src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwCSTab_GeomW3W3_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3W3_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Water3
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW3W3_VF_avx_128_fma_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     int              vdwjidx1A,vdwjidx1B;
  75     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  76     int              vdwjidx2A,vdwjidx2B;
  77     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  78     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  79     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  80     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  81     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  83     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  84     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  85     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  86     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  94     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  95     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  98     real             *vftab;
  99     __m128i          ewitab;
 100     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 101     real             *ewtab;
 102     __m128d          dummy_mask,cutoff_mask;
 103     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 104     __m128d          one     = _mm_set1_pd(1.0);
 105     __m128d          two     = _mm_set1_pd(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_pd(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_vdw->data;
 124     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 125
 126     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 134     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 135     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 136     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 137
 138     jq0              = _mm_set1_pd(charge[inr+0]);
 139     jq1              = _mm_set1_pd(charge[inr+1]);
 140     jq2              = _mm_set1_pd(charge[inr+2]);
 141     vdwjidx0A        = 2*vdwtype[inr+0];
 142     qq00             = _mm_mul_pd(iq0,jq0);
 143     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 144     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 145     qq01             = _mm_mul_pd(iq0,jq1);
 146     qq02             = _mm_mul_pd(iq0,jq2);
 147     qq10             = _mm_mul_pd(iq1,jq0);
 148     qq11             = _mm_mul_pd(iq1,jq1);
 149     qq12             = _mm_mul_pd(iq1,jq2);
 150     qq20             = _mm_mul_pd(iq2,jq0);
 151     qq21             = _mm_mul_pd(iq2,jq1);
 152     qq22             = _mm_mul_pd(iq2,jq2);
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158
 159     outeriter        = 0;
 160     inneriter        = 0;
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 179
 180         fix0             = _mm_setzero_pd();
 181         fiy0             = _mm_setzero_pd();
 182         fiz0             = _mm_setzero_pd();
 183         fix1             = _mm_setzero_pd();
 184         fiy1             = _mm_setzero_pd();
 185         fiz1             = _mm_setzero_pd();
 186         fix2             = _mm_setzero_pd();
 187         fiy2             = _mm_setzero_pd();
 188         fiz2             = _mm_setzero_pd();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm_setzero_pd();
 192         vvdwsum          = _mm_setzero_pd();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm_sub_pd(ix0,jx0);
 210             dy00             = _mm_sub_pd(iy0,jy0);
 211             dz00             = _mm_sub_pd(iz0,jz0);
 212             dx01             = _mm_sub_pd(ix0,jx1);
 213             dy01             = _mm_sub_pd(iy0,jy1);
 214             dz01             = _mm_sub_pd(iz0,jz1);
 215             dx02             = _mm_sub_pd(ix0,jx2);
 216             dy02             = _mm_sub_pd(iy0,jy2);
 217             dz02             = _mm_sub_pd(iz0,jz2);
 218             dx10             = _mm_sub_pd(ix1,jx0);
 219             dy10             = _mm_sub_pd(iy1,jy0);
 220             dz10             = _mm_sub_pd(iz1,jz0);
 221             dx11             = _mm_sub_pd(ix1,jx1);
 222             dy11             = _mm_sub_pd(iy1,jy1);
 223             dz11             = _mm_sub_pd(iz1,jz1);
 224             dx12             = _mm_sub_pd(ix1,jx2);
 225             dy12             = _mm_sub_pd(iy1,jy2);
 226             dz12             = _mm_sub_pd(iz1,jz2);
 227             dx20             = _mm_sub_pd(ix2,jx0);
 228             dy20             = _mm_sub_pd(iy2,jy0);
 229             dz20             = _mm_sub_pd(iz2,jz0);
 230             dx21             = _mm_sub_pd(ix2,jx1);
 231             dy21             = _mm_sub_pd(iy2,jy1);
 232             dz21             = _mm_sub_pd(iz2,jz1);
 233             dx22             = _mm_sub_pd(ix2,jx2);
 234             dy22             = _mm_sub_pd(iy2,jy2);
 235             dz22             = _mm_sub_pd(iz2,jz2);
 236
 237             /* Calculate squared distance and things based on it */
 238             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 239             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 240             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 241             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 242             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 243             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 244             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 245             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 246             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 247
 248             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 249             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 250             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 251             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 252             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 253             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 254             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 255             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 256             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 257
 258             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 259             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 260             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 261             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 262             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 263             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 264             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 265             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 266             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 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,3);
 294
 295             /* EWALD ELECTROSTATICS */
 296
 297             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 298             ewrt             = _mm_mul_pd(r00,ewtabscale);
 299             ewitab           = _mm_cvttpd_epi32(ewrt);
 300 #ifdef __XOP__
 301             eweps            = _mm_frcz_pd(ewrt);
 302 #else
 303             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 304 #endif
 305             twoeweps         = _mm_add_pd(eweps,eweps);
 306             ewitab           = _mm_slli_epi32(ewitab,2);
 307             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 308             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 309             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 310             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 311             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 312             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 313             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 314             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 315             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 316             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 317
 318             /* CUBIC SPLINE TABLE DISPERSION */
 319             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 320             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 321             GMX_MM_TRANSPOSE2_PD(Y,F);
 322             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 323             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 324             GMX_MM_TRANSPOSE2_PD(G,H);
 325             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 326             VV               = _mm_macc_pd(vfeps,Fp,Y);
 327             vvdw6            = _mm_mul_pd(c6_00,VV);
 328             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 329             fvdw6            = _mm_mul_pd(c6_00,FF);
 330
 331             /* CUBIC SPLINE TABLE REPULSION */
 332             vfitab           = _mm_add_epi32(vfitab,ifour);
 333             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 334             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 335             GMX_MM_TRANSPOSE2_PD(Y,F);
 336             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 337             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 338             GMX_MM_TRANSPOSE2_PD(G,H);
 339             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 340             VV               = _mm_macc_pd(vfeps,Fp,Y);
 341             vvdw12           = _mm_mul_pd(c12_00,VV);
 342             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 343             fvdw12           = _mm_mul_pd(c12_00,FF);
 344             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 345             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velecsum         = _mm_add_pd(velecsum,velec);
 349             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 350
 351             fscal            = _mm_add_pd(felec,fvdw);
 352
 353             /* Update vectorial force */
 354             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 355             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 356             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 357
 358             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 359             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 360             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             r01              = _mm_mul_pd(rsq01,rinv01);
 367
 368             /* EWALD ELECTROSTATICS */
 369
 370             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 371             ewrt             = _mm_mul_pd(r01,ewtabscale);
 372             ewitab           = _mm_cvttpd_epi32(ewrt);
 373 #ifdef __XOP__
 374             eweps            = _mm_frcz_pd(ewrt);
 375 #else
 376             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 377 #endif
 378             twoeweps         = _mm_add_pd(eweps,eweps);
 379             ewitab           = _mm_slli_epi32(ewitab,2);
 380             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 381             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 382             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 383             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 384             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 385             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 386             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 387             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 388             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
 389             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velecsum         = _mm_add_pd(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396             /* Update vectorial force */
 397             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 398             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 399             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 400
 401             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 402             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 403             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             r02              = _mm_mul_pd(rsq02,rinv02);
 410
 411             /* EWALD ELECTROSTATICS */
 412
 413             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 414             ewrt             = _mm_mul_pd(r02,ewtabscale);
 415             ewitab           = _mm_cvttpd_epi32(ewrt);
 416 #ifdef __XOP__
 417             eweps            = _mm_frcz_pd(ewrt);
 418 #else
 419             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 420 #endif
 421             twoeweps         = _mm_add_pd(eweps,eweps);
 422             ewitab           = _mm_slli_epi32(ewitab,2);
 423             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 424             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 425             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 426             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 427             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 428             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 429             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 430             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 431             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
