src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwLJSh_GeomW3W3_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3W3_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_GeomW3W3_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     int              vdwjidx1A,vdwjidx1B;
  91     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  92     int              vdwjidx2A,vdwjidx2B;
  93     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  94     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  96     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  97     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  99     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 100     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 102     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 103     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 104     real             *charge;
 105     int              nvdwtype;
 106     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 107     int              *vdwtype;
 108     real             *vdwparam;
 109     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 110     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 111     __m128i          ewitab;
 112     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 113     real             *ewtab;
 114     __m128d          dummy_mask,cutoff_mask;
 115     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 116     __m128d          one     = _mm_set1_pd(1.0);
 117     __m128d          two     = _mm_set1_pd(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm_set1_pd(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134
 135     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 136     ewtab            = fr->ic->tabq_coul_FDV0;
 137     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 138     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 143     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 144     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 145     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 146
 147     jq0              = _mm_set1_pd(charge[inr+0]);
 148     jq1              = _mm_set1_pd(charge[inr+1]);
 149     jq2              = _mm_set1_pd(charge[inr+2]);
 150     vdwjidx0A        = 2*vdwtype[inr+0];
 151     qq00             = _mm_mul_pd(iq0,jq0);
 152     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
 153     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
 154     qq01             = _mm_mul_pd(iq0,jq1);
 155     qq02             = _mm_mul_pd(iq0,jq2);
 156     qq10             = _mm_mul_pd(iq1,jq0);
 157     qq11             = _mm_mul_pd(iq1,jq1);
 158     qq12             = _mm_mul_pd(iq1,jq2);
 159     qq20             = _mm_mul_pd(iq2,jq0);
 160     qq21             = _mm_mul_pd(iq2,jq1);
 161     qq22             = _mm_mul_pd(iq2,jq2);
 162
 163     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 164     rcutoff_scalar   = fr->rcoulomb;
 165     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 166     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 167
 168     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 169     rvdw             = _mm_set1_pd(fr->rvdw);
 170
 171     /* Avoid stupid compiler warnings */
 172     jnrA = jnrB = 0;
 173     j_coord_offsetA = 0;
 174     j_coord_offsetB = 0;
 175
 176     outeriter        = 0;
 177     inneriter        = 0;
 178
 179     /* Start outer loop over neighborlists */
 180     for(iidx=0; iidx<nri; iidx++)
 181     {
 182         /* Load shift vector for this list */
 183         i_shift_offset   = DIM*shiftidx[iidx];
 184
 185         /* Load limits for loop over neighbors */
 186         j_index_start    = jindex[iidx];
 187         j_index_end      = jindex[iidx+1];
 188
 189         /* Get outer coordinate index */
 190         inr              = iinr[iidx];
 191         i_coord_offset   = DIM*inr;
 192
 193         /* Load i particle coords and add shift vector */
 194         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 195                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 196
 197         fix0             = _mm_setzero_pd();
 198         fiy0             = _mm_setzero_pd();
 199         fiz0             = _mm_setzero_pd();
 200         fix1             = _mm_setzero_pd();
 201         fiy1             = _mm_setzero_pd();
 202         fiz1             = _mm_setzero_pd();
 203         fix2             = _mm_setzero_pd();
 204         fiy2             = _mm_setzero_pd();
 205         fiz2             = _mm_setzero_pd();
 206
 207         /* Reset potential sums */
 208         velecsum         = _mm_setzero_pd();
 209         vvdwsum          = _mm_setzero_pd();
 210
 211         /* Start inner kernel loop */
 212         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 213         {
 214
 215             /* Get j neighbor index, and coordinate index */
 216             jnrA             = jjnr[jidx];
 217             jnrB             = jjnr[jidx+1];
 218             j_coord_offsetA  = DIM*jnrA;
 219             j_coord_offsetB  = DIM*jnrB;
 220
 221             /* load j atom coordinates */
 222             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 223                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 224
 225             /* Calculate displacement vector */
 226             dx00             = _mm_sub_pd(ix0,jx0);
 227             dy00             = _mm_sub_pd(iy0,jy0);
 228             dz00             = _mm_sub_pd(iz0,jz0);
 229             dx01             = _mm_sub_pd(ix0,jx1);
 230             dy01             = _mm_sub_pd(iy0,jy1);
 231             dz01             = _mm_sub_pd(iz0,jz1);
 232             dx02             = _mm_sub_pd(ix0,jx2);
 233             dy02             = _mm_sub_pd(iy0,jy2);
 234             dz02             = _mm_sub_pd(iz0,jz2);
 235             dx10             = _mm_sub_pd(ix1,jx0);
 236             dy10             = _mm_sub_pd(iy1,jy0);
 237             dz10             = _mm_sub_pd(iz1,jz0);
 238             dx11             = _mm_sub_pd(ix1,jx1);
 239             dy11             = _mm_sub_pd(iy1,jy1);
 240             dz11             = _mm_sub_pd(iz1,jz1);
 241             dx12             = _mm_sub_pd(ix1,jx2);
 242             dy12             = _mm_sub_pd(iy1,jy2);
 243             dz12             = _mm_sub_pd(iz1,jz2);
 244             dx20             = _mm_sub_pd(ix2,jx0);
 245             dy20             = _mm_sub_pd(iy2,jy0);
 246             dz20             = _mm_sub_pd(iz2,jz0);
 247             dx21             = _mm_sub_pd(ix2,jx1);
 248             dy21             = _mm_sub_pd(iy2,jy1);
 249             dz21             = _mm_sub_pd(iz2,jz1);
 250             dx22             = _mm_sub_pd(ix2,jx2);
 251             dy22             = _mm_sub_pd(iy2,jy2);
 252             dz22             = _mm_sub_pd(iz2,jz2);
 253
 254             /* Calculate squared distance and things based on it */
 255             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 256             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 257             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 258             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 259             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 260             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 261             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 262             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 263             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 264
 265             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 266             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 267             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 268             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 269             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 270             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 271             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 272             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 273             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 274
 275             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 276             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 277             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 278             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 279             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 280             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 281             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 282             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 283             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 284
 285             fjx0             = _mm_setzero_pd();
 286             fjy0             = _mm_setzero_pd();
 287             fjz0             = _mm_setzero_pd();
 288             fjx1             = _mm_setzero_pd();
 289             fjy1             = _mm_setzero_pd();
 290             fjz1             = _mm_setzero_pd();
 291             fjx2             = _mm_setzero_pd();
 292             fjy2             = _mm_setzero_pd();
 293             fjz2             = _mm_setzero_pd();
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             if (gmx_mm_any_lt(rsq00,rcutoff2))
 300             {
 301
 302             r00              = _mm_mul_pd(rsq00,rinv00);
 303
 304             /* EWALD ELECTROSTATICS */
 305
 306             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 307             ewrt             = _mm_mul_pd(r00,ewtabscale);
 308             ewitab           = _mm_cvttpd_epi32(ewrt);
 309 #ifdef __XOP__
 310             eweps            = _mm_frcz_pd(ewrt);
 311 #else
 312             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 313 #endif
 314             twoeweps         = _mm_add_pd(eweps,eweps);
 315             ewitab           = _mm_slli_epi32(ewitab,2);
 316             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 317             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 318             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 319             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 320             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 321             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 322             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 323             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 324             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 325             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 326
 327             /* LENNARD-JONES DISPERSION/REPULSION */
 328
 329             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 330             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 331             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 332             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 333                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 334             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 335
 336             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm_and_pd(velec,cutoff_mask);
 340             velecsum         = _mm_add_pd(velecsum,velec);
 341             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 342             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 343
 344             fscal            = _mm_add_pd(felec,fvdw);
 345
 346             fscal            = _mm_and_pd(fscal,cutoff_mask);
 347
 348             /* Update vectorial force */
 349             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 350             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 351             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 352
 353             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 354             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 355             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 356
 357             }
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             if (gmx_mm_any_lt(rsq01,rcutoff2))
 364             {
 365
 366             r01              = _mm_mul_pd(rsq01,rinv01);
 367
 368             /* EWALD ELECTROSTATICS */
 369
 370             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 371             ewrt             = _mm_mul_pd(r01,ewtabscale);
 372             ewitab           = _mm_cvttpd_epi32(ewrt);
 373 #ifdef __XOP__
 374             eweps            = _mm_frcz_pd(ewrt);
 375 #else
 376             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 377 #endif
 378             twoeweps         = _mm_add_pd(eweps,eweps);
 379             ewitab           = _mm_slli_epi32(ewitab,2);
 380             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 381             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 382             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 383             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 384             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 385             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 386             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 387             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 388             velec            = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
 389             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 390
 391             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velec            = _mm_and_pd(velec,cutoff_mask);
 395             velecsum         = _mm_add_pd(velecsum,velec);
 396
 397             fscal            = felec;
 398
 399             fscal            = _mm_and_pd(fscal,cutoff_mask);
 400
 401             /* Update vectorial force */
 402             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 403             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 404             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 405
 406             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 407             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 408             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 409
 410             }
 411
 412             /**************************
 413              * CALCULATE INTERACTIONS *
 414              **************************/
 415
 416             if (gmx_mm_any_lt(rsq02,rcutoff2))
 417             {
 418
 419             r02              = _mm_mul_pd(rsq02,rinv02);
 420
 421             /* EWALD ELECTROSTATICS */
 422
 423             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 424             ewrt             = _mm_mul_pd(r02,ewtabscale);
 425             ewitab           = _mm_cvttpd_epi32(ewrt);
 426 #ifdef __XOP__
 427             eweps            = _mm_frcz_pd(ewrt);
 428 #else
 429             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 430 #endif
 431             twoeweps         = _mm_add_pd(eweps,eweps);
 432             ewitab           = _mm_slli_epi32(ewitab,2);
