src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJEw_GeomW3W3_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_VdwLJEw_GeomW3W3_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW3W3_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              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     int              vdwjidx1A,vdwjidx1B;
  89     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  90     int              vdwjidx2A,vdwjidx2B;
  91     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  94     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  95     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  97     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  98     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 100     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 101     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 102     real             *charge;
 103     int              nvdwtype;
 104     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 105     int              *vdwtype;
 106     real             *vdwparam;
 107     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 108     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 109     __m128d           c6grid_00;
 110     __m128d           c6grid_01;
 111     __m128d           c6grid_02;
 112     __m128d           c6grid_10;
 113     __m128d           c6grid_11;
 114     __m128d           c6grid_12;
 115     __m128d           c6grid_20;
 116     __m128d           c6grid_21;
 117     __m128d           c6grid_22;
 118     real             *vdwgridparam;
 119     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 120     __m128d           one_half  = _mm_set1_pd(0.5);
 121     __m128d           minus_one = _mm_set1_pd(-1.0);
 122     __m128i          ewitab;
 123     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 124     real             *ewtab;
 125     __m128d          dummy_mask,cutoff_mask;
 126     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 127     __m128d          one     = _mm_set1_pd(1.0);
 128     __m128d          two     = _mm_set1_pd(2.0);
 129     x                = xx[0];
 130     f                = ff[0];
 131
 132     nri              = nlist->nri;
 133     iinr             = nlist->iinr;
 134     jindex           = nlist->jindex;
 135     jjnr             = nlist->jjnr;
 136     shiftidx         = nlist->shift;
 137     gid              = nlist->gid;
 138     shiftvec         = fr->shift_vec[0];
 139     fshift           = fr->fshift[0];
 140     facel            = _mm_set1_pd(fr->epsfac);
 141     charge           = mdatoms->chargeA;
 142     nvdwtype         = fr->ntype;
 143     vdwparam         = fr->nbfp;
 144     vdwtype          = mdatoms->typeA;
 145     vdwgridparam     = fr->ljpme_c6grid;
 146     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 147     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 148     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 149
 150     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 151     ewtab            = fr->ic->tabq_coul_FDV0;
 152     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 153     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 154
 155     /* Setup water-specific parameters */
 156     inr              = nlist->iinr[0];
 157     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 158     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 159     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 160     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 161
 162     jq0              = _mm_set1_pd(charge[inr+0]);
 163     jq1              = _mm_set1_pd(charge[inr+1]);
 164     jq2              = _mm_set1_pd(charge[inr+2]);
 165     vdwjidx0A        = 2*vdwtype[inr+0];
 166     qq00             = _mm_mul_pd(iq0,jq0);
 167     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 168     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 169     c6grid_00        = _mm_set1_pd(vdwgridparam[vdwioffset0+vdwjidx0A]);
 170     qq01             = _mm_mul_pd(iq0,jq1);
 171     qq02             = _mm_mul_pd(iq0,jq2);
 172     qq10             = _mm_mul_pd(iq1,jq0);
 173     qq11             = _mm_mul_pd(iq1,jq1);
 174     qq12             = _mm_mul_pd(iq1,jq2);
 175     qq20             = _mm_mul_pd(iq2,jq0);
 176     qq21             = _mm_mul_pd(iq2,jq1);
 177     qq22             = _mm_mul_pd(iq2,jq2);
 178
 179     /* Avoid stupid compiler warnings */
 180     jnrA = jnrB = 0;
 181     j_coord_offsetA = 0;
 182     j_coord_offsetB = 0;
 183
 184     outeriter        = 0;
 185     inneriter        = 0;
 186
 187     /* Start outer loop over neighborlists */
 188     for(iidx=0; iidx<nri; iidx++)
 189     {
 190         /* Load shift vector for this list */
 191         i_shift_offset   = DIM*shiftidx[iidx];
 192
 193         /* Load limits for loop over neighbors */
 194         j_index_start    = jindex[iidx];
 195         j_index_end      = jindex[iidx+1];
 196
 197         /* Get outer coordinate index */
 198         inr              = iinr[iidx];
 199         i_coord_offset   = DIM*inr;
 200
 201         /* Load i particle coords and add shift vector */
 202         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 203                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 204
 205         fix0             = _mm_setzero_pd();
 206         fiy0             = _mm_setzero_pd();
 207         fiz0             = _mm_setzero_pd();
 208         fix1             = _mm_setzero_pd();
 209         fiy1             = _mm_setzero_pd();
 210         fiz1             = _mm_setzero_pd();
 211         fix2             = _mm_setzero_pd();
 212         fiy2             = _mm_setzero_pd();
 213         fiz2             = _mm_setzero_pd();
 214
 215         /* Reset potential sums */
 216         velecsum         = _mm_setzero_pd();
 217         vvdwsum          = _mm_setzero_pd();
 218
 219         /* Start inner kernel loop */
 220         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 221         {
 222
 223             /* Get j neighbor index, and coordinate index */
 224             jnrA             = jjnr[jidx];
 225             jnrB             = jjnr[jidx+1];
 226             j_coord_offsetA  = DIM*jnrA;
 227             j_coord_offsetB  = DIM*jnrB;
 228
 229             /* load j atom coordinates */
 230             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 231                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 232
 233             /* Calculate displacement vector */
 234             dx00             = _mm_sub_pd(ix0,jx0);
 235             dy00             = _mm_sub_pd(iy0,jy0);
 236             dz00             = _mm_sub_pd(iz0,jz0);
 237             dx01             = _mm_sub_pd(ix0,jx1);
 238             dy01             = _mm_sub_pd(iy0,jy1);
 239             dz01             = _mm_sub_pd(iz0,jz1);
 240             dx02             = _mm_sub_pd(ix0,jx2);
 241             dy02             = _mm_sub_pd(iy0,jy2);
 242             dz02             = _mm_sub_pd(iz0,jz2);
 243             dx10             = _mm_sub_pd(ix1,jx0);
 244             dy10             = _mm_sub_pd(iy1,jy0);
 245             dz10             = _mm_sub_pd(iz1,jz0);
 246             dx11             = _mm_sub_pd(ix1,jx1);
 247             dy11             = _mm_sub_pd(iy1,jy1);
 248             dz11             = _mm_sub_pd(iz1,jz1);
 249             dx12             = _mm_sub_pd(ix1,jx2);
 250             dy12             = _mm_sub_pd(iy1,jy2);
 251             dz12             = _mm_sub_pd(iz1,jz2);
 252             dx20             = _mm_sub_pd(ix2,jx0);
 253             dy20             = _mm_sub_pd(iy2,jy0);
 254             dz20             = _mm_sub_pd(iz2,jz0);
 255             dx21             = _mm_sub_pd(ix2,jx1);
 256             dy21             = _mm_sub_pd(iy2,jy1);
 257             dz21             = _mm_sub_pd(iz2,jz1);
 258             dx22             = _mm_sub_pd(ix2,jx2);
 259             dy22             = _mm_sub_pd(iy2,jy2);
 260             dz22             = _mm_sub_pd(iz2,jz2);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 264             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 265             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 266             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 267             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 268             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 269             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 270             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 271             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 272
 273             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 274             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 275             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 276             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 277             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 278             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 279             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 280             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 281             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 282
 283             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 284             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 285             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 286             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 287             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 288             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 289             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 290             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 291             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 292
 293             fjx0             = _mm_setzero_pd();
 294             fjy0             = _mm_setzero_pd();
 295             fjz0             = _mm_setzero_pd();
 296             fjx1             = _mm_setzero_pd();
 297             fjy1             = _mm_setzero_pd();
 298             fjz1             = _mm_setzero_pd();
 299             fjx2             = _mm_setzero_pd();
 300             fjy2             = _mm_setzero_pd();
 301             fjz2             = _mm_setzero_pd();
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             r00              = _mm_mul_pd(rsq00,rinv00);
 308
 309             /* EWALD ELECTROSTATICS */
 310
 311             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 312             ewrt             = _mm_mul_pd(r00,ewtabscale);
 313             ewitab           = _mm_cvttpd_epi32(ewrt);
 314 #ifdef __XOP__
 315             eweps            = _mm_frcz_pd(ewrt);
 316 #else
 317             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 318 #endif
 319             twoeweps         = _mm_add_pd(eweps,eweps);
 320             ewitab           = _mm_slli_epi32(ewitab,2);
 321             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 322             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 323             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 324             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 325             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 326             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 327             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 328             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 329             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 330             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 331
 332             /* Analytical LJ-PME */
 333             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 334             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 335             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 336             exponent         = gmx_simd_exp_d(ewcljrsq);
 337             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 338             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 339             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 340             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 341             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 342             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 343             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 344             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_pd(velecsum,velec);
 348             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 349