 432             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velecsum         = _mm_add_pd(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439             /* Update vectorial force */
 440             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 441             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 442             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 443
 444             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 445             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 446             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 447
 448             /**************************
 449              * CALCULATE INTERACTIONS *
 450              **************************/
 451
 452             r10              = _mm_mul_pd(rsq10,rinv10);
 453
 454             /* EWALD ELECTROSTATICS */
 455
 456             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 457             ewrt             = _mm_mul_pd(r10,ewtabscale);
 458             ewitab           = _mm_cvttpd_epi32(ewrt);
 459 #ifdef __XOP__
 460             eweps            = _mm_frcz_pd(ewrt);
 461 #else
 462             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 463 #endif
 464             twoeweps         = _mm_add_pd(eweps,eweps);
 465             ewitab           = _mm_slli_epi32(ewitab,2);
 466             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 467             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 468             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 469             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 470             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 471             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 472             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 473             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 474             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 475             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 476
 477             /* Update potential sum for this i atom from the interaction with this j atom. */
 478             velecsum         = _mm_add_pd(velecsum,velec);
 479
 480             fscal            = felec;
 481
 482             /* Update vectorial force */
 483             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 484             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 485             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 486
 487             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 488             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 489             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             r11              = _mm_mul_pd(rsq11,rinv11);
 496
 497             /* EWALD ELECTROSTATICS */
 498
 499             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 500             ewrt             = _mm_mul_pd(r11,ewtabscale);
 501             ewitab           = _mm_cvttpd_epi32(ewrt);
 502 #ifdef __XOP__
 503             eweps            = _mm_frcz_pd(ewrt);
 504 #else
 505             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 506 #endif
 507             twoeweps         = _mm_add_pd(eweps,eweps);
 508             ewitab           = _mm_slli_epi32(ewitab,2);
 509             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 510             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 511             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 512             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 513             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 514             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 515             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 516             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 517             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 518             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velecsum         = _mm_add_pd(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525             /* Update vectorial force */
 526             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 527             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 528             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 529
 530             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 531             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 532             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             r12              = _mm_mul_pd(rsq12,rinv12);
 539
 540             /* EWALD ELECTROSTATICS */
 541
 542             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 543             ewrt             = _mm_mul_pd(r12,ewtabscale);
 544             ewitab           = _mm_cvttpd_epi32(ewrt);
 545 #ifdef __XOP__
 546             eweps            = _mm_frcz_pd(ewrt);
 547 #else
 548             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 549 #endif
 550             twoeweps         = _mm_add_pd(eweps,eweps);
 551             ewitab           = _mm_slli_epi32(ewitab,2);
 552             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 553             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 554             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 555             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 556             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 557             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 558             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 559             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 560             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 561             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             velecsum         = _mm_add_pd(velecsum,velec);
 565
 566             fscal            = felec;
 567
 568             /* Update vectorial force */
 569             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 570             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 571             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 572
 573             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 574             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 575             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             r20              = _mm_mul_pd(rsq20,rinv20);
 582
 583             /* EWALD ELECTROSTATICS */
 584
 585             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 586             ewrt             = _mm_mul_pd(r20,ewtabscale);
 587             ewitab           = _mm_cvttpd_epi32(ewrt);
 588 #ifdef __XOP__
 589             eweps            = _mm_frcz_pd(ewrt);
 590 #else
 591             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 592 #endif
 593             twoeweps         = _mm_add_pd(eweps,eweps);
 594             ewitab           = _mm_slli_epi32(ewitab,2);
 595             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 596             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 597             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 598             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 599             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 600             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 601             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 602             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 603             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 604             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 605
 606             /* Update potential sum for this i atom from the interaction with this j atom. */
 607             velecsum         = _mm_add_pd(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             /* Update vectorial force */
 612             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 613             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 614             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 615
 616             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 617             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 618             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             r21              = _mm_mul_pd(rsq21,rinv21);
 625
 626             /* EWALD ELECTROSTATICS */
 627
 628             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 629             ewrt             = _mm_mul_pd(r21,ewtabscale);
 630             ewitab           = _mm_cvttpd_epi32(ewrt);
 631 #ifdef __XOP__
 632             eweps            = _mm_frcz_pd(ewrt);
 633 #else
 634             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 635 #endif
 636             twoeweps         = _mm_add_pd(eweps,eweps);
 637             ewitab           = _mm_slli_epi32(ewitab,2);
 638             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 639             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 640             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 641             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 642             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 643             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 644             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 645             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 646             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 647             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 648
 649             /* Update potential sum for this i atom from the interaction with this j atom. */
 650             velecsum         = _mm_add_pd(velecsum,velec);
 651
 652             fscal            = felec;
 653
 654             /* Update vectorial force */
 655             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 656             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 657             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 658
 659             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 660             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 661             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 662
 663             /**************************
 664              * CALCULATE INTERACTIONS *
 665              **************************/
 666
 667             r22              = _mm_mul_pd(rsq22,rinv22);
 668
 669             /* EWALD ELECTROSTATICS */
 670
 671             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 672             ewrt             = _mm_mul_pd(r22,ewtabscale);
 673             ewitab           = _mm_cvttpd_epi32(ewrt);
 674 #ifdef __XOP__
 675             eweps            = _mm_frcz_pd(ewrt);
 676 #else
 677             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 678 #endif
 679             twoeweps         = _mm_add_pd(eweps,eweps);
 680             ewitab           = _mm_slli_epi32(ewitab,2);
 681             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 682             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 683             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 684             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 685             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 686             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 687             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 688             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 689             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 690             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 691
 692             /* Update potential sum for this i atom from the interaction with this j atom. */
 693             velecsum         = _mm_add_pd(velecsum,velec);
 694
 695             fscal            = felec;
 696
 697             /* Update vectorial force */
 698             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 699             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 700             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 701
 702             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 703             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 704             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 705
 706             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 707
 708             /* Inner loop uses 430 flops */
 709         }
 710
 711         if(jidx<j_index_end)
 712         {
 713
 714             jnrA             = jjnr[jidx];
 715             j_coord_offsetA  = DIM*jnrA;
 716
 717             /* load j atom coordinates */
 718             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 719                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 720
 721             /* Calculate displacement vector */
 722             dx00             = _mm_sub_pd(ix0,jx0);
 723             dy00             = _mm_sub_pd(iy0,jy0);
 724             dz00             = _mm_sub_pd(iz0,jz0);
 725             dx01             = _mm_sub_pd(ix0,jx1);
 726             dy01             = _mm_sub_pd(iy0,jy1);
 727             dz01             = _mm_sub_pd(iz0,jz1);
 728             dx02             = _mm_sub_pd(ix0,jx2);
 729             dy02             = _mm_sub_pd(iy0,jy2);
