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) 2012,2013,2014, 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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Water4
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwNone_GeomW4W4_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset1;
  81     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  82     int              vdwioffset2;
  83     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  84     int              vdwioffset3;
  85     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  86     int              vdwjidx1A,vdwjidx1B;
  87     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  88     int              vdwjidx2A,vdwjidx2B;
  89     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  90     int              vdwjidx3A,vdwjidx3B;
  91     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  92     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  93     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  94     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  95     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  96     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  97     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  98     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  99     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 100     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 101     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 102     real             *charge;
 103     __m128i          ewitab;
 104     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 105     real             *ewtab;
 106     __m128d          dummy_mask,cutoff_mask;
 107     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 108     __m128d          one     = _mm_set1_pd(1.0);
 109     __m128d          two     = _mm_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123
 124     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 125     ewtab            = fr->ic->tabq_coul_FDV0;
 126     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 127     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 132     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 133     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 134
 135     jq1              = _mm_set1_pd(charge[inr+1]);
 136     jq2              = _mm_set1_pd(charge[inr+2]);
 137     jq3              = _mm_set1_pd(charge[inr+3]);
 138     qq11             = _mm_mul_pd(iq1,jq1);
 139     qq12             = _mm_mul_pd(iq1,jq2);
 140     qq13             = _mm_mul_pd(iq1,jq3);
 141     qq21             = _mm_mul_pd(iq2,jq1);
 142     qq22             = _mm_mul_pd(iq2,jq2);
 143     qq23             = _mm_mul_pd(iq2,jq3);
 144     qq31             = _mm_mul_pd(iq3,jq1);
 145     qq32             = _mm_mul_pd(iq3,jq2);
 146     qq33             = _mm_mul_pd(iq3,jq3);
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152
 153     outeriter        = 0;
 154     inneriter        = 0;
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 172                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 173
 174         fix1             = _mm_setzero_pd();
 175         fiy1             = _mm_setzero_pd();
 176         fiz1             = _mm_setzero_pd();
 177         fix2             = _mm_setzero_pd();
 178         fiy2             = _mm_setzero_pd();
 179         fiz2             = _mm_setzero_pd();
 180         fix3             = _mm_setzero_pd();
 181         fiy3             = _mm_setzero_pd();
 182         fiz3             = _mm_setzero_pd();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_pd();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 199                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 200
 201             /* Calculate displacement vector */
 202             dx11             = _mm_sub_pd(ix1,jx1);
 203             dy11             = _mm_sub_pd(iy1,jy1);
 204             dz11             = _mm_sub_pd(iz1,jz1);
 205             dx12             = _mm_sub_pd(ix1,jx2);
 206             dy12             = _mm_sub_pd(iy1,jy2);
 207             dz12             = _mm_sub_pd(iz1,jz2);
 208             dx13             = _mm_sub_pd(ix1,jx3);
 209             dy13             = _mm_sub_pd(iy1,jy3);
 210             dz13             = _mm_sub_pd(iz1,jz3);
 211             dx21             = _mm_sub_pd(ix2,jx1);
 212             dy21             = _mm_sub_pd(iy2,jy1);
 213             dz21             = _mm_sub_pd(iz2,jz1);
 214             dx22             = _mm_sub_pd(ix2,jx2);
 215             dy22             = _mm_sub_pd(iy2,jy2);
 216             dz22             = _mm_sub_pd(iz2,jz2);
 217             dx23             = _mm_sub_pd(ix2,jx3);
 218             dy23             = _mm_sub_pd(iy2,jy3);
 219             dz23             = _mm_sub_pd(iz2,jz3);
 220             dx31             = _mm_sub_pd(ix3,jx1);
 221             dy31             = _mm_sub_pd(iy3,jy1);
 222             dz31             = _mm_sub_pd(iz3,jz1);
 223             dx32             = _mm_sub_pd(ix3,jx2);
 224             dy32             = _mm_sub_pd(iy3,jy2);
 225             dz32             = _mm_sub_pd(iz3,jz2);
 226             dx33             = _mm_sub_pd(ix3,jx3);
 227             dy33             = _mm_sub_pd(iy3,jy3);
 228             dz33             = _mm_sub_pd(iz3,jz3);
 229
 230             /* Calculate squared distance and things based on it */
 231             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 232             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 233             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 234             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 235             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 236             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 237             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 238             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 239             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 240
 241             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 242             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 243             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 244             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 245             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 246             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 247             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 248             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 249             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 250
 251             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 252             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 253             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 254             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 255             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 256             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 257             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 258             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 259             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 260
 261             fjx1             = _mm_setzero_pd();
 262             fjy1             = _mm_setzero_pd();
 263             fjz1             = _mm_setzero_pd();
 264             fjx2             = _mm_setzero_pd();
 265             fjy2             = _mm_setzero_pd();
 266             fjz2             = _mm_setzero_pd();
 267             fjx3             = _mm_setzero_pd();
 268             fjy3             = _mm_setzero_pd();
 269             fjz3             = _mm_setzero_pd();
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             r11              = _mm_mul_pd(rsq11,rinv11);
 276
 277             /* EWALD ELECTROSTATICS */
 278
 279             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 280             ewrt             = _mm_mul_pd(r11,ewtabscale);
 281             ewitab           = _mm_cvttpd_epi32(ewrt);
 282 #ifdef __XOP__
 283             eweps            = _mm_frcz_pd(ewrt);
 284 #else
 285             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 286 #endif
 287             twoeweps         = _mm_add_pd(eweps,eweps);
 288             ewitab           = _mm_slli_epi32(ewitab,2);
 289             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 290             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 291             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 292             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 293             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 294             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 295             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 296             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 297             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 298             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velecsum         = _mm_add_pd(velecsum,velec);
 302
 303             fscal            = felec;
 304
 305             /* Update vectorial force */
 306             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 307             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 308             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 309
 310             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 311             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 312             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r12              = _mm_mul_pd(rsq12,rinv12);
 319
 320             /* EWALD ELECTROSTATICS */
 321
 322             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 323             ewrt             = _mm_mul_pd(r12,ewtabscale);
 324             ewitab           = _mm_cvttpd_epi32(ewrt);
 325 #ifdef __XOP__
 326             eweps            = _mm_frcz_pd(ewrt);
 327 #else
 328             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 329 #endif
 330             twoeweps         = _mm_add_pd(eweps,eweps);
 331             ewitab           = _mm_slli_epi32(ewitab,2);
 332             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 333             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 334             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 335             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 336             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 337             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 338             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 339             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 340             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 341             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm_add_pd(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             /* Update vectorial force */
 349             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 350             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 351             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 352
 353             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 354             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 355             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             r13              = _mm_mul_pd(rsq13,rinv13);
 362
 363             /* EWALD ELECTROSTATICS */
 364
 365             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 366             ewrt             = _mm_mul_pd(r13,ewtabscale);
 367             ewitab           = _mm_cvttpd_epi32(ewrt);
 368 #ifdef __XOP__
 369             eweps            = _mm_frcz_pd(ewrt);
 370 #else
 371             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 372 #endif
 373             twoeweps         = _mm_add_pd(eweps,eweps);
 374             ewitab           = _mm_slli_epi32(ewitab,2);
 375             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 376             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 377             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 378             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 379             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 380             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 381             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 382             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 383             velec            = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
 384             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velecsum         = _mm_add_pd(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391             /* Update vectorial force */