 350             fscal            = _mm_add_pd(felec,fvdw);
 351
 352             /* Update vectorial force */
 353             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 354             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 355             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 356
 357             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 358             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 359             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 360
 361             /**************************
 362              * CALCULATE INTERACTIONS *
 363              **************************/
 364
 365             r01              = _mm_mul_pd(rsq01,rinv01);
 366
 367             /* EWALD ELECTROSTATICS */
 368
 369             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 370             ewrt             = _mm_mul_pd(r01,ewtabscale);
 371             ewitab           = _mm_cvttpd_epi32(ewrt);
 372 #ifdef __XOP__
 373             eweps            = _mm_frcz_pd(ewrt);
 374 #else
 375             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 376 #endif
 377             twoeweps         = _mm_add_pd(eweps,eweps);
 378             ewitab           = _mm_slli_epi32(ewitab,2);
 379             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 380             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 381             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 382             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 383             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 384             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 385             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 386             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 387             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
 388             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velecsum         = _mm_add_pd(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             /* Update vectorial force */
 396             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 397             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 398             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 399
 400             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 401             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 402             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 403
 404             /**************************
 405              * CALCULATE INTERACTIONS *
 406              **************************/
 407
 408             r02              = _mm_mul_pd(rsq02,rinv02);
 409
 410             /* EWALD ELECTROSTATICS */
 411
 412             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 413             ewrt             = _mm_mul_pd(r02,ewtabscale);
 414             ewitab           = _mm_cvttpd_epi32(ewrt);
 415 #ifdef __XOP__
 416             eweps            = _mm_frcz_pd(ewrt);
 417 #else
 418             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 419 #endif
 420             twoeweps         = _mm_add_pd(eweps,eweps);
 421             ewitab           = _mm_slli_epi32(ewitab,2);
 422             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 423             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 424             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 425             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 426             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 427             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 428             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 429             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 430             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
 431             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velecsum         = _mm_add_pd(velecsum,velec);
 435
 436             fscal            = felec;
 437
 438             /* Update vectorial force */
 439             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 440             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 441             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 442
 443             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 444             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 445             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 446
 447             /**************************
 448              * CALCULATE INTERACTIONS *
 449              **************************/
 450
 451             r10              = _mm_mul_pd(rsq10,rinv10);
 452
 453             /* EWALD ELECTROSTATICS */
 454
 455             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 456             ewrt             = _mm_mul_pd(r10,ewtabscale);
 457             ewitab           = _mm_cvttpd_epi32(ewrt);
 458 #ifdef __XOP__
 459             eweps            = _mm_frcz_pd(ewrt);
 460 #else
 461             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 462 #endif
 463             twoeweps         = _mm_add_pd(eweps,eweps);
 464             ewitab           = _mm_slli_epi32(ewitab,2);
 465             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 466             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 467             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 468             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 469             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 470             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 471             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 472             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 473             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 474             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 475
 476             /* Update potential sum for this i atom from the interaction with this j atom. */
 477             velecsum         = _mm_add_pd(velecsum,velec);
 478
 479             fscal            = felec;
 480
 481             /* Update vectorial force */
 482             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 483             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 484             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 485
 486             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 487             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 488             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             r11              = _mm_mul_pd(rsq11,rinv11);
 495
 496             /* EWALD ELECTROSTATICS */
 497
 498             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 499             ewrt             = _mm_mul_pd(r11,ewtabscale);
 500             ewitab           = _mm_cvttpd_epi32(ewrt);
 501 #ifdef __XOP__
 502             eweps            = _mm_frcz_pd(ewrt);
 503 #else
 504             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 505 #endif
 506             twoeweps         = _mm_add_pd(eweps,eweps);
 507             ewitab           = _mm_slli_epi32(ewitab,2);
 508             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 509             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 510             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 511             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 512             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 513             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 514             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 515             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 516             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 517             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 518
 519             /* Update potential sum for this i atom from the interaction with this j atom. */
 520             velecsum         = _mm_add_pd(velecsum,velec);
 521
 522             fscal            = felec;
 523
 524             /* Update vectorial force */
 525             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 526             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 527             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 528
 529             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 530             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 531             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             r12              = _mm_mul_pd(rsq12,rinv12);
 538
 539             /* EWALD ELECTROSTATICS */
 540
 541             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 542             ewrt             = _mm_mul_pd(r12,ewtabscale);
 543             ewitab           = _mm_cvttpd_epi32(ewrt);
 544 #ifdef __XOP__
 545             eweps            = _mm_frcz_pd(ewrt);
 546 #else
 547             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 548 #endif
 549             twoeweps         = _mm_add_pd(eweps,eweps);
 550             ewitab           = _mm_slli_epi32(ewitab,2);
 551             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 552             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 553             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 554             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 555             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 556             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 557             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 558             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 559             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 560             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             velecsum         = _mm_add_pd(velecsum,velec);
 564
 565             fscal            = felec;
 566
 567             /* Update vectorial force */
 568             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 569             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 570             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 571
 572             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 573             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 574             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 575
 576             /**************************
 577              * CALCULATE INTERACTIONS *
 578              **************************/
 579
 580             r20              = _mm_mul_pd(rsq20,rinv20);
 581
 582             /* EWALD ELECTROSTATICS */
 583
 584             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 585             ewrt             = _mm_mul_pd(r20,ewtabscale);
 586             ewitab           = _mm_cvttpd_epi32(ewrt);
 587 #ifdef __XOP__
 588             eweps            = _mm_frcz_pd(ewrt);
 589 #else
 590             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 591 #endif
 592             twoeweps         = _mm_add_pd(eweps,eweps);
 593             ewitab           = _mm_slli_epi32(ewitab,2);
 594             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 595             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 596             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 597             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 598             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 599             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 600             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 601             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 602             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 603             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velecsum         = _mm_add_pd(velecsum,velec);
 607
 608             fscal            = felec;
 609
 610             /* Update vectorial force */
 611             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 612             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 613             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 614
 615             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 616             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 617             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 618
 619             /**************************
 620              * CALCULATE INTERACTIONS *
 621              **************************/
 622
 623             r21              = _mm_mul_pd(rsq21,rinv21);
 624