 433             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 434             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 435             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 436             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 437             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 438             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 439             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 440             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 441             velec            = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
 442             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 443
 444             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
 445
 446             /* Update potential sum for this i atom from the interaction with this j atom. */
 447             velec            = _mm_and_pd(velec,cutoff_mask);
 448             velecsum         = _mm_add_pd(velecsum,velec);
 449
 450             fscal            = felec;
 451
 452             fscal            = _mm_and_pd(fscal,cutoff_mask);
 453
 454             /* Update vectorial force */
 455             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 456             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 457             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 458
 459             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 460             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 461             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 462
 463             }
 464
 465             /**************************
 466              * CALCULATE INTERACTIONS *
 467              **************************/
 468
 469             if (gmx_mm_any_lt(rsq10,rcutoff2))
 470             {
 471
 472             r10              = _mm_mul_pd(rsq10,rinv10);
 473
 474             /* EWALD ELECTROSTATICS */
 475
 476             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 477             ewrt             = _mm_mul_pd(r10,ewtabscale);
 478             ewitab           = _mm_cvttpd_epi32(ewrt);
 479 #ifdef __XOP__
 480             eweps            = _mm_frcz_pd(ewrt);
 481 #else
 482             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 483 #endif
 484             twoeweps         = _mm_add_pd(eweps,eweps);
 485             ewitab           = _mm_slli_epi32(ewitab,2);
 486             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 487             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 488             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 489             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 490             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 491             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 492             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 493             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 494             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 495             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 496
 497             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 498
 499             /* Update potential sum for this i atom from the interaction with this j atom. */
 500             velec            = _mm_and_pd(velec,cutoff_mask);
 501             velecsum         = _mm_add_pd(velecsum,velec);
 502
 503             fscal            = felec;
 504
 505             fscal            = _mm_and_pd(fscal,cutoff_mask);
 506
 507             /* Update vectorial force */
 508             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 509             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 510             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 511
 512             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 513             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 514             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 515
 516             }
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             if (gmx_mm_any_lt(rsq11,rcutoff2))
 523             {
 524
 525             r11              = _mm_mul_pd(rsq11,rinv11);
 526
 527             /* EWALD ELECTROSTATICS */
 528
 529             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 530             ewrt             = _mm_mul_pd(r11,ewtabscale);
 531             ewitab           = _mm_cvttpd_epi32(ewrt);
 532 #ifdef __XOP__
 533             eweps            = _mm_frcz_pd(ewrt);
 534 #else
 535             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 536 #endif
 537             twoeweps         = _mm_add_pd(eweps,eweps);
 538             ewitab           = _mm_slli_epi32(ewitab,2);
 539             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 540             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 541             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 542             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 543             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 544             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 545             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 546             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 547             velec            = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
 548             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 549
 550             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velec            = _mm_and_pd(velec,cutoff_mask);
 554             velecsum         = _mm_add_pd(velecsum,velec);
 555
 556             fscal            = felec;
 557
 558             fscal            = _mm_and_pd(fscal,cutoff_mask);
 559
 560             /* Update vectorial force */
 561             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 562             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 563             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 564
 565             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 566             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 567             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 568
 569             }
 570
 571             /**************************
 572              * CALCULATE INTERACTIONS *
 573              **************************/
 574
 575             if (gmx_mm_any_lt(rsq12,rcutoff2))
 576             {
 577
 578             r12              = _mm_mul_pd(rsq12,rinv12);
 579
 580             /* EWALD ELECTROSTATICS */
 581
 582             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 583             ewrt             = _mm_mul_pd(r12,ewtabscale);
 584             ewitab           = _mm_cvttpd_epi32(ewrt);
 585 #ifdef __XOP__
 586             eweps            = _mm_frcz_pd(ewrt);
 587 #else
 588             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 589 #endif
 590             twoeweps         = _mm_add_pd(eweps,eweps);
 591             ewitab           = _mm_slli_epi32(ewitab,2);
 592             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 593             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 594             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 595             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 596             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 597             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 598             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 599             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 600             velec            = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
 601             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 602
 603             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm_and_pd(velec,cutoff_mask);
 607             velecsum         = _mm_add_pd(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm_and_pd(fscal,cutoff_mask);
 612
 613             /* Update vectorial force */
 614             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 615             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 616             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 617
 618             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 619             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 620             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 621
 622             }
 623
 624             /**************************
 625              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             if (gmx_mm_any_lt(rsq20,rcutoff2))
 629             {
 630
 631             r20              = _mm_mul_pd(rsq20,rinv20);
 632
 633             /* EWALD ELECTROSTATICS */
 634
 635             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 636             ewrt             = _mm_mul_pd(r20,ewtabscale);
 637             ewitab           = _mm_cvttpd_epi32(ewrt);
 638 #ifdef __XOP__
 639             eweps            = _mm_frcz_pd(ewrt);
 640 #else
 641             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 642 #endif
 643             twoeweps         = _mm_add_pd(eweps,eweps);
 644             ewitab           = _mm_slli_epi32(ewitab,2);
 645             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 646             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 647             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 648             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 649             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 650             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 651             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 652             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 653             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 654             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 655
 656             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 657
 658             /* Update potential sum for this i atom from the interaction with this j atom. */
 659             velec            = _mm_and_pd(velec,cutoff_mask);
 660             velecsum         = _mm_add_pd(velecsum,velec);
 661
 662             fscal            = felec;
 663
 664             fscal            = _mm_and_pd(fscal,cutoff_mask);
 665
 666             /* Update vectorial force */
 667             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 668             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 669             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 670
 671             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 672             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 673             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 674
 675             }
 676
 677             /**************************
 678              * CALCULATE INTERACTIONS *
 679              **************************/
 680
 681             if (gmx_mm_any_lt(rsq21,rcutoff2))
 682             {
 683
 684             r21              = _mm_mul_pd(rsq21,rinv21);
 685
 686             /* EWALD ELECTROSTATICS */
 687
 688             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 689             ewrt             = _mm_mul_pd(r21,ewtabscale);
 690             ewitab           = _mm_cvttpd_epi32(ewrt);
 691 #ifdef __XOP__
 692             eweps            = _mm_frcz_pd(ewrt);
 693 #else
 694             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 695 #endif
 696             twoeweps         = _mm_add_pd(eweps,eweps);
 697             ewitab           = _mm_slli_epi32(ewitab,2);
 698             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 699             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 700             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 701             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 702             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 703             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 704             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 705             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 706             velec            = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
 707             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 708
 709             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
 710
 711             /* Update potential sum for this i atom from the interaction with this j atom. */
 712             velec            = _mm_and_pd(velec,cutoff_mask);
 713             velecsum         = _mm_add_pd(velecsum,velec);
 714
 715             fscal            = felec;
 716
 717             fscal            = _mm_and_pd(fscal,cutoff_mask);
 718
 719             /* Update vectorial force */
 720             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 721             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 722             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 723
 724             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 725             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 726             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 727
 728             }
 729
 730             /**************************
 731              * CALCULATE INTERACTIONS *
 732              **************************/
 733
 734             if (gmx_mm_any_lt(rsq22,rcutoff2))
 735             {
 736
 737             r22              = _mm_mul_pd(rsq22,rinv22);
 738
 739             /* EWALD ELECTROSTATICS */
 740
 741             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 742             ewrt             = _mm_mul_pd(r22,ewtabscale);
 743             ewitab           = _mm_cvttpd_epi32(ewrt);
 744 #ifdef __XOP__
 745             eweps            = _mm_frcz_pd(ewrt);
 746 #else
 747             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 748 #endif
 749             twoeweps         = _mm_add_pd(eweps,eweps);
 750             ewitab           = _mm_slli_epi32(ewitab,2);
 751             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 752             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 753             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 754             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 755             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 756             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 757             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 758             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 759             velec            = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
 760             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 761
 762             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
 763
 764             /* Update potential sum for this i atom from the interaction with this j atom. */
 765             velec            = _mm_and_pd(velec,cutoff_mask);
 766             velecsum         = _mm_add_pd(velecsum,velec);
 767
 768             fscal            = felec;
 769
 770             fscal            = _mm_and_pd(fscal,cutoff_mask);
 771
 772             /* Update vectorial force */