 392             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 393             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 394             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 395
 396             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 397             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 398             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 399
 400             /**************************
 401              * CALCULATE INTERACTIONS *
 402              **************************/
 403
 404             r21              = _mm_mul_pd(rsq21,rinv21);
 405
 406             /* EWALD ELECTROSTATICS */
 407
 408             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 409             ewrt             = _mm_mul_pd(r21,ewtabscale);
 410             ewitab           = _mm_cvttpd_epi32(ewrt);
 411 #ifdef __XOP__
 412             eweps            = _mm_frcz_pd(ewrt);
 413 #else
 414             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 415 #endif
 416             twoeweps         = _mm_add_pd(eweps,eweps);
 417             ewitab           = _mm_slli_epi32(ewitab,2);
 418             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 419             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 420             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 421             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 422             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 423             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 424             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 425             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 426             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 427             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 428
 429             /* Update potential sum for this i atom from the interaction with this j atom. */
 430             velecsum         = _mm_add_pd(velecsum,velec);
 431
 432             fscal            = felec;
 433
 434             /* Update vectorial force */
 435             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 436             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 437             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 438
 439             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 440             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 441             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 442
 443             /**************************
 444              * CALCULATE INTERACTIONS *
 445              **************************/
 446
 447             r22              = _mm_mul_pd(rsq22,rinv22);
 448
 449             /* EWALD ELECTROSTATICS */
 450
 451             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 452             ewrt             = _mm_mul_pd(r22,ewtabscale);
 453             ewitab           = _mm_cvttpd_epi32(ewrt);
 454 #ifdef __XOP__
 455             eweps            = _mm_frcz_pd(ewrt);
 456 #else
 457             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 458 #endif
 459             twoeweps         = _mm_add_pd(eweps,eweps);
 460             ewitab           = _mm_slli_epi32(ewitab,2);
 461             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 462             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 463             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 464             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 465             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 466             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 467             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 468             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 469             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 470             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 471
 472             /* Update potential sum for this i atom from the interaction with this j atom. */
 473             velecsum         = _mm_add_pd(velecsum,velec);
 474
 475             fscal            = felec;
 476
 477             /* Update vectorial force */
 478             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 479             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 480             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 481
 482             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 483             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 484             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             r23              = _mm_mul_pd(rsq23,rinv23);
 491
 492             /* EWALD ELECTROSTATICS */
 493
 494             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 495             ewrt             = _mm_mul_pd(r23,ewtabscale);
 496             ewitab           = _mm_cvttpd_epi32(ewrt);
 497 #ifdef __XOP__
 498             eweps            = _mm_frcz_pd(ewrt);
 499 #else
 500             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 501 #endif
 502             twoeweps         = _mm_add_pd(eweps,eweps);
 503             ewitab           = _mm_slli_epi32(ewitab,2);
 504             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 505             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 506             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 507             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 508             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 509             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 510             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 511             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 512             velec            = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
 513             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
 514
 515             /* Update potential sum for this i atom from the interaction with this j atom. */
 516             velecsum         = _mm_add_pd(velecsum,velec);
 517
 518             fscal            = felec;
 519
 520             /* Update vectorial force */
 521             fix2             = _mm_macc_pd(dx23,fscal,fix2);
 522             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
 523             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
 524
 525             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
 526             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
 527             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             r31              = _mm_mul_pd(rsq31,rinv31);
 534
 535             /* EWALD ELECTROSTATICS */
 536
 537             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 538             ewrt             = _mm_mul_pd(r31,ewtabscale);
 539             ewitab           = _mm_cvttpd_epi32(ewrt);
 540 #ifdef __XOP__
 541             eweps            = _mm_frcz_pd(ewrt);
 542 #else
 543             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 544 #endif
 545             twoeweps         = _mm_add_pd(eweps,eweps);
 546             ewitab           = _mm_slli_epi32(ewitab,2);
 547             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 548             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 549             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 550             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 551             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 552             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 553             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 554             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 555             velec            = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
 556             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             velecsum         = _mm_add_pd(velecsum,velec);
 560
 561             fscal            = felec;
 562
 563             /* Update vectorial force */
 564             fix3             = _mm_macc_pd(dx31,fscal,fix3);
 565             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
 566             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
 567
 568             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
 569             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
 570             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             r32              = _mm_mul_pd(rsq32,rinv32);
 577
 578             /* EWALD ELECTROSTATICS */
 579
 580             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 581             ewrt             = _mm_mul_pd(r32,ewtabscale);
 582             ewitab           = _mm_cvttpd_epi32(ewrt);
 583 #ifdef __XOP__
 584             eweps            = _mm_frcz_pd(ewrt);
 585 #else
 586             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 587 #endif
 588             twoeweps         = _mm_add_pd(eweps,eweps);
 589             ewitab           = _mm_slli_epi32(ewitab,2);
 590             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 591             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 592             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 593             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 594             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 595             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 596             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 597             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 598             velec            = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
 599             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
 600
 601             /* Update potential sum for this i atom from the interaction with this j atom. */
 602             velecsum         = _mm_add_pd(velecsum,velec);
 603
 604             fscal            = felec;
 605
 606             /* Update vectorial force */
 607             fix3             = _mm_macc_pd(dx32,fscal,fix3);
 608             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
 609             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
 610
 611             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
 612             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
 613             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             r33              = _mm_mul_pd(rsq33,rinv33);
 620
 621             /* EWALD ELECTROSTATICS */
 622
 623             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 624             ewrt             = _mm_mul_pd(r33,ewtabscale);
 625             ewitab           = _mm_cvttpd_epi32(ewrt);
 626 #ifdef __XOP__
 627             eweps            = _mm_frcz_pd(ewrt);
 628 #else
 629             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 630 #endif
 631             twoeweps         = _mm_add_pd(eweps,eweps);
 632             ewitab           = _mm_slli_epi32(ewitab,2);
 633             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 634             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 635             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 636             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 637             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 638             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 639             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 640             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 641             velec            = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
 642             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
 643
 644             /* Update potential sum for this i atom from the interaction with this j atom. */
 645             velecsum         = _mm_add_pd(velecsum,velec);
 646
 647             fscal            = felec;
 648
 649             /* Update vectorial force */
 650             fix3             = _mm_macc_pd(dx33,fscal,fix3);
 651             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
 652             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
 653
 654             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
 655             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
 656             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
 657
 658             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);
 659
 660             /* Inner loop uses 396 flops */
 661         }
 662
 663         if(jidx<j_index_end)
 664         {
 665
 666             jnrA             = jjnr[jidx];
 667             j_coord_offsetA  = DIM*jnrA;
 668
 669             /* load j atom coordinates */