 625             /* EWALD ELECTROSTATICS */
 626
 627             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 628             ewrt             = _mm_mul_pd(r21,ewtabscale);
 629             ewitab           = _mm_cvttpd_epi32(ewrt);
 630 #ifdef __XOP__
 631             eweps            = _mm_frcz_pd(ewrt);
 632 #else
 633             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 634 #endif
 635             twoeweps         = _mm_add_pd(eweps,eweps);
 636             ewitab           = _mm_slli_epi32(ewitab,2);
 637             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 638             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 639             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 640             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 641             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 642             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 643             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 644             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 645             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 646             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 647
 648             /* Update potential sum for this i atom from the interaction with this j atom. */
 649             velecsum         = _mm_add_pd(velecsum,velec);
 650
 651             fscal            = felec;
 652
 653             /* Update vectorial force */
 654             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 655             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 656             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 657
 658             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 659             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 660             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 661
 662             /**************************
 663              * CALCULATE INTERACTIONS *
 664              **************************/
 665
 666             r22              = _mm_mul_pd(rsq22,rinv22);
 667
 668             /* EWALD ELECTROSTATICS */
 669
 670             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 671             ewrt             = _mm_mul_pd(r22,ewtabscale);
 672             ewitab           = _mm_cvttpd_epi32(ewrt);
 673 #ifdef __XOP__
 674             eweps            = _mm_frcz_pd(ewrt);
 675 #else
 676             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 677 #endif
 678             twoeweps         = _mm_add_pd(eweps,eweps);
 679             ewitab           = _mm_slli_epi32(ewitab,2);
 680             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 681             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 682             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 683             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 684             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 685             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 686             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 687             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 688             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 689             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 690
 691             /* Update potential sum for this i atom from the interaction with this j atom. */
 692             velecsum         = _mm_add_pd(velecsum,velec);
 693
 694             fscal            = felec;
 695
 696             /* Update vectorial force */
 697             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 698             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 699             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 700
 701             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 702             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 703             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 704
 705             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 706
 707             /* Inner loop uses 420 flops */
 708         }
 709
 710         if(jidx<j_index_end)
 711         {
 712
 713             jnrA             = jjnr[jidx];
 714             j_coord_offsetA  = DIM*jnrA;
 715
 716             /* load j atom coordinates */
 717             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 718                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 719
 720             /* Calculate displacement vector */
 721             dx00             = _mm_sub_pd(ix0,jx0);
 722             dy00             = _mm_sub_pd(iy0,jy0);
 723             dz00             = _mm_sub_pd(iz0,jz0);
 724             dx01             = _mm_sub_pd(ix0,jx1);
 725             dy01             = _mm_sub_pd(iy0,jy1);
 726             dz01             = _mm_sub_pd(iz0,jz1);
 727             dx02             = _mm_sub_pd(ix0,jx2);
 728             dy02             = _mm_sub_pd(iy0,jy2);
 729             dz02             = _mm_sub_pd(iz0,jz2);
 730             dx10             = _mm_sub_pd(ix1,jx0);
 731             dy10             = _mm_sub_pd(iy1,jy0);
 732             dz10             = _mm_sub_pd(iz1,jz0);
 733             dx11             = _mm_sub_pd(ix1,jx1);
 734             dy11             = _mm_sub_pd(iy1,jy1);
 735             dz11             = _mm_sub_pd(iz1,jz1);
 736             dx12             = _mm_sub_pd(ix1,jx2);
 737             dy12             = _mm_sub_pd(iy1,jy2);
 738             dz12             = _mm_sub_pd(iz1,jz2);
 739             dx20             = _mm_sub_pd(ix2,jx0);
 740             dy20             = _mm_sub_pd(iy2,jy0);
 741             dz20             = _mm_sub_pd(iz2,jz0);
 742             dx21             = _mm_sub_pd(ix2,jx1);
 743             dy21             = _mm_sub_pd(iy2,jy1);
 744             dz21             = _mm_sub_pd(iz2,jz1);
 745             dx22             = _mm_sub_pd(ix2,jx2);
 746             dy22             = _mm_sub_pd(iy2,jy2);
 747             dz22             = _mm_sub_pd(iz2,jz2);
 748
 749             /* Calculate squared distance and things based on it */
 750             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 751             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 752             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 753             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 754             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 755             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 756             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 757             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 758             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 759
 760             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 761             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 762             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 763             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 764             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 765             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 766             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 767             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 768             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 769
 770             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 771             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 772             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 773             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 774             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 775             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 776             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 777             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 778             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 779
 780             fjx0             = _mm_setzero_pd();
 781             fjy0             = _mm_setzero_pd();
 782             fjz0             = _mm_setzero_pd();
 783             fjx1             = _mm_setzero_pd();
 784             fjy1             = _mm_setzero_pd();
 785             fjz1             = _mm_setzero_pd();
 786             fjx2             = _mm_setzero_pd();
 787             fjy2             = _mm_setzero_pd();
 788             fjz2             = _mm_setzero_pd();
 789
 790             /**************************
 791              * CALCULATE INTERACTIONS *
 792              **************************/
 793
 794             r00              = _mm_mul_pd(rsq00,rinv00);
 795
 796             /* EWALD ELECTROSTATICS */
 797
 798             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 799             ewrt             = _mm_mul_pd(r00,ewtabscale);
 800             ewitab           = _mm_cvttpd_epi32(ewrt);
 801 #ifdef __XOP__
 802             eweps            = _mm_frcz_pd(ewrt);
 803 #else
 804             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 805 #endif
 806             twoeweps         = _mm_add_pd(eweps,eweps);
 807             ewitab           = _mm_slli_epi32(ewitab,2);
 808             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 809             ewtabD           = _mm_setzero_pd();
 810             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 811             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 812             ewtabFn          = _mm_setzero_pd();
 813             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 814             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 815             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 816             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 817             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 818
 819             /* Analytical LJ-PME */
 820             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 821             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 822             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 823             exponent         = gmx_simd_exp_d(ewcljrsq);
 824             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 825             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 826             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 827             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 828             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 829             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 830             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 831             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 832
 833             /* Update potential sum for this i atom from the interaction with this j atom. */
 834             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 835             velecsum         = _mm_add_pd(velecsum,velec);
 836             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 837             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 838
 839             fscal            = _mm_add_pd(felec,fvdw);
 840
 841             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 842
 843             /* Update vectorial force */
 844             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 845             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 846             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 847
 848             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 849             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 850             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 851
 852             /**************************
 853              * CALCULATE INTERACTIONS *
 854              **************************/
 855
 856             r01              = _mm_mul_pd(rsq01,rinv01);
 857
 858             /* EWALD ELECTROSTATICS */
 859
 860             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 861             ewrt             = _mm_mul_pd(r01,ewtabscale);
 862             ewitab           = _mm_cvttpd_epi32(ewrt);
 863 #ifdef __XOP__
 864             eweps            = _mm_frcz_pd(ewrt);
 865 #else
 866             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 867 #endif
 868             twoeweps         = _mm_add_pd(eweps,eweps);
 869             ewitab           = _mm_slli_epi32(ewitab,2);
 870             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 871             ewtabD           = _mm_setzero_pd();