 730             dz02             = _mm_sub_pd(iz0,jz2);
 731             dx10             = _mm_sub_pd(ix1,jx0);
 732             dy10             = _mm_sub_pd(iy1,jy0);
 733             dz10             = _mm_sub_pd(iz1,jz0);
 734             dx11             = _mm_sub_pd(ix1,jx1);
 735             dy11             = _mm_sub_pd(iy1,jy1);
 736             dz11             = _mm_sub_pd(iz1,jz1);
 737             dx12             = _mm_sub_pd(ix1,jx2);
 738             dy12             = _mm_sub_pd(iy1,jy2);
 739             dz12             = _mm_sub_pd(iz1,jz2);
 740             dx20             = _mm_sub_pd(ix2,jx0);
 741             dy20             = _mm_sub_pd(iy2,jy0);
 742             dz20             = _mm_sub_pd(iz2,jz0);
 743             dx21             = _mm_sub_pd(ix2,jx1);
 744             dy21             = _mm_sub_pd(iy2,jy1);
 745             dz21             = _mm_sub_pd(iz2,jz1);
 746             dx22             = _mm_sub_pd(ix2,jx2);
 747             dy22             = _mm_sub_pd(iy2,jy2);
 748             dz22             = _mm_sub_pd(iz2,jz2);
 749
 750             /* Calculate squared distance and things based on it */
 751             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 752             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 753             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 754             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 755             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 756             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 757             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 758             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 759             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 760
 761             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 762             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 763             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 764             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 765             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 766             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 767             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 768             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 769             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 770
 771             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 772             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 773             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 774             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 775             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 776             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 777             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 778             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 779             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 780
 781             fjx0             = _mm_setzero_pd();
 782             fjy0             = _mm_setzero_pd();
 783             fjz0             = _mm_setzero_pd();
 784             fjx1             = _mm_setzero_pd();
 785             fjy1             = _mm_setzero_pd();
 786             fjz1             = _mm_setzero_pd();
 787             fjx2             = _mm_setzero_pd();
 788             fjy2             = _mm_setzero_pd();
 789             fjz2             = _mm_setzero_pd();
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             r00              = _mm_mul_pd(rsq00,rinv00);
 796
 797             /* Calculate table index by multiplying r with table scale and truncate to integer */
 798             rt               = _mm_mul_pd(r00,vftabscale);
 799             vfitab           = _mm_cvttpd_epi32(rt);
 800 #ifdef __XOP__
 801             vfeps            = _mm_frcz_pd(rt);
 802 #else
 803             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 804 #endif
 805             twovfeps         = _mm_add_pd(vfeps,vfeps);
 806             vfitab           = _mm_slli_epi32(vfitab,3);
 807
 808             /* EWALD ELECTROSTATICS */
 809
 810             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 811             ewrt             = _mm_mul_pd(r00,ewtabscale);
 812             ewitab           = _mm_cvttpd_epi32(ewrt);
 813 #ifdef __XOP__
 814             eweps            = _mm_frcz_pd(ewrt);
 815 #else
 816             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 817 #endif
 818             twoeweps         = _mm_add_pd(eweps,eweps);
 819             ewitab           = _mm_slli_epi32(ewitab,2);
 820             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 821             ewtabD           = _mm_setzero_pd();
 822             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 823             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 824             ewtabFn          = _mm_setzero_pd();
 825             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 826             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 827             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 828             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 829             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 830
 831             /* CUBIC SPLINE TABLE DISPERSION */
 832             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 833             F                = _mm_setzero_pd();
 834             GMX_MM_TRANSPOSE2_PD(Y,F);
 835             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 836             H                = _mm_setzero_pd();
 837             GMX_MM_TRANSPOSE2_PD(G,H);
 838             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 839             VV               = _mm_macc_pd(vfeps,Fp,Y);
 840             vvdw6            = _mm_mul_pd(c6_00,VV);
 841             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 842             fvdw6            = _mm_mul_pd(c6_00,FF);
 843
 844             /* CUBIC SPLINE TABLE REPULSION */
 845             vfitab           = _mm_add_epi32(vfitab,ifour);
 846             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 847             F                = _mm_setzero_pd();
 848             GMX_MM_TRANSPOSE2_PD(Y,F);
 849             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 850             H                = _mm_setzero_pd();
 851             GMX_MM_TRANSPOSE2_PD(G,H);
 852             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 853             VV               = _mm_macc_pd(vfeps,Fp,Y);
 854             vvdw12           = _mm_mul_pd(c12_00,VV);
 855             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 856             fvdw12           = _mm_mul_pd(c12_00,FF);
 857             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 858             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 859
 860             /* Update potential sum for this i atom from the interaction with this j atom. */
 861             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 862             velecsum         = _mm_add_pd(velecsum,velec);
 863             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 864             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 865
 866             fscal            = _mm_add_pd(felec,fvdw);
 867
 868             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 869
 870             /* Update vectorial force */
 871             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 872             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 873             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 874
 875             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 876             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 877             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             r01              = _mm_mul_pd(rsq01,rinv01);
 884
 885             /* EWALD ELECTROSTATICS */
 886
 887             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 888             ewrt             = _mm_mul_pd(r01,ewtabscale);
 889             ewitab           = _mm_cvttpd_epi32(ewrt);
 890 #ifdef __XOP__
 891             eweps            = _mm_frcz_pd(ewrt);
 892 #else
 893             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 894 #endif
 895             twoeweps         = _mm_add_pd(eweps,eweps);
 896             ewitab           = _mm_slli_epi32(ewitab,2);
 897             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 898             ewtabD           = _mm_setzero_pd();
 899             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 900             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 901             ewtabFn          = _mm_setzero_pd();
 902             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 903             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 904             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 905             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
 906             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 907
 908             /* Update potential sum for this i atom from the interaction with this j atom. */
 909             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 910             velecsum         = _mm_add_pd(velecsum,velec);
 911
 912             fscal            = felec;
 913
 914             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 915
 916             /* Update vectorial force */
 917             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 918             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 919             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 920
 921             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 922             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 923             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 924
 925             /**************************
 926              * CALCULATE INTERACTIONS *
 927              **************************/
 928
 929             r02              = _mm_mul_pd(rsq02,rinv02);
 930
 931             /* EWALD ELECTROSTATICS */
 932
 933             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 934             ewrt             = _mm_mul_pd(r02,ewtabscale);
 935             ewitab           = _mm_cvttpd_epi32(ewrt);
 936 #ifdef __XOP__
 937             eweps            = _mm_frcz_pd(ewrt);
 938 #else
 939             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 940 #endif
 941             twoeweps         = _mm_add_pd(eweps,eweps);
 942             ewitab           = _mm_slli_epi32(ewitab,2);
 943             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 944             ewtabD           = _mm_setzero_pd();
 945             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 946             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 947             ewtabFn          = _mm_setzero_pd();
 948             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 949             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 950             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 951             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
 952             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 953
 954             /* Update potential sum for this i atom from the interaction with this j atom. */
 955             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 956             velecsum         = _mm_add_pd(velecsum,velec);
 957
 958             fscal            = felec;
 959
 960             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 961
 962             /* Update vectorial force */
 963             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 964             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 965             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 966
 967             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 968             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 969             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             r10              = _mm_mul_pd(rsq10,rinv10);
 976
 977             /* EWALD ELECTROSTATICS */
 978
 979             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 980             ewrt             = _mm_mul_pd(r10,ewtabscale);
 981             ewitab           = _mm_cvttpd_epi32(ewrt);
 982 #ifdef __XOP__
 983             eweps            = _mm_frcz_pd(ewrt);
 984 #else
 985             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 986 #endif
 987             twoeweps         = _mm_add_pd(eweps,eweps);
 988             ewitab           = _mm_slli_epi32(ewitab,2);
 989             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 990             ewtabD           = _mm_setzero_pd();
 991             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 992             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 993             ewtabFn          = _mm_setzero_pd();
 994             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 995             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 996             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 997             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 998             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 999
1000             /* Update potential sum for this i atom from the interaction with this j atom. */