 773             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 774             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 775             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 776
 777             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 778             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 779             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 780
 781             }
 782
 783             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 784
 785             /* Inner loop uses 459 flops */
 786         }
 787
 788         if(jidx<j_index_end)
 789         {
 790
 791             jnrA             = jjnr[jidx];
 792             j_coord_offsetA  = DIM*jnrA;
 793
 794             /* load j atom coordinates */
 795             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 796                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 797
 798             /* Calculate displacement vector */
 799             dx00             = _mm_sub_pd(ix0,jx0);
 800             dy00             = _mm_sub_pd(iy0,jy0);
 801             dz00             = _mm_sub_pd(iz0,jz0);
 802             dx01             = _mm_sub_pd(ix0,jx1);
 803             dy01             = _mm_sub_pd(iy0,jy1);
 804             dz01             = _mm_sub_pd(iz0,jz1);
 805             dx02             = _mm_sub_pd(ix0,jx2);
 806             dy02             = _mm_sub_pd(iy0,jy2);
 807             dz02             = _mm_sub_pd(iz0,jz2);
 808             dx10             = _mm_sub_pd(ix1,jx0);
 809             dy10             = _mm_sub_pd(iy1,jy0);
 810             dz10             = _mm_sub_pd(iz1,jz0);
 811             dx11             = _mm_sub_pd(ix1,jx1);
 812             dy11             = _mm_sub_pd(iy1,jy1);
 813             dz11             = _mm_sub_pd(iz1,jz1);
 814             dx12             = _mm_sub_pd(ix1,jx2);
 815             dy12             = _mm_sub_pd(iy1,jy2);
 816             dz12             = _mm_sub_pd(iz1,jz2);
 817             dx20             = _mm_sub_pd(ix2,jx0);
 818             dy20             = _mm_sub_pd(iy2,jy0);
 819             dz20             = _mm_sub_pd(iz2,jz0);
 820             dx21             = _mm_sub_pd(ix2,jx1);
 821             dy21             = _mm_sub_pd(iy2,jy1);
 822             dz21             = _mm_sub_pd(iz2,jz1);
 823             dx22             = _mm_sub_pd(ix2,jx2);
 824             dy22             = _mm_sub_pd(iy2,jy2);
 825             dz22             = _mm_sub_pd(iz2,jz2);
 826
 827             /* Calculate squared distance and things based on it */
 828             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 829             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 830             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 831             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 832             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 833             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 834             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 835             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 836             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 837
 838             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 839             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 840             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 841             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 842             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 843             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 844             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 845             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 846             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 847
 848             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 849             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 850             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 851             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 852             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 853             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 854             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 855             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 856             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 857
 858             fjx0             = _mm_setzero_pd();
 859             fjy0             = _mm_setzero_pd();
 860             fjz0             = _mm_setzero_pd();
 861             fjx1             = _mm_setzero_pd();
 862             fjy1             = _mm_setzero_pd();
 863             fjz1             = _mm_setzero_pd();
 864             fjx2             = _mm_setzero_pd();
 865             fjy2             = _mm_setzero_pd();
 866             fjz2             = _mm_setzero_pd();
 867
 868             /**************************
 869              * CALCULATE INTERACTIONS *
 870              **************************/
 871
 872             if (gmx_mm_any_lt(rsq00,rcutoff2))
 873             {
 874
 875             r00              = _mm_mul_pd(rsq00,rinv00);
 876
 877             /* EWALD ELECTROSTATICS */
 878
 879             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 880             ewrt             = _mm_mul_pd(r00,ewtabscale);
 881             ewitab           = _mm_cvttpd_epi32(ewrt);
 882 #ifdef __XOP__
 883             eweps            = _mm_frcz_pd(ewrt);
 884 #else
 885             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 886 #endif
 887             twoeweps         = _mm_add_pd(eweps,eweps);
 888             ewitab           = _mm_slli_epi32(ewitab,2);
 889             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 890             ewtabD           = _mm_setzero_pd();
 891             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 892             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 893             ewtabFn          = _mm_setzero_pd();
 894             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 895             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 896             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 897             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 898             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 899
 900             /* LENNARD-JONES DISPERSION/REPULSION */
 901
 902             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 903             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 904             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 905             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 906                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 907             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 908
 909             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 910
 911             /* Update potential sum for this i atom from the interaction with this j atom. */
 912             velec            = _mm_and_pd(velec,cutoff_mask);
 913             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 914             velecsum         = _mm_add_pd(velecsum,velec);
 915             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 916             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 917             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 918
 919             fscal            = _mm_add_pd(felec,fvdw);
 920
 921             fscal            = _mm_and_pd(fscal,cutoff_mask);
 922
 923             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 924
 925             /* Update vectorial force */
 926             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 927             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 928             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 929
 930             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 931             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 932             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 933
 934             }
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             if (gmx_mm_any_lt(rsq01,rcutoff2))
 941             {
 942
 943             r01              = _mm_mul_pd(rsq01,rinv01);
 944
 945             /* EWALD ELECTROSTATICS */
 946
 947             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 948             ewrt             = _mm_mul_pd(r01,ewtabscale);
 949             ewitab           = _mm_cvttpd_epi32(ewrt);
 950 #ifdef __XOP__
 951             eweps            = _mm_frcz_pd(ewrt);
 952 #else
 953             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 954 #endif
 955             twoeweps         = _mm_add_pd(eweps,eweps);
 956             ewitab           = _mm_slli_epi32(ewitab,2);
 957             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 958             ewtabD           = _mm_setzero_pd();
 959             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 960             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 961             ewtabFn          = _mm_setzero_pd();
 962             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 963             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 964             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 965             velec            = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
 966             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 967
 968             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
 969
 970             /* Update potential sum for this i atom from the interaction with this j atom. */
 971             velec            = _mm_and_pd(velec,cutoff_mask);
 972             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 973             velecsum         = _mm_add_pd(velecsum,velec);
 974
 975             fscal            = felec;
 976
 977             fscal            = _mm_and_pd(fscal,cutoff_mask);
 978
 979             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 980
 981             /* Update vectorial force */
 982             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 983             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 984             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 985
 986             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 987             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 988             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 989
 990             }
 991
 992             /**************************
 993              * CALCULATE INTERACTIONS *
 994              **************************/
 995
 996             if (gmx_mm_any_lt(rsq02,rcutoff2))
 997             {
 998
 999             r02              = _mm_mul_pd(rsq02,rinv02);
1000
1001             /* EWALD ELECTROSTATICS */
1002
1003             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1004             ewrt             = _mm_mul_pd(r02,ewtabscale);
1005             ewitab           = _mm_cvttpd_epi32(ewrt);
1006 #ifdef __XOP__
1007             eweps            = _mm_frcz_pd(ewrt);
1008 #else
1009             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1010 #endif
1011             twoeweps         = _mm_add_pd(eweps,eweps);
1012             ewitab           = _mm_slli_epi32(ewitab,2);
1013             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1014             ewtabD           = _mm_setzero_pd();
1015             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1016             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1017             ewtabFn          = _mm_setzero_pd();
1018             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1019             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1020             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1021             velec            = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
1022             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1023
1024             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
1025
1026             /* Update potential sum for this i atom from the interaction with this j atom. */
1027             velec            = _mm_and_pd(velec,cutoff_mask);
1028             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1029             velecsum         = _mm_add_pd(velecsum,velec);
1030
1031             fscal            = felec;
1032
1033             fscal            = _mm_and_pd(fscal,cutoff_mask);
1034
1035             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1036
1037             /* Update vectorial force */
1038             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1039             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1040             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1041
1042             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1043             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1044             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1045
1046             }
1047
1048             /**************************
1049              * CALCULATE INTERACTIONS *
1050              **************************/
1051
1052             if (gmx_mm_any_lt(rsq10,rcutoff2))
1053             {
1054
1055             r10              = _mm_mul_pd(rsq10,rinv10);
1056
1057             /* EWALD ELECTROSTATICS */
1058
1059             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1060             ewrt             = _mm_mul_pd(r10,ewtabscale);
1061             ewitab           = _mm_cvttpd_epi32(ewrt);
1062 #ifdef __XOP__
1063             eweps            = _mm_frcz_pd(ewrt);
1064 #else
1065             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1066 #endif
1067             twoeweps         = _mm_add_pd(eweps,eweps);
1068             ewitab           = _mm_slli_epi32(ewitab,2);
1069             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1070             ewtabD           = _mm_setzero_pd();
1071             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1072             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1073             ewtabFn          = _mm_setzero_pd();
1074             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1075             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1076             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1077             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
1078             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1079
1080             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1081
1082             /* Update potential sum for this i atom from the interaction with this j atom. */
1083             velec            = _mm_and_pd(velec,cutoff_mask);
1084             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1085             velecsum         = _mm_add_pd(velecsum,velec);
1086
1087             fscal            = felec;
1088
1089             fscal            = _mm_and_pd(fscal,cutoff_mask);
1090
1091             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1092