 670             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
 671                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 672
 673             /* Calculate displacement vector */
 674             dx11             = _mm_sub_pd(ix1,jx1);
 675             dy11             = _mm_sub_pd(iy1,jy1);
 676             dz11             = _mm_sub_pd(iz1,jz1);
 677             dx12             = _mm_sub_pd(ix1,jx2);
 678             dy12             = _mm_sub_pd(iy1,jy2);
 679             dz12             = _mm_sub_pd(iz1,jz2);
 680             dx13             = _mm_sub_pd(ix1,jx3);
 681             dy13             = _mm_sub_pd(iy1,jy3);
 682             dz13             = _mm_sub_pd(iz1,jz3);
 683             dx21             = _mm_sub_pd(ix2,jx1);
 684             dy21             = _mm_sub_pd(iy2,jy1);
 685             dz21             = _mm_sub_pd(iz2,jz1);
 686             dx22             = _mm_sub_pd(ix2,jx2);
 687             dy22             = _mm_sub_pd(iy2,jy2);
 688             dz22             = _mm_sub_pd(iz2,jz2);
 689             dx23             = _mm_sub_pd(ix2,jx3);
 690             dy23             = _mm_sub_pd(iy2,jy3);
 691             dz23             = _mm_sub_pd(iz2,jz3);
 692             dx31             = _mm_sub_pd(ix3,jx1);
 693             dy31             = _mm_sub_pd(iy3,jy1);
 694             dz31             = _mm_sub_pd(iz3,jz1);
 695             dx32             = _mm_sub_pd(ix3,jx2);
 696             dy32             = _mm_sub_pd(iy3,jy2);
 697             dz32             = _mm_sub_pd(iz3,jz2);
 698             dx33             = _mm_sub_pd(ix3,jx3);
 699             dy33             = _mm_sub_pd(iy3,jy3);
 700             dz33             = _mm_sub_pd(iz3,jz3);
 701
 702             /* Calculate squared distance and things based on it */
 703             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 704             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 705             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 706             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 707             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 708             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 709             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 710             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 711             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 712
 713             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 714             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 715             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 716             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 717             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 718             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 719             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 720             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 721             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 722
 723             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 724             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 725             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 726             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 727             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 728             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 729             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 730             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 731             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 732
 733             fjx1             = _mm_setzero_pd();
 734             fjy1             = _mm_setzero_pd();
 735             fjz1             = _mm_setzero_pd();
 736             fjx2             = _mm_setzero_pd();
 737             fjy2             = _mm_setzero_pd();
 738             fjz2             = _mm_setzero_pd();
 739             fjx3             = _mm_setzero_pd();
 740             fjy3             = _mm_setzero_pd();
 741             fjz3             = _mm_setzero_pd();
 742
 743             /**************************
 744              * CALCULATE INTERACTIONS *
 745              **************************/
 746
 747             r11              = _mm_mul_pd(rsq11,rinv11);
 748
 749             /* EWALD ELECTROSTATICS */
 750
 751             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 752             ewrt             = _mm_mul_pd(r11,ewtabscale);
 753             ewitab           = _mm_cvttpd_epi32(ewrt);
 754 #ifdef __XOP__
 755             eweps            = _mm_frcz_pd(ewrt);
 756 #else
 757             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 758 #endif
 759             twoeweps         = _mm_add_pd(eweps,eweps);
 760             ewitab           = _mm_slli_epi32(ewitab,2);
 761             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 762             ewtabD           = _mm_setzero_pd();
 763             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 764             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 765             ewtabFn          = _mm_setzero_pd();
 766             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 767             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 768             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 769             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 770             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 771
 772             /* Update potential sum for this i atom from the interaction with this j atom. */
 773             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 774             velecsum         = _mm_add_pd(velecsum,velec);
 775
 776             fscal            = felec;
 777
 778             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 779
 780             /* Update vectorial force */
 781             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 782             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 783             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 784
 785             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 786             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 787             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 788
 789             /**************************
 790              * CALCULATE INTERACTIONS *
 791              **************************/
 792
 793             r12              = _mm_mul_pd(rsq12,rinv12);
 794
 795             /* EWALD ELECTROSTATICS */
 796
 797             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 798             ewrt             = _mm_mul_pd(r12,ewtabscale);
 799             ewitab           = _mm_cvttpd_epi32(ewrt);
 800 #ifdef __XOP__
 801             eweps            = _mm_frcz_pd(ewrt);
 802 #else
 803             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 804 #endif
 805             twoeweps         = _mm_add_pd(eweps,eweps);
 806             ewitab           = _mm_slli_epi32(ewitab,2);
 807             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 808             ewtabD           = _mm_setzero_pd();
 809             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 810             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 811             ewtabFn          = _mm_setzero_pd();
 812             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 813             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 814             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 815             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 816             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 817
 818             /* Update potential sum for this i atom from the interaction with this j atom. */
 819             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 820             velecsum         = _mm_add_pd(velecsum,velec);
 821
 822             fscal            = felec;
 823
 824             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 825
 826             /* Update vectorial force */
 827             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 828             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 829             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 830
 831             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 832             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 833             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 834
 835             /**************************
 836              * CALCULATE INTERACTIONS *
 837              **************************/
 838
 839             r13              = _mm_mul_pd(rsq13,rinv13);
 840
 841             /* EWALD ELECTROSTATICS */
 842
 843             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 844             ewrt             = _mm_mul_pd(r13,ewtabscale);
 845             ewitab           = _mm_cvttpd_epi32(ewrt);
 846 #ifdef __XOP__
 847             eweps            = _mm_frcz_pd(ewrt);
 848 #else
 849             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 850 #endif
 851             twoeweps         = _mm_add_pd(eweps,eweps);
 852             ewitab           = _mm_slli_epi32(ewitab,2);
 853             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 854             ewtabD           = _mm_setzero_pd();
 855             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 856             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 857             ewtabFn          = _mm_setzero_pd();
 858             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 859             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 860             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 861             velec            = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
 862             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 863
 864             /* Update potential sum for this i atom from the interaction with this j atom. */
 865             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 866             velecsum         = _mm_add_pd(velecsum,velec);
 867
 868             fscal            = felec;
 869
 870             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 871
 872             /* Update vectorial force */
 873             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 874             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 875             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 876
 877             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 878             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 879             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 880
 881             /**************************
 882              * CALCULATE INTERACTIONS *
 883              **************************/
 884
 885             r21              = _mm_mul_pd(rsq21,rinv21);
 886
 887             /* EWALD ELECTROSTATICS */
 888
 889             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 890             ewrt             = _mm_mul_pd(r21,ewtabscale);
 891             ewitab           = _mm_cvttpd_epi32(ewrt);
 892 #ifdef __XOP__
 893             eweps            = _mm_frcz_pd(ewrt);
 894 #else
 895             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 896 #endif
 897             twoeweps         = _mm_add_pd(eweps,eweps);
 898             ewitab           = _mm_slli_epi32(ewitab,2);
 899             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 900             ewtabD           = _mm_setzero_pd();
 901             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 902             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 903             ewtabFn          = _mm_setzero_pd();
 904             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 905             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 906             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 907             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 908             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 909
 910             /* Update potential sum for this i atom from the interaction with this j atom. */
 911             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 912             velecsum         = _mm_add_pd(velecsum,velec);
 913
 914             fscal            = felec;
 915
 916             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 917
 918             /* Update vectorial force */
 919             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 920             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 921             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 922
