src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJ_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_VdwLJ_GeomW3W3_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Water3
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwLJ_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          ewitab;
  96     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     real             *ewtab;
  98     __m128d          dummy_mask,cutoff_mask;
  99     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 100     __m128d          one     = _mm_set1_pd(1.0);
 101     __m128d          two     = _mm_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 127     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 128     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 129     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 130
 131     jq0              = _mm_set1_pd(charge[inr+0]);
 132     jq1              = _mm_set1_pd(charge[inr+1]);
 133     jq2              = _mm_set1_pd(charge[inr+2]);
 134     vdwjidx0A        = 2*vdwtype[inr+0];
 135     qq00             = _mm_mul_pd(iq0,jq0);
 136     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 137     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 138     qq01             = _mm_mul_pd(iq0,jq1);
 139     qq02             = _mm_mul_pd(iq0,jq2);
 140     qq10             = _mm_mul_pd(iq1,jq0);
 141     qq11             = _mm_mul_pd(iq1,jq1);
 142     qq12             = _mm_mul_pd(iq1,jq2);
 143     qq20             = _mm_mul_pd(iq2,jq0);
 144     qq21             = _mm_mul_pd(iq2,jq1);
 145     qq22             = _mm_mul_pd(iq2,jq2);
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 172
 173         fix0             = _mm_setzero_pd();
 174         fiy0             = _mm_setzero_pd();
 175         fiz0             = _mm_setzero_pd();
 176         fix1             = _mm_setzero_pd();
 177         fiy1             = _mm_setzero_pd();
 178         fiz1             = _mm_setzero_pd();
 179         fix2             = _mm_setzero_pd();
 180         fiy2             = _mm_setzero_pd();
 181         fiz2             = _mm_setzero_pd();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm_setzero_pd();
 185         vvdwsum          = _mm_setzero_pd();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 200
 201             /* Calculate displacement vector */
 202             dx00             = _mm_sub_pd(ix0,jx0);
 203             dy00             = _mm_sub_pd(iy0,jy0);
 204             dz00             = _mm_sub_pd(iz0,jz0);
 205             dx01             = _mm_sub_pd(ix0,jx1);
 206             dy01             = _mm_sub_pd(iy0,jy1);
 207             dz01             = _mm_sub_pd(iz0,jz1);
 208             dx02             = _mm_sub_pd(ix0,jx2);
 209             dy02             = _mm_sub_pd(iy0,jy2);
 210             dz02             = _mm_sub_pd(iz0,jz2);
 211             dx10             = _mm_sub_pd(ix1,jx0);
 212             dy10             = _mm_sub_pd(iy1,jy0);
 213             dz10             = _mm_sub_pd(iz1,jz0);
 214             dx11             = _mm_sub_pd(ix1,jx1);
 215             dy11             = _mm_sub_pd(iy1,jy1);
 216             dz11             = _mm_sub_pd(iz1,jz1);
 217             dx12             = _mm_sub_pd(ix1,jx2);
 218             dy12             = _mm_sub_pd(iy1,jy2);
 219             dz12             = _mm_sub_pd(iz1,jz2);
 220             dx20             = _mm_sub_pd(ix2,jx0);
 221             dy20             = _mm_sub_pd(iy2,jy0);
 222             dz20             = _mm_sub_pd(iz2,jz0);
 223             dx21             = _mm_sub_pd(ix2,jx1);
 224             dy21             = _mm_sub_pd(iy2,jy1);
 225             dz21             = _mm_sub_pd(iz2,jz1);
 226             dx22             = _mm_sub_pd(ix2,jx2);
 227             dy22             = _mm_sub_pd(iy2,jy2);
 228             dz22             = _mm_sub_pd(iz2,jz2);
 229
 230             /* Calculate squared distance and things based on it */
 231             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 232             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 233             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 234             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 235             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 236             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 237             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 238             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 239             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 240
 241             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 242             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 243             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 244             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 245             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 246             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 247             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 248             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 249             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 250
 251             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 252             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 253             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 254             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 255             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 256             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 257             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 258             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 259             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 260
 261             fjx0             = _mm_setzero_pd();
 262             fjy0             = _mm_setzero_pd();
 263             fjz0             = _mm_setzero_pd();
 264             fjx1             = _mm_setzero_pd();
 265             fjy1             = _mm_setzero_pd();
 266             fjz1             = _mm_setzero_pd();
 267             fjx2             = _mm_setzero_pd();
 268             fjy2             = _mm_setzero_pd();
 269             fjz2             = _mm_setzero_pd();
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             r00              = _mm_mul_pd(rsq00,rinv00);
 276
 277             /* EWALD ELECTROSTATICS */
 278
 279             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 280             ewrt             = _mm_mul_pd(r00,ewtabscale);
 281             ewitab           = _mm_cvttpd_epi32(ewrt);
 282 #ifdef __XOP__
 283             eweps            = _mm_frcz_pd(ewrt);
 284 #else
 285             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 286 #endif
 287             twoeweps         = _mm_add_pd(eweps,eweps);
 288             ewitab           = _mm_slli_epi32(ewitab,2);
 289             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 290             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 291             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 292             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 293             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 294             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 295             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 296             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 297             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 298             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 299
 300             /* LENNARD-JONES DISPERSION/REPULSION */
 301
 302             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 303             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 304             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 305             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 306             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             velecsum         = _mm_add_pd(velecsum,velec);
 310             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 311
 312             fscal            = _mm_add_pd(felec,fvdw);
 313
 314             /* Update vectorial force */
 315             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 316             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 317             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 318
 319             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 320             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 321             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 322
 323             /**************************
 324              * CALCULATE INTERACTIONS *
 325              **************************/
 326
 327             r01              = _mm_mul_pd(rsq01,rinv01);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 332             ewrt             = _mm_mul_pd(r01,ewtabscale);
 333             ewitab           = _mm_cvttpd_epi32(ewrt);
 334 #ifdef __XOP__
 335             eweps            = _mm_frcz_pd(ewrt);
 336 #else
 337             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 338 #endif
 339             twoeweps         = _mm_add_pd(eweps,eweps);
 340             ewitab           = _mm_slli_epi32(ewitab,2);
 341             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 342             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 343             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 344             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 345             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 346             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 347             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 348             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 349             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
 350             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velecsum         = _mm_add_pd(velecsum,velec);
 354
 355             fscal            = felec;
 356
 357             /* Update vectorial force */
 358             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 359             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 360             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 361
 362             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 363             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 364             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 365
 366             /**************************
 367              * CALCULATE INTERACTIONS *
 368              **************************/
 369
 370             r02              = _mm_mul_pd(rsq02,rinv02);
 371
 372             /* EWALD ELECTROSTATICS */
 373
 374             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 375             ewrt             = _mm_mul_pd(r02,ewtabscale);
 376             ewitab           = _mm_cvttpd_epi32(ewrt);
 377 #ifdef __XOP__
 378             eweps            = _mm_frcz_pd(ewrt);
 379 #else
 380             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 381 #endif
 382             twoeweps         = _mm_add_pd(eweps,eweps);
 383             ewitab           = _mm_slli_epi32(ewitab,2);
 384             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 385             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 386             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 387             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 388             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 389             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 390             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 391             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 392             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
 393             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velecsum         = _mm_add_pd(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             /* Update vectorial force */
 401             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 402             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 403             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 404
 405             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 406             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 407             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 408
 409             /**************************
 410              * CALCULATE INTERACTIONS *
 411              **************************/
 412
 413             r10              = _mm_mul_pd(rsq10,rinv10);
 414
 415             /* EWALD ELECTROSTATICS */