1093             /* Update vectorial force */
1094             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1095             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1096             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1097
1098             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1099             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1100             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1101
1102             }
1103
1104             /**************************
1105              * CALCULATE INTERACTIONS *
1106              **************************/
1107
1108             if (gmx_mm_any_lt(rsq11,rcutoff2))
1109             {
1110
1111             r11              = _mm_mul_pd(rsq11,rinv11);
1112
1113             /* EWALD ELECTROSTATICS */
1114
1115             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1116             ewrt             = _mm_mul_pd(r11,ewtabscale);
1117             ewitab           = _mm_cvttpd_epi32(ewrt);
1118 #ifdef __XOP__
1119             eweps            = _mm_frcz_pd(ewrt);
1120 #else
1121             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1122 #endif
1123             twoeweps         = _mm_add_pd(eweps,eweps);
1124             ewitab           = _mm_slli_epi32(ewitab,2);
1125             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1126             ewtabD           = _mm_setzero_pd();
1127             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1128             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1129             ewtabFn          = _mm_setzero_pd();
1130             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1131             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1132             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1133             velec            = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
1134             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1135
1136             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
1137
1138             /* Update potential sum for this i atom from the interaction with this j atom. */
1139             velec            = _mm_and_pd(velec,cutoff_mask);
1140             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1141             velecsum         = _mm_add_pd(velecsum,velec);
1142
1143             fscal            = felec;
1144
1145             fscal            = _mm_and_pd(fscal,cutoff_mask);
1146
1147             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1148
1149             /* Update vectorial force */
1150             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1151             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1152             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1153
1154             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1155             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1156             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1157
1158             }
1159
1160             /**************************
1161              * CALCULATE INTERACTIONS *
1162              **************************/
1163
1164             if (gmx_mm_any_lt(rsq12,rcutoff2))
1165             {
1166
1167             r12              = _mm_mul_pd(rsq12,rinv12);
1168
1169             /* EWALD ELECTROSTATICS */
1170
1171             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1172             ewrt             = _mm_mul_pd(r12,ewtabscale);
1173             ewitab           = _mm_cvttpd_epi32(ewrt);
1174 #ifdef __XOP__
1175             eweps            = _mm_frcz_pd(ewrt);
1176 #else
1177             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1178 #endif
1179             twoeweps         = _mm_add_pd(eweps,eweps);
1180             ewitab           = _mm_slli_epi32(ewitab,2);
1181             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1182             ewtabD           = _mm_setzero_pd();
1183             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1184             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1185             ewtabFn          = _mm_setzero_pd();
1186             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1187             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1188             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1189             velec            = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
1190             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1191
1192             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
1193
1194             /* Update potential sum for this i atom from the interaction with this j atom. */
1195             velec            = _mm_and_pd(velec,cutoff_mask);
1196             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1197             velecsum         = _mm_add_pd(velecsum,velec);
1198
1199             fscal            = felec;
1200
1201             fscal            = _mm_and_pd(fscal,cutoff_mask);
1202
1203             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1204
1205             /* Update vectorial force */
1206             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1207             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1208             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1209
1210             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1211             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1212             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1213
1214             }
1215
1216             /**************************
1217              * CALCULATE INTERACTIONS *
1218              **************************/
1219
1220             if (gmx_mm_any_lt(rsq20,rcutoff2))
1221             {
1222
1223             r20              = _mm_mul_pd(rsq20,rinv20);
1224
1225             /* EWALD ELECTROSTATICS */
1226
1227             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1228             ewrt             = _mm_mul_pd(r20,ewtabscale);
1229             ewitab           = _mm_cvttpd_epi32(ewrt);
1230 #ifdef __XOP__
1231             eweps            = _mm_frcz_pd(ewrt);
1232 #else
1233             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1234 #endif
1235             twoeweps         = _mm_add_pd(eweps,eweps);
1236             ewitab           = _mm_slli_epi32(ewitab,2);
1237             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1238             ewtabD           = _mm_setzero_pd();
1239             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1240             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1241             ewtabFn          = _mm_setzero_pd();
1242             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1243             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1244             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1245             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
1246             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1247
1248             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1249
1250             /* Update potential sum for this i atom from the interaction with this j atom. */
1251             velec            = _mm_and_pd(velec,cutoff_mask);
1252             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1253             velecsum         = _mm_add_pd(velecsum,velec);
1254
1255             fscal            = felec;
1256
1257             fscal            = _mm_and_pd(fscal,cutoff_mask);
1258
1259             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1260
1261             /* Update vectorial force */
1262             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1263             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1264             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1265
1266             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1267             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1268             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1269
1270             }
1271
1272             /**************************
1273              * CALCULATE INTERACTIONS *
1274              **************************/
1275
1276             if (gmx_mm_any_lt(rsq21,rcutoff2))
1277             {
1278
1279             r21              = _mm_mul_pd(rsq21,rinv21);
1280
1281             /* EWALD ELECTROSTATICS */
1282
1283             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1284             ewrt             = _mm_mul_pd(r21,ewtabscale);
1285             ewitab           = _mm_cvttpd_epi32(ewrt);
1286 #ifdef __XOP__
1287             eweps            = _mm_frcz_pd(ewrt);
1288 #else
1289             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1290 #endif
1291             twoeweps         = _mm_add_pd(eweps,eweps);
1292             ewitab           = _mm_slli_epi32(ewitab,2);
1293             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1294             ewtabD           = _mm_setzero_pd();
1295             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1296             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1297             ewtabFn          = _mm_setzero_pd();
1298             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1299             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1300             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1301             velec            = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
1302             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1303
1304             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
1305
1306             /* Update potential sum for this i atom from the interaction with this j atom. */
1307             velec            = _mm_and_pd(velec,cutoff_mask);
1308             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1309             velecsum         = _mm_add_pd(velecsum,velec);
1310
1311             fscal            = felec;
1312
1313             fscal            = _mm_and_pd(fscal,cutoff_mask);
1314
1315             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1316
1317             /* Update vectorial force */
1318             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1319             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1320             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1321
1322             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1323             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1324             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1325
1326             }
1327
1328             /**************************
1329              * CALCULATE INTERACTIONS *
1330              **************************/
1331
1332             if (gmx_mm_any_lt(rsq22,rcutoff2))
1333             {
1334
1335             r22              = _mm_mul_pd(rsq22,rinv22);
1336
1337             /* EWALD ELECTROSTATICS */
1338
1339             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1340             ewrt             = _mm_mul_pd(r22,ewtabscale);
1341             ewitab           = _mm_cvttpd_epi32(ewrt);
1342 #ifdef __XOP__
1343             eweps            = _mm_frcz_pd(ewrt);
1344 #else
1345             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1346 #endif
1347             twoeweps         = _mm_add_pd(eweps,eweps);
1348             ewitab           = _mm_slli_epi32(ewitab,2);
1349             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1350             ewtabD           = _mm_setzero_pd();
1351             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1352             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1353             ewtabFn          = _mm_setzero_pd();
1354             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1355             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1356             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1357             velec            = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
1358             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1359
1360             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
1361
1362             /* Update potential sum for this i atom from the interaction with this j atom. */
1363             velec            = _mm_and_pd(velec,cutoff_mask);
1364             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1365             velecsum         = _mm_add_pd(velecsum,velec);
1366
1367             fscal            = felec;
1368
1369             fscal            = _mm_and_pd(fscal,cutoff_mask);
1370
1371             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1372
1373             /* Update vectorial force */
1374             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1375             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1376             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1377
1378             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1379             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1380             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1381
1382             }
1383
1384             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1385
1386             /* Inner loop uses 459 flops */
1387         }
1388
1389         /* End of innermost loop */
1390
1391         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1392                                               f+i_coord_offset,fshift+i_shift_offset);
1393
1394         ggid                        = gid[iidx];
1395         /* Update potential energies */
1396         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1397         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1398
1399         /* Increment number of inner iterations */
1400         inneriter                  += j_index_end - j_index_start;
1401
1402         /* Outer loop uses 20 flops */
1403     }
1404
1405     /* Increment number of outer iterations */
1406     outeriter        += nri;
1407
1408     /* Update outer/inner flops */
1409
1410     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*459);
1411 }
1412 /*
1413  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3W3_F_avx_128_fma_double
1414  * Electrostatics interaction: Ewald
1415  * VdW interaction:            LennardJones
1416  * Geometry:                   Water3-Water3
1417  * Calculate force/pot:        Force
1418  */
1419 void
1420 nb_kernel_ElecEwSh_VdwLJSh_GeomW3W3_F_avx_128_fma_double
1421                     (t_nblist                    * gmx_restrict       nlist,
1422                      rvec                        * gmx_restrict          xx,
1423                      rvec                        * gmx_restrict          ff,
1424                      t_forcerec                  * gmx_restrict          fr,
1425                      t_mdatoms                   * gmx_restrict     mdatoms,
1426                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1427                      t_nrnb                      * gmx_restrict        nrnb)
1428 {
1429     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1430      * just 0 for non-waters.