1001             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1002             velecsum         = _mm_add_pd(velecsum,velec);
1003
1004             fscal            = felec;
1005
1006             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1007
1008             /* Update vectorial force */
1009             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1010             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1011             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1012
1013             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1014             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1015             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1016
1017             /**************************
1018              * CALCULATE INTERACTIONS *
1019              **************************/
1020
1021             r11              = _mm_mul_pd(rsq11,rinv11);
1022
1023             /* EWALD ELECTROSTATICS */
1024
1025             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1026             ewrt             = _mm_mul_pd(r11,ewtabscale);
1027             ewitab           = _mm_cvttpd_epi32(ewrt);
1028 #ifdef __XOP__
1029             eweps            = _mm_frcz_pd(ewrt);
1030 #else
1031             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1032 #endif
1033             twoeweps         = _mm_add_pd(eweps,eweps);
1034             ewitab           = _mm_slli_epi32(ewitab,2);
1035             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1036             ewtabD           = _mm_setzero_pd();
1037             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1038             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1039             ewtabFn          = _mm_setzero_pd();
1040             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1041             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1042             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1043             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
1044             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1045
1046             /* Update potential sum for this i atom from the interaction with this j atom. */
1047             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1048             velecsum         = _mm_add_pd(velecsum,velec);
1049
1050             fscal            = felec;
1051
1052             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1053
1054             /* Update vectorial force */
1055             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1056             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1057             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1058
1059             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1060             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1061             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1062
1063             /**************************
1064              * CALCULATE INTERACTIONS *
1065              **************************/
1066
1067             r12              = _mm_mul_pd(rsq12,rinv12);
1068
1069             /* EWALD ELECTROSTATICS */
1070
1071             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1072             ewrt             = _mm_mul_pd(r12,ewtabscale);
1073             ewitab           = _mm_cvttpd_epi32(ewrt);
1074 #ifdef __XOP__
1075             eweps            = _mm_frcz_pd(ewrt);
1076 #else
1077             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1078 #endif
1079             twoeweps         = _mm_add_pd(eweps,eweps);
1080             ewitab           = _mm_slli_epi32(ewitab,2);
1081             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1082             ewtabD           = _mm_setzero_pd();
1083             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1084             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1085             ewtabFn          = _mm_setzero_pd();
1086             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1087             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1088             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1089             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
1090             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1091
1092             /* Update potential sum for this i atom from the interaction with this j atom. */
1093             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1094             velecsum         = _mm_add_pd(velecsum,velec);
1095
1096             fscal            = felec;
1097
1098             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1099
1100             /* Update vectorial force */
1101             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1102             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1103             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1104
1105             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1106             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1107             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             r20              = _mm_mul_pd(rsq20,rinv20);
1114
1115             /* EWALD ELECTROSTATICS */
1116
1117             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1118             ewrt             = _mm_mul_pd(r20,ewtabscale);
1119             ewitab           = _mm_cvttpd_epi32(ewrt);
1120 #ifdef __XOP__
1121             eweps            = _mm_frcz_pd(ewrt);
1122 #else
1123             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1124 #endif
1125             twoeweps         = _mm_add_pd(eweps,eweps);
1126             ewitab           = _mm_slli_epi32(ewitab,2);
1127             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1128             ewtabD           = _mm_setzero_pd();
1129             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1130             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1131             ewtabFn          = _mm_setzero_pd();
1132             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1133             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1134             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1135             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1136             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1137
1138             /* Update potential sum for this i atom from the interaction with this j atom. */
1139             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1140             velecsum         = _mm_add_pd(velecsum,velec);
1141
1142             fscal            = felec;
1143
1144             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1145
1146             /* Update vectorial force */
1147             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1148             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1149             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1150
1151             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1152             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1153             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1154
1155             /**************************
1156              * CALCULATE INTERACTIONS *
1157              **************************/
1158
1159             r21              = _mm_mul_pd(rsq21,rinv21);
1160
1161             /* EWALD ELECTROSTATICS */
1162
1163             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1164             ewrt             = _mm_mul_pd(r21,ewtabscale);
1165             ewitab           = _mm_cvttpd_epi32(ewrt);
1166 #ifdef __XOP__
1167             eweps            = _mm_frcz_pd(ewrt);
1168 #else
1169             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1170 #endif
1171             twoeweps         = _mm_add_pd(eweps,eweps);
1172             ewitab           = _mm_slli_epi32(ewitab,2);
1173             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1174             ewtabD           = _mm_setzero_pd();
1175             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1176             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1177             ewtabFn          = _mm_setzero_pd();
1178             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1179             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1180             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1181             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1182             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1183
1184             /* Update potential sum for this i atom from the interaction with this j atom. */
1185             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1186             velecsum         = _mm_add_pd(velecsum,velec);
1187
1188             fscal            = felec;
1189
1190             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1191
1192             /* Update vectorial force */
1193             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1194             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1195             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1196
1197             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1198             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1199             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             r22              = _mm_mul_pd(rsq22,rinv22);
1206
1207             /* EWALD ELECTROSTATICS */
1208
1209             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1210             ewrt             = _mm_mul_pd(r22,ewtabscale);
1211             ewitab           = _mm_cvttpd_epi32(ewrt);
1212 #ifdef __XOP__
1213             eweps            = _mm_frcz_pd(ewrt);
1214 #else
1215             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1216 #endif
1217             twoeweps         = _mm_add_pd(eweps,eweps);
1218             ewitab           = _mm_slli_epi32(ewitab,2);
1219             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1220             ewtabD           = _mm_setzero_pd();
1221             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1222             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1223             ewtabFn          = _mm_setzero_pd();
1224             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1225             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1226             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1227             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1228             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1229
1230             /* Update potential sum for this i atom from the interaction with this j atom. */
1231             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1232             velecsum         = _mm_add_pd(velecsum,velec);
1233
1234             fscal            = felec;
1235
1236             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1237
1238             /* Update vectorial force */
1239             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1240             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1241             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1242
1243             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1244             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1245             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1246
1247             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1248
1249             /* Inner loop uses 430 flops */
1250         }
1251
1252         /* End of innermost loop */
1253
1254         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1255                                               f+i_coord_offset,fshift+i_shift_offset);
1256
1257         ggid                        = gid[iidx];
1258         /* Update potential energies */
1259         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1260         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1261
1262         /* Increment number of inner iterations */
1263         inneriter                  += j_index_end - j_index_start;
1264
1265         /* Outer loop uses 20 flops */
1266     }
1267
1268     /* Increment number of outer iterations */
1269     outeriter        += nri;
1270
1271     /* Update outer/inner flops */
1272
1273     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*430);
1274 }
1275 /*
1276  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3W3_F_avx_128_fma_double
1277  * Electrostatics interaction: Ewald
1278  * VdW interaction:            CubicSplineTable
1279  * Geometry:                   Water3-Water3
1280  * Calculate force/pot:        Force
1281  */
1282 void
1283 nb_kernel_ElecEw_VdwCSTab_GeomW3W3_F_avx_128_fma_double
1284                     (t_nblist * gmx_restrict                nlist,
1285                      rvec * gmx_restrict                    xx,
1286                      rvec * gmx_restrict                    ff,
1287                      t_forcerec * gmx_restrict              fr,
1288                      t_mdatoms * gmx_restrict               mdatoms,
1289                      nb_kernel_data_t * gmx_restrict        kernel_data,
1290                      t_nrnb * gmx_restrict                  nrnb)
1291 {
1292     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1293      * just 0 for non-waters.
