src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwLJ_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_VF_avx_128_fma_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     int              vdwjidx1A,vdwjidx1B;
  91     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  92     int              vdwjidx2A,vdwjidx2B;
  93     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  94     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  96     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  97     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  99     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 100     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 102     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 103     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 104     real             *charge;
 105     int              nvdwtype;
 106     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 107     int              *vdwtype;
 108     real             *vdwparam;
 109     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 110     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 111     __m128i          vfitab;
 112     __m128i          ifour       = _mm_set1_epi32(4);
 113     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 114     real             *vftab;
 115     __m128d          dummy_mask,cutoff_mask;
 116     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 117     __m128d          one     = _mm_set1_pd(1.0);
 118     __m128d          two     = _mm_set1_pd(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm_set1_pd(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135
 136     vftab            = kernel_data->table_elec->data;
 137     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 142     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 143     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 144     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 145
 146     jq0              = _mm_set1_pd(charge[inr+0]);
 147     jq1              = _mm_set1_pd(charge[inr+1]);
 148     jq2              = _mm_set1_pd(charge[inr+2]);
 149     vdwjidx0A        = 2*vdwtype[inr+0];
 150     qq00             = _mm_mul_pd(iq0,jq0);
 151     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 152     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 153     qq01             = _mm_mul_pd(iq0,jq1);
 154     qq02             = _mm_mul_pd(iq0,jq2);
 155     qq10             = _mm_mul_pd(iq1,jq0);
 156     qq11             = _mm_mul_pd(iq1,jq1);
 157     qq12             = _mm_mul_pd(iq1,jq2);
 158     qq20             = _mm_mul_pd(iq2,jq0);
 159     qq21             = _mm_mul_pd(iq2,jq1);
 160     qq22             = _mm_mul_pd(iq2,jq2);
 161
 162     /* Avoid stupid compiler warnings */
 163     jnrA = jnrB = 0;
 164     j_coord_offsetA = 0;
 165     j_coord_offsetB = 0;
 166
 167     outeriter        = 0;
 168     inneriter        = 0;
 169
 170     /* Start outer loop over neighborlists */
 171     for(iidx=0; iidx<nri; iidx++)
 172     {
 173         /* Load shift vector for this list */
 174         i_shift_offset   = DIM*shiftidx[iidx];
 175
 176         /* Load limits for loop over neighbors */
 177         j_index_start    = jindex[iidx];
 178         j_index_end      = jindex[iidx+1];
 179
 180         /* Get outer coordinate index */
 181         inr              = iinr[iidx];
 182         i_coord_offset   = DIM*inr;
 183
 184         /* Load i particle coords and add shift vector */
 185         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 186                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 187
 188         fix0             = _mm_setzero_pd();
 189         fiy0             = _mm_setzero_pd();
 190         fiz0             = _mm_setzero_pd();
 191         fix1             = _mm_setzero_pd();
 192         fiy1             = _mm_setzero_pd();
 193         fiz1             = _mm_setzero_pd();
 194         fix2             = _mm_setzero_pd();
 195         fiy2             = _mm_setzero_pd();
 196         fiz2             = _mm_setzero_pd();
 197
 198         /* Reset potential sums */
 199         velecsum         = _mm_setzero_pd();
 200         vvdwsum          = _mm_setzero_pd();
 201
 202         /* Start inner kernel loop */
 203         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 204         {
 205
 206             /* Get j neighbor index, and coordinate index */
 207             jnrA             = jjnr[jidx];
 208             jnrB             = jjnr[jidx+1];
 209             j_coord_offsetA  = DIM*jnrA;
 210             j_coord_offsetB  = DIM*jnrB;
 211
 212             /* load j atom coordinates */
 213             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 215
 216             /* Calculate displacement vector */
 217             dx00             = _mm_sub_pd(ix0,jx0);
 218             dy00             = _mm_sub_pd(iy0,jy0);
 219             dz00             = _mm_sub_pd(iz0,jz0);
 220             dx01             = _mm_sub_pd(ix0,jx1);
 221             dy01             = _mm_sub_pd(iy0,jy1);
 222             dz01             = _mm_sub_pd(iz0,jz1);
 223             dx02             = _mm_sub_pd(ix0,jx2);
 224             dy02             = _mm_sub_pd(iy0,jy2);
 225             dz02             = _mm_sub_pd(iz0,jz2);
 226             dx10             = _mm_sub_pd(ix1,jx0);
 227             dy10             = _mm_sub_pd(iy1,jy0);
 228             dz10             = _mm_sub_pd(iz1,jz0);
 229             dx11             = _mm_sub_pd(ix1,jx1);
 230             dy11             = _mm_sub_pd(iy1,jy1);
 231             dz11             = _mm_sub_pd(iz1,jz1);
 232             dx12             = _mm_sub_pd(ix1,jx2);
 233             dy12             = _mm_sub_pd(iy1,jy2);
 234             dz12             = _mm_sub_pd(iz1,jz2);
 235             dx20             = _mm_sub_pd(ix2,jx0);
 236             dy20             = _mm_sub_pd(iy2,jy0);
 237             dz20             = _mm_sub_pd(iz2,jz0);
 238             dx21             = _mm_sub_pd(ix2,jx1);
 239             dy21             = _mm_sub_pd(iy2,jy1);
 240             dz21             = _mm_sub_pd(iz2,jz1);
 241             dx22             = _mm_sub_pd(ix2,jx2);
 242             dy22             = _mm_sub_pd(iy2,jy2);
 243             dz22             = _mm_sub_pd(iz2,jz2);
 244
 245             /* Calculate squared distance and things based on it */
 246             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 247             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 248             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 249             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 250             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 251             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 252             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 253             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 254             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 255
 256             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 257             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 258             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 259             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 260             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 261             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 262             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 263             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 264             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 265
 266             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 267
 268             fjx0             = _mm_setzero_pd();
 269             fjy0             = _mm_setzero_pd();
 270             fjz0             = _mm_setzero_pd();
 271             fjx1             = _mm_setzero_pd();
 272             fjy1             = _mm_setzero_pd();
 273             fjz1             = _mm_setzero_pd();
 274             fjx2             = _mm_setzero_pd();
 275             fjy2             = _mm_setzero_pd();
 276             fjz2             = _mm_setzero_pd();
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             r00              = _mm_mul_pd(rsq00,rinv00);
 283
 284             /* Calculate table index by multiplying r with table scale and truncate to integer */
 285             rt               = _mm_mul_pd(r00,vftabscale);
 286             vfitab           = _mm_cvttpd_epi32(rt);
 287 #ifdef __XOP__
 288             vfeps            = _mm_frcz_pd(rt);
 289 #else
 290             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 291 #endif
 292             twovfeps         = _mm_add_pd(vfeps,vfeps);
 293             vfitab           = _mm_slli_epi32(vfitab,2);
 294
 295             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 296             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 297             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 298             GMX_MM_TRANSPOSE2_PD(Y,F);
 299             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 300             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 301             GMX_MM_TRANSPOSE2_PD(G,H);
 302             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 303             VV               = _mm_macc_pd(vfeps,Fp,Y);
 304             velec            = _mm_mul_pd(qq00,VV);
 305             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 306             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 307
 308             /* LENNARD-JONES DISPERSION/REPULSION */
 309
 310             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 311             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 312             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 313             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 314             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velecsum         = _mm_add_pd(velecsum,velec);
 318             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 319
 320             fscal            = _mm_add_pd(felec,fvdw);
 321
 322             /* Update vectorial force */
 323             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 324             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 325             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 326
 327             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 328             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 329             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             r01              = _mm_mul_pd(rsq01,rinv01);
 336
 337             /* Calculate table index by multiplying r with table scale and truncate to integer */
 338             rt               = _mm_mul_pd(r01,vftabscale);
 339             vfitab           = _mm_cvttpd_epi32(rt);
 340 #ifdef __XOP__
 341             vfeps            = _mm_frcz_pd(rt);
 342 #else
 343             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 344 #endif
 345             twovfeps         = _mm_add_pd(vfeps,vfeps);
 346             vfitab           = _mm_slli_epi32(vfitab,2);
 347
 348             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 349             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 350             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 351             GMX_MM_TRANSPOSE2_PD(Y,F);
 352             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 353             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 354             GMX_MM_TRANSPOSE2_PD(G,H);
 355             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 356             VV               = _mm_macc_pd(vfeps,Fp,Y);
 357             velec            = _mm_mul_pd(qq01,VV);
 358             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 359             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq01,FF),_mm_mul_pd(vftabscale,rinv01)));
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velecsum         = _mm_add_pd(velecsum,velec);
 363
 364             fscal            = felec;
 365
 366             /* Update vectorial force */
 367             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 368             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 369             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 370
 371             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 372             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 373             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             r02              = _mm_mul_pd(rsq02,rinv02);
 380
 381             /* Calculate table index by multiplying r with table scale and truncate to integer */
 382             rt               = _mm_mul_pd(r02,vftabscale);
 383             vfitab           = _mm_cvttpd_epi32(rt);
 384 #ifdef __XOP__
 385             vfeps            = _mm_frcz_pd(rt);
 386 #else
 387             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 388 #endif
 389             twovfeps         = _mm_add_pd(vfeps,vfeps);
 390             vfitab           = _mm_slli_epi32(vfitab,2);
 391
 392             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 393             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 394             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 395             GMX_MM_TRANSPOSE2_PD(Y,F);