 923             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 924             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 925             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             r22              = _mm_mul_pd(rsq22,rinv22);
 932
 933             /* EWALD ELECTROSTATICS */
 934
 935             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 936             ewrt             = _mm_mul_pd(r22,ewtabscale);
 937             ewitab           = _mm_cvttpd_epi32(ewrt);
 938 #ifdef __XOP__
 939             eweps            = _mm_frcz_pd(ewrt);
 940 #else
 941             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 942 #endif
 943             twoeweps         = _mm_add_pd(eweps,eweps);
 944             ewitab           = _mm_slli_epi32(ewitab,2);
 945             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 946             ewtabD           = _mm_setzero_pd();
 947             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 948             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 949             ewtabFn          = _mm_setzero_pd();
 950             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 951             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 952             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 953             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 954             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 955
 956             /* Update potential sum for this i atom from the interaction with this j atom. */
 957             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 958             velecsum         = _mm_add_pd(velecsum,velec);
 959
 960             fscal            = felec;
 961
 962             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 963
 964             /* Update vectorial force */
 965             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 966             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 967             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 968
 969             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 970             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 971             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 972
 973             /**************************
 974              * CALCULATE INTERACTIONS *
 975              **************************/
 976
 977             r23              = _mm_mul_pd(rsq23,rinv23);
 978
 979             /* EWALD ELECTROSTATICS */
 980
 981             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 982             ewrt             = _mm_mul_pd(r23,ewtabscale);
 983             ewitab           = _mm_cvttpd_epi32(ewrt);
 984 #ifdef __XOP__
 985             eweps            = _mm_frcz_pd(ewrt);
 986 #else
 987             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 988 #endif
 989             twoeweps         = _mm_add_pd(eweps,eweps);
 990             ewitab           = _mm_slli_epi32(ewitab,2);
 991             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 992             ewtabD           = _mm_setzero_pd();
 993             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 994             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 995             ewtabFn          = _mm_setzero_pd();
 996             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 997             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 998             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 999             velec            = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
1000             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1001
1002             /* Update potential sum for this i atom from the interaction with this j atom. */
1003             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1004             velecsum         = _mm_add_pd(velecsum,velec);
1005
1006             fscal            = felec;
1007
1008             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1009
1010             /* Update vectorial force */
1011             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1012             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1013             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1014
1015             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1016             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1017             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1018
1019             /**************************
1020              * CALCULATE INTERACTIONS *
1021              **************************/
1022
1023             r31              = _mm_mul_pd(rsq31,rinv31);
1024
1025             /* EWALD ELECTROSTATICS */
1026
1027             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1028             ewrt             = _mm_mul_pd(r31,ewtabscale);
1029             ewitab           = _mm_cvttpd_epi32(ewrt);
1030 #ifdef __XOP__
1031             eweps            = _mm_frcz_pd(ewrt);
1032 #else
1033             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1034 #endif
1035             twoeweps         = _mm_add_pd(eweps,eweps);
1036             ewitab           = _mm_slli_epi32(ewitab,2);
1037             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1038             ewtabD           = _mm_setzero_pd();
1039             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1040             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1041             ewtabFn          = _mm_setzero_pd();
1042             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1043             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1044             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1045             velec            = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
1046             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1047
1048             /* Update potential sum for this i atom from the interaction with this j atom. */
1049             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1050             velecsum         = _mm_add_pd(velecsum,velec);
1051
1052             fscal            = felec;
1053
1054             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1055
1056             /* Update vectorial force */
1057             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1058             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1059             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1060
1061             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1062             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1063             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             r32              = _mm_mul_pd(rsq32,rinv32);
1070
1071             /* EWALD ELECTROSTATICS */
1072
1073             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1074             ewrt             = _mm_mul_pd(r32,ewtabscale);
1075             ewitab           = _mm_cvttpd_epi32(ewrt);
1076 #ifdef __XOP__
1077             eweps            = _mm_frcz_pd(ewrt);
1078 #else
1079             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1080 #endif
1081             twoeweps         = _mm_add_pd(eweps,eweps);
1082             ewitab           = _mm_slli_epi32(ewitab,2);
1083             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1084             ewtabD           = _mm_setzero_pd();
1085             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1086             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1087             ewtabFn          = _mm_setzero_pd();
1088             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1089             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1090             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1091             velec            = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
1092             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1093
1094             /* Update potential sum for this i atom from the interaction with this j atom. */
1095             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1096             velecsum         = _mm_add_pd(velecsum,velec);
1097
1098             fscal            = felec;
1099
1100             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1101
1102             /* Update vectorial force */
1103             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1104             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1105             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1106
1107             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1108             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1109             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             r33              = _mm_mul_pd(rsq33,rinv33);
1116
1117             /* EWALD ELECTROSTATICS */
1118
1119             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1120             ewrt             = _mm_mul_pd(r33,ewtabscale);
1121             ewitab           = _mm_cvttpd_epi32(ewrt);
1122 #ifdef __XOP__
1123             eweps            = _mm_frcz_pd(ewrt);
1124 #else
1125             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1126 #endif
1127             twoeweps         = _mm_add_pd(eweps,eweps);
1128             ewitab           = _mm_slli_epi32(ewitab,2);
1129             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1130             ewtabD           = _mm_setzero_pd();
1131             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1132             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1133             ewtabFn          = _mm_setzero_pd();
1134             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1135             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1136             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1137             velec            = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
1138             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1139
1140             /* Update potential sum for this i atom from the interaction with this j atom. */
1141             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1142             velecsum         = _mm_add_pd(velecsum,velec);
1143
1144             fscal            = felec;
1145
1146             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1147
1148             /* Update vectorial force */
1149             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1150             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1151             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1152
1153             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1154             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1155             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1156
1157             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1158
1159             /* Inner loop uses 396 flops */
1160         }
1161
1162         /* End of innermost loop */
1163
1164         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1165                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1166
1167         ggid                        = gid[iidx];
1168         /* Update potential energies */
1169         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1170
1171         /* Increment number of inner iterations */
1172         inneriter                  += j_index_end - j_index_start;
1173
1174         /* Outer loop uses 19 flops */
1175     }
1176
1177     /* Increment number of outer iterations */
1178     outeriter        += nri;
1179
1180     /* Update outer/inner flops */
1181
1182     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*396);
1183 }
1184 /*
1185  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_128_fma_double
1186  * Electrostatics interaction: Ewald
1187  * VdW interaction:            None
1188  * Geometry:                   Water4-Water4
1189  * Calculate force/pot:        Force
1190  */
1191 void
1192 nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_128_fma_double
1193                     (t_nblist                    * gmx_restrict       nlist,
1194                      rvec                        * gmx_restrict          xx,
1195                      rvec                        * gmx_restrict          ff,
1196                      t_forcerec                  * gmx_restrict          fr,
1197                      t_mdatoms                   * gmx_restrict     mdatoms,
1198                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1199                      t_nrnb                      * gmx_restrict        nrnb)
1200 {
1201     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1202      * just 0 for non-waters.
