2055             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
2056             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
2057
2058             /**************************
2059              * CALCULATE INTERACTIONS *
2060              **************************/
2061
2062             r33              = _mm_mul_pd(rsq33,rinv33);
2063
2064             /* EWALD ELECTROSTATICS */
2065
2066             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2067             ewrt             = _mm_mul_pd(r33,ewtabscale);
2068             ewitab           = _mm_cvttpd_epi32(ewrt);
2069 #ifdef __XOP__
2070             eweps            = _mm_frcz_pd(ewrt);
2071 #else
2072             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2073 #endif
2074             twoeweps         = _mm_add_pd(eweps,eweps);
2075             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2076             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2077             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2078
2079             fscal            = felec;
2080
2081             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2082
2083             /* Update vectorial force */
2084             fix3             = _mm_macc_pd(dx33,fscal,fix3);
2085             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
2086             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
2087
2088             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
2089             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
2090             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
2091
2092             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2093
2094             /* Inner loop uses 351 flops */
2095         }
2096
2097         /* End of innermost loop */
2098
2099         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2100                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
2101
2102         /* Increment number of inner iterations */
2103         inneriter                  += j_index_end - j_index_start;
2104
2105         /* Outer loop uses 18 flops */
2106     }
2107
2108     /* Increment number of outer iterations */
2109     outeriter        += nri;
2110
2111     /* Update outer/inner flops */
2112
2113     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*351);
2114 }