 396             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 397             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 398             GMX_MM_TRANSPOSE2_PD(G,H);
 399             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 400             VV               = _mm_macc_pd(vfeps,Fp,Y);
 401             velec            = _mm_mul_pd(qq02,VV);
 402             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 403             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq02,FF),_mm_mul_pd(vftabscale,rinv02)));
 404
 405             /* Update potential sum for this i atom from the interaction with this j atom. */
 406             velecsum         = _mm_add_pd(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410             /* Update vectorial force */
 411             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 412             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 413             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 414
 415             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 416             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 417             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             r10              = _mm_mul_pd(rsq10,rinv10);
 424
 425             /* Calculate table index by multiplying r with table scale and truncate to integer */
 426             rt               = _mm_mul_pd(r10,vftabscale);
 427             vfitab           = _mm_cvttpd_epi32(rt);
 428 #ifdef __XOP__
 429             vfeps            = _mm_frcz_pd(rt);
 430 #else
 431             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 432 #endif
 433             twovfeps         = _mm_add_pd(vfeps,vfeps);
 434             vfitab           = _mm_slli_epi32(vfitab,2);
 435
 436             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 437             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 438             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 439             GMX_MM_TRANSPOSE2_PD(Y,F);
 440             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 441             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 442             GMX_MM_TRANSPOSE2_PD(G,H);
 443             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 444             VV               = _mm_macc_pd(vfeps,Fp,Y);
 445             velec            = _mm_mul_pd(qq10,VV);
 446             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 447             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 448
 449             /* Update potential sum for this i atom from the interaction with this j atom. */
 450             velecsum         = _mm_add_pd(velecsum,velec);
 451
 452             fscal            = felec;
 453
 454             /* Update vectorial force */
 455             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 456             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 457             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 458
 459             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 460             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 461             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             r11              = _mm_mul_pd(rsq11,rinv11);
 468
 469             /* Calculate table index by multiplying r with table scale and truncate to integer */
 470             rt               = _mm_mul_pd(r11,vftabscale);
 471             vfitab           = _mm_cvttpd_epi32(rt);
 472 #ifdef __XOP__
 473             vfeps            = _mm_frcz_pd(rt);
 474 #else
 475             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 476 #endif
 477             twovfeps         = _mm_add_pd(vfeps,vfeps);
 478             vfitab           = _mm_slli_epi32(vfitab,2);
 479
 480             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 481             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 482             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 483             GMX_MM_TRANSPOSE2_PD(Y,F);
 484             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 485             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 486             GMX_MM_TRANSPOSE2_PD(G,H);
 487             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 488             VV               = _mm_macc_pd(vfeps,Fp,Y);
 489             velec            = _mm_mul_pd(qq11,VV);
 490             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 491             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
 492
 493             /* Update potential sum for this i atom from the interaction with this j atom. */
 494             velecsum         = _mm_add_pd(velecsum,velec);
 495
 496             fscal            = felec;
 497
 498             /* Update vectorial force */
 499             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 500             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 501             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 502
 503             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 504             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 505             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             r12              = _mm_mul_pd(rsq12,rinv12);
 512
 513             /* Calculate table index by multiplying r with table scale and truncate to integer */
 514             rt               = _mm_mul_pd(r12,vftabscale);
 515             vfitab           = _mm_cvttpd_epi32(rt);
 516 #ifdef __XOP__
 517             vfeps            = _mm_frcz_pd(rt);
 518 #else
 519             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 520 #endif
 521             twovfeps         = _mm_add_pd(vfeps,vfeps);
 522             vfitab           = _mm_slli_epi32(vfitab,2);
 523
 524             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 525             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 526             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 527             GMX_MM_TRANSPOSE2_PD(Y,F);
 528             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 529             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 530             GMX_MM_TRANSPOSE2_PD(G,H);
 531             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 532             VV               = _mm_macc_pd(vfeps,Fp,Y);
 533             velec            = _mm_mul_pd(qq12,VV);
 534             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 535             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velecsum         = _mm_add_pd(velecsum,velec);
 539
 540             fscal            = felec;
 541
 542             /* Update vectorial force */
 543             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 544             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 545             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 546
 547             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 548             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 549             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             r20              = _mm_mul_pd(rsq20,rinv20);
 556
 557             /* Calculate table index by multiplying r with table scale and truncate to integer */
 558             rt               = _mm_mul_pd(r20,vftabscale);
 559             vfitab           = _mm_cvttpd_epi32(rt);
 560 #ifdef __XOP__
 561             vfeps            = _mm_frcz_pd(rt);
 562 #else
 563             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 564 #endif
 565             twovfeps         = _mm_add_pd(vfeps,vfeps);
 566             vfitab           = _mm_slli_epi32(vfitab,2);
 567
 568             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 569             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 570             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 571             GMX_MM_TRANSPOSE2_PD(Y,F);
 572             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 573             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 574             GMX_MM_TRANSPOSE2_PD(G,H);
 575             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 576             VV               = _mm_macc_pd(vfeps,Fp,Y);
 577             velec            = _mm_mul_pd(qq20,VV);
 578             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 579             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velecsum         = _mm_add_pd(velecsum,velec);
 583
 584             fscal            = felec;
 585
 586             /* Update vectorial force */
 587             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 588             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 589             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 590
 591             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 592             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 593             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 594
 595             /**************************
 596              * CALCULATE INTERACTIONS *
 597              **************************/
 598
 599             r21              = _mm_mul_pd(rsq21,rinv21);
 600
 601             /* Calculate table index by multiplying r with table scale and truncate to integer */
 602             rt               = _mm_mul_pd(r21,vftabscale);
 603             vfitab           = _mm_cvttpd_epi32(rt);
 604 #ifdef __XOP__
 605             vfeps            = _mm_frcz_pd(rt);
 606 #else
 607             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 608 #endif
 609             twovfeps         = _mm_add_pd(vfeps,vfeps);
 610             vfitab           = _mm_slli_epi32(vfitab,2);
 611
 612             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 613             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 614             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 615             GMX_MM_TRANSPOSE2_PD(Y,F);
 616             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 617             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 618             GMX_MM_TRANSPOSE2_PD(G,H);
 619             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 620             VV               = _mm_macc_pd(vfeps,Fp,Y);
 621             velec            = _mm_mul_pd(qq21,VV);
 622             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 623             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
 624
 625             /* Update potential sum for this i atom from the interaction with this j atom. */
 626             velecsum         = _mm_add_pd(velecsum,velec);
 627
 628             fscal            = felec;
 629
 630             /* Update vectorial force */
 631             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 632             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 633             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 634
 635             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 636             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 637             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 638
 639             /**************************
 640              * CALCULATE INTERACTIONS *
 641              **************************/
 642
 643             r22              = _mm_mul_pd(rsq22,rinv22);
 644
 645             /* Calculate table index by multiplying r with table scale and truncate to integer */
 646             rt               = _mm_mul_pd(r22,vftabscale);
 647             vfitab           = _mm_cvttpd_epi32(rt);
 648 #ifdef __XOP__
 649             vfeps            = _mm_frcz_pd(rt);
 650 #else
 651             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 652 #endif
 653             twovfeps         = _mm_add_pd(vfeps,vfeps);
 654             vfitab           = _mm_slli_epi32(vfitab,2);
 655
 656             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 657             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 658             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 659             GMX_MM_TRANSPOSE2_PD(Y,F);
 660             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 661             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 662             GMX_MM_TRANSPOSE2_PD(G,H);
 663             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 664             VV               = _mm_macc_pd(vfeps,Fp,Y);
 665             velec            = _mm_mul_pd(qq22,VV);
 666             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 667             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
 668
 669             /* Update potential sum for this i atom from the interaction with this j atom. */
 670             velecsum         = _mm_add_pd(velecsum,velec);
 671
 672             fscal            = felec;
 673
 674             /* Update vectorial force */
 675             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 676             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 677             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 678
 679             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 680             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 681             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 682