1203      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1204      * jnr indices corresponding to data put in the four positions in the SIMD register.
1205      */
1206     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1207     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1208     int              jnrA,jnrB;
1209     int              j_coord_offsetA,j_coord_offsetB;
1210     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1211     real             rcutoff_scalar;
1212     real             *shiftvec,*fshift,*x,*f;
1213     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1214     int              vdwioffset1;
1215     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1216     int              vdwioffset2;
1217     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1218     int              vdwioffset3;
1219     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1220     int              vdwjidx1A,vdwjidx1B;
1221     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1222     int              vdwjidx2A,vdwjidx2B;
1223     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1224     int              vdwjidx3A,vdwjidx3B;
1225     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1226     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1227     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1228     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1229     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1230     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1231     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1232     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1233     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1234     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1235     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1236     real             *charge;
1237     __m128i          ewitab;
1238     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1239     real             *ewtab;
1240     __m128d          dummy_mask,cutoff_mask;
1241     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1242     __m128d          one     = _mm_set1_pd(1.0);
1243     __m128d          two     = _mm_set1_pd(2.0);
1244     x                = xx[0];
1245     f                = ff[0];
1246
1247     nri              = nlist->nri;
1248     iinr             = nlist->iinr;
1249     jindex           = nlist->jindex;
1250     jjnr             = nlist->jjnr;
1251     shiftidx         = nlist->shift;
1252     gid              = nlist->gid;
1253     shiftvec         = fr->shift_vec[0];
1254     fshift           = fr->fshift[0];
1255     facel            = _mm_set1_pd(fr->epsfac);
1256     charge           = mdatoms->chargeA;
1257
1258     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1259     ewtab            = fr->ic->tabq_coul_F;
1260     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1261     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1262
1263     /* Setup water-specific parameters */
1264     inr              = nlist->iinr[0];
1265     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1266     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1267     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1268
1269     jq1              = _mm_set1_pd(charge[inr+1]);
1270     jq2              = _mm_set1_pd(charge[inr+2]);
1271     jq3              = _mm_set1_pd(charge[inr+3]);
1272     qq11             = _mm_mul_pd(iq1,jq1);
1273     qq12             = _mm_mul_pd(iq1,jq2);
1274     qq13             = _mm_mul_pd(iq1,jq3);
1275     qq21             = _mm_mul_pd(iq2,jq1);
1276     qq22             = _mm_mul_pd(iq2,jq2);
1277     qq23             = _mm_mul_pd(iq2,jq3);
1278     qq31             = _mm_mul_pd(iq3,jq1);
1279     qq32             = _mm_mul_pd(iq3,jq2);
1280     qq33             = _mm_mul_pd(iq3,jq3);
1281
1282     /* Avoid stupid compiler warnings */
1283     jnrA = jnrB = 0;
1284     j_coord_offsetA = 0;
1285     j_coord_offsetB = 0;
1286
1287     outeriter        = 0;
1288     inneriter        = 0;
1289
1290     /* Start outer loop over neighborlists */
1291     for(iidx=0; iidx<nri; iidx++)
1292     {
1293         /* Load shift vector for this list */
1294         i_shift_offset   = DIM*shiftidx[iidx];
1295
1296         /* Load limits for loop over neighbors */
1297         j_index_start    = jindex[iidx];
1298         j_index_end      = jindex[iidx+1];
1299
1300         /* Get outer coordinate index */
1301         inr              = iinr[iidx];
1302         i_coord_offset   = DIM*inr;
1303
1304         /* Load i particle coords and add shift vector */
1305         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1306                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1307
1308         fix1             = _mm_setzero_pd();
1309         fiy1             = _mm_setzero_pd();
1310         fiz1             = _mm_setzero_pd();
1311         fix2             = _mm_setzero_pd();
1312         fiy2             = _mm_setzero_pd();
1313         fiz2             = _mm_setzero_pd();
1314         fix3             = _mm_setzero_pd();
1315         fiy3             = _mm_setzero_pd();
1316         fiz3             = _mm_setzero_pd();
1317
1318         /* Start inner kernel loop */
1319         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1320         {
1321
1322             /* Get j neighbor index, and coordinate index */
1323             jnrA             = jjnr[jidx];
1324             jnrB             = jjnr[jidx+1];
1325             j_coord_offsetA  = DIM*jnrA;
1326             j_coord_offsetB  = DIM*jnrB;
1327
1328             /* load j atom coordinates */
1329             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1330                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1331
1332             /* Calculate displacement vector */
1333             dx11             = _mm_sub_pd(ix1,jx1);
1334             dy11             = _mm_sub_pd(iy1,jy1);
1335             dz11             = _mm_sub_pd(iz1,jz1);
1336             dx12             = _mm_sub_pd(ix1,jx2);
1337             dy12             = _mm_sub_pd(iy1,jy2);
1338             dz12             = _mm_sub_pd(iz1,jz2);
1339             dx13             = _mm_sub_pd(ix1,jx3);
1340             dy13             = _mm_sub_pd(iy1,jy3);
1341             dz13             = _mm_sub_pd(iz1,jz3);
1342             dx21             = _mm_sub_pd(ix2,jx1);
1343             dy21             = _mm_sub_pd(iy2,jy1);
1344             dz21             = _mm_sub_pd(iz2,jz1);
1345             dx22             = _mm_sub_pd(ix2,jx2);
1346             dy22             = _mm_sub_pd(iy2,jy2);
1347             dz22             = _mm_sub_pd(iz2,jz2);
1348             dx23             = _mm_sub_pd(ix2,jx3);
1349             dy23             = _mm_sub_pd(iy2,jy3);
1350             dz23             = _mm_sub_pd(iz2,jz3);
1351             dx31             = _mm_sub_pd(ix3,jx1);
1352             dy31             = _mm_sub_pd(iy3,jy1);
1353             dz31             = _mm_sub_pd(iz3,jz1);
1354             dx32             = _mm_sub_pd(ix3,jx2);
1355             dy32             = _mm_sub_pd(iy3,jy2);
1356             dz32             = _mm_sub_pd(iz3,jz2);
1357             dx33             = _mm_sub_pd(ix3,jx3);
1358             dy33             = _mm_sub_pd(iy3,jy3);
1359             dz33             = _mm_sub_pd(iz3,jz3);
1360
1361             /* Calculate squared distance and things based on it */
1362             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1363             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1364             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1365             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1366             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1367             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1368             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1369             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1370             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1371
1372             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1373             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1374             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1375             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1376             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1377             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1378             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1379             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1380             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1381
1382             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1383             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1384             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1385             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1386             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1387             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1388             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1389             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1390             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1391
1392             fjx1             = _mm_setzero_pd();
1393             fjy1             = _mm_setzero_pd();
1394             fjz1             = _mm_setzero_pd();
1395             fjx2             = _mm_setzero_pd();
1396             fjy2             = _mm_setzero_pd();
1397             fjz2             = _mm_setzero_pd();
1398             fjx3             = _mm_setzero_pd();
1399             fjy3             = _mm_setzero_pd();
1400             fjz3             = _mm_setzero_pd();
1401
1402             /**************************
1403              * CALCULATE INTERACTIONS *
1404              **************************/
1405
1406             r11              = _mm_mul_pd(rsq11,rinv11);
1407
1408             /* EWALD ELECTROSTATICS */
1409
1410             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1411             ewrt             = _mm_mul_pd(r11,ewtabscale);
1412             ewitab           = _mm_cvttpd_epi32(ewrt);
1413 #ifdef __XOP__
1414             eweps            = _mm_frcz_pd(ewrt);
1415 #else
1416             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1417 #endif