 416
 417             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 418             ewrt             = _mm_mul_pd(r10,ewtabscale);
 419             ewitab           = _mm_cvttpd_epi32(ewrt);
 420 #ifdef __XOP__
 421             eweps            = _mm_frcz_pd(ewrt);
 422 #else
 423             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 424 #endif
 425             twoeweps         = _mm_add_pd(eweps,eweps);
 426             ewitab           = _mm_slli_epi32(ewitab,2);
 427             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 428             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 429             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 430             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 431             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 432             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 433             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 434             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 435             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 436             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 437
 438             /* Update potential sum for this i atom from the interaction with this j atom. */
 439             velecsum         = _mm_add_pd(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443             /* Update vectorial force */
 444             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 445             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 446             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 447
 448             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 449             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 450             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             r11              = _mm_mul_pd(rsq11,rinv11);
 457
 458             /* EWALD ELECTROSTATICS */
 459
 460             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 461             ewrt             = _mm_mul_pd(r11,ewtabscale);
 462             ewitab           = _mm_cvttpd_epi32(ewrt);
 463 #ifdef __XOP__
 464             eweps            = _mm_frcz_pd(ewrt);
 465 #else
 466             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 467 #endif
 468             twoeweps         = _mm_add_pd(eweps,eweps);
 469             ewitab           = _mm_slli_epi32(ewitab,2);
 470             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 471             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 472             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 473             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 474             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 475             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 476             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 477             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 478             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 479             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 480
 481             /* Update potential sum for this i atom from the interaction with this j atom. */
 482             velecsum         = _mm_add_pd(velecsum,velec);
 483
 484             fscal            = felec;
 485
 486             /* Update vectorial force */
 487             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 488             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 489             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 490
 491             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 492             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 493             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             r12              = _mm_mul_pd(rsq12,rinv12);
 500
 501             /* EWALD ELECTROSTATICS */
 502
 503             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 504             ewrt             = _mm_mul_pd(r12,ewtabscale);
 505             ewitab           = _mm_cvttpd_epi32(ewrt);
 506 #ifdef __XOP__
 507             eweps            = _mm_frcz_pd(ewrt);
 508 #else
 509             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 510 #endif
 511             twoeweps         = _mm_add_pd(eweps,eweps);
 512             ewitab           = _mm_slli_epi32(ewitab,2);
 513             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 514             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 515             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 516             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 517             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 518             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 519             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 520             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 521             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 522             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 523
 524             /* Update potential sum for this i atom from the interaction with this j atom. */
 525             velecsum         = _mm_add_pd(velecsum,velec);
 526
 527             fscal            = felec;
 528
 529             /* Update vectorial force */
 530             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 531             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 532             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 533
 534             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 535             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 536             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             r20              = _mm_mul_pd(rsq20,rinv20);
 543
 544             /* EWALD ELECTROSTATICS */
 545
 546             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 547             ewrt             = _mm_mul_pd(r20,ewtabscale);
 548             ewitab           = _mm_cvttpd_epi32(ewrt);
 549 #ifdef __XOP__
 550             eweps            = _mm_frcz_pd(ewrt);
 551 #else
 552             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 553 #endif
 554             twoeweps         = _mm_add_pd(eweps,eweps);
 555             ewitab           = _mm_slli_epi32(ewitab,2);
 556             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 557             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 558             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 559             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 560             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 561             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 562             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 563             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 564             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 565             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 566
 567             /* Update potential sum for this i atom from the interaction with this j atom. */
 568             velecsum         = _mm_add_pd(velecsum,velec);
 569
 570             fscal            = felec;
 571
 572             /* Update vectorial force */
 573             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 574             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 575             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 576
 577             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 578             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 579             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 580
 581             /**************************
 582              * CALCULATE INTERACTIONS *
 583              **************************/
 584
 585             r21              = _mm_mul_pd(rsq21,rinv21);
 586
 587             /* EWALD ELECTROSTATICS */
 588
 589             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 590             ewrt             = _mm_mul_pd(r21,ewtabscale);
 591             ewitab           = _mm_cvttpd_epi32(ewrt);
 592 #ifdef __XOP__
 593             eweps            = _mm_frcz_pd(ewrt);
 594 #else
 595             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 596 #endif
 597             twoeweps         = _mm_add_pd(eweps,eweps);
 598             ewitab           = _mm_slli_epi32(ewitab,2);
 599             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 600             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 601             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 602             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 603             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 604             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 605             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 606             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 607             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 608             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 609
 610             /* Update potential sum for this i atom from the interaction with this j atom. */
 611             velecsum         = _mm_add_pd(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615             /* Update vectorial force */
 616             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 617             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 618             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 619
 620             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 621             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 622             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 623
 624             /**************************
 625              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             r22              = _mm_mul_pd(rsq22,rinv22);
 629
 630             /* EWALD ELECTROSTATICS */
 631
 632             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 633             ewrt             = _mm_mul_pd(r22,ewtabscale);
 634             ewitab           = _mm_cvttpd_epi32(ewrt);
 635 #ifdef __XOP__
 636             eweps            = _mm_frcz_pd(ewrt);
 637 #else
 638             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 639 #endif
 640             twoeweps         = _mm_add_pd(eweps,eweps);
 641             ewitab           = _mm_slli_epi32(ewitab,2);
 642             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 643             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 644             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 645             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 646             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 647             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 648             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 649             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 650             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 651             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 652
 653             /* Update potential sum for this i atom from the interaction with this j atom. */
 654             velecsum         = _mm_add_pd(velecsum,velec);
 655
 656             fscal            = felec;
 657
 658             /* Update vectorial force */
 659             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 660             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 661             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 662
 663             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 664             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 665             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 666
 667             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 668
 669             /* Inner loop uses 408 flops */
 670         }
 671
 672         if(jidx<j_index_end)
 673         {
 674
 675             jnrA             = jjnr[jidx];
 676             j_coord_offsetA  = DIM*jnrA;
 677
 678             /* load j atom coordinates */
 679             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 680                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 681
 682             /* Calculate displacement vector */
 683             dx00             = _mm_sub_pd(ix0,jx0);
 684             dy00             = _mm_sub_pd(iy0,jy0);
 685             dz00             = _mm_sub_pd(iz0,jz0);
 686             dx01             = _mm_sub_pd(ix0,jx1);
 687             dy01             = _mm_sub_pd(iy0,jy1);
 688             dz01             = _mm_sub_pd(iz0,jz1);
 689             dx02             = _mm_sub_pd(ix0,jx2);