 683             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 684
 685             /* Inner loop uses 427 flops */
 686         }
 687
 688         if(jidx<j_index_end)
 689         {
 690
 691             jnrA             = jjnr[jidx];
 692             j_coord_offsetA  = DIM*jnrA;
 693
 694             /* load j atom coordinates */
 695             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 696                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 697
 698             /* Calculate displacement vector */
 699             dx00             = _mm_sub_pd(ix0,jx0);
 700             dy00             = _mm_sub_pd(iy0,jy0);
 701             dz00             = _mm_sub_pd(iz0,jz0);
 702             dx01             = _mm_sub_pd(ix0,jx1);
 703             dy01             = _mm_sub_pd(iy0,jy1);
 704             dz01             = _mm_sub_pd(iz0,jz1);
 705             dx02             = _mm_sub_pd(ix0,jx2);
 706             dy02             = _mm_sub_pd(iy0,jy2);
 707             dz02             = _mm_sub_pd(iz0,jz2);
 708             dx10             = _mm_sub_pd(ix1,jx0);
 709             dy10             = _mm_sub_pd(iy1,jy0);
 710             dz10             = _mm_sub_pd(iz1,jz0);
 711             dx11             = _mm_sub_pd(ix1,jx1);
 712             dy11             = _mm_sub_pd(iy1,jy1);
 713             dz11             = _mm_sub_pd(iz1,jz1);
 714             dx12             = _mm_sub_pd(ix1,jx2);
 715             dy12             = _mm_sub_pd(iy1,jy2);
 716             dz12             = _mm_sub_pd(iz1,jz2);
 717             dx20             = _mm_sub_pd(ix2,jx0);
 718             dy20             = _mm_sub_pd(iy2,jy0);
 719             dz20             = _mm_sub_pd(iz2,jz0);
 720             dx21             = _mm_sub_pd(ix2,jx1);
 721             dy21             = _mm_sub_pd(iy2,jy1);
 722             dz21             = _mm_sub_pd(iz2,jz1);
 723             dx22             = _mm_sub_pd(ix2,jx2);
 724             dy22             = _mm_sub_pd(iy2,jy2);
 725             dz22             = _mm_sub_pd(iz2,jz2);
 726
 727             /* Calculate squared distance and things based on it */
 728             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 729             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 730             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 731             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 732             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 733             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 734             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 735             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 736             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 737
 738             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 739             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 740             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 741             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 742             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 743             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 744             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 745             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 746             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 747
 748             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 749
 750             fjx0             = _mm_setzero_pd();
 751             fjy0             = _mm_setzero_pd();
 752             fjz0             = _mm_setzero_pd();
 753             fjx1             = _mm_setzero_pd();
 754             fjy1             = _mm_setzero_pd();
 755             fjz1             = _mm_setzero_pd();
 756             fjx2             = _mm_setzero_pd();
 757             fjy2             = _mm_setzero_pd();
 758             fjz2             = _mm_setzero_pd();
 759
 760             /**************************
 761              * CALCULATE INTERACTIONS *
 762              **************************/
 763
 764             r00              = _mm_mul_pd(rsq00,rinv00);
 765
 766             /* Calculate table index by multiplying r with table scale and truncate to integer */
 767             rt               = _mm_mul_pd(r00,vftabscale);
 768             vfitab           = _mm_cvttpd_epi32(rt);
 769 #ifdef __XOP__
 770             vfeps            = _mm_frcz_pd(rt);
 771 #else
 772             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 773 #endif
 774             twovfeps         = _mm_add_pd(vfeps,vfeps);
 775             vfitab           = _mm_slli_epi32(vfitab,2);
 776
 777             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 778             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 779             F                = _mm_setzero_pd();
 780             GMX_MM_TRANSPOSE2_PD(Y,F);
 781             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 782             H                = _mm_setzero_pd();
 783             GMX_MM_TRANSPOSE2_PD(G,H);
 784             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 785             VV               = _mm_macc_pd(vfeps,Fp,Y);
 786             velec            = _mm_mul_pd(qq00,VV);
 787             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 788             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 789
 790             /* LENNARD-JONES DISPERSION/REPULSION */
 791
 792             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 793             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 794             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 795             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 796             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 797
 798             /* Update potential sum for this i atom from the interaction with this j atom. */
 799             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 800             velecsum         = _mm_add_pd(velecsum,velec);
 801             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 802             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 803
 804             fscal            = _mm_add_pd(felec,fvdw);
 805
 806             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 807
 808             /* Update vectorial force */
 809             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 810             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 811             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 812
 813             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 814             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 815             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 816
 817             /**************************
 818              * CALCULATE INTERACTIONS *
 819              **************************/
 820
 821             r01              = _mm_mul_pd(rsq01,rinv01);
 822
 823             /* Calculate table index by multiplying r with table scale and truncate to integer */
 824             rt               = _mm_mul_pd(r01,vftabscale);
 825             vfitab           = _mm_cvttpd_epi32(rt);
 826 #ifdef __XOP__
 827             vfeps            = _mm_frcz_pd(rt);
 828 #else
 829             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 830 #endif
 831             twovfeps         = _mm_add_pd(vfeps,vfeps);
 832             vfitab           = _mm_slli_epi32(vfitab,2);
 833
 834             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 835             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 836             F                = _mm_setzero_pd();
 837             GMX_MM_TRANSPOSE2_PD(Y,F);
 838             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 839             H                = _mm_setzero_pd();
 840             GMX_MM_TRANSPOSE2_PD(G,H);
 841             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 842             VV               = _mm_macc_pd(vfeps,Fp,Y);
 843             velec            = _mm_mul_pd(qq01,VV);
 844             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 845             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq01,FF),_mm_mul_pd(vftabscale,rinv01)));
 846
 847             /* Update potential sum for this i atom from the interaction with this j atom. */
 848             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 849             velecsum         = _mm_add_pd(velecsum,velec);
 850
 851             fscal            = felec;
 852
 853             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 854
 855             /* Update vectorial force */
 856             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 857             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 858             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 859
 860             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 861             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 862             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 863
 864             /**************************
 865              * CALCULATE INTERACTIONS *
 866              **************************/
 867
 868             r02              = _mm_mul_pd(rsq02,rinv02);
 869
 870             /* Calculate table index by multiplying r with table scale and truncate to integer */
 871             rt               = _mm_mul_pd(r02,vftabscale);
 872             vfitab           = _mm_cvttpd_epi32(rt);
 873 #ifdef __XOP__
 874             vfeps            = _mm_frcz_pd(rt);
 875 #else
 876             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 877 #endif
 878             twovfeps         = _mm_add_pd(vfeps,vfeps);
 879             vfitab           = _mm_slli_epi32(vfitab,2);
 880
 881             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 882             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 883             F                = _mm_setzero_pd();
 884             GMX_MM_TRANSPOSE2_PD(Y,F);
 885             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 886             H                = _mm_setzero_pd();
 887             GMX_MM_TRANSPOSE2_PD(G,H);
 888             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 889             VV               = _mm_macc_pd(vfeps,Fp,Y);
 890             velec            = _mm_mul_pd(qq02,VV);
 891             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 892             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq02,FF),_mm_mul_pd(vftabscale,rinv02)));
 893
 894             /* Update potential sum for this i atom from the interaction with this j atom. */
 895             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 896             velecsum         = _mm_add_pd(velecsum,velec);
 897
 898             fscal            = felec;
 899
 900             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 901
 902             /* Update vectorial force */
 903             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 904             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 905             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 906
 907             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 908             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 909             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             r10              = _mm_mul_pd(rsq10,rinv10);
 916
 917             /* Calculate table index by multiplying r with table scale and truncate to integer */
 918             rt               = _mm_mul_pd(r10,vftabscale);
 919             vfitab           = _mm_cvttpd_epi32(rt);
 920 #ifdef __XOP__
 921             vfeps            = _mm_frcz_pd(rt);
 922 #else
 923             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 924 #endif
 925             twovfeps         = _mm_add_pd(vfeps,vfeps);
 926             vfitab           = _mm_slli_epi32(vfitab,2);
 927
 928             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 929             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 930             F                = _mm_setzero_pd();
 931             GMX_MM_TRANSPOSE2_PD(Y,F);
 932             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 933             H                = _mm_setzero_pd();
 934             GMX_MM_TRANSPOSE2_PD(G,H);
 935             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 936             VV               = _mm_macc_pd(vfeps,Fp,Y);
 937             velec            = _mm_mul_pd(qq10,VV);
 938             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 939             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 940
 941             /* Update potential sum for this i atom from the interaction with this j atom. */
 942             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 943             velecsum         = _mm_add_pd(velecsum,velec);
 944
 945             fscal            = felec;
 946
 947             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 948
 949             /* Update vectorial force */
 950             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 951             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 952             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 953