1294      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1295      * jnr indices corresponding to data put in the four positions in the SIMD register.
1296      */
1297     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1298     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1299     int              jnrA,jnrB;
1300     int              j_coord_offsetA,j_coord_offsetB;
1301     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1302     real             rcutoff_scalar;
1303     real             *shiftvec,*fshift,*x,*f;
1304     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1305     int              vdwioffset0;
1306     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1307     int              vdwioffset1;
1308     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1309     int              vdwioffset2;
1310     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1311     int              vdwjidx0A,vdwjidx0B;
1312     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1313     int              vdwjidx1A,vdwjidx1B;
1314     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1315     int              vdwjidx2A,vdwjidx2B;
1316     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1317     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1318     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1319     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1320     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1321     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1322     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1323     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1324     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1325     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1326     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1327     real             *charge;
1328     int              nvdwtype;
1329     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1330     int              *vdwtype;
1331     real             *vdwparam;
1332     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1333     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1334     __m128i          vfitab;
1335     __m128i          ifour       = _mm_set1_epi32(4);
1336     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
1337     real             *vftab;
1338     __m128i          ewitab;
1339     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1340     real             *ewtab;
1341     __m128d          dummy_mask,cutoff_mask;
1342     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1343     __m128d          one     = _mm_set1_pd(1.0);
1344     __m128d          two     = _mm_set1_pd(2.0);
1345     x                = xx[0];
1346     f                = ff[0];
1347
1348     nri              = nlist->nri;
1349     iinr             = nlist->iinr;
1350     jindex           = nlist->jindex;
1351     jjnr             = nlist->jjnr;
1352     shiftidx         = nlist->shift;
1353     gid              = nlist->gid;
1354     shiftvec         = fr->shift_vec[0];
1355     fshift           = fr->fshift[0];
1356     facel            = _mm_set1_pd(fr->epsfac);
1357     charge           = mdatoms->chargeA;
1358     nvdwtype         = fr->ntype;
1359     vdwparam         = fr->nbfp;
1360     vdwtype          = mdatoms->typeA;
1361
1362     vftab            = kernel_data->table_vdw->data;
1363     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
1364
1365     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1366     ewtab            = fr->ic->tabq_coul_F;
1367     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1368     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1369
1370     /* Setup water-specific parameters */
1371     inr              = nlist->iinr[0];
1372     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1373     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1374     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1375     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1376
1377     jq0              = _mm_set1_pd(charge[inr+0]);
1378     jq1              = _mm_set1_pd(charge[inr+1]);
1379     jq2              = _mm_set1_pd(charge[inr+2]);
1380     vdwjidx0A        = 2*vdwtype[inr+0];
1381     qq00             = _mm_mul_pd(iq0,jq0);
1382     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1383     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1384     qq01             = _mm_mul_pd(iq0,jq1);
1385     qq02             = _mm_mul_pd(iq0,jq2);
1386     qq10             = _mm_mul_pd(iq1,jq0);
1387     qq11             = _mm_mul_pd(iq1,jq1);
1388     qq12             = _mm_mul_pd(iq1,jq2);
1389     qq20             = _mm_mul_pd(iq2,jq0);
1390     qq21             = _mm_mul_pd(iq2,jq1);
1391     qq22             = _mm_mul_pd(iq2,jq2);
1392
1393     /* Avoid stupid compiler warnings */
1394     jnrA = jnrB = 0;
1395     j_coord_offsetA = 0;
1396     j_coord_offsetB = 0;
1397
1398     outeriter        = 0;
1399     inneriter        = 0;
1400
1401     /* Start outer loop over neighborlists */
1402     for(iidx=0; iidx<nri; iidx++)
1403     {
1404         /* Load shift vector for this list */
1405         i_shift_offset   = DIM*shiftidx[iidx];
1406
1407         /* Load limits for loop over neighbors */
1408         j_index_start    = jindex[iidx];
1409         j_index_end      = jindex[iidx+1];
1410
1411         /* Get outer coordinate index */
1412         inr              = iinr[iidx];
1413         i_coord_offset   = DIM*inr;
1414
1415         /* Load i particle coords and add shift vector */
1416         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1417                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1418
1419         fix0             = _mm_setzero_pd();
1420         fiy0             = _mm_setzero_pd();
1421         fiz0             = _mm_setzero_pd();
1422         fix1             = _mm_setzero_pd();
1423         fiy1             = _mm_setzero_pd();
1424         fiz1             = _mm_setzero_pd();
1425         fix2             = _mm_setzero_pd();
1426         fiy2             = _mm_setzero_pd();
1427         fiz2             = _mm_setzero_pd();
1428
1429         /* Start inner kernel loop */
1430         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1431         {
1432
1433             /* Get j neighbor index, and coordinate index */
1434             jnrA             = jjnr[jidx];
1435             jnrB             = jjnr[jidx+1];
1436             j_coord_offsetA  = DIM*jnrA;
1437             j_coord_offsetB  = DIM*jnrB;
1438
1439             /* load j atom coordinates */
1440             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1441                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1442
1443             /* Calculate displacement vector */
1444             dx00             = _mm_sub_pd(ix0,jx0);
1445             dy00             = _mm_sub_pd(iy0,jy0);
1446             dz00             = _mm_sub_pd(iz0,jz0);
1447             dx01             = _mm_sub_pd(ix0,jx1);
1448             dy01             = _mm_sub_pd(iy0,jy1);
1449             dz01             = _mm_sub_pd(iz0,jz1);
1450             dx02             = _mm_sub_pd(ix0,jx2);
1451             dy02             = _mm_sub_pd(iy0,jy2);
1452             dz02             = _mm_sub_pd(iz0,jz2);
1453             dx10             = _mm_sub_pd(ix1,jx0);
1454             dy10             = _mm_sub_pd(iy1,jy0);
1455             dz10             = _mm_sub_pd(iz1,jz0);
1456             dx11             = _mm_sub_pd(ix1,jx1);
1457             dy11             = _mm_sub_pd(iy1,jy1);
1458             dz11             = _mm_sub_pd(iz1,jz1);
1459             dx12             = _mm_sub_pd(ix1,jx2);
1460             dy12             = _mm_sub_pd(iy1,jy2);
1461             dz12             = _mm_sub_pd(iz1,jz2);
1462             dx20             = _mm_sub_pd(ix2,jx0);
1463             dy20             = _mm_sub_pd(iy2,jy0);
1464             dz20             = _mm_sub_pd(iz2,jz0);
1465             dx21             = _mm_sub_pd(ix2,jx1);
1466             dy21             = _mm_sub_pd(iy2,jy1);
1467             dz21             = _mm_sub_pd(iz2,jz1);
1468             dx22             = _mm_sub_pd(ix2,jx2);
1469             dy22             = _mm_sub_pd(iy2,jy2);
1470             dz22             = _mm_sub_pd(iz2,jz2);
1471
1472             /* Calculate squared distance and things based on it */
1473             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1474             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1475             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1476             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1477             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1478             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1479             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1480             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1481             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1482
1483             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1484             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1485             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1486             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1487             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1488             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1489             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1490             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1491             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1492
1493             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1494             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1495             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1496             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1497             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1498             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1499             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1500             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1501             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1502
1503             fjx0             = _mm_setzero_pd();
1504             fjy0             = _mm_setzero_pd();
1505             fjz0             = _mm_setzero_pd();
1506             fjx1             = _mm_setzero_pd();
1507             fjy1             = _mm_setzero_pd();
1508             fjz1             = _mm_setzero_pd();
1509             fjx2             = _mm_setzero_pd();
1510             fjy2             = _mm_setzero_pd();
1511             fjz2             = _mm_setzero_pd();
1512
1513             /**************************
1514              * CALCULATE INTERACTIONS *
1515              **************************/
1516
1517             r00              = _mm_mul_pd(rsq00,rinv00);
1518
1519             /* Calculate table index by multiplying r with table scale and truncate to integer */
1520             rt               = _mm_mul_pd(r00,vftabscale);
1521             vfitab           = _mm_cvttpd_epi32(rt);
1522 #ifdef __XOP__
1523             vfeps            = _mm_frcz_pd(rt);
1524 #else
1525             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1526 #endif
1527             twovfeps         = _mm_add_pd(vfeps,vfeps);
1528             vfitab           = _mm_slli_epi32(vfitab,3);
1529
1530             /* EWALD ELECTROSTATICS */
1531