 690             dy02             = _mm_sub_pd(iy0,jy2);
 691             dz02             = _mm_sub_pd(iz0,jz2);
 692             dx10             = _mm_sub_pd(ix1,jx0);
 693             dy10             = _mm_sub_pd(iy1,jy0);
 694             dz10             = _mm_sub_pd(iz1,jz0);
 695             dx11             = _mm_sub_pd(ix1,jx1);
 696             dy11             = _mm_sub_pd(iy1,jy1);
 697             dz11             = _mm_sub_pd(iz1,jz1);
 698             dx12             = _mm_sub_pd(ix1,jx2);
 699             dy12             = _mm_sub_pd(iy1,jy2);
 700             dz12             = _mm_sub_pd(iz1,jz2);
 701             dx20             = _mm_sub_pd(ix2,jx0);
 702             dy20             = _mm_sub_pd(iy2,jy0);
 703             dz20             = _mm_sub_pd(iz2,jz0);
 704             dx21             = _mm_sub_pd(ix2,jx1);
 705             dy21             = _mm_sub_pd(iy2,jy1);
 706             dz21             = _mm_sub_pd(iz2,jz1);
 707             dx22             = _mm_sub_pd(ix2,jx2);
 708             dy22             = _mm_sub_pd(iy2,jy2);
 709             dz22             = _mm_sub_pd(iz2,jz2);
 710
 711             /* Calculate squared distance and things based on it */
 712             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 713             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 714             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 715             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 716             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 717             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 718             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 719             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 720             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 721
 722             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 723             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 724             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 725             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 726             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 727             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 728             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 729             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 730             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 731
 732             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 733             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 734             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 735             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 736             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 737             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 738             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 739             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 740             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 741
 742             fjx0             = _mm_setzero_pd();
 743             fjy0             = _mm_setzero_pd();
 744             fjz0             = _mm_setzero_pd();
 745             fjx1             = _mm_setzero_pd();
 746             fjy1             = _mm_setzero_pd();
 747             fjz1             = _mm_setzero_pd();
 748             fjx2             = _mm_setzero_pd();
 749             fjy2             = _mm_setzero_pd();
 750             fjz2             = _mm_setzero_pd();
 751
 752             /**************************
 753              * CALCULATE INTERACTIONS *
 754              **************************/
 755
 756             r00              = _mm_mul_pd(rsq00,rinv00);
 757
 758             /* EWALD ELECTROSTATICS */
 759
 760             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 761             ewrt             = _mm_mul_pd(r00,ewtabscale);
 762             ewitab           = _mm_cvttpd_epi32(ewrt);
 763 #ifdef __XOP__
 764             eweps            = _mm_frcz_pd(ewrt);
 765 #else
 766             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 767 #endif
 768             twoeweps         = _mm_add_pd(eweps,eweps);
 769             ewitab           = _mm_slli_epi32(ewitab,2);
 770             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 771             ewtabD           = _mm_setzero_pd();
 772             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 773             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 774             ewtabFn          = _mm_setzero_pd();
 775             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 776             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 777             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 778             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 779             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 780
 781             /* LENNARD-JONES DISPERSION/REPULSION */
 782
 783             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 784             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 785             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 786             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 787             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 788
 789             /* Update potential sum for this i atom from the interaction with this j atom. */
 790             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 791             velecsum         = _mm_add_pd(velecsum,velec);
 792             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 793             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 794
 795             fscal            = _mm_add_pd(felec,fvdw);
 796
 797             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 798
 799             /* Update vectorial force */
 800             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 801             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 802             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 803
 804             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 805             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 806             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 807
 808             /**************************
 809              * CALCULATE INTERACTIONS *
 810              **************************/
 811
 812             r01              = _mm_mul_pd(rsq01,rinv01);
 813
 814             /* EWALD ELECTROSTATICS */
 815
 816             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 817             ewrt             = _mm_mul_pd(r01,ewtabscale);
 818             ewitab           = _mm_cvttpd_epi32(ewrt);
 819 #ifdef __XOP__
 820             eweps            = _mm_frcz_pd(ewrt);
 821 #else
 822             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 823 #endif
 824             twoeweps         = _mm_add_pd(eweps,eweps);
 825             ewitab           = _mm_slli_epi32(ewitab,2);
 826             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 827             ewtabD           = _mm_setzero_pd();
 828             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 829             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 830             ewtabFn          = _mm_setzero_pd();
 831             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 832             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 833             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 834             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
 835             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 836
 837             /* Update potential sum for this i atom from the interaction with this j atom. */
 838             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 839             velecsum         = _mm_add_pd(velecsum,velec);
 840
 841             fscal            = felec;
 842
 843             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 844
 845             /* Update vectorial force */
 846             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 847             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 848             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 849
 850             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 851             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 852             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 853
 854             /**************************
 855              * CALCULATE INTERACTIONS *
 856              **************************/
 857
 858             r02              = _mm_mul_pd(rsq02,rinv02);
 859
 860             /* EWALD ELECTROSTATICS */
 861
 862             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 863             ewrt             = _mm_mul_pd(r02,ewtabscale);
 864             ewitab           = _mm_cvttpd_epi32(ewrt);
 865 #ifdef __XOP__
 866             eweps            = _mm_frcz_pd(ewrt);
 867 #else
 868             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 869 #endif
 870             twoeweps         = _mm_add_pd(eweps,eweps);
 871             ewitab           = _mm_slli_epi32(ewitab,2);
 872             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 873             ewtabD           = _mm_setzero_pd();
 874             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 875             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 876             ewtabFn          = _mm_setzero_pd();
 877             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 878             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 879             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 880             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
 881             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 882
 883             /* Update potential sum for this i atom from the interaction with this j atom. */
 884             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 885             velecsum         = _mm_add_pd(velecsum,velec);
 886
 887             fscal            = felec;
 888
 889             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 890
 891             /* Update vectorial force */
 892             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 893             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 894             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 895
 896             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 897             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 898             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             r10              = _mm_mul_pd(rsq10,rinv10);
 905
 906             /* EWALD ELECTROSTATICS */
 907
 908             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 909             ewrt             = _mm_mul_pd(r10,ewtabscale);
 910             ewitab           = _mm_cvttpd_epi32(ewrt);
 911 #ifdef __XOP__
 912             eweps            = _mm_frcz_pd(ewrt);
 913 #else
 914             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 915 #endif
 916             twoeweps         = _mm_add_pd(eweps,eweps);
 917             ewitab           = _mm_slli_epi32(ewitab,2);
 918             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 919             ewtabD           = _mm_setzero_pd();
 920             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 921             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 922             ewtabFn          = _mm_setzero_pd();
 923             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 924             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 925             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 926             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 927             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 928
 929             /* Update potential sum for this i atom from the interaction with this j atom. */
 930             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 931             velecsum         = _mm_add_pd(velecsum,velec);
 932
 933             fscal            = felec;
 934
 935             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 936
 937             /* Update vectorial force */
 938             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 939             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 940             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 941
 942             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 943             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 944             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 945