 954             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 955             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 956             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 957
 958             /**************************
 959              * CALCULATE INTERACTIONS *
 960              **************************/
 961
 962             r11              = _mm_mul_pd(rsq11,rinv11);
 963
 964             /* Calculate table index by multiplying r with table scale and truncate to integer */
 965             rt               = _mm_mul_pd(r11,vftabscale);
 966             vfitab           = _mm_cvttpd_epi32(rt);
 967 #ifdef __XOP__
 968             vfeps            = _mm_frcz_pd(rt);
 969 #else
 970             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 971 #endif
 972             twovfeps         = _mm_add_pd(vfeps,vfeps);
 973             vfitab           = _mm_slli_epi32(vfitab,2);
 974
 975             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 976             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 977             F                = _mm_setzero_pd();
 978             GMX_MM_TRANSPOSE2_PD(Y,F);
 979             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 980             H                = _mm_setzero_pd();
 981             GMX_MM_TRANSPOSE2_PD(G,H);
 982             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 983             VV               = _mm_macc_pd(vfeps,Fp,Y);
 984             velec            = _mm_mul_pd(qq11,VV);
 985             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 986             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
 987
 988             /* Update potential sum for this i atom from the interaction with this j atom. */
 989             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 990             velecsum         = _mm_add_pd(velecsum,velec);
 991
 992             fscal            = felec;
 993
 994             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 995
 996             /* Update vectorial force */
 997             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 998             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 999             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1000
1001             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1002             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1003             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             r12              = _mm_mul_pd(rsq12,rinv12);
1010
1011             /* Calculate table index by multiplying r with table scale and truncate to integer */
1012             rt               = _mm_mul_pd(r12,vftabscale);
1013             vfitab           = _mm_cvttpd_epi32(rt);
1014 #ifdef __XOP__
1015             vfeps            = _mm_frcz_pd(rt);
1016 #else
1017             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1018 #endif
1019             twovfeps         = _mm_add_pd(vfeps,vfeps);
1020             vfitab           = _mm_slli_epi32(vfitab,2);
1021
1022             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1023             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1024             F                = _mm_setzero_pd();
1025             GMX_MM_TRANSPOSE2_PD(Y,F);
1026             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1027             H                = _mm_setzero_pd();
1028             GMX_MM_TRANSPOSE2_PD(G,H);
1029             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1030             VV               = _mm_macc_pd(vfeps,Fp,Y);
1031             velec            = _mm_mul_pd(qq12,VV);
1032             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1033             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
1034
1035             /* Update potential sum for this i atom from the interaction with this j atom. */
1036             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1037             velecsum         = _mm_add_pd(velecsum,velec);
1038
1039             fscal            = felec;
1040
1041             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1042
1043             /* Update vectorial force */
1044             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1045             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1046             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1047
1048             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1049             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1050             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1051
1052             /**************************
1053              * CALCULATE INTERACTIONS *
1054              **************************/
1055
1056             r20              = _mm_mul_pd(rsq20,rinv20);
1057
1058             /* Calculate table index by multiplying r with table scale and truncate to integer */
1059             rt               = _mm_mul_pd(r20,vftabscale);
1060             vfitab           = _mm_cvttpd_epi32(rt);
1061 #ifdef __XOP__
1062             vfeps            = _mm_frcz_pd(rt);
1063 #else
1064             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1065 #endif
1066             twovfeps         = _mm_add_pd(vfeps,vfeps);
1067             vfitab           = _mm_slli_epi32(vfitab,2);
1068
1069             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1070             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1071             F                = _mm_setzero_pd();
1072             GMX_MM_TRANSPOSE2_PD(Y,F);
1073             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1074             H                = _mm_setzero_pd();
1075             GMX_MM_TRANSPOSE2_PD(G,H);
1076             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1077             VV               = _mm_macc_pd(vfeps,Fp,Y);
1078             velec            = _mm_mul_pd(qq20,VV);
1079             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1080             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1081
1082             /* Update potential sum for this i atom from the interaction with this j atom. */
1083             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1084             velecsum         = _mm_add_pd(velecsum,velec);
1085
1086             fscal            = felec;
1087
1088             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1089
1090             /* Update vectorial force */
1091             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1092             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1093             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1094
1095             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1096             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1097             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1098
1099             /**************************
1100              * CALCULATE INTERACTIONS *
1101              **************************/
1102
1103             r21              = _mm_mul_pd(rsq21,rinv21);
1104
1105             /* Calculate table index by multiplying r with table scale and truncate to integer */
1106             rt               = _mm_mul_pd(r21,vftabscale);
1107             vfitab           = _mm_cvttpd_epi32(rt);
1108 #ifdef __XOP__
1109             vfeps            = _mm_frcz_pd(rt);
1110 #else
1111             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1112 #endif
1113             twovfeps         = _mm_add_pd(vfeps,vfeps);
1114             vfitab           = _mm_slli_epi32(vfitab,2);
1115
1116             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1117             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1118             F                = _mm_setzero_pd();
1119             GMX_MM_TRANSPOSE2_PD(Y,F);
1120             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1121             H                = _mm_setzero_pd();
1122             GMX_MM_TRANSPOSE2_PD(G,H);
1123             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1124             VV               = _mm_macc_pd(vfeps,Fp,Y);
1125             velec            = _mm_mul_pd(qq21,VV);
1126             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1127             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
1128
1129             /* Update potential sum for this i atom from the interaction with this j atom. */
1130             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1131             velecsum         = _mm_add_pd(velecsum,velec);
1132
1133             fscal            = felec;
1134
1135             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1136
1137             /* Update vectorial force */
1138             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1139             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1140             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1141
1142             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1143             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1144             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1145
1146             /**************************
1147              * CALCULATE INTERACTIONS *
1148              **************************/
1149
1150             r22              = _mm_mul_pd(rsq22,rinv22);
1151
1152             /* Calculate table index by multiplying r with table scale and truncate to integer */
1153             rt               = _mm_mul_pd(r22,vftabscale);
1154             vfitab           = _mm_cvttpd_epi32(rt);
1155 #ifdef __XOP__
1156             vfeps            = _mm_frcz_pd(rt);
1157 #else
1158             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1159 #endif
1160             twovfeps         = _mm_add_pd(vfeps,vfeps);
1161             vfitab           = _mm_slli_epi32(vfitab,2);
1162
1163             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1164             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1165             F                = _mm_setzero_pd();
1166             GMX_MM_TRANSPOSE2_PD(Y,F);
1167             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1168             H                = _mm_setzero_pd();
1169             GMX_MM_TRANSPOSE2_PD(G,H);
1170             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1171             VV               = _mm_macc_pd(vfeps,Fp,Y);
1172             velec            = _mm_mul_pd(qq22,VV);
1173             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1174             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
1175
1176             /* Update potential sum for this i atom from the interaction with this j atom. */
1177             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1178             velecsum         = _mm_add_pd(velecsum,velec);
1179
1180             fscal            = felec;
1181
1182             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1183
1184             /* Update vectorial force */
1185             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1186             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1187             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1188
1189             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1190             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1191             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1192
1193             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1194
1195             /* Inner loop uses 427 flops */
1196         }
1197
1198         /* End of innermost loop */
1199
1200         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1201                                               f+i_coord_offset,fshift+i_shift_offset);
1202
1203         ggid                        = gid[iidx];
1204         /* Update potential energies */
1205         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1206         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1207
1208         /* Increment number of inner iterations */
1209         inneriter                  += j_index_end - j_index_start;
1210
1211         /* Outer loop uses 20 flops */
1212     }
1213
1214     /* Increment number of outer iterations */
1215     outeriter        += nri;
1216
1217     /* Update outer/inner flops */
1218
1219     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*427);
1220 }
1221 /*
1222  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_F_avx_128_fma_double
1223  * Electrostatics interaction: CubicSplineTable
1224  * VdW interaction:            LennardJones
1225  * Geometry:                   Water3-Water3
1226  * Calculate force/pot:        Force
1227  */
1228 void
1229 nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_F_avx_128_fma_double
1230                     (t_nblist                    * gmx_restrict       nlist,
1231                      rvec                        * gmx_restrict          xx,
1232                      rvec                        * gmx_restrict          ff,
1233                      t_forcerec                  * gmx_restrict          fr,
1234                      t_mdatoms                   * gmx_restrict     mdatoms,
1235                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1236                      t_nrnb                      * gmx_restrict        nrnb)
1237 {
1238     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1239      * just 0 for non-waters.
