src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSw_VdwNone_GeomW3W3_sse2_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 sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3W3_VF_sse2_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwNone_GeomW3W3_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     int              vdwjidx1A,vdwjidx1B;
  89     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  90     int              vdwjidx2A,vdwjidx2B;
  91     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  94     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  95     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  97     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  98     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 100     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 101     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 102     real             *charge;
 103     __m128i          ewitab;
 104     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 105     real             *ewtab;
 106     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 107     real             rswitch_scalar,d_scalar;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125
 126     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 134     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 135     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 136
 137     jq0              = _mm_set1_pd(charge[inr+0]);
 138     jq1              = _mm_set1_pd(charge[inr+1]);
 139     jq2              = _mm_set1_pd(charge[inr+2]);
 140     qq00             = _mm_mul_pd(iq0,jq0);
 141     qq01             = _mm_mul_pd(iq0,jq1);
 142     qq02             = _mm_mul_pd(iq0,jq2);
 143     qq10             = _mm_mul_pd(iq1,jq0);
 144     qq11             = _mm_mul_pd(iq1,jq1);
 145     qq12             = _mm_mul_pd(iq1,jq2);
 146     qq20             = _mm_mul_pd(iq2,jq0);
 147     qq21             = _mm_mul_pd(iq2,jq1);
 148     qq22             = _mm_mul_pd(iq2,jq2);
 149
 150     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 151     rcutoff_scalar   = fr->rcoulomb;
 152     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 153     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 154
 155     rswitch_scalar   = fr->rcoulomb_switch;
 156     rswitch          = _mm_set1_pd(rswitch_scalar);
 157     /* Setup switch parameters */
 158     d_scalar         = rcutoff_scalar-rswitch_scalar;
 159     d                = _mm_set1_pd(d_scalar);
 160     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 161     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 162     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 163     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 164     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 165     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 166
 167     /* Avoid stupid compiler warnings */
 168     jnrA = jnrB = 0;
 169     j_coord_offsetA = 0;
 170     j_coord_offsetB = 0;
 171
 172     outeriter        = 0;
 173     inneriter        = 0;
 174
 175     /* Start outer loop over neighborlists */
 176     for(iidx=0; iidx<nri; iidx++)
 177     {
 178         /* Load shift vector for this list */
 179         i_shift_offset   = DIM*shiftidx[iidx];
 180
 181         /* Load limits for loop over neighbors */
 182         j_index_start    = jindex[iidx];
 183         j_index_end      = jindex[iidx+1];
 184
 185         /* Get outer coordinate index */
 186         inr              = iinr[iidx];
 187         i_coord_offset   = DIM*inr;
 188
 189         /* Load i particle coords and add shift vector */
 190         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 191                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 192
 193         fix0             = _mm_setzero_pd();
 194         fiy0             = _mm_setzero_pd();
 195         fiz0             = _mm_setzero_pd();
 196         fix1             = _mm_setzero_pd();
 197         fiy1             = _mm_setzero_pd();
 198         fiz1             = _mm_setzero_pd();
 199         fix2             = _mm_setzero_pd();
 200         fiy2             = _mm_setzero_pd();
 201         fiz2             = _mm_setzero_pd();
 202
 203         /* Reset potential sums */
 204         velecsum         = _mm_setzero_pd();
 205
 206         /* Start inner kernel loop */
 207         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 208         {
 209
 210             /* Get j neighbor index, and coordinate index */
 211             jnrA             = jjnr[jidx];
 212             jnrB             = jjnr[jidx+1];
 213             j_coord_offsetA  = DIM*jnrA;
 214             j_coord_offsetB  = DIM*jnrB;
 215
 216             /* load j atom coordinates */
 217             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 218                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 219
 220             /* Calculate displacement vector */
 221             dx00             = _mm_sub_pd(ix0,jx0);
 222             dy00             = _mm_sub_pd(iy0,jy0);
 223             dz00             = _mm_sub_pd(iz0,jz0);
 224             dx01             = _mm_sub_pd(ix0,jx1);
 225             dy01             = _mm_sub_pd(iy0,jy1);
 226             dz01             = _mm_sub_pd(iz0,jz1);
 227             dx02             = _mm_sub_pd(ix0,jx2);
 228             dy02             = _mm_sub_pd(iy0,jy2);
 229             dz02             = _mm_sub_pd(iz0,jz2);
 230             dx10             = _mm_sub_pd(ix1,jx0);
 231             dy10             = _mm_sub_pd(iy1,jy0);
 232             dz10             = _mm_sub_pd(iz1,jz0);
 233             dx11             = _mm_sub_pd(ix1,jx1);
 234             dy11             = _mm_sub_pd(iy1,jy1);
 235             dz11             = _mm_sub_pd(iz1,jz1);
 236             dx12             = _mm_sub_pd(ix1,jx2);
 237             dy12             = _mm_sub_pd(iy1,jy2);
 238             dz12             = _mm_sub_pd(iz1,jz2);
 239             dx20             = _mm_sub_pd(ix2,jx0);
 240             dy20             = _mm_sub_pd(iy2,jy0);
 241             dz20             = _mm_sub_pd(iz2,jz0);
 242             dx21             = _mm_sub_pd(ix2,jx1);
 243             dy21             = _mm_sub_pd(iy2,jy1);
 244             dz21             = _mm_sub_pd(iz2,jz1);
 245             dx22             = _mm_sub_pd(ix2,jx2);
 246             dy22             = _mm_sub_pd(iy2,jy2);
 247             dz22             = _mm_sub_pd(iz2,jz2);
 248
 249             /* Calculate squared distance and things based on it */
 250             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 251             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 252             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 253             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 254             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 255             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 256             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 257             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 258             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 259
 260             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 261             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 262             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 263             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 264             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 265             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 266             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 267             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 268             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 269
 270             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 271             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 272             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 273             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 274             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 275             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 276             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 277             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 278             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 279
 280             fjx0             = _mm_setzero_pd();
 281             fjy0             = _mm_setzero_pd();
 282             fjz0             = _mm_setzero_pd();
 283             fjx1             = _mm_setzero_pd();
 284             fjy1             = _mm_setzero_pd();
 285             fjz1             = _mm_setzero_pd();
 286             fjx2             = _mm_setzero_pd();
 287             fjy2             = _mm_setzero_pd();
 288             fjz2             = _mm_setzero_pd();
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             if (gmx_mm_any_lt(rsq00,rcutoff2))
 295             {
 296
 297             r00              = _mm_mul_pd(rsq00,rinv00);
 298
 299             /* EWALD ELECTROSTATICS */
 300
 301             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 302             ewrt             = _mm_mul_pd(r00,ewtabscale);
 303             ewitab           = _mm_cvttpd_epi32(ewrt);
 304             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 305             ewitab           = _mm_slli_epi32(ewitab,2);
 306             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 307             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 308             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 309             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 310             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 311             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 312             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 313             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 314             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 315             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 316
 317             d                = _mm_sub_pd(r00,rswitch);
 318             d                = _mm_max_pd(d,_mm_setzero_pd());
 319             d2               = _mm_mul_pd(d,d);
 320             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 321
 322             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 323
 324             /* Evaluate switch function */
 325             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 326             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 327             velec            = _mm_mul_pd(velec,sw);