1532             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1533             ewrt             = _mm_mul_pd(r00,ewtabscale);
1534             ewitab           = _mm_cvttpd_epi32(ewrt);
1535 #ifdef __XOP__
1536             eweps            = _mm_frcz_pd(ewrt);
1537 #else
1538             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1539 #endif
1540             twoeweps         = _mm_add_pd(eweps,eweps);
1541             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1542                                          &ewtabF,&ewtabFn);
1543             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1544             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1545
1546             /* CUBIC SPLINE TABLE DISPERSION */
1547             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1548             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1549             GMX_MM_TRANSPOSE2_PD(Y,F);
1550             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1551             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1552             GMX_MM_TRANSPOSE2_PD(G,H);
1553             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1554             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1555             fvdw6            = _mm_mul_pd(c6_00,FF);
1556
1557             /* CUBIC SPLINE TABLE REPULSION */
1558             vfitab           = _mm_add_epi32(vfitab,ifour);
1559             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1560             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1561             GMX_MM_TRANSPOSE2_PD(Y,F);
1562             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1563             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1564             GMX_MM_TRANSPOSE2_PD(G,H);
1565             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1566             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1567             fvdw12           = _mm_mul_pd(c12_00,FF);
1568             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1569
1570             fscal            = _mm_add_pd(felec,fvdw);
1571
1572             /* Update vectorial force */
1573             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1574             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1575             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1576
1577             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1578             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1579             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1580
1581             /**************************
1582              * CALCULATE INTERACTIONS *
1583              **************************/
1584
1585             r01              = _mm_mul_pd(rsq01,rinv01);
1586
1587             /* EWALD ELECTROSTATICS */
1588
1589             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1590             ewrt             = _mm_mul_pd(r01,ewtabscale);
1591             ewitab           = _mm_cvttpd_epi32(ewrt);
1592 #ifdef __XOP__
1593             eweps            = _mm_frcz_pd(ewrt);
1594 #else
1595             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1596 #endif
1597             twoeweps         = _mm_add_pd(eweps,eweps);
1598             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1599                                          &ewtabF,&ewtabFn);
1600             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1601             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1602
1603             fscal            = felec;
1604
1605             /* Update vectorial force */
1606             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1607             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1608             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1609
1610             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1611             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1612             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1613
1614             /**************************
1615              * CALCULATE INTERACTIONS *
1616              **************************/
1617
1618             r02              = _mm_mul_pd(rsq02,rinv02);
1619
1620             /* EWALD ELECTROSTATICS */
1621
1622             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1623             ewrt             = _mm_mul_pd(r02,ewtabscale);
1624             ewitab           = _mm_cvttpd_epi32(ewrt);
1625 #ifdef __XOP__
1626             eweps            = _mm_frcz_pd(ewrt);
1627 #else
1628             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1629 #endif
1630             twoeweps         = _mm_add_pd(eweps,eweps);
1631             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1632                                          &ewtabF,&ewtabFn);
1633             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1634             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1635
1636             fscal            = felec;
1637
1638             /* Update vectorial force */
1639             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1640             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1641             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1642
1643             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1644             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1645             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1646
1647             /**************************
1648              * CALCULATE INTERACTIONS *
1649              **************************/
1650
1651             r10              = _mm_mul_pd(rsq10,rinv10);
1652
1653             /* EWALD ELECTROSTATICS */
1654
1655             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1656             ewrt             = _mm_mul_pd(r10,ewtabscale);
1657             ewitab           = _mm_cvttpd_epi32(ewrt);
1658 #ifdef __XOP__
1659             eweps            = _mm_frcz_pd(ewrt);
1660 #else
1661             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1662 #endif
1663             twoeweps         = _mm_add_pd(eweps,eweps);
1664             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1665                                          &ewtabF,&ewtabFn);
1666             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1667             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1668
1669             fscal            = felec;
1670
1671             /* Update vectorial force */
1672             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1673             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1674             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1675
1676             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1677             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1678             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1679
1680             /**************************
1681              * CALCULATE INTERACTIONS *
1682              **************************/
1683
1684             r11              = _mm_mul_pd(rsq11,rinv11);
1685
1686             /* EWALD ELECTROSTATICS */
1687
1688             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1689             ewrt             = _mm_mul_pd(r11,ewtabscale);
1690             ewitab           = _mm_cvttpd_epi32(ewrt);
1691 #ifdef __XOP__
1692             eweps            = _mm_frcz_pd(ewrt);
1693 #else
1694             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1695 #endif
1696             twoeweps         = _mm_add_pd(eweps,eweps);
1697             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1698                                          &ewtabF,&ewtabFn);
1699             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1700             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1701
1702             fscal            = felec;
1703
1704             /* Update vectorial force */
1705             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1706             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1707             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1708
1709             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1710             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1711             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1712
1713             /**************************
1714              * CALCULATE INTERACTIONS *
1715              **************************/
1716
1717             r12              = _mm_mul_pd(rsq12,rinv12);
1718
1719             /* EWALD ELECTROSTATICS */
1720
1721             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1722             ewrt             = _mm_mul_pd(r12,ewtabscale);
1723             ewitab           = _mm_cvttpd_epi32(ewrt);
1724 #ifdef __XOP__
1725             eweps            = _mm_frcz_pd(ewrt);
1726 #else
1727             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1728 #endif
1729             twoeweps         = _mm_add_pd(eweps,eweps);
1730             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1731                                          &ewtabF,&ewtabFn);
1732             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1733             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1734
1735             fscal            = felec;
1736
1737             /* Update vectorial force */
1738             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1739             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1740             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1741
1742             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1743             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1744             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1745
1746             /**************************
1747              * CALCULATE INTERACTIONS *
1748              **************************/
1749
1750             r20              = _mm_mul_pd(rsq20,rinv20);
1751
1752             /* EWALD ELECTROSTATICS */
1753
1754             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1755             ewrt             = _mm_mul_pd(r20,ewtabscale);
1756             ewitab           = _mm_cvttpd_epi32(ewrt);
1757 #ifdef __XOP__
1758             eweps            = _mm_frcz_pd(ewrt);
1759 #else
1760             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1761 #endif
1762             twoeweps         = _mm_add_pd(eweps,eweps);
1763             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1764                                          &ewtabF,&ewtabFn);
1765             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1766             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1767
1768             fscal            = felec;
1769
1770             /* Update vectorial force */
1771             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1772             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1773             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1774
1775             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1776             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1777             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1778
1779             /**************************
1780              * CALCULATE INTERACTIONS *
1781              **************************/
1782
1783             r21              = _mm_mul_pd(rsq21,rinv21);