1240      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1241      * jnr indices corresponding to data put in the four positions in the SIMD register.
1242      */
1243     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1244     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1245     int              jnrA,jnrB;
1246     int              j_coord_offsetA,j_coord_offsetB;
1247     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1248     real             rcutoff_scalar;
1249     real             *shiftvec,*fshift,*x,*f;
1250     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1251     int              vdwioffset0;
1252     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1253     int              vdwioffset1;
1254     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1255     int              vdwioffset2;
1256     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1257     int              vdwjidx0A,vdwjidx0B;
1258     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1259     int              vdwjidx1A,vdwjidx1B;
1260     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1261     int              vdwjidx2A,vdwjidx2B;
1262     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1263     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1264     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1265     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1266     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1267     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1268     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1269     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1270     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1271     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1272     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1273     real             *charge;
1274     int              nvdwtype;
1275     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1276     int              *vdwtype;
1277     real             *vdwparam;
1278     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1279     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1280     __m128i          vfitab;
1281     __m128i          ifour       = _mm_set1_epi32(4);
1282     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
1283     real             *vftab;
1284     __m128d          dummy_mask,cutoff_mask;
1285     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1286     __m128d          one     = _mm_set1_pd(1.0);
1287     __m128d          two     = _mm_set1_pd(2.0);
1288     x                = xx[0];
1289     f                = ff[0];
1290
1291     nri              = nlist->nri;
1292     iinr             = nlist->iinr;
1293     jindex           = nlist->jindex;
1294     jjnr             = nlist->jjnr;
1295     shiftidx         = nlist->shift;
1296     gid              = nlist->gid;
1297     shiftvec         = fr->shift_vec[0];
1298     fshift           = fr->fshift[0];
1299     facel            = _mm_set1_pd(fr->epsfac);
1300     charge           = mdatoms->chargeA;
1301     nvdwtype         = fr->ntype;
1302     vdwparam         = fr->nbfp;
1303     vdwtype          = mdatoms->typeA;
1304
1305     vftab            = kernel_data->table_elec->data;
1306     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
1307
1308     /* Setup water-specific parameters */
1309     inr              = nlist->iinr[0];
1310     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1311     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1312     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1313     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1314
1315     jq0              = _mm_set1_pd(charge[inr+0]);
1316     jq1              = _mm_set1_pd(charge[inr+1]);
1317     jq2              = _mm_set1_pd(charge[inr+2]);
1318     vdwjidx0A        = 2*vdwtype[inr+0];
1319     qq00             = _mm_mul_pd(iq0,jq0);
1320     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1321     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1322     qq01             = _mm_mul_pd(iq0,jq1);
1323     qq02             = _mm_mul_pd(iq0,jq2);
1324     qq10             = _mm_mul_pd(iq1,jq0);
1325     qq11             = _mm_mul_pd(iq1,jq1);
1326     qq12             = _mm_mul_pd(iq1,jq2);
1327     qq20             = _mm_mul_pd(iq2,jq0);
1328     qq21             = _mm_mul_pd(iq2,jq1);
1329     qq22             = _mm_mul_pd(iq2,jq2);
1330
1331     /* Avoid stupid compiler warnings */
1332     jnrA = jnrB = 0;
1333     j_coord_offsetA = 0;
1334     j_coord_offsetB = 0;
1335
1336     outeriter        = 0;
1337     inneriter        = 0;
1338
1339     /* Start outer loop over neighborlists */
1340     for(iidx=0; iidx<nri; iidx++)
1341     {
1342         /* Load shift vector for this list */
1343         i_shift_offset   = DIM*shiftidx[iidx];
1344
1345         /* Load limits for loop over neighbors */
1346         j_index_start    = jindex[iidx];
1347         j_index_end      = jindex[iidx+1];
1348
1349         /* Get outer coordinate index */
1350         inr              = iinr[iidx];
1351         i_coord_offset   = DIM*inr;
1352
1353         /* Load i particle coords and add shift vector */
1354         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1355                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1356
1357         fix0             = _mm_setzero_pd();
1358         fiy0             = _mm_setzero_pd();
1359         fiz0             = _mm_setzero_pd();
1360         fix1             = _mm_setzero_pd();
1361         fiy1             = _mm_setzero_pd();
1362         fiz1             = _mm_setzero_pd();
1363         fix2             = _mm_setzero_pd();
1364         fiy2             = _mm_setzero_pd();
1365         fiz2             = _mm_setzero_pd();
1366
1367         /* Start inner kernel loop */
1368         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1369         {
1370
1371             /* Get j neighbor index, and coordinate index */
1372             jnrA             = jjnr[jidx];
1373             jnrB             = jjnr[jidx+1];
1374             j_coord_offsetA  = DIM*jnrA;
1375             j_coord_offsetB  = DIM*jnrB;
1376
1377             /* load j atom coordinates */
1378             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1379                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1380
1381             /* Calculate displacement vector */
1382             dx00             = _mm_sub_pd(ix0,jx0);
1383             dy00             = _mm_sub_pd(iy0,jy0);
1384             dz00             = _mm_sub_pd(iz0,jz0);
1385             dx01             = _mm_sub_pd(ix0,jx1);
1386             dy01             = _mm_sub_pd(iy0,jy1);
1387             dz01             = _mm_sub_pd(iz0,jz1);
1388             dx02             = _mm_sub_pd(ix0,jx2);
1389             dy02             = _mm_sub_pd(iy0,jy2);
1390             dz02             = _mm_sub_pd(iz0,jz2);
1391             dx10             = _mm_sub_pd(ix1,jx0);
1392             dy10             = _mm_sub_pd(iy1,jy0);
1393             dz10             = _mm_sub_pd(iz1,jz0);
1394             dx11             = _mm_sub_pd(ix1,jx1);
1395             dy11             = _mm_sub_pd(iy1,jy1);
1396             dz11             = _mm_sub_pd(iz1,jz1);
1397             dx12             = _mm_sub_pd(ix1,jx2);
1398             dy12             = _mm_sub_pd(iy1,jy2);
1399             dz12             = _mm_sub_pd(iz1,jz2);
1400             dx20             = _mm_sub_pd(ix2,jx0);
1401             dy20             = _mm_sub_pd(iy2,jy0);
1402             dz20             = _mm_sub_pd(iz2,jz0);
1403             dx21             = _mm_sub_pd(ix2,jx1);
1404             dy21             = _mm_sub_pd(iy2,jy1);
1405             dz21             = _mm_sub_pd(iz2,jz1);
1406             dx22             = _mm_sub_pd(ix2,jx2);
1407             dy22             = _mm_sub_pd(iy2,jy2);
1408             dz22             = _mm_sub_pd(iz2,jz2);
1409
1410             /* Calculate squared distance and things based on it */
1411             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1412             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1413             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1414             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1415             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1416             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1417             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1418             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1419             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1420
1421             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1422             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1423             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1424             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1425             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1426             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1427             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1428             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1429             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1430
1431             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1432
1433             fjx0             = _mm_setzero_pd();
1434             fjy0             = _mm_setzero_pd();
1435             fjz0             = _mm_setzero_pd();
1436             fjx1             = _mm_setzero_pd();
1437             fjy1             = _mm_setzero_pd();
1438             fjz1             = _mm_setzero_pd();
1439             fjx2             = _mm_setzero_pd();
1440             fjy2             = _mm_setzero_pd();
1441             fjz2             = _mm_setzero_pd();
1442
1443             /**************************
1444              * CALCULATE INTERACTIONS *
1445              **************************/
1446
1447             r00              = _mm_mul_pd(rsq00,rinv00);
1448
1449             /* Calculate table index by multiplying r with table scale and truncate to integer */
1450             rt               = _mm_mul_pd(r00,vftabscale);
1451             vfitab           = _mm_cvttpd_epi32(rt);
1452 #ifdef __XOP__
1453             vfeps            = _mm_frcz_pd(rt);
1454 #else
1455             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1456 #endif
1457             twovfeps         = _mm_add_pd(vfeps,vfeps);
1458             vfitab           = _mm_slli_epi32(vfitab,2);
1459
1460             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1461             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1462             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1463             GMX_MM_TRANSPOSE2_PD(Y,F);
1464             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1465             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1466             GMX_MM_TRANSPOSE2_PD(G,H);
1467             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1468             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1469             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1470
1471             /* LENNARD-JONES DISPERSION/REPULSION */
1472
1473             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1474             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1475
1476             fscal            = _mm_add_pd(felec,fvdw);
1477
1478             /* Update vectorial force */
1479             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1480             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1481             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1482
1483             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1484             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1485             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1486
1487             /**************************
1488              * CALCULATE INTERACTIONS *
1489              **************************/
1490
1491             r01              = _mm_mul_pd(rsq01,rinv01);
1492
1493             /* Calculate table index by multiplying r with table scale and truncate to integer */
1494             rt               = _mm_mul_pd(r01,vftabscale);
1495             vfitab           = _mm_cvttpd_epi32(rt);