 328             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velec            = _mm_and_pd(velec,cutoff_mask);
 332             velecsum         = _mm_add_pd(velecsum,velec);
 333
 334             fscal            = felec;
 335
 336             fscal            = _mm_and_pd(fscal,cutoff_mask);
 337
 338             /* Calculate temporary vectorial force */
 339             tx               = _mm_mul_pd(fscal,dx00);
 340             ty               = _mm_mul_pd(fscal,dy00);
 341             tz               = _mm_mul_pd(fscal,dz00);
 342
 343             /* Update vectorial force */
 344             fix0             = _mm_add_pd(fix0,tx);
 345             fiy0             = _mm_add_pd(fiy0,ty);
 346             fiz0             = _mm_add_pd(fiz0,tz);
 347
 348             fjx0             = _mm_add_pd(fjx0,tx);
 349             fjy0             = _mm_add_pd(fjy0,ty);
 350             fjz0             = _mm_add_pd(fjz0,tz);
 351
 352             }
 353
 354             /**************************
 355              * CALCULATE INTERACTIONS *
 356              **************************/
 357
 358             if (gmx_mm_any_lt(rsq01,rcutoff2))
 359             {
 360
 361             r01              = _mm_mul_pd(rsq01,rinv01);
 362
 363             /* EWALD ELECTROSTATICS */
 364
 365             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 366             ewrt             = _mm_mul_pd(r01,ewtabscale);
 367             ewitab           = _mm_cvttpd_epi32(ewrt);
 368             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 369             ewitab           = _mm_slli_epi32(ewitab,2);
 370             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 371             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 372             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 373             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 374             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 375             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 376             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 377             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 378             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
 379             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 380
 381             d                = _mm_sub_pd(r01,rswitch);
 382             d                = _mm_max_pd(d,_mm_setzero_pd());
 383             d2               = _mm_mul_pd(d,d);
 384             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 385
 386             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 387
 388             /* Evaluate switch function */
 389             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 390             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv01,_mm_mul_pd(velec,dsw)) );
 391             velec            = _mm_mul_pd(velec,sw);
 392             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velec            = _mm_and_pd(velec,cutoff_mask);
 396             velecsum         = _mm_add_pd(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             fscal            = _mm_and_pd(fscal,cutoff_mask);
 401
 402             /* Calculate temporary vectorial force */
 403             tx               = _mm_mul_pd(fscal,dx01);
 404             ty               = _mm_mul_pd(fscal,dy01);
 405             tz               = _mm_mul_pd(fscal,dz01);
 406
 407             /* Update vectorial force */
 408             fix0             = _mm_add_pd(fix0,tx);
 409             fiy0             = _mm_add_pd(fiy0,ty);
 410             fiz0             = _mm_add_pd(fiz0,tz);
 411
 412             fjx1             = _mm_add_pd(fjx1,tx);
 413             fjy1             = _mm_add_pd(fjy1,ty);
 414             fjz1             = _mm_add_pd(fjz1,tz);
 415
 416             }
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             if (gmx_mm_any_lt(rsq02,rcutoff2))
 423             {
 424
 425             r02              = _mm_mul_pd(rsq02,rinv02);
 426
 427             /* EWALD ELECTROSTATICS */
 428
 429             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 430             ewrt             = _mm_mul_pd(r02,ewtabscale);
 431             ewitab           = _mm_cvttpd_epi32(ewrt);
 432             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 433             ewitab           = _mm_slli_epi32(ewitab,2);
 434             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 435             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 436             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 437             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 438             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 439             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 440             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 441             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 442             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
 443             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 444
 445             d                = _mm_sub_pd(r02,rswitch);
 446             d                = _mm_max_pd(d,_mm_setzero_pd());
 447             d2               = _mm_mul_pd(d,d);
 448             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 449
 450             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 451
 452             /* Evaluate switch function */
 453             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 454             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv02,_mm_mul_pd(velec,dsw)) );
 455             velec            = _mm_mul_pd(velec,sw);
 456             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velec            = _mm_and_pd(velec,cutoff_mask);
 460             velecsum         = _mm_add_pd(velecsum,velec);
 461
 462             fscal            = felec;
 463
 464             fscal            = _mm_and_pd(fscal,cutoff_mask);
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm_mul_pd(fscal,dx02);
 468             ty               = _mm_mul_pd(fscal,dy02);
 469             tz               = _mm_mul_pd(fscal,dz02);
 470
 471             /* Update vectorial force */
 472             fix0             = _mm_add_pd(fix0,tx);
 473             fiy0             = _mm_add_pd(fiy0,ty);
 474             fiz0             = _mm_add_pd(fiz0,tz);
 475
 476             fjx2             = _mm_add_pd(fjx2,tx);
 477             fjy2             = _mm_add_pd(fjy2,ty);
 478             fjz2             = _mm_add_pd(fjz2,tz);
 479
 480             }
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             if (gmx_mm_any_lt(rsq10,rcutoff2))
 487             {
 488
 489             r10              = _mm_mul_pd(rsq10,rinv10);
 490
 491             /* EWALD ELECTROSTATICS */
 492
 493             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 494             ewrt             = _mm_mul_pd(r10,ewtabscale);
 495             ewitab           = _mm_cvttpd_epi32(ewrt);
 496             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 497             ewitab           = _mm_slli_epi32(ewitab,2);
 498             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 499             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 500             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 501             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 502             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 503             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 504             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 505             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 506             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 507             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 508
 509             d                = _mm_sub_pd(r10,rswitch);
 510             d                = _mm_max_pd(d,_mm_setzero_pd());
 511             d2               = _mm_mul_pd(d,d);
 512             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 513
 514             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 515
 516             /* Evaluate switch function */
 517             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 518             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 519             velec            = _mm_mul_pd(velec,sw);
 520             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             velec            = _mm_and_pd(velec,cutoff_mask);
 524             velecsum         = _mm_add_pd(velecsum,velec);
 525
 526             fscal            = felec;
 527
 528             fscal            = _mm_and_pd(fscal,cutoff_mask);
 529
 530             /* Calculate temporary vectorial force */
 531             tx               = _mm_mul_pd(fscal,dx10);
 532             ty               = _mm_mul_pd(fscal,dy10);
 533             tz               = _mm_mul_pd(fscal,dz10);
 534
 535             /* Update vectorial force */
 536             fix1             = _mm_add_pd(fix1,tx);
 537             fiy1             = _mm_add_pd(fiy1,ty);
 538             fiz1             = _mm_add_pd(fiz1,tz);
 539
 540             fjx0             = _mm_add_pd(fjx0,tx);
 541             fjy0             = _mm_add_pd(fjy0,ty);
 542             fjz0             = _mm_add_pd(fjz0,tz);
 543
 544             }
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             if (gmx_mm_any_lt(rsq11,rcutoff2))
 551             {
 552
 553             r11              = _mm_mul_pd(rsq11,rinv11);
 554
 555             /* EWALD ELECTROSTATICS */
 556
 557             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 558             ewrt             = _mm_mul_pd(r11,ewtabscale);
 559             ewitab           = _mm_cvttpd_epi32(ewrt);
 560             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 561             ewitab           = _mm_slli_epi32(ewitab,2);
 562             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 563             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 564             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 565             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 566             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 567             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 568             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 569             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 570             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 571             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 572