1784
1785             /* EWALD ELECTROSTATICS */
1786
1787             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1788             ewrt             = _mm_mul_pd(r21,ewtabscale);
1789             ewitab           = _mm_cvttpd_epi32(ewrt);
1790 #ifdef __XOP__
1791             eweps            = _mm_frcz_pd(ewrt);
1792 #else
1793             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1794 #endif
1795             twoeweps         = _mm_add_pd(eweps,eweps);
1796             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1797                                          &ewtabF,&ewtabFn);
1798             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1799             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1800
1801             fscal            = felec;
1802
1803             /* Update vectorial force */
1804             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1805             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1806             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1807
1808             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1809             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1810             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1811
1812             /**************************
1813              * CALCULATE INTERACTIONS *
1814              **************************/
1815
1816             r22              = _mm_mul_pd(rsq22,rinv22);
1817
1818             /* EWALD ELECTROSTATICS */
1819
1820             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1821             ewrt             = _mm_mul_pd(r22,ewtabscale);
1822             ewitab           = _mm_cvttpd_epi32(ewrt);
1823 #ifdef __XOP__
1824             eweps            = _mm_frcz_pd(ewrt);
1825 #else
1826             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1827 #endif
1828             twoeweps         = _mm_add_pd(eweps,eweps);
1829             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1830                                          &ewtabF,&ewtabFn);
1831             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1832             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1833
1834             fscal            = felec;
1835
1836             /* Update vectorial force */
1837             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1838             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1839             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1840
1841             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1842             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1843             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1844
1845             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1846
1847             /* Inner loop uses 377 flops */
1848         }
1849
1850         if(jidx<j_index_end)
1851         {
1852
1853             jnrA             = jjnr[jidx];
1854             j_coord_offsetA  = DIM*jnrA;
1855
1856             /* load j atom coordinates */
1857             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1858                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1859
1860             /* Calculate displacement vector */
1861             dx00             = _mm_sub_pd(ix0,jx0);
1862             dy00             = _mm_sub_pd(iy0,jy0);
1863             dz00             = _mm_sub_pd(iz0,jz0);
1864             dx01             = _mm_sub_pd(ix0,jx1);
1865             dy01             = _mm_sub_pd(iy0,jy1);
1866             dz01             = _mm_sub_pd(iz0,jz1);
1867             dx02             = _mm_sub_pd(ix0,jx2);
1868             dy02             = _mm_sub_pd(iy0,jy2);
1869             dz02             = _mm_sub_pd(iz0,jz2);
1870             dx10             = _mm_sub_pd(ix1,jx0);
1871             dy10             = _mm_sub_pd(iy1,jy0);
1872             dz10             = _mm_sub_pd(iz1,jz0);
1873             dx11             = _mm_sub_pd(ix1,jx1);
1874             dy11             = _mm_sub_pd(iy1,jy1);
1875             dz11             = _mm_sub_pd(iz1,jz1);
1876             dx12             = _mm_sub_pd(ix1,jx2);
1877             dy12             = _mm_sub_pd(iy1,jy2);
1878             dz12             = _mm_sub_pd(iz1,jz2);
1879             dx20             = _mm_sub_pd(ix2,jx0);
1880             dy20             = _mm_sub_pd(iy2,jy0);
1881             dz20             = _mm_sub_pd(iz2,jz0);
1882             dx21             = _mm_sub_pd(ix2,jx1);
1883             dy21             = _mm_sub_pd(iy2,jy1);
1884             dz21             = _mm_sub_pd(iz2,jz1);
1885             dx22             = _mm_sub_pd(ix2,jx2);
1886             dy22             = _mm_sub_pd(iy2,jy2);
1887             dz22             = _mm_sub_pd(iz2,jz2);
1888
1889             /* Calculate squared distance and things based on it */
1890             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1891             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1892             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1893             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1894             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1895             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1896             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1897             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1898             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1899
1900             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1901             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1902             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1903             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1904             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1905             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1906             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1907             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1908             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1909
1910             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1911             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1912             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1913             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1914             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1915             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1916             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1917             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1918             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1919
1920             fjx0             = _mm_setzero_pd();
1921             fjy0             = _mm_setzero_pd();
1922             fjz0             = _mm_setzero_pd();
1923             fjx1             = _mm_setzero_pd();
1924             fjy1             = _mm_setzero_pd();
1925             fjz1             = _mm_setzero_pd();
1926             fjx2             = _mm_setzero_pd();
1927             fjy2             = _mm_setzero_pd();
1928             fjz2             = _mm_setzero_pd();
1929
1930             /**************************
1931              * CALCULATE INTERACTIONS *
1932              **************************/
1933
1934             r00              = _mm_mul_pd(rsq00,rinv00);
1935
1936             /* Calculate table index by multiplying r with table scale and truncate to integer */
1937             rt               = _mm_mul_pd(r00,vftabscale);
1938             vfitab           = _mm_cvttpd_epi32(rt);
1939 #ifdef __XOP__
1940             vfeps            = _mm_frcz_pd(rt);
1941 #else
1942             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1943 #endif
1944             twovfeps         = _mm_add_pd(vfeps,vfeps);
1945             vfitab           = _mm_slli_epi32(vfitab,3);
1946
1947             /* EWALD ELECTROSTATICS */
1948
1949             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1950             ewrt             = _mm_mul_pd(r00,ewtabscale);
1951             ewitab           = _mm_cvttpd_epi32(ewrt);
1952 #ifdef __XOP__
1953             eweps            = _mm_frcz_pd(ewrt);
1954 #else
1955             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1956 #endif
1957             twoeweps         = _mm_add_pd(eweps,eweps);
1958             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1959             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1960             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1961
1962             /* CUBIC SPLINE TABLE DISPERSION */
1963             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1964             F                = _mm_setzero_pd();
1965             GMX_MM_TRANSPOSE2_PD(Y,F);
1966             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1967             H                = _mm_setzero_pd();
1968             GMX_MM_TRANSPOSE2_PD(G,H);
1969             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1970             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1971             fvdw6            = _mm_mul_pd(c6_00,FF);
1972
1973             /* CUBIC SPLINE TABLE REPULSION */
1974             vfitab           = _mm_add_epi32(vfitab,ifour);
1975             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1976             F                = _mm_setzero_pd();
1977             GMX_MM_TRANSPOSE2_PD(Y,F);
1978             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1979             H                = _mm_setzero_pd();
1980             GMX_MM_TRANSPOSE2_PD(G,H);
1981             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1982             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1983             fvdw12           = _mm_mul_pd(c12_00,FF);
1984             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1985
1986             fscal            = _mm_add_pd(felec,fvdw);
1987
1988             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1989
1990             /* Update vectorial force */
1991             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1992             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1993             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1994
1995             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1996             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1997             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1998
1999             /**************************
2000              * CALCULATE INTERACTIONS *
2001              **************************/
2002
2003             r01              = _mm_mul_pd(rsq01,rinv01);
2004
2005             /* EWALD ELECTROSTATICS */
2006
2007             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2008             ewrt             = _mm_mul_pd(r01,ewtabscale);
2009             ewitab           = _mm_cvttpd_epi32(ewrt);
2010 #ifdef __XOP__
2011             eweps            = _mm_frcz_pd(ewrt);
2012 #else
2013             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2014 #endif
2015             twoeweps         = _mm_add_pd(eweps,eweps);
2016             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2017             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2018             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
2019
2020             fscal            = felec;
2021
2022             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2023