1496 #ifdef __XOP__
1497             vfeps            = _mm_frcz_pd(rt);
1498 #else
1499             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1500 #endif
1501             twovfeps         = _mm_add_pd(vfeps,vfeps);
1502             vfitab           = _mm_slli_epi32(vfitab,2);
1503
1504             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1505             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1506             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1507             GMX_MM_TRANSPOSE2_PD(Y,F);
1508             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1509             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1510             GMX_MM_TRANSPOSE2_PD(G,H);
1511             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1512             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1513             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq01,FF),_mm_mul_pd(vftabscale,rinv01)));
1514
1515             fscal            = felec;
1516
1517             /* Update vectorial force */
1518             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1519             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1520             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1521
1522             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1523             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1524             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1525
1526             /**************************
1527              * CALCULATE INTERACTIONS *
1528              **************************/
1529
1530             r02              = _mm_mul_pd(rsq02,rinv02);
1531
1532             /* Calculate table index by multiplying r with table scale and truncate to integer */
1533             rt               = _mm_mul_pd(r02,vftabscale);
1534             vfitab           = _mm_cvttpd_epi32(rt);
1535 #ifdef __XOP__
1536             vfeps            = _mm_frcz_pd(rt);
1537 #else
1538             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1539 #endif
1540             twovfeps         = _mm_add_pd(vfeps,vfeps);
1541             vfitab           = _mm_slli_epi32(vfitab,2);
1542
1543             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1544             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1545             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1546             GMX_MM_TRANSPOSE2_PD(Y,F);
1547             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1548             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1549             GMX_MM_TRANSPOSE2_PD(G,H);
1550             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1551             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1552             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq02,FF),_mm_mul_pd(vftabscale,rinv02)));
1553
1554             fscal            = felec;
1555
1556             /* Update vectorial force */
1557             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1558             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1559             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1560
1561             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1562             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1563             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1564
1565             /**************************
1566              * CALCULATE INTERACTIONS *
1567              **************************/
1568
1569             r10              = _mm_mul_pd(rsq10,rinv10);
1570
1571             /* Calculate table index by multiplying r with table scale and truncate to integer */
1572             rt               = _mm_mul_pd(r10,vftabscale);
1573             vfitab           = _mm_cvttpd_epi32(rt);
1574 #ifdef __XOP__
1575             vfeps            = _mm_frcz_pd(rt);
1576 #else
1577             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1578 #endif
1579             twovfeps         = _mm_add_pd(vfeps,vfeps);
1580             vfitab           = _mm_slli_epi32(vfitab,2);
1581
1582             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1583             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1584             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1585             GMX_MM_TRANSPOSE2_PD(Y,F);
1586             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1587             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1588             GMX_MM_TRANSPOSE2_PD(G,H);
1589             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1590             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1591             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1592
1593             fscal            = felec;
1594
1595             /* Update vectorial force */
1596             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1597             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1598             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1599
1600             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1601             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1602             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1603
1604             /**************************
1605              * CALCULATE INTERACTIONS *
1606              **************************/
1607
1608             r11              = _mm_mul_pd(rsq11,rinv11);
1609
1610             /* Calculate table index by multiplying r with table scale and truncate to integer */
1611             rt               = _mm_mul_pd(r11,vftabscale);
1612             vfitab           = _mm_cvttpd_epi32(rt);
1613 #ifdef __XOP__
1614             vfeps            = _mm_frcz_pd(rt);
1615 #else
1616             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1617 #endif
1618             twovfeps         = _mm_add_pd(vfeps,vfeps);
1619             vfitab           = _mm_slli_epi32(vfitab,2);
1620
1621             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1622             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1623             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1624             GMX_MM_TRANSPOSE2_PD(Y,F);
1625             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1626             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1627             GMX_MM_TRANSPOSE2_PD(G,H);
1628             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1629             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1630             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
1631
1632             fscal            = felec;
1633
1634             /* Update vectorial force */
1635             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1636             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1637             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1638
1639             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1640             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1641             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1642
1643             /**************************
1644              * CALCULATE INTERACTIONS *
1645              **************************/
1646
1647             r12              = _mm_mul_pd(rsq12,rinv12);
1648
1649             /* Calculate table index by multiplying r with table scale and truncate to integer */
1650             rt               = _mm_mul_pd(r12,vftabscale);
1651             vfitab           = _mm_cvttpd_epi32(rt);
1652 #ifdef __XOP__
1653             vfeps            = _mm_frcz_pd(rt);
1654 #else
1655             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1656 #endif
1657             twovfeps         = _mm_add_pd(vfeps,vfeps);
1658             vfitab           = _mm_slli_epi32(vfitab,2);
1659
1660             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1661             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1662             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1663             GMX_MM_TRANSPOSE2_PD(Y,F);
1664             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1665             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1666             GMX_MM_TRANSPOSE2_PD(G,H);
1667             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1668             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1669             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
1670
1671             fscal            = felec;
1672
1673             /* Update vectorial force */
1674             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1675             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1676             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1677
1678             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1679             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1680             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1681
1682             /**************************
1683              * CALCULATE INTERACTIONS *
1684              **************************/
1685
1686             r20              = _mm_mul_pd(rsq20,rinv20);
1687
1688             /* Calculate table index by multiplying r with table scale and truncate to integer */
1689             rt               = _mm_mul_pd(r20,vftabscale);
1690             vfitab           = _mm_cvttpd_epi32(rt);
1691 #ifdef __XOP__
1692             vfeps            = _mm_frcz_pd(rt);
1693 #else
1694             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1695 #endif
1696             twovfeps         = _mm_add_pd(vfeps,vfeps);
1697             vfitab           = _mm_slli_epi32(vfitab,2);
1698
1699             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1700             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1701             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1702             GMX_MM_TRANSPOSE2_PD(Y,F);
1703             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1704             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1705             GMX_MM_TRANSPOSE2_PD(G,H);
1706             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1707             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1708             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1709
1710             fscal            = felec;
1711
1712             /* Update vectorial force */
1713             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1714             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1715             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1716
1717             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1718             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1719             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1720
1721             /**************************
1722              * CALCULATE INTERACTIONS *
1723              **************************/
1724
1725             r21              = _mm_mul_pd(rsq21,rinv21);
1726
1727             /* Calculate table index by multiplying r with table scale and truncate to integer */
1728             rt               = _mm_mul_pd(r21,vftabscale);
1729             vfitab           = _mm_cvttpd_epi32(rt);
1730 #ifdef __XOP__
1731             vfeps            = _mm_frcz_pd(rt);
1732 #else
1733             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1734 #endif
1735             twovfeps         = _mm_add_pd(vfeps,vfeps);
1736             vfitab           = _mm_slli_epi32(vfitab,2);
1737
1738             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1739             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1740             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1741             GMX_MM_TRANSPOSE2_PD(Y,F);
1742             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1743             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1744             GMX_MM_TRANSPOSE2_PD(G,H);
1745             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1746             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1747             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
1748
1749             fscal            = felec;