 573             d                = _mm_sub_pd(r11,rswitch);
 574             d                = _mm_max_pd(d,_mm_setzero_pd());
 575             d2               = _mm_mul_pd(d,d);
 576             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 577
 578             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 579
 580             /* Evaluate switch function */
 581             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 582             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
 583             velec            = _mm_mul_pd(velec,sw);
 584             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velec            = _mm_and_pd(velec,cutoff_mask);
 588             velecsum         = _mm_add_pd(velecsum,velec);
 589
 590             fscal            = felec;
 591
 592             fscal            = _mm_and_pd(fscal,cutoff_mask);
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm_mul_pd(fscal,dx11);
 596             ty               = _mm_mul_pd(fscal,dy11);
 597             tz               = _mm_mul_pd(fscal,dz11);
 598
 599             /* Update vectorial force */
 600             fix1             = _mm_add_pd(fix1,tx);
 601             fiy1             = _mm_add_pd(fiy1,ty);
 602             fiz1             = _mm_add_pd(fiz1,tz);
 603
 604             fjx1             = _mm_add_pd(fjx1,tx);
 605             fjy1             = _mm_add_pd(fjy1,ty);
 606             fjz1             = _mm_add_pd(fjz1,tz);
 607
 608             }
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             if (gmx_mm_any_lt(rsq12,rcutoff2))
 615             {
 616
 617             r12              = _mm_mul_pd(rsq12,rinv12);
 618
 619             /* EWALD ELECTROSTATICS */
 620
 621             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 622             ewrt             = _mm_mul_pd(r12,ewtabscale);
 623             ewitab           = _mm_cvttpd_epi32(ewrt);
 624             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 625             ewitab           = _mm_slli_epi32(ewitab,2);
 626             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 627             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 628             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 629             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 630             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 631             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 632             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 633             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 634             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 635             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 636
 637             d                = _mm_sub_pd(r12,rswitch);
 638             d                = _mm_max_pd(d,_mm_setzero_pd());
 639             d2               = _mm_mul_pd(d,d);
 640             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 641
 642             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 643
 644             /* Evaluate switch function */
 645             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 646             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
 647             velec            = _mm_mul_pd(velec,sw);
 648             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm_and_pd(velec,cutoff_mask);
 652             velecsum         = _mm_add_pd(velecsum,velec);
 653
 654             fscal            = felec;
 655
 656             fscal            = _mm_and_pd(fscal,cutoff_mask);
 657
 658             /* Calculate temporary vectorial force */
 659             tx               = _mm_mul_pd(fscal,dx12);
 660             ty               = _mm_mul_pd(fscal,dy12);
 661             tz               = _mm_mul_pd(fscal,dz12);
 662
 663             /* Update vectorial force */
 664             fix1             = _mm_add_pd(fix1,tx);
 665             fiy1             = _mm_add_pd(fiy1,ty);
 666             fiz1             = _mm_add_pd(fiz1,tz);
 667
 668             fjx2             = _mm_add_pd(fjx2,tx);
 669             fjy2             = _mm_add_pd(fjy2,ty);
 670             fjz2             = _mm_add_pd(fjz2,tz);
 671
 672             }
 673
 674             /**************************
 675              * CALCULATE INTERACTIONS *
 676              **************************/
 677
 678             if (gmx_mm_any_lt(rsq20,rcutoff2))
 679             {
 680
 681             r20              = _mm_mul_pd(rsq20,rinv20);
 682
 683             /* EWALD ELECTROSTATICS */
 684
 685             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 686             ewrt             = _mm_mul_pd(r20,ewtabscale);
 687             ewitab           = _mm_cvttpd_epi32(ewrt);
 688             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 689             ewitab           = _mm_slli_epi32(ewitab,2);
 690             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 691             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 692             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 693             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 694             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 695             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 696             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 697             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 698             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 699             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 700
 701             d                = _mm_sub_pd(r20,rswitch);
 702             d                = _mm_max_pd(d,_mm_setzero_pd());
 703             d2               = _mm_mul_pd(d,d);
 704             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 705
 706             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 707
 708             /* Evaluate switch function */
 709             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 710             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 711             velec            = _mm_mul_pd(velec,sw);
 712             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 713
 714             /* Update potential sum for this i atom from the interaction with this j atom. */
 715             velec            = _mm_and_pd(velec,cutoff_mask);
 716             velecsum         = _mm_add_pd(velecsum,velec);
 717
 718             fscal            = felec;
 719
 720             fscal            = _mm_and_pd(fscal,cutoff_mask);
 721
 722             /* Calculate temporary vectorial force */
 723             tx               = _mm_mul_pd(fscal,dx20);
 724             ty               = _mm_mul_pd(fscal,dy20);
 725             tz               = _mm_mul_pd(fscal,dz20);
 726
 727             /* Update vectorial force */
 728             fix2             = _mm_add_pd(fix2,tx);
 729             fiy2             = _mm_add_pd(fiy2,ty);
 730             fiz2             = _mm_add_pd(fiz2,tz);
 731
 732             fjx0             = _mm_add_pd(fjx0,tx);
 733             fjy0             = _mm_add_pd(fjy0,ty);
 734             fjz0             = _mm_add_pd(fjz0,tz);
 735
 736             }
 737
 738             /**************************
 739              * CALCULATE INTERACTIONS *
 740              **************************/
 741
 742             if (gmx_mm_any_lt(rsq21,rcutoff2))
 743             {
 744
 745             r21              = _mm_mul_pd(rsq21,rinv21);
 746
 747             /* EWALD ELECTROSTATICS */
 748
 749             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 750             ewrt             = _mm_mul_pd(r21,ewtabscale);
 751             ewitab           = _mm_cvttpd_epi32(ewrt);
 752             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 753             ewitab           = _mm_slli_epi32(ewitab,2);
 754             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 755             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 756             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 757             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 758             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 759             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 760             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 761             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 762             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 763             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 764
 765             d                = _mm_sub_pd(r21,rswitch);
 766             d                = _mm_max_pd(d,_mm_setzero_pd());
 767             d2               = _mm_mul_pd(d,d);
 768             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 769
 770             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 771
 772             /* Evaluate switch function */
 773             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 774             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
 775             velec            = _mm_mul_pd(velec,sw);
 776             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
 777
 778             /* Update potential sum for this i atom from the interaction with this j atom. */
 779             velec            = _mm_and_pd(velec,cutoff_mask);
 780             velecsum         = _mm_add_pd(velecsum,velec);
 781
 782             fscal            = felec;
 783
 784             fscal            = _mm_and_pd(fscal,cutoff_mask);
 785
 786             /* Calculate temporary vectorial force */
 787             tx               = _mm_mul_pd(fscal,dx21);
 788             ty               = _mm_mul_pd(fscal,dy21);
 789             tz               = _mm_mul_pd(fscal,dz21);
 790
 791             /* Update vectorial force */
 792             fix2             = _mm_add_pd(fix2,tx);
 793             fiy2             = _mm_add_pd(fiy2,ty);
 794             fiz2             = _mm_add_pd(fiz2,tz);
 795
 796             fjx1             = _mm_add_pd(fjx1,tx);
 797             fjy1             = _mm_add_pd(fjy1,ty);
 798             fjz1             = _mm_add_pd(fjz1,tz);
 799
 800             }
 801
 802             /**************************
 803              * CALCULATE INTERACTIONS *
 804              **************************/
 805
 806             if (gmx_mm_any_lt(rsq22,rcutoff2))
 807             {
 808
 809             r22              = _mm_mul_pd(rsq22,rinv22);