 946             /**************************
 947              * CALCULATE INTERACTIONS *
 948              **************************/
 949
 950             r11              = _mm_mul_pd(rsq11,rinv11);
 951
 952             /* EWALD ELECTROSTATICS */
 953
 954             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 955             ewrt             = _mm_mul_pd(r11,ewtabscale);
 956             ewitab           = _mm_cvttpd_epi32(ewrt);
 957 #ifdef __XOP__
 958             eweps            = _mm_frcz_pd(ewrt);
 959 #else
 960             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 961 #endif
 962             twoeweps         = _mm_add_pd(eweps,eweps);
 963             ewitab           = _mm_slli_epi32(ewitab,2);
 964             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 965             ewtabD           = _mm_setzero_pd();
 966             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 967             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 968             ewtabFn          = _mm_setzero_pd();
 969             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 970             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 971             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 972             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 973             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 974
 975             /* Update potential sum for this i atom from the interaction with this j atom. */
 976             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 977             velecsum         = _mm_add_pd(velecsum,velec);
 978
 979             fscal            = felec;
 980
 981             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 982
 983             /* Update vectorial force */
 984             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 985             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 986             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 987
 988             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 989             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 990             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 991
 992             /**************************
 993              * CALCULATE INTERACTIONS *
 994              **************************/
 995
 996             r12              = _mm_mul_pd(rsq12,rinv12);
 997
 998             /* EWALD ELECTROSTATICS */
 999
1000             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1001             ewrt             = _mm_mul_pd(r12,ewtabscale);
1002             ewitab           = _mm_cvttpd_epi32(ewrt);
1003 #ifdef __XOP__
1004             eweps            = _mm_frcz_pd(ewrt);
1005 #else
1006             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1007 #endif
1008             twoeweps         = _mm_add_pd(eweps,eweps);
1009             ewitab           = _mm_slli_epi32(ewitab,2);
1010             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1011             ewtabD           = _mm_setzero_pd();
1012             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1013             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1014             ewtabFn          = _mm_setzero_pd();
1015             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1016             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1017             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1018             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
1019             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1020
1021             /* Update potential sum for this i atom from the interaction with this j atom. */
1022             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1023             velecsum         = _mm_add_pd(velecsum,velec);
1024
1025             fscal            = felec;
1026
1027             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1028
1029             /* Update vectorial force */
1030             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1031             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1032             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1033
1034             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1035             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1036             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             r20              = _mm_mul_pd(rsq20,rinv20);
1043
1044             /* EWALD ELECTROSTATICS */
1045
1046             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1047             ewrt             = _mm_mul_pd(r20,ewtabscale);
1048             ewitab           = _mm_cvttpd_epi32(ewrt);
1049 #ifdef __XOP__
1050             eweps            = _mm_frcz_pd(ewrt);
1051 #else
1052             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1053 #endif
1054             twoeweps         = _mm_add_pd(eweps,eweps);
1055             ewitab           = _mm_slli_epi32(ewitab,2);
1056             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1057             ewtabD           = _mm_setzero_pd();
1058             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1059             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1060             ewtabFn          = _mm_setzero_pd();
1061             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1062             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1063             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1064             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1065             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1066
1067             /* Update potential sum for this i atom from the interaction with this j atom. */
1068             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1069             velecsum         = _mm_add_pd(velecsum,velec);
1070
1071             fscal            = felec;
1072
1073             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1074
1075             /* Update vectorial force */
1076             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1077             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1078             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1079
1080             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1081             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1082             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1083
1084             /**************************
1085              * CALCULATE INTERACTIONS *
1086              **************************/
1087
1088             r21              = _mm_mul_pd(rsq21,rinv21);
1089
1090             /* EWALD ELECTROSTATICS */
1091
1092             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1093             ewrt             = _mm_mul_pd(r21,ewtabscale);
1094             ewitab           = _mm_cvttpd_epi32(ewrt);
1095 #ifdef __XOP__
1096             eweps            = _mm_frcz_pd(ewrt);
1097 #else
1098             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1099 #endif
1100             twoeweps         = _mm_add_pd(eweps,eweps);
1101             ewitab           = _mm_slli_epi32(ewitab,2);
1102             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1103             ewtabD           = _mm_setzero_pd();
1104             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1105             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1106             ewtabFn          = _mm_setzero_pd();
1107             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1108             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1109             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1110             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1111             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1112
1113             /* Update potential sum for this i atom from the interaction with this j atom. */
1114             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1115             velecsum         = _mm_add_pd(velecsum,velec);
1116
1117             fscal            = felec;
1118
1119             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1120
1121             /* Update vectorial force */
1122             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1123             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1124             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1125
1126             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1127             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1128             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1129
1130             /**************************
1131              * CALCULATE INTERACTIONS *
1132              **************************/
1133
1134             r22              = _mm_mul_pd(rsq22,rinv22);
1135
1136             /* EWALD ELECTROSTATICS */
1137
1138             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1139             ewrt             = _mm_mul_pd(r22,ewtabscale);
1140             ewitab           = _mm_cvttpd_epi32(ewrt);
1141 #ifdef __XOP__
1142             eweps            = _mm_frcz_pd(ewrt);
1143 #else
1144             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1145 #endif
1146             twoeweps         = _mm_add_pd(eweps,eweps);
1147             ewitab           = _mm_slli_epi32(ewitab,2);
1148             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1149             ewtabD           = _mm_setzero_pd();
1150             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1151             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1152             ewtabFn          = _mm_setzero_pd();
1153             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1154             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1155             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1156             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1157             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1158
1159             /* Update potential sum for this i atom from the interaction with this j atom. */
1160             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1161             velecsum         = _mm_add_pd(velecsum,velec);
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1166
1167             /* Update vectorial force */
1168             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1169             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1170             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1171
1172             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1173             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1174             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1175
1176             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1177
1178             /* Inner loop uses 408 flops */
1179         }
1180
1181         /* End of innermost loop */
1182
1183         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1184                                               f+i_coord_offset,fshift+i_shift_offset);
1185
1186         ggid                        = gid[iidx];
1187         /* Update potential energies */
1188         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1189         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1190
1191         /* Increment number of inner iterations */
1192         inneriter                  += j_index_end - j_index_start;
1193
1194         /* Outer loop uses 20 flops */
1195     }
1196
1197     /* Increment number of outer iterations */
1198     outeriter        += nri;
1199
1200     /* Update outer/inner flops */
1201
1202     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*408);
1203 }
1204 /*
1205  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3W3_F_avx_128_fma_double
1206  * Electrostatics interaction: Ewald
1207  * VdW interaction:            LennardJones
1208  * Geometry:                   Water3-Water3
1209  * Calculate force/pot:        Force
1210  */
1211 void
1212 nb_kernel_ElecEw_VdwLJ_GeomW3W3_F_avx_128_fma_double
1213                     (t_nblist * gmx_restrict                nlist,
1214                      rvec * gmx_restrict                    xx,
1215                      rvec * gmx_restrict                    ff,
1216                      t_forcerec * gmx_restrict              fr,
1217                      t_mdatoms * gmx_restrict               mdatoms,
1218                      nb_kernel_data_t * gmx_restrict        kernel_data,
1219                      t_nrnb * gmx_restrict                  nrnb)
1220 {
1221     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1222      * just 0 for non-waters.