1750
1751             /* Update vectorial force */
1752             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1753             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1754             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1755
1756             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1757             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1758             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1759
1760             /**************************
1761              * CALCULATE INTERACTIONS *
1762              **************************/
1763
1764             r22              = _mm_mul_pd(rsq22,rinv22);
1765
1766             /* Calculate table index by multiplying r with table scale and truncate to integer */
1767             rt               = _mm_mul_pd(r22,vftabscale);
1768             vfitab           = _mm_cvttpd_epi32(rt);
1769 #ifdef __XOP__
1770             vfeps            = _mm_frcz_pd(rt);
1771 #else
1772             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1773 #endif
1774             twovfeps         = _mm_add_pd(vfeps,vfeps);
1775             vfitab           = _mm_slli_epi32(vfitab,2);
1776
1777             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1778             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1779             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1780             GMX_MM_TRANSPOSE2_PD(Y,F);
1781             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1782             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1783             GMX_MM_TRANSPOSE2_PD(G,H);
1784             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1785             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1786             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
1787
1788             fscal            = felec;
1789
1790             /* Update vectorial force */
1791             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1792             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1793             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1794
1795             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1796             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1797             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1798
1799             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1800
1801             /* Inner loop uses 386 flops */
1802         }
1803
1804         if(jidx<j_index_end)
1805         {
1806
1807             jnrA             = jjnr[jidx];
1808             j_coord_offsetA  = DIM*jnrA;
1809
1810             /* load j atom coordinates */
1811             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1812                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1813
1814             /* Calculate displacement vector */
1815             dx00             = _mm_sub_pd(ix0,jx0);
1816             dy00             = _mm_sub_pd(iy0,jy0);
1817             dz00             = _mm_sub_pd(iz0,jz0);
1818             dx01             = _mm_sub_pd(ix0,jx1);
1819             dy01             = _mm_sub_pd(iy0,jy1);
1820             dz01             = _mm_sub_pd(iz0,jz1);
1821             dx02             = _mm_sub_pd(ix0,jx2);
1822             dy02             = _mm_sub_pd(iy0,jy2);
1823             dz02             = _mm_sub_pd(iz0,jz2);
1824             dx10             = _mm_sub_pd(ix1,jx0);
1825             dy10             = _mm_sub_pd(iy1,jy0);
1826             dz10             = _mm_sub_pd(iz1,jz0);
1827             dx11             = _mm_sub_pd(ix1,jx1);
1828             dy11             = _mm_sub_pd(iy1,jy1);
1829             dz11             = _mm_sub_pd(iz1,jz1);
1830             dx12             = _mm_sub_pd(ix1,jx2);
1831             dy12             = _mm_sub_pd(iy1,jy2);
1832             dz12             = _mm_sub_pd(iz1,jz2);
1833             dx20             = _mm_sub_pd(ix2,jx0);
1834             dy20             = _mm_sub_pd(iy2,jy0);
1835             dz20             = _mm_sub_pd(iz2,jz0);
1836             dx21             = _mm_sub_pd(ix2,jx1);
1837             dy21             = _mm_sub_pd(iy2,jy1);
1838             dz21             = _mm_sub_pd(iz2,jz1);
1839             dx22             = _mm_sub_pd(ix2,jx2);
1840             dy22             = _mm_sub_pd(iy2,jy2);
1841             dz22             = _mm_sub_pd(iz2,jz2);
1842
1843             /* Calculate squared distance and things based on it */
1844             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1845             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1846             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1847             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1848             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1849             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1850             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1851             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1852             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1853
1854             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1855             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1856             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1857             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1858             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1859             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1860             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1861             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1862             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1863
1864             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1865
1866             fjx0             = _mm_setzero_pd();
1867             fjy0             = _mm_setzero_pd();
1868             fjz0             = _mm_setzero_pd();
1869             fjx1             = _mm_setzero_pd();
1870             fjy1             = _mm_setzero_pd();
1871             fjz1             = _mm_setzero_pd();
1872             fjx2             = _mm_setzero_pd();
1873             fjy2             = _mm_setzero_pd();
1874             fjz2             = _mm_setzero_pd();
1875
1876             /**************************
1877              * CALCULATE INTERACTIONS *
1878              **************************/
1879
1880             r00              = _mm_mul_pd(rsq00,rinv00);
1881
1882             /* Calculate table index by multiplying r with table scale and truncate to integer */
1883             rt               = _mm_mul_pd(r00,vftabscale);
1884             vfitab           = _mm_cvttpd_epi32(rt);
1885 #ifdef __XOP__
1886             vfeps            = _mm_frcz_pd(rt);
1887 #else
1888             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1889 #endif
1890             twovfeps         = _mm_add_pd(vfeps,vfeps);
1891             vfitab           = _mm_slli_epi32(vfitab,2);
1892
1893             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1894             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1895             F                = _mm_setzero_pd();
1896             GMX_MM_TRANSPOSE2_PD(Y,F);
1897             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1898             H                = _mm_setzero_pd();
1899             GMX_MM_TRANSPOSE2_PD(G,H);
1900             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1901             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1902             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1903
1904             /* LENNARD-JONES DISPERSION/REPULSION */
1905
1906             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1907             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1908
1909             fscal            = _mm_add_pd(felec,fvdw);
1910
1911             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1912
1913             /* Update vectorial force */
1914             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1915             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1916             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1917
1918             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1919             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1920             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1921
1922             /**************************
1923              * CALCULATE INTERACTIONS *
1924              **************************/
1925
1926             r01              = _mm_mul_pd(rsq01,rinv01);
1927
1928             /* Calculate table index by multiplying r with table scale and truncate to integer */
1929             rt               = _mm_mul_pd(r01,vftabscale);
1930             vfitab           = _mm_cvttpd_epi32(rt);
1931 #ifdef __XOP__
1932             vfeps            = _mm_frcz_pd(rt);
1933 #else
1934             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1935 #endif
1936             twovfeps         = _mm_add_pd(vfeps,vfeps);
1937             vfitab           = _mm_slli_epi32(vfitab,2);
1938
1939             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1940             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1941             F                = _mm_setzero_pd();
1942             GMX_MM_TRANSPOSE2_PD(Y,F);
1943             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1944             H                = _mm_setzero_pd();
1945             GMX_MM_TRANSPOSE2_PD(G,H);
1946             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1947             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1948             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq01,FF),_mm_mul_pd(vftabscale,rinv01)));
1949
1950             fscal            = felec;
1951
1952             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1953
1954             /* Update vectorial force */
1955             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1956             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1957             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1958
1959             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1960             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1961             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1962
1963             /**************************
1964              * CALCULATE INTERACTIONS *
1965              **************************/
1966
1967             r02              = _mm_mul_pd(rsq02,rinv02);
1968
1969             /* Calculate table index by multiplying r with table scale and truncate to integer */
1970             rt               = _mm_mul_pd(r02,vftabscale);
1971             vfitab           = _mm_cvttpd_epi32(rt);
1972 #ifdef __XOP__
1973             vfeps            = _mm_frcz_pd(rt);
1974 #else
1975             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1976 #endif
1977             twovfeps         = _mm_add_pd(vfeps,vfeps);
1978             vfitab           = _mm_slli_epi32(vfitab,2);
1979
1980             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1981             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1982             F                = _mm_setzero_pd();
1983             GMX_MM_TRANSPOSE2_PD(Y,F);
1984             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1985             H                = _mm_setzero_pd();
1986             GMX_MM_TRANSPOSE2_PD(G,H);
1987             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1988             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1989             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq02,FF),_mm_mul_pd(vftabscale,rinv02)));
1990
1991             fscal            = felec;
1992
1993             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1994
1995             /* Update vectorial force */
1996             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1997             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1998             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1999
2000             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
2001             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