 872             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 873             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 874             ewtabFn          = _mm_setzero_pd();
 875             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 876             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 877             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 878             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
 879             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 880
 881             /* Update potential sum for this i atom from the interaction with this j atom. */
 882             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 883             velecsum         = _mm_add_pd(velecsum,velec);
 884
 885             fscal            = felec;
 886
 887             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 888
 889             /* Update vectorial force */
 890             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 891             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 892             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 893
 894             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 895             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 896             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             r02              = _mm_mul_pd(rsq02,rinv02);
 903
 904             /* EWALD ELECTROSTATICS */
 905
 906             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 907             ewrt             = _mm_mul_pd(r02,ewtabscale);
 908             ewitab           = _mm_cvttpd_epi32(ewrt);
 909 #ifdef __XOP__
 910             eweps            = _mm_frcz_pd(ewrt);
 911 #else
 912             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 913 #endif
 914             twoeweps         = _mm_add_pd(eweps,eweps);
 915             ewitab           = _mm_slli_epi32(ewitab,2);
 916             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 917             ewtabD           = _mm_setzero_pd();
 918             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 919             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 920             ewtabFn          = _mm_setzero_pd();
 921             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 922             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 923             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 924             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
 925             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 926
 927             /* Update potential sum for this i atom from the interaction with this j atom. */
 928             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 929             velecsum         = _mm_add_pd(velecsum,velec);
 930
 931             fscal            = felec;
 932
 933             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 934
 935             /* Update vectorial force */
 936             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 937             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 938             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 939
 940             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 941             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 942             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             r10              = _mm_mul_pd(rsq10,rinv10);
 949
 950             /* EWALD ELECTROSTATICS */
 951
 952             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 953             ewrt             = _mm_mul_pd(r10,ewtabscale);
 954             ewitab           = _mm_cvttpd_epi32(ewrt);
 955 #ifdef __XOP__
 956             eweps            = _mm_frcz_pd(ewrt);
 957 #else
 958             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 959 #endif
 960             twoeweps         = _mm_add_pd(eweps,eweps);
 961             ewitab           = _mm_slli_epi32(ewitab,2);
 962             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 963             ewtabD           = _mm_setzero_pd();
 964             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 965             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 966             ewtabFn          = _mm_setzero_pd();
 967             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 968             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 969             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 970             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 971             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 972
 973             /* Update potential sum for this i atom from the interaction with this j atom. */
 974             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 975             velecsum         = _mm_add_pd(velecsum,velec);
 976
 977             fscal            = felec;
 978
 979             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 980
 981             /* Update vectorial force */
 982             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 983             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 984             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 985
 986             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 987             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 988             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 989
 990             /**************************
 991              * CALCULATE INTERACTIONS *
 992              **************************/
 993
 994             r11              = _mm_mul_pd(rsq11,rinv11);
 995
 996             /* EWALD ELECTROSTATICS */
 997
 998             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 999             ewrt             = _mm_mul_pd(r11,ewtabscale);
1000             ewitab           = _mm_cvttpd_epi32(ewrt);
1001 #ifdef __XOP__
1002             eweps            = _mm_frcz_pd(ewrt);
1003 #else
1004             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1005 #endif
1006             twoeweps         = _mm_add_pd(eweps,eweps);
1007             ewitab           = _mm_slli_epi32(ewitab,2);
1008             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1009             ewtabD           = _mm_setzero_pd();
1010             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1011             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1012             ewtabFn          = _mm_setzero_pd();
1013             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1014             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1015             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1016             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
1017             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1018
1019             /* Update potential sum for this i atom from the interaction with this j atom. */
1020             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1021             velecsum         = _mm_add_pd(velecsum,velec);
1022
1023             fscal            = felec;
1024
1025             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1026
1027             /* Update vectorial force */
1028             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1029             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1030             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1031
1032             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1033             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1034             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1035
1036             /**************************
1037              * CALCULATE INTERACTIONS *
1038              **************************/
1039
1040             r12              = _mm_mul_pd(rsq12,rinv12);
1041
1042             /* EWALD ELECTROSTATICS */
1043
1044             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1045             ewrt             = _mm_mul_pd(r12,ewtabscale);
1046             ewitab           = _mm_cvttpd_epi32(ewrt);
1047 #ifdef __XOP__
1048             eweps            = _mm_frcz_pd(ewrt);
1049 #else
1050             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1051 #endif
1052             twoeweps         = _mm_add_pd(eweps,eweps);
1053             ewitab           = _mm_slli_epi32(ewitab,2);
1054             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1055             ewtabD           = _mm_setzero_pd();
1056             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1057             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1058             ewtabFn          = _mm_setzero_pd();
1059             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1060             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1061             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1062             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
1063             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1064
1065             /* Update potential sum for this i atom from the interaction with this j atom. */
1066             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1067             velecsum         = _mm_add_pd(velecsum,velec);
1068
1069             fscal            = felec;
1070
1071             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1072
1073             /* Update vectorial force */
1074             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1075             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1076             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1077
1078             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1079             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1080             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1081
1082             /**************************
1083              * CALCULATE INTERACTIONS *
1084              **************************/
1085
1086             r20              = _mm_mul_pd(rsq20,rinv20);
1087
1088             /* EWALD ELECTROSTATICS */
1089
1090             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1091             ewrt             = _mm_mul_pd(r20,ewtabscale);
1092             ewitab           = _mm_cvttpd_epi32(ewrt);
1093 #ifdef __XOP__
1094             eweps            = _mm_frcz_pd(ewrt);
1095 #else
1096             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1097 #endif
1098             twoeweps         = _mm_add_pd(eweps,eweps);
1099             ewitab           = _mm_slli_epi32(ewitab,2);
1100             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1101             ewtabD           = _mm_setzero_pd();
1102             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1103             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1104             ewtabFn          = _mm_setzero_pd();
1105             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1106             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1107             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1108             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1109             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1110
1111             /* Update potential sum for this i atom from the interaction with this j atom. */
1112             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1113             velecsum         = _mm_add_pd(velecsum,velec);
1114
1115             fscal            = felec;
1116
1117             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1118
1119             /* Update vectorial force */
1120             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1121             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1122             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1123
1124             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1125             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1126             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1127
1128             /**************************
1129              * CALCULATE INTERACTIONS *
1130              **************************/
1131
1132             r21              = _mm_mul_pd(rsq21,rinv21);
1133
1134             /* EWALD ELECTROSTATICS */
1135
1136             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1137             ewrt             = _mm_mul_pd(r21,ewtabscale);
1138             ewitab           = _mm_cvttpd_epi32(ewrt);
1139 #ifdef __XOP__
1140             eweps            = _mm_frcz_pd(ewrt);