1223      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1224      * jnr indices corresponding to data put in the four positions in the SIMD register.
1225      */
1226     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1227     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1228     int              jnrA,jnrB;
1229     int              j_coord_offsetA,j_coord_offsetB;
1230     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1231     real             rcutoff_scalar;
1232     real             *shiftvec,*fshift,*x,*f;
1233     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1234     int              vdwioffset0;
1235     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1236     int              vdwioffset1;
1237     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1238     int              vdwioffset2;
1239     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1240     int              vdwjidx0A,vdwjidx0B;
1241     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1242     int              vdwjidx1A,vdwjidx1B;
1243     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1244     int              vdwjidx2A,vdwjidx2B;
1245     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1246     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1247     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1248     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1249     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1250     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1251     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1252     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1253     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1254     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1255     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1256     real             *charge;
1257     int              nvdwtype;
1258     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1259     int              *vdwtype;
1260     real             *vdwparam;
1261     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1262     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1263     __m128i          ewitab;
1264     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1265     real             *ewtab;
1266     __m128d          dummy_mask,cutoff_mask;
1267     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1268     __m128d          one     = _mm_set1_pd(1.0);
1269     __m128d          two     = _mm_set1_pd(2.0);
1270     x                = xx[0];
1271     f                = ff[0];
1272
1273     nri              = nlist->nri;
1274     iinr             = nlist->iinr;
1275     jindex           = nlist->jindex;
1276     jjnr             = nlist->jjnr;
1277     shiftidx         = nlist->shift;
1278     gid              = nlist->gid;
1279     shiftvec         = fr->shift_vec[0];
1280     fshift           = fr->fshift[0];
1281     facel            = _mm_set1_pd(fr->epsfac);
1282     charge           = mdatoms->chargeA;
1283     nvdwtype         = fr->ntype;
1284     vdwparam         = fr->nbfp;
1285     vdwtype          = mdatoms->typeA;
1286
1287     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1288     ewtab            = fr->ic->tabq_coul_F;
1289     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1290     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1291
1292     /* Setup water-specific parameters */
1293     inr              = nlist->iinr[0];
1294     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1295     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1296     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1297     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1298
1299     jq0              = _mm_set1_pd(charge[inr+0]);
1300     jq1              = _mm_set1_pd(charge[inr+1]);
1301     jq2              = _mm_set1_pd(charge[inr+2]);
1302     vdwjidx0A        = 2*vdwtype[inr+0];
1303     qq00             = _mm_mul_pd(iq0,jq0);
1304     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1305     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1306     qq01             = _mm_mul_pd(iq0,jq1);
1307     qq02             = _mm_mul_pd(iq0,jq2);
1308     qq10             = _mm_mul_pd(iq1,jq0);
1309     qq11             = _mm_mul_pd(iq1,jq1);
1310     qq12             = _mm_mul_pd(iq1,jq2);
1311     qq20             = _mm_mul_pd(iq2,jq0);
1312     qq21             = _mm_mul_pd(iq2,jq1);
1313     qq22             = _mm_mul_pd(iq2,jq2);
1314
1315     /* Avoid stupid compiler warnings */
1316     jnrA = jnrB = 0;
1317     j_coord_offsetA = 0;
1318     j_coord_offsetB = 0;
1319
1320     outeriter        = 0;
1321     inneriter        = 0;
1322
1323     /* Start outer loop over neighborlists */
1324     for(iidx=0; iidx<nri; iidx++)
1325     {
1326         /* Load shift vector for this list */
1327         i_shift_offset   = DIM*shiftidx[iidx];
1328
1329         /* Load limits for loop over neighbors */
1330         j_index_start    = jindex[iidx];
1331         j_index_end      = jindex[iidx+1];
1332
1333         /* Get outer coordinate index */
1334         inr              = iinr[iidx];
1335         i_coord_offset   = DIM*inr;
1336
1337         /* Load i particle coords and add shift vector */
1338         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1339                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1340
1341         fix0             = _mm_setzero_pd();
1342         fiy0             = _mm_setzero_pd();
1343         fiz0             = _mm_setzero_pd();
1344         fix1             = _mm_setzero_pd();
1345         fiy1             = _mm_setzero_pd();
1346         fiz1             = _mm_setzero_pd();
1347         fix2             = _mm_setzero_pd();
1348         fiy2             = _mm_setzero_pd();
1349         fiz2             = _mm_setzero_pd();
1350
1351         /* Start inner kernel loop */
1352         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1353         {
1354
1355             /* Get j neighbor index, and coordinate index */
1356             jnrA             = jjnr[jidx];
1357             jnrB             = jjnr[jidx+1];
1358             j_coord_offsetA  = DIM*jnrA;
1359             j_coord_offsetB  = DIM*jnrB;
1360
1361             /* load j atom coordinates */
1362             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1363                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1364
1365             /* Calculate displacement vector */
1366             dx00             = _mm_sub_pd(ix0,jx0);
1367             dy00             = _mm_sub_pd(iy0,jy0);
1368             dz00             = _mm_sub_pd(iz0,jz0);
1369             dx01             = _mm_sub_pd(ix0,jx1);
1370             dy01             = _mm_sub_pd(iy0,jy1);
1371             dz01             = _mm_sub_pd(iz0,jz1);
1372             dx02             = _mm_sub_pd(ix0,jx2);
1373             dy02             = _mm_sub_pd(iy0,jy2);
1374             dz02             = _mm_sub_pd(iz0,jz2);
1375             dx10             = _mm_sub_pd(ix1,jx0);
1376             dy10             = _mm_sub_pd(iy1,jy0);
1377             dz10             = _mm_sub_pd(iz1,jz0);
1378             dx11             = _mm_sub_pd(ix1,jx1);
1379             dy11             = _mm_sub_pd(iy1,jy1);
1380             dz11             = _mm_sub_pd(iz1,jz1);
1381             dx12             = _mm_sub_pd(ix1,jx2);
1382             dy12             = _mm_sub_pd(iy1,jy2);
1383             dz12             = _mm_sub_pd(iz1,jz2);
1384             dx20             = _mm_sub_pd(ix2,jx0);
1385             dy20             = _mm_sub_pd(iy2,jy0);
1386             dz20             = _mm_sub_pd(iz2,jz0);
1387             dx21             = _mm_sub_pd(ix2,jx1);
1388             dy21             = _mm_sub_pd(iy2,jy1);
1389             dz21             = _mm_sub_pd(iz2,jz1);
1390             dx22             = _mm_sub_pd(ix2,jx2);
1391             dy22             = _mm_sub_pd(iy2,jy2);
1392             dz22             = _mm_sub_pd(iz2,jz2);
1393
1394             /* Calculate squared distance and things based on it */
1395             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1396             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1397             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1398             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1399             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1400             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1401             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1402             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1403             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1404
1405             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1406             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1407             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1408             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1409             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1410             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1411             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1412             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1413             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1414
1415             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1416             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1417             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1418             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1419             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1420             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1421             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1422             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1423             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1424
1425             fjx0             = _mm_setzero_pd();
1426             fjy0             = _mm_setzero_pd();
1427             fjz0             = _mm_setzero_pd();
1428             fjx1             = _mm_setzero_pd();
1429             fjy1             = _mm_setzero_pd();
1430             fjz1             = _mm_setzero_pd();
1431             fjx2             = _mm_setzero_pd();
1432             fjy2             = _mm_setzero_pd();
1433             fjz2             = _mm_setzero_pd();
1434
1435             /**************************
1436              * CALCULATE INTERACTIONS *
1437              **************************/
1438
1439             r00              = _mm_mul_pd(rsq00,rinv00);
1440