1431      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1432      * jnr indices corresponding to data put in the four positions in the SIMD register.
1433      */
1434     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1435     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1436     int              jnrA,jnrB;
1437     int              j_coord_offsetA,j_coord_offsetB;
1438     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1439     real             rcutoff_scalar;
1440     real             *shiftvec,*fshift,*x,*f;
1441     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1442     int              vdwioffset0;
1443     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1444     int              vdwioffset1;
1445     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1446     int              vdwioffset2;
1447     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1448     int              vdwjidx0A,vdwjidx0B;
1449     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1450     int              vdwjidx1A,vdwjidx1B;
1451     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1452     int              vdwjidx2A,vdwjidx2B;
1453     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1454     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1455     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1456     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1457     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1458     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1459     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1460     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1461     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1462     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1463     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1464     real             *charge;
1465     int              nvdwtype;
1466     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1467     int              *vdwtype;
1468     real             *vdwparam;
1469     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
1470     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
1471     __m128i          ewitab;
1472     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1473     real             *ewtab;
1474     __m128d          dummy_mask,cutoff_mask;
1475     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1476     __m128d          one     = _mm_set1_pd(1.0);
1477     __m128d          two     = _mm_set1_pd(2.0);
1478     x                = xx[0];
1479     f                = ff[0];
1480
1481     nri              = nlist->nri;
1482     iinr             = nlist->iinr;
1483     jindex           = nlist->jindex;
1484     jjnr             = nlist->jjnr;
1485     shiftidx         = nlist->shift;
1486     gid              = nlist->gid;
1487     shiftvec         = fr->shift_vec[0];
1488     fshift           = fr->fshift[0];
1489     facel            = _mm_set1_pd(fr->epsfac);
1490     charge           = mdatoms->chargeA;
1491     nvdwtype         = fr->ntype;
1492     vdwparam         = fr->nbfp;
1493     vdwtype          = mdatoms->typeA;
1494
1495     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1496     ewtab            = fr->ic->tabq_coul_F;
1497     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1498     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1499
1500     /* Setup water-specific parameters */
1501     inr              = nlist->iinr[0];
1502     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1503     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1504     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1505     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1506
1507     jq0              = _mm_set1_pd(charge[inr+0]);
1508     jq1              = _mm_set1_pd(charge[inr+1]);
1509     jq2              = _mm_set1_pd(charge[inr+2]);
1510     vdwjidx0A        = 2*vdwtype[inr+0];
1511     qq00             = _mm_mul_pd(iq0,jq0);
1512     c6_00            = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A]);
1513     c12_00           = _mm_set1_pd(vdwparam[vdwioffset0+vdwjidx0A+1]);
1514     qq01             = _mm_mul_pd(iq0,jq1);
1515     qq02             = _mm_mul_pd(iq0,jq2);
1516     qq10             = _mm_mul_pd(iq1,jq0);
1517     qq11             = _mm_mul_pd(iq1,jq1);
1518     qq12             = _mm_mul_pd(iq1,jq2);
1519     qq20             = _mm_mul_pd(iq2,jq0);
1520     qq21             = _mm_mul_pd(iq2,jq1);
1521     qq22             = _mm_mul_pd(iq2,jq2);
1522
1523     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1524     rcutoff_scalar   = fr->rcoulomb;
1525     rcutoff          = _mm_set1_pd(rcutoff_scalar);
1526     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
1527
1528     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
1529     rvdw             = _mm_set1_pd(fr->rvdw);
1530
1531     /* Avoid stupid compiler warnings */
1532     jnrA = jnrB = 0;
1533     j_coord_offsetA = 0;
1534     j_coord_offsetB = 0;
1535
1536     outeriter        = 0;
1537     inneriter        = 0;
1538
1539     /* Start outer loop over neighborlists */
1540     for(iidx=0; iidx<nri; iidx++)
1541     {
1542         /* Load shift vector for this list */
1543         i_shift_offset   = DIM*shiftidx[iidx];
1544
1545         /* Load limits for loop over neighbors */
1546         j_index_start    = jindex[iidx];
1547         j_index_end      = jindex[iidx+1];
1548
1549         /* Get outer coordinate index */
1550         inr              = iinr[iidx];
1551         i_coord_offset   = DIM*inr;
1552
1553         /* Load i particle coords and add shift vector */
1554         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1555                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1556
1557         fix0             = _mm_setzero_pd();
1558         fiy0             = _mm_setzero_pd();
1559         fiz0             = _mm_setzero_pd();
1560         fix1             = _mm_setzero_pd();
1561         fiy1             = _mm_setzero_pd();
1562         fiz1             = _mm_setzero_pd();
1563         fix2             = _mm_setzero_pd();
1564         fiy2             = _mm_setzero_pd();
1565         fiz2             = _mm_setzero_pd();
1566
1567         /* Start inner kernel loop */
1568         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1569         {
1570
1571             /* Get j neighbor index, and coordinate index */
1572             jnrA             = jjnr[jidx];
1573             jnrB             = jjnr[jidx+1];
1574             j_coord_offsetA  = DIM*jnrA;
1575             j_coord_offsetB  = DIM*jnrB;
1576
1577             /* load j atom coordinates */
1578             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1579                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1580
1581             /* Calculate displacement vector */
1582             dx00             = _mm_sub_pd(ix0,jx0);
1583             dy00             = _mm_sub_pd(iy0,jy0);
1584             dz00             = _mm_sub_pd(iz0,jz0);
1585             dx01             = _mm_sub_pd(ix0,jx1);
1586             dy01             = _mm_sub_pd(iy0,jy1);
1587             dz01             = _mm_sub_pd(iz0,jz1);
1588             dx02             = _mm_sub_pd(ix0,jx2);
1589             dy02             = _mm_sub_pd(iy0,jy2);
1590             dz02             = _mm_sub_pd(iz0,jz2);
1591             dx10             = _mm_sub_pd(ix1,jx0);
1592             dy10             = _mm_sub_pd(iy1,jy0);
1593             dz10             = _mm_sub_pd(iz1,jz0);
1594             dx11             = _mm_sub_pd(ix1,jx1);
1595             dy11             = _mm_sub_pd(iy1,jy1);
1596             dz11             = _mm_sub_pd(iz1,jz1);
1597             dx12             = _mm_sub_pd(ix1,jx2);
1598             dy12             = _mm_sub_pd(iy1,jy2);
1599             dz12             = _mm_sub_pd(iz1,jz2);
1600             dx20             = _mm_sub_pd(ix2,jx0);
1601             dy20             = _mm_sub_pd(iy2,jy0);
1602             dz20             = _mm_sub_pd(iz2,jz0);
1603             dx21             = _mm_sub_pd(ix2,jx1);
1604             dy21             = _mm_sub_pd(iy2,jy1);
1605             dz21             = _mm_sub_pd(iz2,jz1);
1606             dx22             = _mm_sub_pd(ix2,jx2);
1607             dy22             = _mm_sub_pd(iy2,jy2);
1608             dz22             = _mm_sub_pd(iz2,jz2);
1609
1610             /* Calculate squared distance and things based on it */
1611             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1612             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1613             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1614             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1615             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1616             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1617             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1618             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1619             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1620
1621             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1622             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1623             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1624             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1625             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1626             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1627             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1628             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1629             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1630
1631             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1632             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1633             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1634             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1635             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1636             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1637             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1638             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1639             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1640
1641             fjx0             = _mm_setzero_pd();
1642             fjy0             = _mm_setzero_pd();
1643             fjz0             = _mm_setzero_pd();
1644             fjx1             = _mm_setzero_pd();
1645             fjy1             = _mm_setzero_pd();
1646             fjz1             = _mm_setzero_pd();
1647             fjx2             = _mm_setzero_pd();
1648             fjy2             = _mm_setzero_pd();
1649             fjz2             = _mm_setzero_pd();
1650
1651             /**************************
1652              * CALCULATE INTERACTIONS *
1653              **************************/
1654
1655             if (gmx_mm_any_lt(rsq00,rcutoff2))
1656             {
1657
1658             r00              = _mm_mul_pd(rsq00,rinv00);
1659
1660             /* EWALD ELECTROSTATICS */
1661
1662             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1663             ewrt             = _mm_mul_pd(r00,ewtabscale);
1664             ewitab           = _mm_cvttpd_epi32(ewrt);
1665 #ifdef __XOP__
1666             eweps            = _mm_frcz_pd(ewrt);
1667 #else
1668             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1669 #endif
1670             twoeweps         = _mm_add_pd(eweps,eweps);
1671             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1672                                          &ewtabF,&ewtabFn);
1673             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1674             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1675