 810
 811             /* EWALD ELECTROSTATICS */
 812
 813             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 814             ewrt             = _mm_mul_pd(r22,ewtabscale);
 815             ewitab           = _mm_cvttpd_epi32(ewrt);
 816             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 817             ewitab           = _mm_slli_epi32(ewitab,2);
 818             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 819             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 820             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 821             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 822             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 823             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 824             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 825             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 826             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 827             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 828
 829             d                = _mm_sub_pd(r22,rswitch);
 830             d                = _mm_max_pd(d,_mm_setzero_pd());
 831             d2               = _mm_mul_pd(d,d);
 832             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 833
 834             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 835
 836             /* Evaluate switch function */
 837             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 838             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
 839             velec            = _mm_mul_pd(velec,sw);
 840             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
 841
 842             /* Update potential sum for this i atom from the interaction with this j atom. */
 843             velec            = _mm_and_pd(velec,cutoff_mask);
 844             velecsum         = _mm_add_pd(velecsum,velec);
 845
 846             fscal            = felec;
 847
 848             fscal            = _mm_and_pd(fscal,cutoff_mask);
 849
 850             /* Calculate temporary vectorial force */
 851             tx               = _mm_mul_pd(fscal,dx22);
 852             ty               = _mm_mul_pd(fscal,dy22);
 853             tz               = _mm_mul_pd(fscal,dz22);
 854
 855             /* Update vectorial force */
 856             fix2             = _mm_add_pd(fix2,tx);
 857             fiy2             = _mm_add_pd(fiy2,ty);
 858             fiz2             = _mm_add_pd(fiz2,tz);
 859
 860             fjx2             = _mm_add_pd(fjx2,tx);
 861             fjy2             = _mm_add_pd(fjy2,ty);
 862             fjz2             = _mm_add_pd(fjz2,tz);
 863
 864             }
 865
 866             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 867
 868             /* Inner loop uses 585 flops */
 869         }
 870
 871         if(jidx<j_index_end)
 872         {
 873
 874             jnrA             = jjnr[jidx];
 875             j_coord_offsetA  = DIM*jnrA;
 876
 877             /* load j atom coordinates */
 878             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 879                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 880
 881             /* Calculate displacement vector */
 882             dx00             = _mm_sub_pd(ix0,jx0);
 883             dy00             = _mm_sub_pd(iy0,jy0);
 884             dz00             = _mm_sub_pd(iz0,jz0);
 885             dx01             = _mm_sub_pd(ix0,jx1);
 886             dy01             = _mm_sub_pd(iy0,jy1);
 887             dz01             = _mm_sub_pd(iz0,jz1);
 888             dx02             = _mm_sub_pd(ix0,jx2);
 889             dy02             = _mm_sub_pd(iy0,jy2);
 890             dz02             = _mm_sub_pd(iz0,jz2);
 891             dx10             = _mm_sub_pd(ix1,jx0);
 892             dy10             = _mm_sub_pd(iy1,jy0);
 893             dz10             = _mm_sub_pd(iz1,jz0);
 894             dx11             = _mm_sub_pd(ix1,jx1);
 895             dy11             = _mm_sub_pd(iy1,jy1);
 896             dz11             = _mm_sub_pd(iz1,jz1);
 897             dx12             = _mm_sub_pd(ix1,jx2);
 898             dy12             = _mm_sub_pd(iy1,jy2);
 899             dz12             = _mm_sub_pd(iz1,jz2);
 900             dx20             = _mm_sub_pd(ix2,jx0);
 901             dy20             = _mm_sub_pd(iy2,jy0);
 902             dz20             = _mm_sub_pd(iz2,jz0);
 903             dx21             = _mm_sub_pd(ix2,jx1);
 904             dy21             = _mm_sub_pd(iy2,jy1);
 905             dz21             = _mm_sub_pd(iz2,jz1);
 906             dx22             = _mm_sub_pd(ix2,jx2);
 907             dy22             = _mm_sub_pd(iy2,jy2);
 908             dz22             = _mm_sub_pd(iz2,jz2);
 909
 910             /* Calculate squared distance and things based on it */
 911             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 912             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 913             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 914             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 915             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 916             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 917             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 918             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 919             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 920
 921             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 922             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 923             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 924             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 925             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 926             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 927             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 928             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 929             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 930
 931             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 932             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 933             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 934             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 935             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 936             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 937             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 938             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 939             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 940
 941             fjx0             = _mm_setzero_pd();
 942             fjy0             = _mm_setzero_pd();
 943             fjz0             = _mm_setzero_pd();
 944             fjx1             = _mm_setzero_pd();
 945             fjy1             = _mm_setzero_pd();
 946             fjz1             = _mm_setzero_pd();
 947             fjx2             = _mm_setzero_pd();
 948             fjy2             = _mm_setzero_pd();
 949             fjz2             = _mm_setzero_pd();
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             if (gmx_mm_any_lt(rsq00,rcutoff2))
 956             {
 957
 958             r00              = _mm_mul_pd(rsq00,rinv00);
 959
 960             /* EWALD ELECTROSTATICS */
 961
 962             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 963             ewrt             = _mm_mul_pd(r00,ewtabscale);
 964             ewitab           = _mm_cvttpd_epi32(ewrt);
 965             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 966             ewitab           = _mm_slli_epi32(ewitab,2);
 967             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 968             ewtabD           = _mm_setzero_pd();
 969             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 970             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 971             ewtabFn          = _mm_setzero_pd();
 972             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 973             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 974             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 975             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 976             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 977
 978             d                = _mm_sub_pd(r00,rswitch);
 979             d                = _mm_max_pd(d,_mm_setzero_pd());
 980             d2               = _mm_mul_pd(d,d);
 981             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 982
 983             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 984
 985             /* Evaluate switch function */
 986             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 987             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 988             velec            = _mm_mul_pd(velec,sw);
 989             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 990
 991             /* Update potential sum for this i atom from the interaction with this j atom. */
 992             velec            = _mm_and_pd(velec,cutoff_mask);
 993             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 994             velecsum         = _mm_add_pd(velecsum,velec);
 995
 996             fscal            = felec;
 997
 998             fscal            = _mm_and_pd(fscal,cutoff_mask);
 999
1000             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1001
1002             /* Calculate temporary vectorial force */
1003             tx               = _mm_mul_pd(fscal,dx00);
1004             ty               = _mm_mul_pd(fscal,dy00);
1005             tz               = _mm_mul_pd(fscal,dz00);
1006
1007             /* Update vectorial force */
1008             fix0             = _mm_add_pd(fix0,tx);
1009             fiy0             = _mm_add_pd(fiy0,ty);
1010             fiz0             = _mm_add_pd(fiz0,tz);
1011
1012             fjx0             = _mm_add_pd(fjx0,tx);
1013             fjy0             = _mm_add_pd(fjy0,ty);
1014             fjz0             = _mm_add_pd(fjz0,tz);
1015
1016             }
1017
1018             /**************************
1019              * CALCULATE INTERACTIONS *
1020              **************************/
1021
1022             if (gmx_mm_any_lt(rsq01,rcutoff2))
1023             {
1024
1025             r01              = _mm_mul_pd(rsq01,rinv01);
1026
1027             /* EWALD ELECTROSTATICS */
1028