1441             /* EWALD ELECTROSTATICS */
1442
1443             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1444             ewrt             = _mm_mul_pd(r00,ewtabscale);
1445             ewitab           = _mm_cvttpd_epi32(ewrt);
1446 #ifdef __XOP__
1447             eweps            = _mm_frcz_pd(ewrt);
1448 #else
1449             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1450 #endif
1451             twoeweps         = _mm_add_pd(eweps,eweps);
1452             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1453                                          &ewtabF,&ewtabFn);
1454             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1455             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1456
1457             /* LENNARD-JONES DISPERSION/REPULSION */
1458
1459             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1460             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1461
1462             fscal            = _mm_add_pd(felec,fvdw);
1463
1464             /* Update vectorial force */
1465             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1466             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1467             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1468
1469             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1470             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1471             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1472
1473             /**************************
1474              * CALCULATE INTERACTIONS *
1475              **************************/
1476
1477             r01              = _mm_mul_pd(rsq01,rinv01);
1478
1479             /* EWALD ELECTROSTATICS */
1480
1481             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1482             ewrt             = _mm_mul_pd(r01,ewtabscale);
1483             ewitab           = _mm_cvttpd_epi32(ewrt);
1484 #ifdef __XOP__
1485             eweps            = _mm_frcz_pd(ewrt);
1486 #else
1487             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1488 #endif
1489             twoeweps         = _mm_add_pd(eweps,eweps);
1490             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1491                                          &ewtabF,&ewtabFn);
1492             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1493             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1494
1495             fscal            = felec;
1496
1497             /* Update vectorial force */
1498             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1499             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1500             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1501
1502             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1503             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1504             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1505
1506             /**************************
1507              * CALCULATE INTERACTIONS *
1508              **************************/
1509
1510             r02              = _mm_mul_pd(rsq02,rinv02);
1511
1512             /* EWALD ELECTROSTATICS */
1513
1514             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1515             ewrt             = _mm_mul_pd(r02,ewtabscale);
1516             ewitab           = _mm_cvttpd_epi32(ewrt);
1517 #ifdef __XOP__
1518             eweps            = _mm_frcz_pd(ewrt);
1519 #else
1520             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1521 #endif
1522             twoeweps         = _mm_add_pd(eweps,eweps);
1523             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1524                                          &ewtabF,&ewtabFn);
1525             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1526             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1527
1528             fscal            = felec;
1529
1530             /* Update vectorial force */
1531             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1532             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1533             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1534
1535             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1536             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1537             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1538
1539             /**************************
1540              * CALCULATE INTERACTIONS *
1541              **************************/
1542
1543             r10              = _mm_mul_pd(rsq10,rinv10);
1544
1545             /* EWALD ELECTROSTATICS */
1546
1547             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1548             ewrt             = _mm_mul_pd(r10,ewtabscale);
1549             ewitab           = _mm_cvttpd_epi32(ewrt);
1550 #ifdef __XOP__
1551             eweps            = _mm_frcz_pd(ewrt);
1552 #else
1553             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1554 #endif
1555             twoeweps         = _mm_add_pd(eweps,eweps);
1556             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1557                                          &ewtabF,&ewtabFn);
1558             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1559             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1560
1561             fscal            = felec;
1562
1563             /* Update vectorial force */
1564             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1565             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1566             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1567
1568             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1569             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1570             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1571
1572             /**************************
1573              * CALCULATE INTERACTIONS *
1574              **************************/
1575
1576             r11              = _mm_mul_pd(rsq11,rinv11);
1577
1578             /* EWALD ELECTROSTATICS */
1579
1580             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1581             ewrt             = _mm_mul_pd(r11,ewtabscale);
1582             ewitab           = _mm_cvttpd_epi32(ewrt);
1583 #ifdef __XOP__
1584             eweps            = _mm_frcz_pd(ewrt);
1585 #else
1586             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1587 #endif
1588             twoeweps         = _mm_add_pd(eweps,eweps);
1589             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1590                                          &ewtabF,&ewtabFn);
1591             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1592             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1593
1594             fscal            = felec;
1595
1596             /* Update vectorial force */
1597             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1598             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1599             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1600
1601             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1602             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1603             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1604
1605             /**************************
1606              * CALCULATE INTERACTIONS *
1607              **************************/
1608
1609             r12              = _mm_mul_pd(rsq12,rinv12);
1610
1611             /* EWALD ELECTROSTATICS */
1612
1613             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1614             ewrt             = _mm_mul_pd(r12,ewtabscale);
1615             ewitab           = _mm_cvttpd_epi32(ewrt);
1616 #ifdef __XOP__
1617             eweps            = _mm_frcz_pd(ewrt);
1618 #else
1619             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1620 #endif
1621             twoeweps         = _mm_add_pd(eweps,eweps);
1622             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1623                                          &ewtabF,&ewtabFn);
1624             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1625             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1626
1627             fscal            = felec;
1628
1629             /* Update vectorial force */
1630             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1631             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1632             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1633
1634             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1635             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1636             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1637
1638             /**************************
1639              * CALCULATE INTERACTIONS *
1640              **************************/
1641
1642             r20              = _mm_mul_pd(rsq20,rinv20);
1643
1644             /* EWALD ELECTROSTATICS */
1645
1646             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1647             ewrt             = _mm_mul_pd(r20,ewtabscale);
1648             ewitab           = _mm_cvttpd_epi32(ewrt);
1649 #ifdef __XOP__
1650             eweps            = _mm_frcz_pd(ewrt);
1651 #else
1652             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1653 #endif
1654             twoeweps         = _mm_add_pd(eweps,eweps);
1655             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1656                                          &ewtabF,&ewtabFn);
1657             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1658             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1659
1660             fscal            = felec;
1661
1662             /* Update vectorial force */
1663             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1664             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1665             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1666
1667             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1668             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1669             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1670
1671             /**************************
1672              * CALCULATE INTERACTIONS *
1673              **************************/
1674
1675             r21              = _mm_mul_pd(rsq21,rinv21);
1676
1677             /* EWALD ELECTROSTATICS */
1678
1679             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1680             ewrt             = _mm_mul_pd(r21,ewtabscale);