1141 #else
1142             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1143 #endif
1144             twoeweps         = _mm_add_pd(eweps,eweps);
1145             ewitab           = _mm_slli_epi32(ewitab,2);
1146             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1147             ewtabD           = _mm_setzero_pd();
1148             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1149             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1150             ewtabFn          = _mm_setzero_pd();
1151             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1152             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1153             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1154             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1155             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1156
1157             /* Update potential sum for this i atom from the interaction with this j atom. */
1158             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1159             velecsum         = _mm_add_pd(velecsum,velec);
1160
1161             fscal            = felec;
1162
1163             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1164
1165             /* Update vectorial force */
1166             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1167             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1168             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1169
1170             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1171             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1172             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1173
1174             /**************************
1175              * CALCULATE INTERACTIONS *
1176              **************************/
1177
1178             r22              = _mm_mul_pd(rsq22,rinv22);
1179
1180             /* EWALD ELECTROSTATICS */
1181
1182             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1183             ewrt             = _mm_mul_pd(r22,ewtabscale);
1184             ewitab           = _mm_cvttpd_epi32(ewrt);
1185 #ifdef __XOP__
1186             eweps            = _mm_frcz_pd(ewrt);
1187 #else
1188             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1189 #endif
1190             twoeweps         = _mm_add_pd(eweps,eweps);
1191             ewitab           = _mm_slli_epi32(ewitab,2);
1192             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1193             ewtabD           = _mm_setzero_pd();
1194             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1195             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1196             ewtabFn          = _mm_setzero_pd();
1197             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1198             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1199             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1200             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1201             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1202
1203             /* Update potential sum for this i atom from the interaction with this j atom. */
1204             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1205             velecsum         = _mm_add_pd(velecsum,velec);
1206
1207             fscal            = felec;
1208
1209             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1210
1211             /* Update vectorial force */
1212             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1213             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1214             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1215
1216             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1217             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1218             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1219
1220             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1221
1222             /* Inner loop uses 420 flops */
1223         }
1224
1225         /* End of innermost loop */
1226
1227         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1228                                               f+i_coord_offset,fshift+i_shift_offset);
1229
1230         ggid                        = gid[iidx];
1231         /* Update potential energies */
1232         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1233         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1234
1235         /* Increment number of inner iterations */
1236         inneriter                  += j_index_end - j_index_start;
1237
1238         /* Outer loop uses 20 flops */
1239     }
1240
1241     /* Increment number of outer iterations */
1242     outeriter        += nri;
1243
1244     /* Update outer/inner flops */
1245
1246     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*420);
1247 }
1248 /*
1249  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3W3_F_avx_128_fma_double
1250  * Electrostatics interaction: Ewald
1251  * VdW interaction:            LJEwald
1252  * Geometry:                   Water3-Water3
1253  * Calculate force/pot:        Force
1254  */
1255 void
1256 nb_kernel_ElecEw_VdwLJEw_GeomW3W3_F_avx_128_fma_double
1257                     (t_nblist                    * gmx_restrict       nlist,
1258                      rvec                        * gmx_restrict          xx,
1259                      rvec                        * gmx_restrict          ff,
1260                      t_forcerec                  * gmx_restrict          fr,
1261                      t_mdatoms                   * gmx_restrict     mdatoms,
1262                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1263                      t_nrnb                      * gmx_restrict        nrnb)
1264 {
1265     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1266      * just 0 for non-waters.
1267      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1268      * jnr indices corresponding to data put in the four positions in the SIMD register.
1269      */
1270     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1271     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1272     int              jnrA,jnrB;
1273     int              j_coord_offsetA,j_coord_offsetB;
1274     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1275     real             rcutoff_scalar;
1276     real             *shiftvec,*fshift,*x,*f;
1277     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1278     int              vdwioffset0;
1279     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1280     int              vdwioffset1;
1281     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1282     int              vdwioffset2;
1283     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1284     int              vdwjidx0A,vdwjidx0B;
1285     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1286     int              vdwjidx1A,vdwjidx1B;
1287     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1288     int              vdwjidx2A,vdwjidx2B;
1289     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1290     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1291     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1292     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1293     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1294     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1295     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1296     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1297     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1298     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1299     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1300     real             *charge;
1301     int              nvdwtype;
1302     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1303     int              *vdwtype;
1304     real             *vdwparam;
1305     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1306     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1307     __m128d           c6grid_00;
1308     __m128d           c6grid_01;
1309     __m128d           c6grid_02;
1310     __m128d           c6grid_10;
1311     __m128d           c6grid_11;
1312     __m128d           c6grid_12;
1313     __m128d           c6grid_20;
1314     __m128d           c6grid_21;
1315     __m128d           c6grid_22;
1316     real             *vdwgridparam;
1317     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1318     __m128d           one_half  = _mm_set1_pd(0.5);
1319     __m128d           minus_one = _mm_set1_pd(-1.0);
1320     __m128i          ewitab;
1321     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1322     real             *ewtab;
1323     __m128d          dummy_mask,cutoff_mask;
1324     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1325     __m128d          one     = _mm_set1_pd(1.0);
1326     __m128d          two     = _mm_set1_pd(2.0);
1327     x                = xx[0];
1328     f                = ff[0];
1329
1330     nri              = nlist->nri;
1331     iinr             = nlist->iinr;
1332     jindex           = nlist->jindex;
1333     jjnr             = nlist->jjnr;
1334     shiftidx         = nlist->shift;
1335     gid              = nlist->gid;
1336     shiftvec         = fr->shift_vec[0];
1337     fshift           = fr->fshift[0];
1338     facel            = _mm_set1_pd(fr->epsfac);
1339     charge           = mdatoms->chargeA;
1340     nvdwtype         = fr->ntype;
1341     vdwparam         = fr->nbfp;
1342     vdwtype          = mdatoms->typeA;
1343     vdwgridparam     = fr->ljpme_c6grid;
1344     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
1345     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
1346     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
1347
1348     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1349     ewtab            = fr->ic->tabq_coul_F;
1350     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1351     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1352
1353     /* Setup water-specific parameters */
1354     inr              = nlist->iinr[0];
1355     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1356     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1357     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1358     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1359
1360     jq0              = _mm_set1_pd(charge[inr+0]);
1361     jq1              = _mm_set1_pd(charge[inr+1]);
1362     jq2              = _mm_set1_pd(charge[inr+2]);
1363     vdwjidx0A        = 2*vdwtype[inr+0];
1364     qq00             = _mm_mul_pd(iq0,jq0);
1365     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1366     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1367     c6grid_00        = _mm_set1_pd(vdwgridparam[vdwioffset0+vdwjidx0A]);
1368     qq01             = _mm_mul_pd(iq0,jq1);
1369     qq02             = _mm_mul_pd(iq0,jq2);
1370     qq10             = _mm_mul_pd(iq1,jq0);
1371     qq11             = _mm_mul_pd(iq1,jq1);
1372     qq12             = _mm_mul_pd(iq1,jq2);
1373     qq20             = _mm_mul_pd(iq2,jq0);
1374     qq21             = _mm_mul_pd(iq2,jq1);
1375     qq22             = _mm_mul_pd(iq2,jq2);
1376
1377     /* Avoid stupid compiler warnings */
1378     jnrA = jnrB = 0;
1379     j_coord_offsetA = 0;
1380     j_coord_offsetB = 0;
1381
1382     outeriter        = 0;
1383     inneriter        = 0;
1384
1385     /* Start outer loop over neighborlists */
1386     for(iidx=0; iidx<nri; iidx++)
1387     {
1388         /* Load shift vector for this list */
1389         i_shift_offset   = DIM*shiftidx[iidx];
1390
1391         /* Load limits for loop over neighbors */
1392         j_index_start    = jindex[iidx];
1393         j_index_end      = jindex[iidx+1];
1394
1395         /* Get outer coordinate index */
1396         inr              = iinr[iidx];
1397         i_coord_offset   = DIM*inr;
1398
1399         /* Load i particle coords and add shift vector */
1400         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1401                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1402
1403         fix0             = _mm_setzero_pd();
1404         fiy0             = _mm_setzero_pd();
1405         fiz0             = _mm_setzero_pd();
1406         fix1             = _mm_setzero_pd();
1407         fiy1             = _mm_setzero_pd();
1408         fiz1             = _mm_setzero_pd();
1409         fix2             = _mm_setzero_pd();
1410         fiy2             = _mm_setzero_pd();
1411         fiz2             = _mm_setzero_pd();
1412
1413         /* Start inner kernel loop */