1029             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1030             ewrt             = _mm_mul_pd(r01,ewtabscale);
1031             ewitab           = _mm_cvttpd_epi32(ewrt);
1032             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1033             ewitab           = _mm_slli_epi32(ewitab,2);
1034             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1035             ewtabD           = _mm_setzero_pd();
1036             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1037             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1038             ewtabFn          = _mm_setzero_pd();
1039             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1040             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1041             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1042             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
1043             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1044
1045             d                = _mm_sub_pd(r01,rswitch);
1046             d                = _mm_max_pd(d,_mm_setzero_pd());
1047             d2               = _mm_mul_pd(d,d);
1048             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1049
1050             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1051
1052             /* Evaluate switch function */
1053             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1054             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv01,_mm_mul_pd(velec,dsw)) );
1055             velec            = _mm_mul_pd(velec,sw);
1056             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
1057
1058             /* Update potential sum for this i atom from the interaction with this j atom. */
1059             velec            = _mm_and_pd(velec,cutoff_mask);
1060             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1061             velecsum         = _mm_add_pd(velecsum,velec);
1062
1063             fscal            = felec;
1064
1065             fscal            = _mm_and_pd(fscal,cutoff_mask);
1066
1067             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1068
1069             /* Calculate temporary vectorial force */
1070             tx               = _mm_mul_pd(fscal,dx01);
1071             ty               = _mm_mul_pd(fscal,dy01);
1072             tz               = _mm_mul_pd(fscal,dz01);
1073
1074             /* Update vectorial force */
1075             fix0             = _mm_add_pd(fix0,tx);
1076             fiy0             = _mm_add_pd(fiy0,ty);
1077             fiz0             = _mm_add_pd(fiz0,tz);
1078
1079             fjx1             = _mm_add_pd(fjx1,tx);
1080             fjy1             = _mm_add_pd(fjy1,ty);
1081             fjz1             = _mm_add_pd(fjz1,tz);
1082
1083             }
1084
1085             /**************************
1086              * CALCULATE INTERACTIONS *
1087              **************************/
1088
1089             if (gmx_mm_any_lt(rsq02,rcutoff2))
1090             {
1091
1092             r02              = _mm_mul_pd(rsq02,rinv02);
1093
1094             /* EWALD ELECTROSTATICS */
1095
1096             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1097             ewrt             = _mm_mul_pd(r02,ewtabscale);
1098             ewitab           = _mm_cvttpd_epi32(ewrt);
1099             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1100             ewitab           = _mm_slli_epi32(ewitab,2);
1101             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1102             ewtabD           = _mm_setzero_pd();
1103             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1104             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1105             ewtabFn          = _mm_setzero_pd();
1106             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1107             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1108             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1109             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
1110             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1111
1112             d                = _mm_sub_pd(r02,rswitch);
1113             d                = _mm_max_pd(d,_mm_setzero_pd());
1114             d2               = _mm_mul_pd(d,d);
1115             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1116
1117             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1118
1119             /* Evaluate switch function */
1120             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1121             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv02,_mm_mul_pd(velec,dsw)) );
1122             velec            = _mm_mul_pd(velec,sw);
1123             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
1124
1125             /* Update potential sum for this i atom from the interaction with this j atom. */
1126             velec            = _mm_and_pd(velec,cutoff_mask);
1127             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1128             velecsum         = _mm_add_pd(velecsum,velec);
1129
1130             fscal            = felec;
1131
1132             fscal            = _mm_and_pd(fscal,cutoff_mask);
1133
1134             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1135
1136             /* Calculate temporary vectorial force */
1137             tx               = _mm_mul_pd(fscal,dx02);
1138             ty               = _mm_mul_pd(fscal,dy02);
1139             tz               = _mm_mul_pd(fscal,dz02);
1140
1141             /* Update vectorial force */
1142             fix0             = _mm_add_pd(fix0,tx);
1143             fiy0             = _mm_add_pd(fiy0,ty);
1144             fiz0             = _mm_add_pd(fiz0,tz);
1145
1146             fjx2             = _mm_add_pd(fjx2,tx);
1147             fjy2             = _mm_add_pd(fjy2,ty);
1148             fjz2             = _mm_add_pd(fjz2,tz);
1149
1150             }
1151
1152             /**************************
1153              * CALCULATE INTERACTIONS *
1154              **************************/
1155
1156             if (gmx_mm_any_lt(rsq10,rcutoff2))
1157             {
1158
1159             r10              = _mm_mul_pd(rsq10,rinv10);
1160
1161             /* EWALD ELECTROSTATICS */
1162
1163             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1164             ewrt             = _mm_mul_pd(r10,ewtabscale);
1165             ewitab           = _mm_cvttpd_epi32(ewrt);
1166             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1167             ewitab           = _mm_slli_epi32(ewitab,2);
1168             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1169             ewtabD           = _mm_setzero_pd();
1170             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1171             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1172             ewtabFn          = _mm_setzero_pd();
1173             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1174             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1175             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1176             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1177             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1178
1179             d                = _mm_sub_pd(r10,rswitch);
1180             d                = _mm_max_pd(d,_mm_setzero_pd());
1181             d2               = _mm_mul_pd(d,d);
1182             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1183
1184             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1185
1186             /* Evaluate switch function */
1187             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1188             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1189             velec            = _mm_mul_pd(velec,sw);
1190             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1191
1192             /* Update potential sum for this i atom from the interaction with this j atom. */
1193             velec            = _mm_and_pd(velec,cutoff_mask);
1194             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1195             velecsum         = _mm_add_pd(velecsum,velec);
1196
1197             fscal            = felec;
1198
1199             fscal            = _mm_and_pd(fscal,cutoff_mask);
1200
1201             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1202
1203             /* Calculate temporary vectorial force */
1204             tx               = _mm_mul_pd(fscal,dx10);
1205             ty               = _mm_mul_pd(fscal,dy10);
1206             tz               = _mm_mul_pd(fscal,dz10);
1207
1208             /* Update vectorial force */
1209             fix1             = _mm_add_pd(fix1,tx);
1210             fiy1             = _mm_add_pd(fiy1,ty);
1211             fiz1             = _mm_add_pd(fiz1,tz);
1212
1213             fjx0             = _mm_add_pd(fjx0,tx);
1214             fjy0             = _mm_add_pd(fjy0,ty);
1215             fjz0             = _mm_add_pd(fjz0,tz);
1216
1217             }
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             if (gmx_mm_any_lt(rsq11,rcutoff2))
1224             {
1225
1226             r11              = _mm_mul_pd(rsq11,rinv11);
1227
1228             /* EWALD ELECTROSTATICS */
1229
1230             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1231             ewrt             = _mm_mul_pd(r11,ewtabscale);
1232             ewitab           = _mm_cvttpd_epi32(ewrt);
1233             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1234             ewitab           = _mm_slli_epi32(ewitab,2);
1235             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1236             ewtabD           = _mm_setzero_pd();
1237             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1238             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1239             ewtabFn          = _mm_setzero_pd();
1240             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1241             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1242             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1243             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
1244             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1245
1246             d                = _mm_sub_pd(r11,rswitch);
1247             d                = _mm_max_pd(d,_mm_setzero_pd());
1248             d2               = _mm_mul_pd(d,d);
1249             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1250
1251             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1252
1253             /* Evaluate switch function */