1681             ewitab           = _mm_cvttpd_epi32(ewrt);
1682 #ifdef __XOP__
1683             eweps            = _mm_frcz_pd(ewrt);
1684 #else
1685             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1686 #endif
1687             twoeweps         = _mm_add_pd(eweps,eweps);
1688             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1689                                          &ewtabF,&ewtabFn);
1690             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1691             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1692
1693             fscal            = felec;
1694
1695             /* Update vectorial force */
1696             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1697             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1698             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1699
1700             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1701             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1702             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1703
1704             /**************************
1705              * CALCULATE INTERACTIONS *
1706              **************************/
1707
1708             r22              = _mm_mul_pd(rsq22,rinv22);
1709
1710             /* EWALD ELECTROSTATICS */
1711
1712             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1713             ewrt             = _mm_mul_pd(r22,ewtabscale);
1714             ewitab           = _mm_cvttpd_epi32(ewrt);
1715 #ifdef __XOP__
1716             eweps            = _mm_frcz_pd(ewrt);
1717 #else
1718             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1719 #endif
1720             twoeweps         = _mm_add_pd(eweps,eweps);
1721             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1722                                          &ewtabF,&ewtabFn);
1723             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1724             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1725
1726             fscal            = felec;
1727
1728             /* Update vectorial force */
1729             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1730             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1731             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1732
1733             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1734             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1735             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1736
1737             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1738
1739             /* Inner loop uses 358 flops */
1740         }
1741
1742         if(jidx<j_index_end)
1743         {
1744
1745             jnrA             = jjnr[jidx];
1746             j_coord_offsetA  = DIM*jnrA;
1747
1748             /* load j atom coordinates */
1749             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1750                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1751
1752             /* Calculate displacement vector */
1753             dx00             = _mm_sub_pd(ix0,jx0);
1754             dy00             = _mm_sub_pd(iy0,jy0);
1755             dz00             = _mm_sub_pd(iz0,jz0);
1756             dx01             = _mm_sub_pd(ix0,jx1);
1757             dy01             = _mm_sub_pd(iy0,jy1);
1758             dz01             = _mm_sub_pd(iz0,jz1);
1759             dx02             = _mm_sub_pd(ix0,jx2);
1760             dy02             = _mm_sub_pd(iy0,jy2);
1761             dz02             = _mm_sub_pd(iz0,jz2);
1762             dx10             = _mm_sub_pd(ix1,jx0);
1763             dy10             = _mm_sub_pd(iy1,jy0);
1764             dz10             = _mm_sub_pd(iz1,jz0);
1765             dx11             = _mm_sub_pd(ix1,jx1);
1766             dy11             = _mm_sub_pd(iy1,jy1);
1767             dz11             = _mm_sub_pd(iz1,jz1);
1768             dx12             = _mm_sub_pd(ix1,jx2);
1769             dy12             = _mm_sub_pd(iy1,jy2);
1770             dz12             = _mm_sub_pd(iz1,jz2);
1771             dx20             = _mm_sub_pd(ix2,jx0);
1772             dy20             = _mm_sub_pd(iy2,jy0);
1773             dz20             = _mm_sub_pd(iz2,jz0);
1774             dx21             = _mm_sub_pd(ix2,jx1);
1775             dy21             = _mm_sub_pd(iy2,jy1);
1776             dz21             = _mm_sub_pd(iz2,jz1);
1777             dx22             = _mm_sub_pd(ix2,jx2);
1778             dy22             = _mm_sub_pd(iy2,jy2);
1779             dz22             = _mm_sub_pd(iz2,jz2);
1780
1781             /* Calculate squared distance and things based on it */
1782             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1783             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1784             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1785             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1786             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1787             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1788             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1789             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1790             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1791
1792             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1793             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1794             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1795             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1796             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1797             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1798             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1799             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1800             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1801
1802             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1803             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1804             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1805             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1806             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1807             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1808             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1809             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1810             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1811
1812             fjx0             = _mm_setzero_pd();
1813             fjy0             = _mm_setzero_pd();
1814             fjz0             = _mm_setzero_pd();
1815             fjx1             = _mm_setzero_pd();
1816             fjy1             = _mm_setzero_pd();
1817             fjz1             = _mm_setzero_pd();
1818             fjx2             = _mm_setzero_pd();
1819             fjy2             = _mm_setzero_pd();
1820             fjz2             = _mm_setzero_pd();
1821
1822             /**************************
1823              * CALCULATE INTERACTIONS *
1824              **************************/
1825
1826             r00              = _mm_mul_pd(rsq00,rinv00);
1827
1828             /* EWALD ELECTROSTATICS */
1829
1830             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1831             ewrt             = _mm_mul_pd(r00,ewtabscale);
1832             ewitab           = _mm_cvttpd_epi32(ewrt);
1833 #ifdef __XOP__
1834             eweps            = _mm_frcz_pd(ewrt);
1835 #else
1836             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1837 #endif
1838             twoeweps         = _mm_add_pd(eweps,eweps);
1839             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1840             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1841             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1842
1843             /* LENNARD-JONES DISPERSION/REPULSION */
1844
1845             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1846             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1847
1848             fscal            = _mm_add_pd(felec,fvdw);
1849
1850             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1851
1852             /* Update vectorial force */
1853             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1854             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1855             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1856
1857             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1858             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1859             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1860
1861             /**************************
1862              * CALCULATE INTERACTIONS *
1863              **************************/
1864
1865             r01              = _mm_mul_pd(rsq01,rinv01);
1866
1867             /* EWALD ELECTROSTATICS */
1868
1869             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1870             ewrt             = _mm_mul_pd(r01,ewtabscale);
1871             ewitab           = _mm_cvttpd_epi32(ewrt);
1872 #ifdef __XOP__
1873             eweps            = _mm_frcz_pd(ewrt);
1874 #else
1875             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1876 #endif
1877             twoeweps         = _mm_add_pd(eweps,eweps);
1878             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1879             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1880             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1881
1882             fscal            = felec;
1883
1884             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1885
1886             /* Update vectorial force */
1887             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1888             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1889             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1890
1891             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1892             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1893             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1894
1895             /**************************
1896              * CALCULATE INTERACTIONS *
1897              **************************/
1898
1899             r02              = _mm_mul_pd(rsq02,rinv02);
1900
1901             /* EWALD ELECTROSTATICS */
1902
1903             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1904             ewrt             = _mm_mul_pd(r02,ewtabscale);
1905             ewitab           = _mm_cvttpd_epi32(ewrt);