1414         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1415         {
1416
1417             /* Get j neighbor index, and coordinate index */
1418             jnrA             = jjnr[jidx];
1419             jnrB             = jjnr[jidx+1];
1420             j_coord_offsetA  = DIM*jnrA;
1421             j_coord_offsetB  = DIM*jnrB;
1422
1423             /* load j atom coordinates */
1424             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1425                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1426
1427             /* Calculate displacement vector */
1428             dx00             = _mm_sub_pd(ix0,jx0);
1429             dy00             = _mm_sub_pd(iy0,jy0);
1430             dz00             = _mm_sub_pd(iz0,jz0);
1431             dx01             = _mm_sub_pd(ix0,jx1);
1432             dy01             = _mm_sub_pd(iy0,jy1);
1433             dz01             = _mm_sub_pd(iz0,jz1);
1434             dx02             = _mm_sub_pd(ix0,jx2);
1435             dy02             = _mm_sub_pd(iy0,jy2);
1436             dz02             = _mm_sub_pd(iz0,jz2);
1437             dx10             = _mm_sub_pd(ix1,jx0);
1438             dy10             = _mm_sub_pd(iy1,jy0);
1439             dz10             = _mm_sub_pd(iz1,jz0);
1440             dx11             = _mm_sub_pd(ix1,jx1);
1441             dy11             = _mm_sub_pd(iy1,jy1);
1442             dz11             = _mm_sub_pd(iz1,jz1);
1443             dx12             = _mm_sub_pd(ix1,jx2);
1444             dy12             = _mm_sub_pd(iy1,jy2);
1445             dz12             = _mm_sub_pd(iz1,jz2);
1446             dx20             = _mm_sub_pd(ix2,jx0);
1447             dy20             = _mm_sub_pd(iy2,jy0);
1448             dz20             = _mm_sub_pd(iz2,jz0);
1449             dx21             = _mm_sub_pd(ix2,jx1);
1450             dy21             = _mm_sub_pd(iy2,jy1);
1451             dz21             = _mm_sub_pd(iz2,jz1);
1452             dx22             = _mm_sub_pd(ix2,jx2);
1453             dy22             = _mm_sub_pd(iy2,jy2);
1454             dz22             = _mm_sub_pd(iz2,jz2);
1455
1456             /* Calculate squared distance and things based on it */
1457             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1458             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1459             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1460             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1461             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1462             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1463             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1464             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1465             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1466
1467             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1468             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1469             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1470             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1471             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1472             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1473             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1474             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1475             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1476
1477             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1478             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1479             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1480             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1481             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1482             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1483             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1484             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1485             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1486
1487             fjx0             = _mm_setzero_pd();
1488             fjy0             = _mm_setzero_pd();
1489             fjz0             = _mm_setzero_pd();
1490             fjx1             = _mm_setzero_pd();
1491             fjy1             = _mm_setzero_pd();
1492             fjz1             = _mm_setzero_pd();
1493             fjx2             = _mm_setzero_pd();
1494             fjy2             = _mm_setzero_pd();
1495             fjz2             = _mm_setzero_pd();
1496
1497             /**************************
1498              * CALCULATE INTERACTIONS *
1499              **************************/
1500
1501             r00              = _mm_mul_pd(rsq00,rinv00);
1502
1503             /* EWALD ELECTROSTATICS */
1504
1505             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1506             ewrt             = _mm_mul_pd(r00,ewtabscale);
1507             ewitab           = _mm_cvttpd_epi32(ewrt);
1508 #ifdef __XOP__
1509             eweps            = _mm_frcz_pd(ewrt);
1510 #else
1511             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1512 #endif
1513             twoeweps         = _mm_add_pd(eweps,eweps);
1514             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1515                                          &ewtabF,&ewtabFn);
1516             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1517             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1518
1519             /* Analytical LJ-PME */
1520             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1521             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1522             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1523             exponent         = gmx_simd_exp_d(ewcljrsq);
1524             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1525             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1526             /* f6A = 6 * C6grid * (1 - poly) */
1527             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1528             /* f6B = C6grid * exponent * beta^6 */
1529             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1530             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1531             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1532
1533             fscal            = _mm_add_pd(felec,fvdw);
1534
1535             /* Update vectorial force */
1536             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1537             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1538             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1539
1540             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1541             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1542             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1543
1544             /**************************
1545              * CALCULATE INTERACTIONS *
1546              **************************/
1547
1548             r01              = _mm_mul_pd(rsq01,rinv01);
1549
1550             /* EWALD ELECTROSTATICS */
1551
1552             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1553             ewrt             = _mm_mul_pd(r01,ewtabscale);
1554             ewitab           = _mm_cvttpd_epi32(ewrt);
1555 #ifdef __XOP__
1556             eweps            = _mm_frcz_pd(ewrt);
1557 #else
1558             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1559 #endif
1560             twoeweps         = _mm_add_pd(eweps,eweps);
1561             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1562                                          &ewtabF,&ewtabFn);
1563             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1564             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1565
1566             fscal            = felec;
1567
1568             /* Update vectorial force */
1569             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1570             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1571             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1572
1573             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1574             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1575             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1576
1577             /**************************
1578              * CALCULATE INTERACTIONS *
1579              **************************/
1580
1581             r02              = _mm_mul_pd(rsq02,rinv02);
1582
1583             /* EWALD ELECTROSTATICS */
1584
1585             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1586             ewrt             = _mm_mul_pd(r02,ewtabscale);
1587             ewitab           = _mm_cvttpd_epi32(ewrt);
1588 #ifdef __XOP__
1589             eweps            = _mm_frcz_pd(ewrt);
1590 #else
1591             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1592 #endif
1593             twoeweps         = _mm_add_pd(eweps,eweps);
1594             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1595                                          &ewtabF,&ewtabFn);
1596             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1597             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1598
1599             fscal            = felec;
1600
1601             /* Update vectorial force */
1602             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1603             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1604             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1605
1606             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1607             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1608             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1609
1610             /**************************
1611              * CALCULATE INTERACTIONS *
1612              **************************/
1613
1614             r10              = _mm_mul_pd(rsq10,rinv10);
1615
1616             /* EWALD ELECTROSTATICS */
1617
1618             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1619             ewrt             = _mm_mul_pd(r10,ewtabscale);
1620             ewitab           = _mm_cvttpd_epi32(ewrt);
1621 #ifdef __XOP__
1622             eweps            = _mm_frcz_pd(ewrt);
1623 #else
1624             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1625 #endif
1626             twoeweps         = _mm_add_pd(eweps,eweps);
1627             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1628                                          &ewtabF,&ewtabFn);
1629             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1630             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1631
1632             fscal            = felec;
1633
1634             /* Update vectorial force */
1635             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1636             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1637             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1638
1639             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1640             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1641             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1642
1643             /**************************
1644              * CALCULATE INTERACTIONS *
1645              **************************/
1646
1647             r11              = _mm_mul_pd(rsq11,rinv11);
1648
1649             /* EWALD ELECTROSTATICS */
1650
1651             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1652             ewrt             = _mm_mul_pd(r11,ewtabscale);
1653             ewitab           = _mm_cvttpd_epi32(ewrt);
1654 #ifdef __XOP__
1655             eweps            = _mm_frcz_pd(ewrt);
1656 #else
1657             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1658 #endif
1659             twoeweps         = _mm_add_pd(eweps,eweps);
1660             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1661                                          &ewtabF,&ewtabFn);
1662             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1663             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1664
1665             fscal            = felec;
1666
1667             /* Update vectorial force */
1668             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1669             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1670             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1671
1672             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1673             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1674             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1675