1254             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1255             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
1256             velec            = _mm_mul_pd(velec,sw);
1257             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
1258
1259             /* Update potential sum for this i atom from the interaction with this j atom. */
1260             velec            = _mm_and_pd(velec,cutoff_mask);
1261             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1262             velecsum         = _mm_add_pd(velecsum,velec);
1263
1264             fscal            = felec;
1265
1266             fscal            = _mm_and_pd(fscal,cutoff_mask);
1267
1268             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1269
1270             /* Calculate temporary vectorial force */
1271             tx               = _mm_mul_pd(fscal,dx11);
1272             ty               = _mm_mul_pd(fscal,dy11);
1273             tz               = _mm_mul_pd(fscal,dz11);
1274
1275             /* Update vectorial force */
1276             fix1             = _mm_add_pd(fix1,tx);
1277             fiy1             = _mm_add_pd(fiy1,ty);
1278             fiz1             = _mm_add_pd(fiz1,tz);
1279
1280             fjx1             = _mm_add_pd(fjx1,tx);
1281             fjy1             = _mm_add_pd(fjy1,ty);
1282             fjz1             = _mm_add_pd(fjz1,tz);
1283
1284             }
1285
1286             /**************************
1287              * CALCULATE INTERACTIONS *
1288              **************************/
1289
1290             if (gmx_mm_any_lt(rsq12,rcutoff2))
1291             {
1292
1293             r12              = _mm_mul_pd(rsq12,rinv12);
1294
1295             /* EWALD ELECTROSTATICS */
1296
1297             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1298             ewrt             = _mm_mul_pd(r12,ewtabscale);
1299             ewitab           = _mm_cvttpd_epi32(ewrt);
1300             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1301             ewitab           = _mm_slli_epi32(ewitab,2);
1302             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1303             ewtabD           = _mm_setzero_pd();
1304             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1305             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1306             ewtabFn          = _mm_setzero_pd();
1307             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1308             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1309             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1310             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
1311             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1312
1313             d                = _mm_sub_pd(r12,rswitch);
1314             d                = _mm_max_pd(d,_mm_setzero_pd());
1315             d2               = _mm_mul_pd(d,d);
1316             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1317
1318             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1319
1320             /* Evaluate switch function */
1321             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1322             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
1323             velec            = _mm_mul_pd(velec,sw);
1324             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
1325
1326             /* Update potential sum for this i atom from the interaction with this j atom. */
1327             velec            = _mm_and_pd(velec,cutoff_mask);
1328             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1329             velecsum         = _mm_add_pd(velecsum,velec);
1330
1331             fscal            = felec;
1332
1333             fscal            = _mm_and_pd(fscal,cutoff_mask);
1334
1335             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1336
1337             /* Calculate temporary vectorial force */
1338             tx               = _mm_mul_pd(fscal,dx12);
1339             ty               = _mm_mul_pd(fscal,dy12);
1340             tz               = _mm_mul_pd(fscal,dz12);
1341
1342             /* Update vectorial force */
1343             fix1             = _mm_add_pd(fix1,tx);
1344             fiy1             = _mm_add_pd(fiy1,ty);
1345             fiz1             = _mm_add_pd(fiz1,tz);
1346
1347             fjx2             = _mm_add_pd(fjx2,tx);
1348             fjy2             = _mm_add_pd(fjy2,ty);
1349             fjz2             = _mm_add_pd(fjz2,tz);
1350
1351             }
1352
1353             /**************************
1354              * CALCULATE INTERACTIONS *
1355              **************************/
1356
1357             if (gmx_mm_any_lt(rsq20,rcutoff2))
1358             {
1359
1360             r20              = _mm_mul_pd(rsq20,rinv20);
1361
1362             /* EWALD ELECTROSTATICS */
1363
1364             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1365             ewrt             = _mm_mul_pd(r20,ewtabscale);
1366             ewitab           = _mm_cvttpd_epi32(ewrt);
1367             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1368             ewitab           = _mm_slli_epi32(ewitab,2);
1369             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1370             ewtabD           = _mm_setzero_pd();
1371             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1372             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1373             ewtabFn          = _mm_setzero_pd();
1374             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1375             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1376             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1377             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1378             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1379
1380             d                = _mm_sub_pd(r20,rswitch);
1381             d                = _mm_max_pd(d,_mm_setzero_pd());
1382             d2               = _mm_mul_pd(d,d);
1383             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1384
1385             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1386
1387             /* Evaluate switch function */
1388             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1389             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1390             velec            = _mm_mul_pd(velec,sw);
1391             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1392
1393             /* Update potential sum for this i atom from the interaction with this j atom. */
1394             velec            = _mm_and_pd(velec,cutoff_mask);
1395             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1396             velecsum         = _mm_add_pd(velecsum,velec);
1397
1398             fscal            = felec;
1399
1400             fscal            = _mm_and_pd(fscal,cutoff_mask);
1401
1402             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1403
1404             /* Calculate temporary vectorial force */
1405             tx               = _mm_mul_pd(fscal,dx20);
1406             ty               = _mm_mul_pd(fscal,dy20);
1407             tz               = _mm_mul_pd(fscal,dz20);
1408
1409             /* Update vectorial force */
1410             fix2             = _mm_add_pd(fix2,tx);
1411             fiy2             = _mm_add_pd(fiy2,ty);
1412             fiz2             = _mm_add_pd(fiz2,tz);
1413
1414             fjx0             = _mm_add_pd(fjx0,tx);
1415             fjy0             = _mm_add_pd(fjy0,ty);
1416             fjz0             = _mm_add_pd(fjz0,tz);
1417
1418             }
1419
1420             /**************************
1421              * CALCULATE INTERACTIONS *
1422              **************************/
1423
1424             if (gmx_mm_any_lt(rsq21,rcutoff2))
1425             {
1426
1427             r21              = _mm_mul_pd(rsq21,rinv21);
1428
1429             /* EWALD ELECTROSTATICS */
1430
1431             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1432             ewrt             = _mm_mul_pd(r21,ewtabscale);
1433             ewitab           = _mm_cvttpd_epi32(ewrt);
1434             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1435             ewitab           = _mm_slli_epi32(ewitab,2);
1436             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1437             ewtabD           = _mm_setzero_pd();
1438             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1439             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1440             ewtabFn          = _mm_setzero_pd();
1441             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1442             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1443             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1444             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1445             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1446
1447             d                = _mm_sub_pd(r21,rswitch);
1448             d                = _mm_max_pd(d,_mm_setzero_pd());
1449             d2               = _mm_mul_pd(d,d);
1450             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1451
1452             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1453
1454             /* Evaluate switch function */
1455             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1456             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
1457             velec            = _mm_mul_pd(velec,sw);
1458             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
1459
1460             /* Update potential sum for this i atom from the interaction with this j atom. */
1461             velec            = _mm_and_pd(velec,cutoff_mask);
1462             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1463             velecsum         = _mm_add_pd(velecsum,velec);
1464
1465             fscal            = felec;
1466
1467             fscal            = _mm_and_pd(fscal,cutoff_mask);
1468
1469             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1470
1471             /* Calculate temporary vectorial force */
1472             tx               = _mm_mul_pd(fscal,dx21);
1473             ty               = _mm_mul_pd(fscal,dy21);
1474             tz               = _mm_mul_pd(fscal,dz21);
1475
1476             /* Update vectorial force */
1477             fix2             = _mm_add_pd(fix2,tx);
1478             fiy2             = _mm_add_pd(fiy2,ty);
1479             fiz2             = _mm_add_pd(fiz2,tz);
1480
1481             fjx1             = _mm_add_pd(fjx1,tx);
1482             fjy1             = _mm_add_pd(fjy1,ty);