1676             /* LENNARD-JONES DISPERSION/REPULSION */
1677
1678             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1679             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1680
1681             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1682
1683             fscal            = _mm_add_pd(felec,fvdw);
1684
1685             fscal            = _mm_and_pd(fscal,cutoff_mask);
1686
1687             /* Update vectorial force */
1688             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1689             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1690             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1691
1692             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1693             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1694             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1695
1696             }
1697
1698             /**************************
1699              * CALCULATE INTERACTIONS *
1700              **************************/
1701
1702             if (gmx_mm_any_lt(rsq01,rcutoff2))
1703             {
1704
1705             r01              = _mm_mul_pd(rsq01,rinv01);
1706
1707             /* EWALD ELECTROSTATICS */
1708
1709             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1710             ewrt             = _mm_mul_pd(r01,ewtabscale);
1711             ewitab           = _mm_cvttpd_epi32(ewrt);
1712 #ifdef __XOP__
1713             eweps            = _mm_frcz_pd(ewrt);
1714 #else
1715             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1716 #endif
1717             twoeweps         = _mm_add_pd(eweps,eweps);
1718             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1719                                          &ewtabF,&ewtabFn);
1720             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1721             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1722
1723             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
1724
1725             fscal            = felec;
1726
1727             fscal            = _mm_and_pd(fscal,cutoff_mask);
1728
1729             /* Update vectorial force */
1730             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1731             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1732             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1733
1734             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1735             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1736             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1737
1738             }
1739
1740             /**************************
1741              * CALCULATE INTERACTIONS *
1742              **************************/
1743
1744             if (gmx_mm_any_lt(rsq02,rcutoff2))
1745             {
1746
1747             r02              = _mm_mul_pd(rsq02,rinv02);
1748
1749             /* EWALD ELECTROSTATICS */
1750
1751             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1752             ewrt             = _mm_mul_pd(r02,ewtabscale);
1753             ewitab           = _mm_cvttpd_epi32(ewrt);
1754 #ifdef __XOP__
1755             eweps            = _mm_frcz_pd(ewrt);
1756 #else
1757             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1758 #endif
1759             twoeweps         = _mm_add_pd(eweps,eweps);
1760             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1761                                          &ewtabF,&ewtabFn);
1762             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1763             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1764
1765             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
1766
1767             fscal            = felec;
1768
1769             fscal            = _mm_and_pd(fscal,cutoff_mask);
1770
1771             /* Update vectorial force */
1772             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1773             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1774             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1775
1776             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1777             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1778             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1779
1780             }
1781
1782             /**************************
1783              * CALCULATE INTERACTIONS *
1784              **************************/
1785
1786             if (gmx_mm_any_lt(rsq10,rcutoff2))
1787             {
1788
1789             r10              = _mm_mul_pd(rsq10,rinv10);
1790
1791             /* EWALD ELECTROSTATICS */
1792
1793             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1794             ewrt             = _mm_mul_pd(r10,ewtabscale);
1795             ewitab           = _mm_cvttpd_epi32(ewrt);
1796 #ifdef __XOP__
1797             eweps            = _mm_frcz_pd(ewrt);
1798 #else
1799             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1800 #endif
1801             twoeweps         = _mm_add_pd(eweps,eweps);
1802             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1803                                          &ewtabF,&ewtabFn);
1804             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1805             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1806
1807             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1808
1809             fscal            = felec;
1810
1811             fscal            = _mm_and_pd(fscal,cutoff_mask);
1812
1813             /* Update vectorial force */
1814             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1815             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1816             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1817
1818             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1819             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1820             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1821
1822             }
1823
1824             /**************************
1825              * CALCULATE INTERACTIONS *
1826              **************************/
1827
1828             if (gmx_mm_any_lt(rsq11,rcutoff2))
1829             {
1830
1831             r11              = _mm_mul_pd(rsq11,rinv11);
1832
1833             /* EWALD ELECTROSTATICS */
1834
1835             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1836             ewrt             = _mm_mul_pd(r11,ewtabscale);
1837             ewitab           = _mm_cvttpd_epi32(ewrt);
1838 #ifdef __XOP__
1839             eweps            = _mm_frcz_pd(ewrt);
1840 #else
1841             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1842 #endif
1843             twoeweps         = _mm_add_pd(eweps,eweps);
1844             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1845                                          &ewtabF,&ewtabFn);
1846             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1847             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1848
1849             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
1850
1851             fscal            = felec;
1852
1853             fscal            = _mm_and_pd(fscal,cutoff_mask);
1854
1855             /* Update vectorial force */
1856             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1857             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1858             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1859
1860             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1861             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1862             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1863
1864             }
1865
1866             /**************************
1867              * CALCULATE INTERACTIONS *
1868              **************************/
1869
1870             if (gmx_mm_any_lt(rsq12,rcutoff2))
1871             {
1872
1873             r12              = _mm_mul_pd(rsq12,rinv12);
1874
1875             /* EWALD ELECTROSTATICS */
1876
1877             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1878             ewrt             = _mm_mul_pd(r12,ewtabscale);
1879             ewitab           = _mm_cvttpd_epi32(ewrt);
1880 #ifdef __XOP__
1881             eweps            = _mm_frcz_pd(ewrt);
1882 #else
1883             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1884 #endif
1885             twoeweps         = _mm_add_pd(eweps,eweps);
1886             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1887                                          &ewtabF,&ewtabFn);
1888             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1889             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1890
1891             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
1892
1893             fscal            = felec;
1894
1895             fscal            = _mm_and_pd(fscal,cutoff_mask);
1896
1897             /* Update vectorial force */
1898             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1899             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1900             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1901
1902             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1903             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1904             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1905
1906             }
1907
1908             /**************************
1909              * CALCULATE INTERACTIONS *
1910              **************************/
1911
1912             if (gmx_mm_any_lt(rsq20,rcutoff2))
1913             {
1914
1915             r20              = _mm_mul_pd(rsq20,rinv20);
1916
1917             /* EWALD ELECTROSTATICS */
1918
1919             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1920             ewrt             = _mm_mul_pd(r20,ewtabscale);
1921             ewitab           = _mm_cvttpd_epi32(ewrt);
1922 #ifdef __XOP__
1923             eweps            = _mm_frcz_pd(ewrt);
1924 #else
1925             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1926 #endif
1927             twoeweps         = _mm_add_pd(eweps,eweps);
1928             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1929                                          &ewtabF,&ewtabFn);
1930             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1931             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1932
1933             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1934
1935             fscal            = felec;
1936
1937             fscal            = _mm_and_pd(fscal,cutoff_mask);
1938
1939             /* Update vectorial force */
1940             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1941             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1942             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1943
1944             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1945             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1946             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1947
1948             }
1949
1950             /**************************
1951              * CALCULATE INTERACTIONS *
1952              **************************/
1953
1954             if (gmx_mm_any_lt(rsq21,rcutoff2))
1955             {
1956
1957             r21              = _mm_mul_pd(rsq21,rinv21);
1958
1959             /* EWALD ELECTROSTATICS */
1960
1961             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1962             ewrt             = _mm_mul_pd(r21,ewtabscale);
1963             ewitab           = _mm_cvttpd_epi32(ewrt);
1964 #ifdef __XOP__
1965             eweps            = _mm_frcz_pd(ewrt);
1966 #else
1967             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1968 #endif