1676             /**************************
1677              * CALCULATE INTERACTIONS *
1678              **************************/
1679
1680             r12              = _mm_mul_pd(rsq12,rinv12);
1681
1682             /* EWALD ELECTROSTATICS */
1683
1684             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1685             ewrt             = _mm_mul_pd(r12,ewtabscale);
1686             ewitab           = _mm_cvttpd_epi32(ewrt);
1687 #ifdef __XOP__
1688             eweps            = _mm_frcz_pd(ewrt);
1689 #else
1690             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1691 #endif
1692             twoeweps         = _mm_add_pd(eweps,eweps);
1693             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1694                                          &ewtabF,&ewtabFn);
1695             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1696             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1697
1698             fscal            = felec;
1699
1700             /* Update vectorial force */
1701             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1702             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1703             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1704
1705             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1706             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1707             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1708
1709             /**************************
1710              * CALCULATE INTERACTIONS *
1711              **************************/
1712
1713             r20              = _mm_mul_pd(rsq20,rinv20);
1714
1715             /* EWALD ELECTROSTATICS */
1716
1717             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1718             ewrt             = _mm_mul_pd(r20,ewtabscale);
1719             ewitab           = _mm_cvttpd_epi32(ewrt);
1720 #ifdef __XOP__
1721             eweps            = _mm_frcz_pd(ewrt);
1722 #else
1723             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1724 #endif
1725             twoeweps         = _mm_add_pd(eweps,eweps);
1726             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1727                                          &ewtabF,&ewtabFn);
1728             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1729             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1730
1731             fscal            = felec;
1732
1733             /* Update vectorial force */
1734             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1735             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1736             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1737
1738             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1739             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1740             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1741
1742             /**************************
1743              * CALCULATE INTERACTIONS *
1744              **************************/
1745
1746             r21              = _mm_mul_pd(rsq21,rinv21);
1747
1748             /* EWALD ELECTROSTATICS */
1749
1750             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1751             ewrt             = _mm_mul_pd(r21,ewtabscale);
1752             ewitab           = _mm_cvttpd_epi32(ewrt);
1753 #ifdef __XOP__
1754             eweps            = _mm_frcz_pd(ewrt);
1755 #else
1756             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1757 #endif
1758             twoeweps         = _mm_add_pd(eweps,eweps);
1759             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1760                                          &ewtabF,&ewtabFn);
1761             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1762             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1763
1764             fscal            = felec;
1765
1766             /* Update vectorial force */
1767             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1768             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1769             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1770
1771             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1772             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1773             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1774
1775             /**************************
1776              * CALCULATE INTERACTIONS *
1777              **************************/
1778
1779             r22              = _mm_mul_pd(rsq22,rinv22);
1780
1781             /* EWALD ELECTROSTATICS */
1782
1783             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1784             ewrt             = _mm_mul_pd(r22,ewtabscale);
1785             ewitab           = _mm_cvttpd_epi32(ewrt);
1786 #ifdef __XOP__
1787             eweps            = _mm_frcz_pd(ewrt);
1788 #else
1789             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1790 #endif
1791             twoeweps         = _mm_add_pd(eweps,eweps);
1792             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1793                                          &ewtabF,&ewtabFn);
1794             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1795             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1796
1797             fscal            = felec;
1798
1799             /* Update vectorial force */
1800             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1801             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1802             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1803
1804             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1805             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1806             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1807
1808             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1809
1810             /* Inner loop uses 372 flops */
1811         }
1812
1813         if(jidx<j_index_end)
1814         {
1815
1816             jnrA             = jjnr[jidx];
1817             j_coord_offsetA  = DIM*jnrA;
1818
1819             /* load j atom coordinates */
1820             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1821                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1822
1823             /* Calculate displacement vector */
1824             dx00             = _mm_sub_pd(ix0,jx0);
1825             dy00             = _mm_sub_pd(iy0,jy0);
1826             dz00             = _mm_sub_pd(iz0,jz0);
1827             dx01             = _mm_sub_pd(ix0,jx1);
1828             dy01             = _mm_sub_pd(iy0,jy1);
1829             dz01             = _mm_sub_pd(iz0,jz1);
1830             dx02             = _mm_sub_pd(ix0,jx2);
1831             dy02             = _mm_sub_pd(iy0,jy2);
1832             dz02             = _mm_sub_pd(iz0,jz2);
1833             dx10             = _mm_sub_pd(ix1,jx0);
1834             dy10             = _mm_sub_pd(iy1,jy0);
1835             dz10             = _mm_sub_pd(iz1,jz0);
1836             dx11             = _mm_sub_pd(ix1,jx1);
1837             dy11             = _mm_sub_pd(iy1,jy1);
1838             dz11             = _mm_sub_pd(iz1,jz1);
1839             dx12             = _mm_sub_pd(ix1,jx2);
1840             dy12             = _mm_sub_pd(iy1,jy2);
1841             dz12             = _mm_sub_pd(iz1,jz2);
1842             dx20             = _mm_sub_pd(ix2,jx0);
1843             dy20             = _mm_sub_pd(iy2,jy0);
1844             dz20             = _mm_sub_pd(iz2,jz0);
1845             dx21             = _mm_sub_pd(ix2,jx1);
1846             dy21             = _mm_sub_pd(iy2,jy1);
1847             dz21             = _mm_sub_pd(iz2,jz1);
1848             dx22             = _mm_sub_pd(ix2,jx2);
1849             dy22             = _mm_sub_pd(iy2,jy2);
1850             dz22             = _mm_sub_pd(iz2,jz2);
1851
1852             /* Calculate squared distance and things based on it */
1853             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1854             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1855             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1856             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1857             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1858             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1859             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1860             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1861             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1862
1863             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1864             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1865             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1866             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1867             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1868             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1869             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1870             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1871             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1872
1873             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1874             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1875             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1876             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1877             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1878             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1879             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1880             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1881             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1882
1883             fjx0             = _mm_setzero_pd();
1884             fjy0             = _mm_setzero_pd();
1885             fjz0             = _mm_setzero_pd();
1886             fjx1             = _mm_setzero_pd();
1887             fjy1             = _mm_setzero_pd();
1888             fjz1             = _mm_setzero_pd();
1889             fjx2             = _mm_setzero_pd();
1890             fjy2             = _mm_setzero_pd();
1891             fjz2             = _mm_setzero_pd();
1892
1893             /**************************
1894              * CALCULATE INTERACTIONS *
1895              **************************/
1896
1897             r00              = _mm_mul_pd(rsq00,rinv00);
1898
1899             /* EWALD ELECTROSTATICS */
1900
1901             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1902             ewrt             = _mm_mul_pd(r00,ewtabscale);
1903             ewitab           = _mm_cvttpd_epi32(ewrt);
1904 #ifdef __XOP__
1905             eweps            = _mm_frcz_pd(ewrt);
1906 #else
1907             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1908 #endif
1909             twoeweps         = _mm_add_pd(eweps,eweps);
1910             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1911             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1912             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1913
1914             /* Analytical LJ-PME */
1915             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1916             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1917             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1918             exponent         = gmx_simd_exp_d(ewcljrsq);
1919             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1920             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1921             /* f6A = 6 * C6grid * (1 - poly) */
1922             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1923             /* f6B = C6grid * exponent * beta^6 */
1924             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1925             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1926             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1927
1928             fscal            = _mm_add_pd(felec,fvdw);
1929
1930             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1931
1932             /* Update vectorial force */
1933             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1934             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1935             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1936