1418             twoeweps         = _mm_add_pd(eweps,eweps);
1419             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1420                                          &ewtabF,&ewtabFn);
1421             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1422             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1423
1424             fscal            = felec;
1425
1426             /* Update vectorial force */
1427             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1428             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1429             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1430
1431             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1432             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1433             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1434
1435             /**************************
1436              * CALCULATE INTERACTIONS *
1437              **************************/
1438
1439             r12              = _mm_mul_pd(rsq12,rinv12);
1440
1441             /* EWALD ELECTROSTATICS */
1442
1443             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1444             ewrt             = _mm_mul_pd(r12,ewtabscale);
1445             ewitab           = _mm_cvttpd_epi32(ewrt);
1446 #ifdef __XOP__
1447             eweps            = _mm_frcz_pd(ewrt);
1448 #else
1449             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1450 #endif
1451             twoeweps         = _mm_add_pd(eweps,eweps);
1452             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1453                                          &ewtabF,&ewtabFn);
1454             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1455             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1456
1457             fscal            = felec;
1458
1459             /* Update vectorial force */
1460             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1461             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1462             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1463
1464             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1465             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1466             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1467
1468             /**************************
1469              * CALCULATE INTERACTIONS *
1470              **************************/
1471
1472             r13              = _mm_mul_pd(rsq13,rinv13);
1473
1474             /* EWALD ELECTROSTATICS */
1475
1476             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1477             ewrt             = _mm_mul_pd(r13,ewtabscale);
1478             ewitab           = _mm_cvttpd_epi32(ewrt);
1479 #ifdef __XOP__
1480             eweps            = _mm_frcz_pd(ewrt);
1481 #else
1482             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1483 #endif
1484             twoeweps         = _mm_add_pd(eweps,eweps);
1485             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1486                                          &ewtabF,&ewtabFn);
1487             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1488             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1489
1490             fscal            = felec;
1491
1492             /* Update vectorial force */
1493             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1494             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1495             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1496
1497             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1498             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1499             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1500
1501             /**************************
1502              * CALCULATE INTERACTIONS *
1503              **************************/
1504
1505             r21              = _mm_mul_pd(rsq21,rinv21);
1506
1507             /* EWALD ELECTROSTATICS */
1508
1509             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1510             ewrt             = _mm_mul_pd(r21,ewtabscale);
1511             ewitab           = _mm_cvttpd_epi32(ewrt);
1512 #ifdef __XOP__
1513             eweps            = _mm_frcz_pd(ewrt);
1514 #else
1515             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1516 #endif
1517             twoeweps         = _mm_add_pd(eweps,eweps);
1518             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1519                                          &ewtabF,&ewtabFn);
1520             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1521             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1522
1523             fscal            = felec;
1524
1525             /* Update vectorial force */
1526             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1527             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1528             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1529
1530             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1531             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1532             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1533
1534             /**************************
1535              * CALCULATE INTERACTIONS *
1536              **************************/
1537
1538             r22              = _mm_mul_pd(rsq22,rinv22);
1539
1540             /* EWALD ELECTROSTATICS */
1541
1542             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1543             ewrt             = _mm_mul_pd(r22,ewtabscale);
1544             ewitab           = _mm_cvttpd_epi32(ewrt);
1545 #ifdef __XOP__
1546             eweps            = _mm_frcz_pd(ewrt);
1547 #else
1548             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1549 #endif
1550             twoeweps         = _mm_add_pd(eweps,eweps);
1551             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1552                                          &ewtabF,&ewtabFn);
1553             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1554             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1555
1556             fscal            = felec;
1557
1558             /* Update vectorial force */
1559             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1560             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1561             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1562
1563             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1564             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1565             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1566
1567             /**************************
1568              * CALCULATE INTERACTIONS *
1569              **************************/
1570
1571             r23              = _mm_mul_pd(rsq23,rinv23);
1572
1573             /* EWALD ELECTROSTATICS */
1574
1575             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1576             ewrt             = _mm_mul_pd(r23,ewtabscale);
1577             ewitab           = _mm_cvttpd_epi32(ewrt);
1578 #ifdef __XOP__
1579             eweps            = _mm_frcz_pd(ewrt);
1580 #else
1581             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1582 #endif
1583             twoeweps         = _mm_add_pd(eweps,eweps);
1584             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1585                                          &ewtabF,&ewtabFn);
1586             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1587             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1588
1589             fscal            = felec;
1590
1591             /* Update vectorial force */
1592             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1593             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1594             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1595
1596             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1597             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1598             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1599
1600             /**************************
1601              * CALCULATE INTERACTIONS *
1602              **************************/
1603
1604             r31              = _mm_mul_pd(rsq31,rinv31);
1605
1606             /* EWALD ELECTROSTATICS */
1607
1608             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1609             ewrt             = _mm_mul_pd(r31,ewtabscale);
1610             ewitab           = _mm_cvttpd_epi32(ewrt);
1611 #ifdef __XOP__
1612             eweps            = _mm_frcz_pd(ewrt);
1613 #else
1614             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1615 #endif
1616             twoeweps         = _mm_add_pd(eweps,eweps);
1617             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1618                                          &ewtabF,&ewtabFn);
1619             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1620             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1621
1622             fscal            = felec;
1623
1624             /* Update vectorial force */
1625             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1626             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1627             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1628
1629             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1630             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1631             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1632
1633             /**************************
1634              * CALCULATE INTERACTIONS *
1635              **************************/
1636
1637             r32              = _mm_mul_pd(rsq32,rinv32);
1638
1639             /* EWALD ELECTROSTATICS */
1640
1641             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1642             ewrt             = _mm_mul_pd(r32,ewtabscale);
1643             ewitab           = _mm_cvttpd_epi32(ewrt);
1644 #ifdef __XOP__
1645             eweps            = _mm_frcz_pd(ewrt);
1646 #else
1647             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1648 #endif
1649             twoeweps         = _mm_add_pd(eweps,eweps);