1483             fjz1             = _mm_add_pd(fjz1,tz);
1484
1485             }
1486
1487             /**************************
1488              * CALCULATE INTERACTIONS *
1489              **************************/
1490
1491             if (gmx_mm_any_lt(rsq22,rcutoff2))
1492             {
1493
1494             r22              = _mm_mul_pd(rsq22,rinv22);
1495
1496             /* EWALD ELECTROSTATICS */
1497
1498             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1499             ewrt             = _mm_mul_pd(r22,ewtabscale);
1500             ewitab           = _mm_cvttpd_epi32(ewrt);
1501             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1502             ewitab           = _mm_slli_epi32(ewitab,2);
1503             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1504             ewtabD           = _mm_setzero_pd();
1505             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1506             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1507             ewtabFn          = _mm_setzero_pd();
1508             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1509             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1510             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1511             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1512             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1513
1514             d                = _mm_sub_pd(r22,rswitch);
1515             d                = _mm_max_pd(d,_mm_setzero_pd());
1516             d2               = _mm_mul_pd(d,d);
1517             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1518
1519             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1520
1521             /* Evaluate switch function */
1522             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1523             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
1524             velec            = _mm_mul_pd(velec,sw);
1525             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
1526
1527             /* Update potential sum for this i atom from the interaction with this j atom. */
1528             velec            = _mm_and_pd(velec,cutoff_mask);
1529             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1530             velecsum         = _mm_add_pd(velecsum,velec);
1531
1532             fscal            = felec;
1533
1534             fscal            = _mm_and_pd(fscal,cutoff_mask);
1535
1536             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1537
1538             /* Calculate temporary vectorial force */
1539             tx               = _mm_mul_pd(fscal,dx22);
1540             ty               = _mm_mul_pd(fscal,dy22);
1541             tz               = _mm_mul_pd(fscal,dz22);
1542
1543             /* Update vectorial force */
1544             fix2             = _mm_add_pd(fix2,tx);
1545             fiy2             = _mm_add_pd(fiy2,ty);
1546             fiz2             = _mm_add_pd(fiz2,tz);
1547
1548             fjx2             = _mm_add_pd(fjx2,tx);
1549             fjy2             = _mm_add_pd(fjy2,ty);
1550             fjz2             = _mm_add_pd(fjz2,tz);
1551
1552             }
1553
1554             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1555
1556             /* Inner loop uses 585 flops */
1557         }
1558
1559         /* End of innermost loop */
1560
1561         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1562                                               f+i_coord_offset,fshift+i_shift_offset);
1563
1564         ggid                        = gid[iidx];
1565         /* Update potential energies */
1566         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1567
1568         /* Increment number of inner iterations */
1569         inneriter                  += j_index_end - j_index_start;
1570
1571         /* Outer loop uses 19 flops */
1572     }
1573
1574     /* Increment number of outer iterations */
1575     outeriter        += nri;
1576
1577     /* Update outer/inner flops */
1578
1579     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*585);
1580 }
1581 /*
1582  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3W3_F_sse2_double
1583  * Electrostatics interaction: Ewald
1584  * VdW interaction:            None
1585  * Geometry:                   Water3-Water3
1586  * Calculate force/pot:        Force
1587  */
1588 void
1589 nb_kernel_ElecEwSw_VdwNone_GeomW3W3_F_sse2_double
1590                     (t_nblist                    * gmx_restrict       nlist,
1591                      rvec                        * gmx_restrict          xx,
1592                      rvec                        * gmx_restrict          ff,
1593                      t_forcerec                  * gmx_restrict          fr,
1594                      t_mdatoms                   * gmx_restrict     mdatoms,
1595                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1596                      t_nrnb                      * gmx_restrict        nrnb)
1597 {
1598     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1599      * just 0 for non-waters.
1600      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1601      * jnr indices corresponding to data put in the four positions in the SIMD register.
1602      */
1603     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1604     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1605     int              jnrA,jnrB;
1606     int              j_coord_offsetA,j_coord_offsetB;
1607     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1608     real             rcutoff_scalar;
1609     real             *shiftvec,*fshift,*x,*f;
1610     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1611     int              vdwioffset0;
1612     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1613     int              vdwioffset1;
1614     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1615     int              vdwioffset2;
1616     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1617     int              vdwjidx0A,vdwjidx0B;
1618     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1619     int              vdwjidx1A,vdwjidx1B;
1620     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1621     int              vdwjidx2A,vdwjidx2B;
1622     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1623     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1624     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1625     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1626     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1627     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1628     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1629     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1630     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1631     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1632     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1633     real             *charge;
1634     __m128i          ewitab;
1635     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1636     real             *ewtab;
1637     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
1638     real             rswitch_scalar,d_scalar;
1639     __m128d          dummy_mask,cutoff_mask;
1640     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1641     __m128d          one     = _mm_set1_pd(1.0);
1642     __m128d          two     = _mm_set1_pd(2.0);
1643     x                = xx[0];
1644     f                = ff[0];
1645
1646     nri              = nlist->nri;
1647     iinr             = nlist->iinr;
1648     jindex           = nlist->jindex;
1649     jjnr             = nlist->jjnr;
1650     shiftidx         = nlist->shift;
1651     gid              = nlist->gid;
1652     shiftvec         = fr->shift_vec[0];
1653     fshift           = fr->fshift[0];
1654     facel            = _mm_set1_pd(fr->epsfac);
1655     charge           = mdatoms->chargeA;
1656
1657     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1658     ewtab            = fr->ic->tabq_coul_FDV0;
1659     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1660     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1661
1662     /* Setup water-specific parameters */
1663     inr              = nlist->iinr[0];
1664     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1665     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1666     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1667
1668     jq0              = _mm_set1_pd(charge[inr+0]);
1669     jq1              = _mm_set1_pd(charge[inr+1]);
1670     jq2              = _mm_set1_pd(charge[inr+2]);
1671     qq00             = _mm_mul_pd(iq0,jq0);
1672     qq01             = _mm_mul_pd(iq0,jq1);
1673     qq02             = _mm_mul_pd(iq0,jq2);
1674     qq10             = _mm_mul_pd(iq1,jq0);
1675     qq11             = _mm_mul_pd(iq1,jq1);
1676     qq12             = _mm_mul_pd(iq1,jq2);
1677     qq20             = _mm_mul_pd(iq2,jq0);
1678     qq21             = _mm_mul_pd(iq2,jq1);
1679     qq22             = _mm_mul_pd(iq2,jq2);
1680
1681     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1682     rcutoff_scalar   = fr->rcoulomb;
1683     rcutoff          = _mm_set1_pd(rcutoff_scalar);
1684     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
1685
1686     rswitch_scalar   = fr->rcoulomb_switch;
1687     rswitch          = _mm_set1_pd(rswitch_scalar);
1688     /* Setup switch parameters */
1689     d_scalar         = rcutoff_scalar-rswitch_scalar;
1690     d                = _mm_set1_pd(d_scalar);
1691     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
1692     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1693     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1694     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
1695     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1696     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1697
1698     /* Avoid stupid compiler warnings */
1699     jnrA = jnrB = 0;
1700     j_coord_offsetA = 0;
1701     j_coord_offsetB = 0;
1702
1703     outeriter        = 0;
1704     inneriter        = 0;
1705
1706     /* Start outer loop over neighborlists */
1707     for(iidx=0; iidx<nri; iidx++)
1708     {
1709         /* Load shift vector for this list */
1710         i_shift_offset   = DIM*shiftidx[iidx];
1711
1712         /* Load limits for loop over neighbors */
1713         j_index_start    = jindex[iidx];
1714         j_index_end      = jindex[iidx+1];
1715
1716         /* Get outer coordinate index */
1717         inr              = iinr[iidx];
1718         i_coord_offset   = DIM*inr;