2024             /* Update vectorial force */
2025             fix0             = _mm_macc_pd(dx01,fscal,fix0);
2026             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
2027             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
2028
2029             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
2030             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
2031             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
2032
2033             /**************************
2034              * CALCULATE INTERACTIONS *
2035              **************************/
2036
2037             r02              = _mm_mul_pd(rsq02,rinv02);
2038
2039             /* EWALD ELECTROSTATICS */
2040
2041             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2042             ewrt             = _mm_mul_pd(r02,ewtabscale);
2043             ewitab           = _mm_cvttpd_epi32(ewrt);
2044 #ifdef __XOP__
2045             eweps            = _mm_frcz_pd(ewrt);
2046 #else
2047             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2048 #endif
2049             twoeweps         = _mm_add_pd(eweps,eweps);
2050             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2051             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2052             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
2053
2054             fscal            = felec;
2055
2056             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2057
2058             /* Update vectorial force */
2059             fix0             = _mm_macc_pd(dx02,fscal,fix0);
2060             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
2061             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
2062
2063             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
2064             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
2065             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
2066
2067             /**************************
2068              * CALCULATE INTERACTIONS *
2069              **************************/
2070
2071             r10              = _mm_mul_pd(rsq10,rinv10);
2072
2073             /* EWALD ELECTROSTATICS */
2074
2075             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2076             ewrt             = _mm_mul_pd(r10,ewtabscale);
2077             ewitab           = _mm_cvttpd_epi32(ewrt);
2078 #ifdef __XOP__
2079             eweps            = _mm_frcz_pd(ewrt);
2080 #else
2081             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2082 #endif
2083             twoeweps         = _mm_add_pd(eweps,eweps);
2084             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2085             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2086             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2087
2088             fscal            = felec;
2089
2090             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2091
2092             /* Update vectorial force */
2093             fix1             = _mm_macc_pd(dx10,fscal,fix1);
2094             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
2095             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
2096
2097             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
2098             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
2099             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
2100
2101             /**************************
2102              * CALCULATE INTERACTIONS *
2103              **************************/
2104
2105             r11              = _mm_mul_pd(rsq11,rinv11);
2106
2107             /* EWALD ELECTROSTATICS */
2108
2109             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2110             ewrt             = _mm_mul_pd(r11,ewtabscale);
2111             ewitab           = _mm_cvttpd_epi32(ewrt);
2112 #ifdef __XOP__
2113             eweps            = _mm_frcz_pd(ewrt);
2114 #else
2115             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2116 #endif
2117             twoeweps         = _mm_add_pd(eweps,eweps);
2118             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2119             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2120             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2121
2122             fscal            = felec;
2123
2124             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2125
2126             /* Update vectorial force */
2127             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2128             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2129             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2130
2131             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2132             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2133             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2134
2135             /**************************
2136              * CALCULATE INTERACTIONS *
2137              **************************/
2138
2139             r12              = _mm_mul_pd(rsq12,rinv12);
2140
2141             /* EWALD ELECTROSTATICS */
2142
2143             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2144             ewrt             = _mm_mul_pd(r12,ewtabscale);
2145             ewitab           = _mm_cvttpd_epi32(ewrt);
2146 #ifdef __XOP__
2147             eweps            = _mm_frcz_pd(ewrt);
2148 #else
2149             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2150 #endif
2151             twoeweps         = _mm_add_pd(eweps,eweps);
2152             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2153             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2154             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2155
2156             fscal            = felec;
2157
2158             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2159
2160             /* Update vectorial force */
2161             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2162             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2163             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2164
2165             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2166             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2167             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2168
2169             /**************************
2170              * CALCULATE INTERACTIONS *
2171              **************************/
2172
2173             r20              = _mm_mul_pd(rsq20,rinv20);
2174
2175             /* EWALD ELECTROSTATICS */
2176
2177             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2178             ewrt             = _mm_mul_pd(r20,ewtabscale);
2179             ewitab           = _mm_cvttpd_epi32(ewrt);
2180 #ifdef __XOP__
2181             eweps            = _mm_frcz_pd(ewrt);
2182 #else
2183             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2184 #endif
2185             twoeweps         = _mm_add_pd(eweps,eweps);
2186             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2187             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2188             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2189
2190             fscal            = felec;
2191
2192             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2193
2194             /* Update vectorial force */
2195             fix2             = _mm_macc_pd(dx20,fscal,fix2);
2196             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
2197             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
2198
2199             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2200             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2201             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
2202
2203             /**************************
2204              * CALCULATE INTERACTIONS *
2205              **************************/
2206
2207             r21              = _mm_mul_pd(rsq21,rinv21);
2208
2209             /* EWALD ELECTROSTATICS */
2210
2211             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2212             ewrt             = _mm_mul_pd(r21,ewtabscale);
2213             ewitab           = _mm_cvttpd_epi32(ewrt);
2214 #ifdef __XOP__
2215             eweps            = _mm_frcz_pd(ewrt);
2216 #else
2217             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2218 #endif
2219             twoeweps         = _mm_add_pd(eweps,eweps);
2220             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2221             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2222             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2223
2224             fscal            = felec;
2225
2226             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2227
2228             /* Update vectorial force */
2229             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2230             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2231             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2232
2233             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2234             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2235             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2236
2237             /**************************
2238              * CALCULATE INTERACTIONS *
2239              **************************/
2240
2241             r22              = _mm_mul_pd(rsq22,rinv22);
2242
2243             /* EWALD ELECTROSTATICS */
2244
2245             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2246             ewrt             = _mm_mul_pd(r22,ewtabscale);
2247             ewitab           = _mm_cvttpd_epi32(ewrt);
2248 #ifdef __XOP__
2249             eweps            = _mm_frcz_pd(ewrt);
2250 #else
2251             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2252 #endif
2253             twoeweps         = _mm_add_pd(eweps,eweps);
2254             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2255             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2256             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2257
2258             fscal            = felec;
2259
2260             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2261
2262             /* Update vectorial force */
2263             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2264             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2265             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2266
2267             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2268             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2269             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2270
2271             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2272
2273             /* Inner loop uses 377 flops */
2274         }
2275
2276         /* End of innermost loop */
2277
2278         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2279                                               f+i_coord_offset,fshift+i_shift_offset);
2280
2281         /* Increment number of inner iterations */
2282         inneriter                  += j_index_end - j_index_start;
2283
2284         /* Outer loop uses 18 flops */
2285     }
2286
2287     /* Increment number of outer iterations */
2288     outeriter        += nri;
2289
2290     /* Update outer/inner flops */
2291
2292     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*377);
2293 }