1937             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1938             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1939             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1940
1941             /**************************
1942              * CALCULATE INTERACTIONS *
1943              **************************/
1944
1945             r01              = _mm_mul_pd(rsq01,rinv01);
1946
1947             /* EWALD ELECTROSTATICS */
1948
1949             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1950             ewrt             = _mm_mul_pd(r01,ewtabscale);
1951             ewitab           = _mm_cvttpd_epi32(ewrt);
1952 #ifdef __XOP__
1953             eweps            = _mm_frcz_pd(ewrt);
1954 #else
1955             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1956 #endif
1957             twoeweps         = _mm_add_pd(eweps,eweps);
1958             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1959             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1960             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1961
1962             fscal            = felec;
1963
1964             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1965
1966             /* Update vectorial force */
1967             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1968             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1969             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1970
1971             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1972             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1973             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1974
1975             /**************************
1976              * CALCULATE INTERACTIONS *
1977              **************************/
1978
1979             r02              = _mm_mul_pd(rsq02,rinv02);
1980
1981             /* EWALD ELECTROSTATICS */
1982
1983             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1984             ewrt             = _mm_mul_pd(r02,ewtabscale);
1985             ewitab           = _mm_cvttpd_epi32(ewrt);
1986 #ifdef __XOP__
1987             eweps            = _mm_frcz_pd(ewrt);
1988 #else
1989             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1990 #endif
1991             twoeweps         = _mm_add_pd(eweps,eweps);
1992             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1993             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1994             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1995
1996             fscal            = felec;
1997
1998             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1999
2000             /* Update vectorial force */
2001             fix0             = _mm_macc_pd(dx02,fscal,fix0);
2002             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
2003             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
2004
2005             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
2006             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
2007             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
2008
2009             /**************************
2010              * CALCULATE INTERACTIONS *
2011              **************************/
2012
2013             r10              = _mm_mul_pd(rsq10,rinv10);
2014
2015             /* EWALD ELECTROSTATICS */
2016
2017             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2018             ewrt             = _mm_mul_pd(r10,ewtabscale);
2019             ewitab           = _mm_cvttpd_epi32(ewrt);
2020 #ifdef __XOP__
2021             eweps            = _mm_frcz_pd(ewrt);
2022 #else
2023             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2024 #endif
2025             twoeweps         = _mm_add_pd(eweps,eweps);
2026             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2027             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2028             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2029
2030             fscal            = felec;
2031
2032             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2033
2034             /* Update vectorial force */
2035             fix1             = _mm_macc_pd(dx10,fscal,fix1);
2036             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
2037             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
2038
2039             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
2040             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
2041             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
2042
2043             /**************************
2044              * CALCULATE INTERACTIONS *
2045              **************************/
2046
2047             r11              = _mm_mul_pd(rsq11,rinv11);
2048
2049             /* EWALD ELECTROSTATICS */
2050
2051             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2052             ewrt             = _mm_mul_pd(r11,ewtabscale);
2053             ewitab           = _mm_cvttpd_epi32(ewrt);
2054 #ifdef __XOP__
2055             eweps            = _mm_frcz_pd(ewrt);
2056 #else
2057             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2058 #endif
2059             twoeweps         = _mm_add_pd(eweps,eweps);
2060             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2061             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2062             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2063
2064             fscal            = felec;
2065
2066             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2067
2068             /* Update vectorial force */
2069             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2070             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2071             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2072
2073             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2074             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2075             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2076
2077             /**************************
2078              * CALCULATE INTERACTIONS *
2079              **************************/
2080
2081             r12              = _mm_mul_pd(rsq12,rinv12);
2082
2083             /* EWALD ELECTROSTATICS */
2084
2085             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2086             ewrt             = _mm_mul_pd(r12,ewtabscale);
2087             ewitab           = _mm_cvttpd_epi32(ewrt);
2088 #ifdef __XOP__
2089             eweps            = _mm_frcz_pd(ewrt);
2090 #else
2091             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2092 #endif
2093             twoeweps         = _mm_add_pd(eweps,eweps);
2094             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2095             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2096             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2097
2098             fscal            = felec;
2099
2100             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2101
2102             /* Update vectorial force */
2103             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2104             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2105             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2106
2107             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2108             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2109             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2110
2111             /**************************
2112              * CALCULATE INTERACTIONS *
2113              **************************/
2114
2115             r20              = _mm_mul_pd(rsq20,rinv20);
2116
2117             /* EWALD ELECTROSTATICS */
2118
2119             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2120             ewrt             = _mm_mul_pd(r20,ewtabscale);
2121             ewitab           = _mm_cvttpd_epi32(ewrt);
2122 #ifdef __XOP__
2123             eweps            = _mm_frcz_pd(ewrt);
2124 #else
2125             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2126 #endif
2127             twoeweps         = _mm_add_pd(eweps,eweps);
2128             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2129             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2130             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2131
2132             fscal            = felec;
2133
2134             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2135
2136             /* Update vectorial force */
2137             fix2             = _mm_macc_pd(dx20,fscal,fix2);
2138             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
2139             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
2140
2141             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2142             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2143             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
2144
2145             /**************************
2146              * CALCULATE INTERACTIONS *
2147              **************************/
2148
2149             r21              = _mm_mul_pd(rsq21,rinv21);
2150
2151             /* EWALD ELECTROSTATICS */
2152
2153             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2154             ewrt             = _mm_mul_pd(r21,ewtabscale);
2155             ewitab           = _mm_cvttpd_epi32(ewrt);
2156 #ifdef __XOP__
2157             eweps            = _mm_frcz_pd(ewrt);
2158 #else
2159             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2160 #endif
2161             twoeweps         = _mm_add_pd(eweps,eweps);
2162             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2163             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2164             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2165
2166             fscal            = felec;
2167
2168             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2169
2170             /* Update vectorial force */
2171             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2172             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2173             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2174
2175             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2176             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2177             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2178
2179             /**************************
2180              * CALCULATE INTERACTIONS *
2181              **************************/
2182
2183             r22              = _mm_mul_pd(rsq22,rinv22);
2184
2185             /* EWALD ELECTROSTATICS */
2186
2187             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2188             ewrt             = _mm_mul_pd(r22,ewtabscale);
2189             ewitab           = _mm_cvttpd_epi32(ewrt);
2190 #ifdef __XOP__
2191             eweps            = _mm_frcz_pd(ewrt);
2192 #else
2193             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2194 #endif
2195             twoeweps         = _mm_add_pd(eweps,eweps);
2196             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2197             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2198             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2199
2200             fscal            = felec;
2201
2202             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2203
2204             /* Update vectorial force */
2205             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2206             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2207             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2208
2209             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2210             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2211             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2212
2213             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2214
2215             /* Inner loop uses 372 flops */
2216         }
2217
2218         /* End of innermost loop */
2219
2220         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2221                                               f+i_coord_offset,fshift+i_shift_offset);
2222
2223         /* Increment number of inner iterations */
2224         inneriter                  += j_index_end - j_index_start;
2225
2226         /* Outer loop uses 18 flops */
2227     }
2228
2229     /* Increment number of outer iterations */
2230     outeriter        += nri;
2231
2232     /* Update outer/inner flops */
2233
2234     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*372);
2235 }