1650             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1651                                          &ewtabF,&ewtabFn);
1652             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1653             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1654
1655             fscal            = felec;
1656
1657             /* Update vectorial force */
1658             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1659             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1660             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1661
1662             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1663             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1664             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1665
1666             /**************************
1667              * CALCULATE INTERACTIONS *
1668              **************************/
1669
1670             r33              = _mm_mul_pd(rsq33,rinv33);
1671
1672             /* EWALD ELECTROSTATICS */
1673
1674             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1675             ewrt             = _mm_mul_pd(r33,ewtabscale);
1676             ewitab           = _mm_cvttpd_epi32(ewrt);
1677 #ifdef __XOP__
1678             eweps            = _mm_frcz_pd(ewrt);
1679 #else
1680             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1681 #endif
1682             twoeweps         = _mm_add_pd(eweps,eweps);
1683             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1684                                          &ewtabF,&ewtabFn);
1685             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1686             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1687
1688             fscal            = felec;
1689
1690             /* Update vectorial force */
1691             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1692             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1693             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1694
1695             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1696             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1697             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1698
1699             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);
1700
1701             /* Inner loop uses 351 flops */
1702         }
1703
1704         if(jidx<j_index_end)
1705         {
1706
1707             jnrA             = jjnr[jidx];
1708             j_coord_offsetA  = DIM*jnrA;
1709
1710             /* load j atom coordinates */
1711             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
1712                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1713
1714             /* Calculate displacement vector */
1715             dx11             = _mm_sub_pd(ix1,jx1);
1716             dy11             = _mm_sub_pd(iy1,jy1);
1717             dz11             = _mm_sub_pd(iz1,jz1);
1718             dx12             = _mm_sub_pd(ix1,jx2);
1719             dy12             = _mm_sub_pd(iy1,jy2);
1720             dz12             = _mm_sub_pd(iz1,jz2);
1721             dx13             = _mm_sub_pd(ix1,jx3);
1722             dy13             = _mm_sub_pd(iy1,jy3);
1723             dz13             = _mm_sub_pd(iz1,jz3);
1724             dx21             = _mm_sub_pd(ix2,jx1);
1725             dy21             = _mm_sub_pd(iy2,jy1);
1726             dz21             = _mm_sub_pd(iz2,jz1);
1727             dx22             = _mm_sub_pd(ix2,jx2);
1728             dy22             = _mm_sub_pd(iy2,jy2);
1729             dz22             = _mm_sub_pd(iz2,jz2);
1730             dx23             = _mm_sub_pd(ix2,jx3);
1731             dy23             = _mm_sub_pd(iy2,jy3);
1732             dz23             = _mm_sub_pd(iz2,jz3);
1733             dx31             = _mm_sub_pd(ix3,jx1);
1734             dy31             = _mm_sub_pd(iy3,jy1);
1735             dz31             = _mm_sub_pd(iz3,jz1);
1736             dx32             = _mm_sub_pd(ix3,jx2);
1737             dy32             = _mm_sub_pd(iy3,jy2);
1738             dz32             = _mm_sub_pd(iz3,jz2);
1739             dx33             = _mm_sub_pd(ix3,jx3);
1740             dy33             = _mm_sub_pd(iy3,jy3);
1741             dz33             = _mm_sub_pd(iz3,jz3);
1742
1743             /* Calculate squared distance and things based on it */
1744             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1745             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1746             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1747             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1748             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1749             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1750             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1751             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1752             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1753
1754             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1755             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1756             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1757             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1758             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1759             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1760             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1761             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1762             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1763
1764             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1765             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1766             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1767             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1768             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1769             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1770             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1771             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1772             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1773
1774             fjx1             = _mm_setzero_pd();
1775             fjy1             = _mm_setzero_pd();
1776             fjz1             = _mm_setzero_pd();
1777             fjx2             = _mm_setzero_pd();
1778             fjy2             = _mm_setzero_pd();
1779             fjz2             = _mm_setzero_pd();
1780             fjx3             = _mm_setzero_pd();
1781             fjy3             = _mm_setzero_pd();
1782             fjz3             = _mm_setzero_pd();
1783
1784             /**************************
1785              * CALCULATE INTERACTIONS *
1786              **************************/
1787
1788             r11              = _mm_mul_pd(rsq11,rinv11);
1789
1790             /* EWALD ELECTROSTATICS */
1791
1792             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1793             ewrt             = _mm_mul_pd(r11,ewtabscale);
1794             ewitab           = _mm_cvttpd_epi32(ewrt);
1795 #ifdef __XOP__
1796             eweps            = _mm_frcz_pd(ewrt);
1797 #else
1798             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1799 #endif
1800             twoeweps         = _mm_add_pd(eweps,eweps);
1801             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1802             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1803             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1804
1805             fscal            = felec;
1806
1807             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1808
1809             /* Update vectorial force */
1810             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1811             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1812             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1813
1814             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1815             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1816             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1817
1818             /**************************
1819              * CALCULATE INTERACTIONS *
1820              **************************/
1821
1822             r12              = _mm_mul_pd(rsq12,rinv12);
1823
1824             /* EWALD ELECTROSTATICS */
1825
1826             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1827             ewrt             = _mm_mul_pd(r12,ewtabscale);
1828             ewitab           = _mm_cvttpd_epi32(ewrt);
1829 #ifdef __XOP__
1830             eweps            = _mm_frcz_pd(ewrt);
1831 #else
1832             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1833 #endif
1834             twoeweps         = _mm_add_pd(eweps,eweps);
1835             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1836             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1837             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1838
1839             fscal            = felec;
1840
1841             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1842
1843             /* Update vectorial force */
1844             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1845             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1846             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1847
1848             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1849             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1850             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1851
1852             /**************************
1853              * CALCULATE INTERACTIONS *
1854              **************************/
1855
1856             r13              = _mm_mul_pd(rsq13,rinv13);
1857
1858             /* EWALD ELECTROSTATICS */
1859
1860             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1861             ewrt             = _mm_mul_pd(r13,ewtabscale);
1862             ewitab           = _mm_cvttpd_epi32(ewrt);
1863 #ifdef __XOP__
1864             eweps            = _mm_frcz_pd(ewrt);
1865 #else
1866             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1867 #endif
1868             twoeweps         = _mm_add_pd(eweps,eweps);