1969             twoeweps         = _mm_add_pd(eweps,eweps);
1970             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1971                                          &ewtabF,&ewtabFn);
1972             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1973             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1974
1975             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
1976
1977             fscal            = felec;
1978
1979             fscal            = _mm_and_pd(fscal,cutoff_mask);
1980
1981             /* Update vectorial force */
1982             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1983             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1984             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1985
1986             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1987             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1988             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1989
1990             }
1991
1992             /**************************
1993              * CALCULATE INTERACTIONS *
1994              **************************/
1995
1996             if (gmx_mm_any_lt(rsq22,rcutoff2))
1997             {
1998
1999             r22              = _mm_mul_pd(rsq22,rinv22);
2000
2001             /* EWALD ELECTROSTATICS */
2002
2003             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2004             ewrt             = _mm_mul_pd(r22,ewtabscale);
2005             ewitab           = _mm_cvttpd_epi32(ewrt);
2006 #ifdef __XOP__
2007             eweps            = _mm_frcz_pd(ewrt);
2008 #else
2009             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2010 #endif
2011             twoeweps         = _mm_add_pd(eweps,eweps);
2012             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
2013                                          &ewtabF,&ewtabFn);
2014             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2015             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2016
2017             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
2018
2019             fscal            = felec;
2020
2021             fscal            = _mm_and_pd(fscal,cutoff_mask);
2022
2023             /* Update vectorial force */
2024             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2025             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2026             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2027
2028             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2029             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2030             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2031
2032             }
2033
2034             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2035
2036             /* Inner loop uses 385 flops */
2037         }
2038
2039         if(jidx<j_index_end)
2040         {
2041
2042             jnrA             = jjnr[jidx];
2043             j_coord_offsetA  = DIM*jnrA;
2044
2045             /* load j atom coordinates */
2046             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
2047                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2048
2049             /* Calculate displacement vector */
2050             dx00             = _mm_sub_pd(ix0,jx0);
2051             dy00             = _mm_sub_pd(iy0,jy0);
2052             dz00             = _mm_sub_pd(iz0,jz0);
2053             dx01             = _mm_sub_pd(ix0,jx1);
2054             dy01             = _mm_sub_pd(iy0,jy1);
2055             dz01             = _mm_sub_pd(iz0,jz1);
2056             dx02             = _mm_sub_pd(ix0,jx2);
2057             dy02             = _mm_sub_pd(iy0,jy2);
2058             dz02             = _mm_sub_pd(iz0,jz2);
2059             dx10             = _mm_sub_pd(ix1,jx0);
2060             dy10             = _mm_sub_pd(iy1,jy0);
2061             dz10             = _mm_sub_pd(iz1,jz0);
2062             dx11             = _mm_sub_pd(ix1,jx1);
2063             dy11             = _mm_sub_pd(iy1,jy1);
2064             dz11             = _mm_sub_pd(iz1,jz1);
2065             dx12             = _mm_sub_pd(ix1,jx2);
2066             dy12             = _mm_sub_pd(iy1,jy2);
2067             dz12             = _mm_sub_pd(iz1,jz2);
2068             dx20             = _mm_sub_pd(ix2,jx0);
2069             dy20             = _mm_sub_pd(iy2,jy0);
2070             dz20             = _mm_sub_pd(iz2,jz0);
2071             dx21             = _mm_sub_pd(ix2,jx1);
2072             dy21             = _mm_sub_pd(iy2,jy1);
2073             dz21             = _mm_sub_pd(iz2,jz1);
2074             dx22             = _mm_sub_pd(ix2,jx2);
2075             dy22             = _mm_sub_pd(iy2,jy2);
2076             dz22             = _mm_sub_pd(iz2,jz2);
2077
2078             /* Calculate squared distance and things based on it */
2079             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2080             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
2081             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
2082             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
2083             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2084             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2085             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
2086             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2087             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2088
2089             rinv00           = gmx_mm_invsqrt_pd(rsq00);
2090             rinv01           = gmx_mm_invsqrt_pd(rsq01);
2091             rinv02           = gmx_mm_invsqrt_pd(rsq02);
2092             rinv10           = gmx_mm_invsqrt_pd(rsq10);
2093             rinv11           = gmx_mm_invsqrt_pd(rsq11);
2094             rinv12           = gmx_mm_invsqrt_pd(rsq12);
2095             rinv20           = gmx_mm_invsqrt_pd(rsq20);
2096             rinv21           = gmx_mm_invsqrt_pd(rsq21);
2097             rinv22           = gmx_mm_invsqrt_pd(rsq22);
2098
2099             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
2100             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
2101             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
2102             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
2103             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
2104             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
2105             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
2106             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
2107             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
2108
2109             fjx0             = _mm_setzero_pd();
2110             fjy0             = _mm_setzero_pd();
2111             fjz0             = _mm_setzero_pd();
2112             fjx1             = _mm_setzero_pd();
2113             fjy1             = _mm_setzero_pd();
2114             fjz1             = _mm_setzero_pd();
2115             fjx2             = _mm_setzero_pd();
2116             fjy2             = _mm_setzero_pd();
2117             fjz2             = _mm_setzero_pd();
2118
2119             /**************************
2120              * CALCULATE INTERACTIONS *
2121              **************************/
2122
2123             if (gmx_mm_any_lt(rsq00,rcutoff2))
2124             {
2125
2126             r00              = _mm_mul_pd(rsq00,rinv00);
2127
2128             /* EWALD ELECTROSTATICS */
2129
2130             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2131             ewrt             = _mm_mul_pd(r00,ewtabscale);
2132             ewitab           = _mm_cvttpd_epi32(ewrt);
2133 #ifdef __XOP__
2134             eweps            = _mm_frcz_pd(ewrt);
2135 #else
2136             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2137 #endif
2138             twoeweps         = _mm_add_pd(eweps,eweps);
2139             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2140             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2141             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
2142
2143             /* LENNARD-JONES DISPERSION/REPULSION */
2144
2145             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2146             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
2147
2148             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
2149
2150             fscal            = _mm_add_pd(felec,fvdw);
2151
2152             fscal            = _mm_and_pd(fscal,cutoff_mask);
2153
2154             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2155
2156             /* Update vectorial force */
2157             fix0             = _mm_macc_pd(dx00,fscal,fix0);
2158             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
2159             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
2160
2161             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
2162             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
2163             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
2164
2165             }
2166
2167             /**************************
2168              * CALCULATE INTERACTIONS *
2169              **************************/
2170
2171             if (gmx_mm_any_lt(rsq01,rcutoff2))
2172             {
2173
2174             r01              = _mm_mul_pd(rsq01,rinv01);
2175
2176             /* EWALD ELECTROSTATICS */
2177
2178             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2179             ewrt             = _mm_mul_pd(r01,ewtabscale);
2180             ewitab           = _mm_cvttpd_epi32(ewrt);
2181 #ifdef __XOP__
2182             eweps            = _mm_frcz_pd(ewrt);
2183 #else
2184             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2185 #endif
2186             twoeweps         = _mm_add_pd(eweps,eweps);
2187             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2188             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2189             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
2190
2191             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
2192
2193             fscal            = felec;
2194
2195             fscal            = _mm_and_pd(fscal,cutoff_mask);
2196
2197             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2198
2199             /* Update vectorial force */
2200             fix0             = _mm_macc_pd(dx01,fscal,fix0);
2201             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
2202             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
2203
2204             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
2205             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
2206             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
2207
2208             }
2209
2210             /**************************
2211              * CALCULATE INTERACTIONS *
2212              **************************/
2213
2214             if (gmx_mm_any_lt(rsq02,rcutoff2))
2215             {
2216
2217             r02              = _mm_mul_pd(rsq02,rinv02);
2218
2219             /* EWALD ELECTROSTATICS */
2220
2221             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2222             ewrt             = _mm_mul_pd(r02,ewtabscale);
2223             ewitab           = _mm_cvttpd_epi32(ewrt);
2224 #ifdef __XOP__
2225             eweps            = _mm_frcz_pd(ewrt);
2226 #else
2227             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2228 #endif
2229             twoeweps         = _mm_add_pd(eweps,eweps);
2230             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2231             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2232             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
2233