1906 #ifdef __XOP__
1907             eweps            = _mm_frcz_pd(ewrt);
1908 #else
1909             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1910 #endif
1911             twoeweps         = _mm_add_pd(eweps,eweps);
1912             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1913             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1914             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1915
1916             fscal            = felec;
1917
1918             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1919
1920             /* Update vectorial force */
1921             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1922             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1923             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1924
1925             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1926             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1927             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1928
1929             /**************************
1930              * CALCULATE INTERACTIONS *
1931              **************************/
1932
1933             r10              = _mm_mul_pd(rsq10,rinv10);
1934
1935             /* EWALD ELECTROSTATICS */
1936
1937             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1938             ewrt             = _mm_mul_pd(r10,ewtabscale);
1939             ewitab           = _mm_cvttpd_epi32(ewrt);
1940 #ifdef __XOP__
1941             eweps            = _mm_frcz_pd(ewrt);
1942 #else
1943             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1944 #endif
1945             twoeweps         = _mm_add_pd(eweps,eweps);
1946             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1947             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1948             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1949
1950             fscal            = felec;
1951
1952             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1953
1954             /* Update vectorial force */
1955             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1956             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1957             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1958
1959             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1960             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1961             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1962
1963             /**************************
1964              * CALCULATE INTERACTIONS *
1965              **************************/
1966
1967             r11              = _mm_mul_pd(rsq11,rinv11);
1968
1969             /* EWALD ELECTROSTATICS */
1970
1971             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1972             ewrt             = _mm_mul_pd(r11,ewtabscale);
1973             ewitab           = _mm_cvttpd_epi32(ewrt);
1974 #ifdef __XOP__
1975             eweps            = _mm_frcz_pd(ewrt);
1976 #else
1977             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1978 #endif
1979             twoeweps         = _mm_add_pd(eweps,eweps);
1980             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1981             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1982             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1983
1984             fscal            = felec;
1985
1986             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1987
1988             /* Update vectorial force */
1989             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1990             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1991             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1992
1993             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1994             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1995             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1996
1997             /**************************
1998              * CALCULATE INTERACTIONS *
1999              **************************/
2000
2001             r12              = _mm_mul_pd(rsq12,rinv12);
2002
2003             /* EWALD ELECTROSTATICS */
2004
2005             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2006             ewrt             = _mm_mul_pd(r12,ewtabscale);
2007             ewitab           = _mm_cvttpd_epi32(ewrt);
2008 #ifdef __XOP__
2009             eweps            = _mm_frcz_pd(ewrt);
2010 #else
2011             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2012 #endif
2013             twoeweps         = _mm_add_pd(eweps,eweps);
2014             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2015             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2016             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2017
2018             fscal            = felec;
2019
2020             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2021
2022             /* Update vectorial force */
2023             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2024             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2025             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2026
2027             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2028             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2029             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2030
2031             /**************************
2032              * CALCULATE INTERACTIONS *
2033              **************************/
2034
2035             r20              = _mm_mul_pd(rsq20,rinv20);
2036
2037             /* EWALD ELECTROSTATICS */
2038
2039             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2040             ewrt             = _mm_mul_pd(r20,ewtabscale);
2041             ewitab           = _mm_cvttpd_epi32(ewrt);
2042 #ifdef __XOP__
2043             eweps            = _mm_frcz_pd(ewrt);
2044 #else
2045             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2046 #endif
2047             twoeweps         = _mm_add_pd(eweps,eweps);
2048             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2049             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2050             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2051
2052             fscal            = felec;
2053
2054             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2055
2056             /* Update vectorial force */
2057             fix2             = _mm_macc_pd(dx20,fscal,fix2);
2058             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
2059             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
2060
2061             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2062             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2063             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
2064
2065             /**************************
2066              * CALCULATE INTERACTIONS *
2067              **************************/
2068
2069             r21              = _mm_mul_pd(rsq21,rinv21);
2070
2071             /* EWALD ELECTROSTATICS */
2072
2073             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2074             ewrt             = _mm_mul_pd(r21,ewtabscale);
2075             ewitab           = _mm_cvttpd_epi32(ewrt);
2076 #ifdef __XOP__
2077             eweps            = _mm_frcz_pd(ewrt);
2078 #else
2079             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2080 #endif
2081             twoeweps         = _mm_add_pd(eweps,eweps);
2082             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2083             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2084             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2085
2086             fscal            = felec;
2087
2088             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2089
2090             /* Update vectorial force */
2091             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2092             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2093             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2094
2095             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2096             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2097             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2098
2099             /**************************
2100              * CALCULATE INTERACTIONS *
2101              **************************/
2102
2103             r22              = _mm_mul_pd(rsq22,rinv22);
2104
2105             /* EWALD ELECTROSTATICS */
2106
2107             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2108             ewrt             = _mm_mul_pd(r22,ewtabscale);
2109             ewitab           = _mm_cvttpd_epi32(ewrt);
2110 #ifdef __XOP__
2111             eweps            = _mm_frcz_pd(ewrt);
2112 #else
2113             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2114 #endif
2115             twoeweps         = _mm_add_pd(eweps,eweps);
2116             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2117             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2118             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2119
2120             fscal            = felec;
2121
2122             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2123
2124             /* Update vectorial force */
2125             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2126             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2127             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2128
2129             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2130             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2131             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2132
2133             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2134
2135             /* Inner loop uses 358 flops */
2136         }
2137
2138         /* End of innermost loop */
2139
2140         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2141                                               f+i_coord_offset,fshift+i_shift_offset);
2142
2143         /* Increment number of inner iterations */
2144         inneriter                  += j_index_end - j_index_start;
2145
2146         /* Outer loop uses 18 flops */
2147     }
2148
2149     /* Increment number of outer iterations */
2150     outeriter        += nri;
2151
2152     /* Update outer/inner flops */
2153
2154     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*358);
2155 }