1869             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1870             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1871             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1872
1873             fscal            = felec;
1874
1875             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1876
1877             /* Update vectorial force */
1878             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1879             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1880             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1881
1882             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1883             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1884             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1885
1886             /**************************
1887              * CALCULATE INTERACTIONS *
1888              **************************/
1889
1890             r21              = _mm_mul_pd(rsq21,rinv21);
1891
1892             /* EWALD ELECTROSTATICS */
1893
1894             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1895             ewrt             = _mm_mul_pd(r21,ewtabscale);
1896             ewitab           = _mm_cvttpd_epi32(ewrt);
1897 #ifdef __XOP__
1898             eweps            = _mm_frcz_pd(ewrt);
1899 #else
1900             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1901 #endif
1902             twoeweps         = _mm_add_pd(eweps,eweps);
1903             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1904             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1905             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1906
1907             fscal            = felec;
1908
1909             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1910
1911             /* Update vectorial force */
1912             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1913             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1914             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1915
1916             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1917             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1918             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1919
1920             /**************************
1921              * CALCULATE INTERACTIONS *
1922              **************************/
1923
1924             r22              = _mm_mul_pd(rsq22,rinv22);
1925
1926             /* EWALD ELECTROSTATICS */
1927
1928             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1929             ewrt             = _mm_mul_pd(r22,ewtabscale);
1930             ewitab           = _mm_cvttpd_epi32(ewrt);
1931 #ifdef __XOP__
1932             eweps            = _mm_frcz_pd(ewrt);
1933 #else
1934             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1935 #endif
1936             twoeweps         = _mm_add_pd(eweps,eweps);
1937             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1938             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1939             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1940
1941             fscal            = felec;
1942
1943             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1944
1945             /* Update vectorial force */
1946             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1947             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1948             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1949
1950             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1951             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1952             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1953
1954             /**************************
1955              * CALCULATE INTERACTIONS *
1956              **************************/
1957
1958             r23              = _mm_mul_pd(rsq23,rinv23);
1959
1960             /* EWALD ELECTROSTATICS */
1961
1962             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1963             ewrt             = _mm_mul_pd(r23,ewtabscale);
1964             ewitab           = _mm_cvttpd_epi32(ewrt);
1965 #ifdef __XOP__
1966             eweps            = _mm_frcz_pd(ewrt);
1967 #else
1968             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1969 #endif
1970             twoeweps         = _mm_add_pd(eweps,eweps);
1971             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1972             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1973             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1974
1975             fscal            = felec;
1976
1977             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1978
1979             /* Update vectorial force */
1980             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1981             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1982             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1983
1984             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1985             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1986             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1987
1988             /**************************
1989              * CALCULATE INTERACTIONS *
1990              **************************/
1991
1992             r31              = _mm_mul_pd(rsq31,rinv31);
1993
1994             /* EWALD ELECTROSTATICS */
1995
1996             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1997             ewrt             = _mm_mul_pd(r31,ewtabscale);
1998             ewitab           = _mm_cvttpd_epi32(ewrt);
1999 #ifdef __XOP__
2000             eweps            = _mm_frcz_pd(ewrt);
2001 #else
2002             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2003 #endif
2004             twoeweps         = _mm_add_pd(eweps,eweps);
2005             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2006             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2007             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
2008
2009             fscal            = felec;
2010
2011             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2012
2013             /* Update vectorial force */
2014             fix3             = _mm_macc_pd(dx31,fscal,fix3);
2015             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
2016             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
2017
2018             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
2019             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
2020             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
2021
2022             /**************************
2023              * CALCULATE INTERACTIONS *
2024              **************************/
2025
2026             r32              = _mm_mul_pd(rsq32,rinv32);
2027
2028             /* EWALD ELECTROSTATICS */
2029
2030             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2031             ewrt             = _mm_mul_pd(r32,ewtabscale);
2032             ewitab           = _mm_cvttpd_epi32(ewrt);
2033 #ifdef __XOP__
2034             eweps            = _mm_frcz_pd(ewrt);
2035 #else
2036             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2037 #endif
2038             twoeweps         = _mm_add_pd(eweps,eweps);
2039             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2040             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2041             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2042
2043             fscal            = felec;
2044
2045             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2046
2047             /* Update vectorial force */
2048             fix3             = _mm_macc_pd(dx32,fscal,fix3);
2049             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
2050             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
2051
2052             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
2053             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
2054             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
2055
2056             /**************************
2057              * CALCULATE INTERACTIONS *
2058              **************************/
2059
2060             r33              = _mm_mul_pd(rsq33,rinv33);
2061
2062             /* EWALD ELECTROSTATICS */
2063
2064             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2065             ewrt             = _mm_mul_pd(r33,ewtabscale);
2066             ewitab           = _mm_cvttpd_epi32(ewrt);
2067 #ifdef __XOP__
2068             eweps            = _mm_frcz_pd(ewrt);
2069 #else
2070             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2071 #endif
2072             twoeweps         = _mm_add_pd(eweps,eweps);
2073             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2074             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2075             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2076
2077             fscal            = felec;
2078
2079             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2080
2081             /* Update vectorial force */
2082             fix3             = _mm_macc_pd(dx33,fscal,fix3);
2083             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
2084             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
2085
2086             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
2087             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
2088             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
2089
2090             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2091
2092             /* Inner loop uses 351 flops */
2093         }
2094
2095         /* End of innermost loop */
2096
2097         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2098                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
2099
2100         /* Increment number of inner iterations */
2101         inneriter                  += j_index_end - j_index_start;
2102
2103         /* Outer loop uses 18 flops */
2104     }
2105
2106     /* Increment number of outer iterations */
2107     outeriter        += nri;
2108
2109     /* Update outer/inner flops */
2110
2111     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*351);
2112 }