2002             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
2003
2004             /**************************
2005              * CALCULATE INTERACTIONS *
2006              **************************/
2007
2008             r10              = _mm_mul_pd(rsq10,rinv10);
2009
2010             /* Calculate table index by multiplying r with table scale and truncate to integer */
2011             rt               = _mm_mul_pd(r10,vftabscale);
2012             vfitab           = _mm_cvttpd_epi32(rt);
2013 #ifdef __XOP__
2014             vfeps            = _mm_frcz_pd(rt);
2015 #else
2016             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2017 #endif
2018             twovfeps         = _mm_add_pd(vfeps,vfeps);
2019             vfitab           = _mm_slli_epi32(vfitab,2);
2020
2021             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2022             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2023             F                = _mm_setzero_pd();
2024             GMX_MM_TRANSPOSE2_PD(Y,F);
2025             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2026             H                = _mm_setzero_pd();
2027             GMX_MM_TRANSPOSE2_PD(G,H);
2028             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2029             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2030             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
2031
2032             fscal            = felec;
2033
2034             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2035
2036             /* Update vectorial force */
2037             fix1             = _mm_macc_pd(dx10,fscal,fix1);
2038             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
2039             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
2040
2041             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
2042             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
2043             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
2044
2045             /**************************
2046              * CALCULATE INTERACTIONS *
2047              **************************/
2048
2049             r11              = _mm_mul_pd(rsq11,rinv11);
2050
2051             /* Calculate table index by multiplying r with table scale and truncate to integer */
2052             rt               = _mm_mul_pd(r11,vftabscale);
2053             vfitab           = _mm_cvttpd_epi32(rt);
2054 #ifdef __XOP__
2055             vfeps            = _mm_frcz_pd(rt);
2056 #else
2057             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2058 #endif
2059             twovfeps         = _mm_add_pd(vfeps,vfeps);
2060             vfitab           = _mm_slli_epi32(vfitab,2);
2061
2062             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2063             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2064             F                = _mm_setzero_pd();
2065             GMX_MM_TRANSPOSE2_PD(Y,F);
2066             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2067             H                = _mm_setzero_pd();
2068             GMX_MM_TRANSPOSE2_PD(G,H);
2069             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2070             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2071             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
2072
2073             fscal            = felec;
2074
2075             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2076
2077             /* Update vectorial force */
2078             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2079             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2080             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2081
2082             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2083             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2084             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2085
2086             /**************************
2087              * CALCULATE INTERACTIONS *
2088              **************************/
2089
2090             r12              = _mm_mul_pd(rsq12,rinv12);
2091
2092             /* Calculate table index by multiplying r with table scale and truncate to integer */
2093             rt               = _mm_mul_pd(r12,vftabscale);
2094             vfitab           = _mm_cvttpd_epi32(rt);
2095 #ifdef __XOP__
2096             vfeps            = _mm_frcz_pd(rt);
2097 #else
2098             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2099 #endif
2100             twovfeps         = _mm_add_pd(vfeps,vfeps);
2101             vfitab           = _mm_slli_epi32(vfitab,2);
2102
2103             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2104             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2105             F                = _mm_setzero_pd();
2106             GMX_MM_TRANSPOSE2_PD(Y,F);
2107             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2108             H                = _mm_setzero_pd();
2109             GMX_MM_TRANSPOSE2_PD(G,H);
2110             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2111             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2112             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
2113
2114             fscal            = felec;
2115
2116             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2117
2118             /* Update vectorial force */
2119             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2120             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2121             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2122
2123             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2124             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2125             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2126
2127             /**************************
2128              * CALCULATE INTERACTIONS *
2129              **************************/
2130
2131             r20              = _mm_mul_pd(rsq20,rinv20);
2132
2133             /* Calculate table index by multiplying r with table scale and truncate to integer */
2134             rt               = _mm_mul_pd(r20,vftabscale);
2135             vfitab           = _mm_cvttpd_epi32(rt);
2136 #ifdef __XOP__
2137             vfeps            = _mm_frcz_pd(rt);
2138 #else
2139             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2140 #endif
2141             twovfeps         = _mm_add_pd(vfeps,vfeps);
2142             vfitab           = _mm_slli_epi32(vfitab,2);
2143
2144             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2145             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2146             F                = _mm_setzero_pd();
2147             GMX_MM_TRANSPOSE2_PD(Y,F);
2148             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2149             H                = _mm_setzero_pd();
2150             GMX_MM_TRANSPOSE2_PD(G,H);
2151             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2152             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2153             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
2154
2155             fscal            = felec;
2156
2157             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2158
2159             /* Update vectorial force */
2160             fix2             = _mm_macc_pd(dx20,fscal,fix2);
2161             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
2162             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
2163
2164             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2165             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2166             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
2167
2168             /**************************
2169              * CALCULATE INTERACTIONS *
2170              **************************/
2171
2172             r21              = _mm_mul_pd(rsq21,rinv21);
2173
2174             /* Calculate table index by multiplying r with table scale and truncate to integer */
2175             rt               = _mm_mul_pd(r21,vftabscale);
2176             vfitab           = _mm_cvttpd_epi32(rt);
2177 #ifdef __XOP__
2178             vfeps            = _mm_frcz_pd(rt);
2179 #else
2180             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2181 #endif
2182             twovfeps         = _mm_add_pd(vfeps,vfeps);
2183             vfitab           = _mm_slli_epi32(vfitab,2);
2184
2185             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2186             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2187             F                = _mm_setzero_pd();
2188             GMX_MM_TRANSPOSE2_PD(Y,F);
2189             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2190             H                = _mm_setzero_pd();
2191             GMX_MM_TRANSPOSE2_PD(G,H);
2192             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2193             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2194             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
2195
2196             fscal            = felec;
2197
2198             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2199
2200             /* Update vectorial force */
2201             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2202             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2203             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2204
2205             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2206             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2207             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2208
2209             /**************************
2210              * CALCULATE INTERACTIONS *
2211              **************************/
2212
2213             r22              = _mm_mul_pd(rsq22,rinv22);
2214
2215             /* Calculate table index by multiplying r with table scale and truncate to integer */
2216             rt               = _mm_mul_pd(r22,vftabscale);
2217             vfitab           = _mm_cvttpd_epi32(rt);
2218 #ifdef __XOP__
2219             vfeps            = _mm_frcz_pd(rt);
2220 #else
2221             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2222 #endif
2223             twovfeps         = _mm_add_pd(vfeps,vfeps);
2224             vfitab           = _mm_slli_epi32(vfitab,2);
2225
2226             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2227             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2228             F                = _mm_setzero_pd();
2229             GMX_MM_TRANSPOSE2_PD(Y,F);
2230             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2231             H                = _mm_setzero_pd();
2232             GMX_MM_TRANSPOSE2_PD(G,H);
2233             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2234             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2235             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
2236
2237             fscal            = felec;
2238
2239             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2240
2241             /* Update vectorial force */
2242             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2243             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2244             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2245
2246             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2247             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2248             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2249
2250             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2251
2252             /* Inner loop uses 386 flops */
2253         }
2254
2255         /* End of innermost loop */
2256
2257         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2258                                               f+i_coord_offset,fshift+i_shift_offset);
2259
2260         /* Increment number of inner iterations */
2261         inneriter                  += j_index_end - j_index_start;
2262
2263         /* Outer loop uses 18 flops */
2264     }
2265
2266     /* Increment number of outer iterations */
2267     outeriter        += nri;
2268
2269     /* Update outer/inner flops */
2270
2271     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*386);
2272 }