2234             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
2235
2236             fscal            = felec;
2237
2238             fscal            = _mm_and_pd(fscal,cutoff_mask);
2239
2240             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2241
2242             /* Update vectorial force */
2243             fix0             = _mm_macc_pd(dx02,fscal,fix0);
2244             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
2245             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
2246
2247             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
2248             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
2249             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
2250
2251             }
2252
2253             /**************************
2254              * CALCULATE INTERACTIONS *
2255              **************************/
2256
2257             if (gmx_mm_any_lt(rsq10,rcutoff2))
2258             {
2259
2260             r10              = _mm_mul_pd(rsq10,rinv10);
2261
2262             /* EWALD ELECTROSTATICS */
2263
2264             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2265             ewrt             = _mm_mul_pd(r10,ewtabscale);
2266             ewitab           = _mm_cvttpd_epi32(ewrt);
2267 #ifdef __XOP__
2268             eweps            = _mm_frcz_pd(ewrt);
2269 #else
2270             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2271 #endif
2272             twoeweps         = _mm_add_pd(eweps,eweps);
2273             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2274             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2275             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2276
2277             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
2278
2279             fscal            = felec;
2280
2281             fscal            = _mm_and_pd(fscal,cutoff_mask);
2282
2283             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2284
2285             /* Update vectorial force */
2286             fix1             = _mm_macc_pd(dx10,fscal,fix1);
2287             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
2288             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
2289
2290             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
2291             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
2292             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
2293
2294             }
2295
2296             /**************************
2297              * CALCULATE INTERACTIONS *
2298              **************************/
2299
2300             if (gmx_mm_any_lt(rsq11,rcutoff2))
2301             {
2302
2303             r11              = _mm_mul_pd(rsq11,rinv11);
2304
2305             /* EWALD ELECTROSTATICS */
2306
2307             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2308             ewrt             = _mm_mul_pd(r11,ewtabscale);
2309             ewitab           = _mm_cvttpd_epi32(ewrt);
2310 #ifdef __XOP__
2311             eweps            = _mm_frcz_pd(ewrt);
2312 #else
2313             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2314 #endif
2315             twoeweps         = _mm_add_pd(eweps,eweps);
2316             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2317             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2318             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2319
2320             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
2321
2322             fscal            = felec;
2323
2324             fscal            = _mm_and_pd(fscal,cutoff_mask);
2325
2326             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2327
2328             /* Update vectorial force */
2329             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2330             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2331             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2332
2333             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2334             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2335             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2336
2337             }
2338
2339             /**************************
2340              * CALCULATE INTERACTIONS *
2341              **************************/
2342
2343             if (gmx_mm_any_lt(rsq12,rcutoff2))
2344             {
2345
2346             r12              = _mm_mul_pd(rsq12,rinv12);
2347
2348             /* EWALD ELECTROSTATICS */
2349
2350             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2351             ewrt             = _mm_mul_pd(r12,ewtabscale);
2352             ewitab           = _mm_cvttpd_epi32(ewrt);
2353 #ifdef __XOP__
2354             eweps            = _mm_frcz_pd(ewrt);
2355 #else
2356             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2357 #endif
2358             twoeweps         = _mm_add_pd(eweps,eweps);
2359             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2360             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2361             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2362
2363             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
2364
2365             fscal            = felec;
2366
2367             fscal            = _mm_and_pd(fscal,cutoff_mask);
2368
2369             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2370
2371             /* Update vectorial force */
2372             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2373             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2374             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2375
2376             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2377             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2378             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2379
2380             }
2381
2382             /**************************
2383              * CALCULATE INTERACTIONS *
2384              **************************/
2385
2386             if (gmx_mm_any_lt(rsq20,rcutoff2))
2387             {
2388
2389             r20              = _mm_mul_pd(rsq20,rinv20);
2390
2391             /* EWALD ELECTROSTATICS */
2392
2393             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2394             ewrt             = _mm_mul_pd(r20,ewtabscale);
2395             ewitab           = _mm_cvttpd_epi32(ewrt);
2396 #ifdef __XOP__
2397             eweps            = _mm_frcz_pd(ewrt);
2398 #else
2399             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2400 #endif
2401             twoeweps         = _mm_add_pd(eweps,eweps);
2402             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2403             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2404             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2405
2406             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
2407
2408             fscal            = felec;
2409
2410             fscal            = _mm_and_pd(fscal,cutoff_mask);
2411
2412             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2413
2414             /* Update vectorial force */
2415             fix2             = _mm_macc_pd(dx20,fscal,fix2);
2416             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
2417             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
2418
2419             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2420             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2421             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
2422
2423             }
2424
2425             /**************************
2426              * CALCULATE INTERACTIONS *
2427              **************************/
2428
2429             if (gmx_mm_any_lt(rsq21,rcutoff2))
2430             {
2431
2432             r21              = _mm_mul_pd(rsq21,rinv21);
2433
2434             /* EWALD ELECTROSTATICS */
2435
2436             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2437             ewrt             = _mm_mul_pd(r21,ewtabscale);
2438             ewitab           = _mm_cvttpd_epi32(ewrt);
2439 #ifdef __XOP__
2440             eweps            = _mm_frcz_pd(ewrt);
2441 #else
2442             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2443 #endif
2444             twoeweps         = _mm_add_pd(eweps,eweps);
2445             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2446             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2447             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2448
2449             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
2450
2451             fscal            = felec;
2452
2453             fscal            = _mm_and_pd(fscal,cutoff_mask);
2454
2455             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2456
2457             /* Update vectorial force */
2458             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2459             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2460             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2461
2462             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2463             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2464             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2465
2466             }
2467
2468             /**************************
2469              * CALCULATE INTERACTIONS *
2470              **************************/
2471
2472             if (gmx_mm_any_lt(rsq22,rcutoff2))
2473             {
2474
2475             r22              = _mm_mul_pd(rsq22,rinv22);
2476
2477             /* EWALD ELECTROSTATICS */
2478
2479             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2480             ewrt             = _mm_mul_pd(r22,ewtabscale);
2481             ewitab           = _mm_cvttpd_epi32(ewrt);
2482 #ifdef __XOP__
2483             eweps            = _mm_frcz_pd(ewrt);
2484 #else
2485             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2486 #endif
2487             twoeweps         = _mm_add_pd(eweps,eweps);
2488             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2489             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2490             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2491
2492             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
2493
2494             fscal            = felec;
2495
2496             fscal            = _mm_and_pd(fscal,cutoff_mask);
2497
2498             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2499
2500             /* Update vectorial force */
2501             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2502             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2503             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2504
2505             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2506             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2507             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2508
2509             }
2510
2511             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2512
2513             /* Inner loop uses 385 flops */
2514         }
2515
2516         /* End of innermost loop */
2517
2518         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2519                                               f+i_coord_offset,fshift+i_shift_offset);
2520
2521         /* Increment number of inner iterations */
2522         inneriter                  += j_index_end - j_index_start;
2523
2524         /* Outer loop uses 18 flops */
2525     }
2526
2527     /* Increment number of outer iterations */
2528     outeriter        += nri;
2529
2530     /* Update outer/inner flops */
2531
2532     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*385);
2533 }