1719
1720         /* Load i particle coords and add shift vector */
1721         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1722                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1723
1724         fix0             = _mm_setzero_pd();
1725         fiy0             = _mm_setzero_pd();
1726         fiz0             = _mm_setzero_pd();
1727         fix1             = _mm_setzero_pd();
1728         fiy1             = _mm_setzero_pd();
1729         fiz1             = _mm_setzero_pd();
1730         fix2             = _mm_setzero_pd();
1731         fiy2             = _mm_setzero_pd();
1732         fiz2             = _mm_setzero_pd();
1733
1734         /* Start inner kernel loop */
1735         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1736         {
1737
1738             /* Get j neighbor index, and coordinate index */
1739             jnrA             = jjnr[jidx];
1740             jnrB             = jjnr[jidx+1];
1741             j_coord_offsetA  = DIM*jnrA;
1742             j_coord_offsetB  = DIM*jnrB;
1743
1744             /* load j atom coordinates */
1745             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1746                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1747
1748             /* Calculate displacement vector */
1749             dx00             = _mm_sub_pd(ix0,jx0);
1750             dy00             = _mm_sub_pd(iy0,jy0);
1751             dz00             = _mm_sub_pd(iz0,jz0);
1752             dx01             = _mm_sub_pd(ix0,jx1);
1753             dy01             = _mm_sub_pd(iy0,jy1);
1754             dz01             = _mm_sub_pd(iz0,jz1);
1755             dx02             = _mm_sub_pd(ix0,jx2);
1756             dy02             = _mm_sub_pd(iy0,jy2);
1757             dz02             = _mm_sub_pd(iz0,jz2);
1758             dx10             = _mm_sub_pd(ix1,jx0);
1759             dy10             = _mm_sub_pd(iy1,jy0);
1760             dz10             = _mm_sub_pd(iz1,jz0);
1761             dx11             = _mm_sub_pd(ix1,jx1);
1762             dy11             = _mm_sub_pd(iy1,jy1);
1763             dz11             = _mm_sub_pd(iz1,jz1);
1764             dx12             = _mm_sub_pd(ix1,jx2);
1765             dy12             = _mm_sub_pd(iy1,jy2);
1766             dz12             = _mm_sub_pd(iz1,jz2);
1767             dx20             = _mm_sub_pd(ix2,jx0);
1768             dy20             = _mm_sub_pd(iy2,jy0);
1769             dz20             = _mm_sub_pd(iz2,jz0);
1770             dx21             = _mm_sub_pd(ix2,jx1);
1771             dy21             = _mm_sub_pd(iy2,jy1);
1772             dz21             = _mm_sub_pd(iz2,jz1);
1773             dx22             = _mm_sub_pd(ix2,jx2);
1774             dy22             = _mm_sub_pd(iy2,jy2);
1775             dz22             = _mm_sub_pd(iz2,jz2);
1776
1777             /* Calculate squared distance and things based on it */
1778             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1779             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1780             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1781             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1782             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1783             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1784             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1785             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1786             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1787
1788             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1789             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1790             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1791             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1792             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1793             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1794             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1795             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1796             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1797
1798             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1799             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1800             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1801             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1802             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1803             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1804             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1805             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1806             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1807
1808             fjx0             = _mm_setzero_pd();
1809             fjy0             = _mm_setzero_pd();
1810             fjz0             = _mm_setzero_pd();
1811             fjx1             = _mm_setzero_pd();
1812             fjy1             = _mm_setzero_pd();
1813             fjz1             = _mm_setzero_pd();
1814             fjx2             = _mm_setzero_pd();
1815             fjy2             = _mm_setzero_pd();
1816             fjz2             = _mm_setzero_pd();
1817
1818             /**************************
1819              * CALCULATE INTERACTIONS *
1820              **************************/
1821
1822             if (gmx_mm_any_lt(rsq00,rcutoff2))
1823             {
1824
1825             r00              = _mm_mul_pd(rsq00,rinv00);
1826
1827             /* EWALD ELECTROSTATICS */
1828
1829             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1830             ewrt             = _mm_mul_pd(r00,ewtabscale);
1831             ewitab           = _mm_cvttpd_epi32(ewrt);
1832             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1833             ewitab           = _mm_slli_epi32(ewitab,2);
1834             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1835             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1836             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1837             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1838             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1839             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1840             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1841             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1842             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
1843             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1844
1845             d                = _mm_sub_pd(r00,rswitch);
1846             d                = _mm_max_pd(d,_mm_setzero_pd());
1847             d2               = _mm_mul_pd(d,d);
1848             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1849
1850             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1851
1852             /* Evaluate switch function */
1853             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1854             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
1855             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1856
1857             fscal            = felec;
1858
1859             fscal            = _mm_and_pd(fscal,cutoff_mask);
1860
1861             /* Calculate temporary vectorial force */
1862             tx               = _mm_mul_pd(fscal,dx00);
1863             ty               = _mm_mul_pd(fscal,dy00);
1864             tz               = _mm_mul_pd(fscal,dz00);
1865
1866             /* Update vectorial force */
1867             fix0             = _mm_add_pd(fix0,tx);
1868             fiy0             = _mm_add_pd(fiy0,ty);
1869             fiz0             = _mm_add_pd(fiz0,tz);
1870
1871             fjx0             = _mm_add_pd(fjx0,tx);
1872             fjy0             = _mm_add_pd(fjy0,ty);
1873             fjz0             = _mm_add_pd(fjz0,tz);
1874
1875             }
1876
1877             /**************************
1878              * CALCULATE INTERACTIONS *
1879              **************************/
1880
1881             if (gmx_mm_any_lt(rsq01,rcutoff2))
1882             {
1883
1884             r01              = _mm_mul_pd(rsq01,rinv01);
1885
1886             /* EWALD ELECTROSTATICS */
1887
1888             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1889             ewrt             = _mm_mul_pd(r01,ewtabscale);
1890             ewitab           = _mm_cvttpd_epi32(ewrt);
1891             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1892             ewitab           = _mm_slli_epi32(ewitab,2);
1893             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1894             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1895             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1896             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1897             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1898             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1899             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1900             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1901             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
1902             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1903
1904             d                = _mm_sub_pd(r01,rswitch);
1905             d                = _mm_max_pd(d,_mm_setzero_pd());
1906             d2               = _mm_mul_pd(d,d);
1907             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1908
1909             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1910
1911             /* Evaluate switch function */
1912             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1913             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv01,_mm_mul_pd(velec,dsw)) );
1914             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
1915
1916             fscal            = felec;
1917
1918             fscal            = _mm_and_pd(fscal,cutoff_mask);
1919
1920             /* Calculate temporary vectorial force */
1921             tx               = _mm_mul_pd(fscal,dx01);
1922             ty               = _mm_mul_pd(fscal,dy01);
1923             tz               = _mm_mul_pd(fscal,dz01);
1924
1925             /* Update vectorial force */
1926             fix0             = _mm_add_pd(fix0,tx);
1927             fiy0             = _mm_add_pd(fiy0,ty);
1928             fiz0             = _mm_add_pd(fiz0,tz);
1929
1930             fjx1             = _mm_add_pd(fjx1,tx);
1931             fjy1             = _mm_add_pd(fjy1,ty);
1932             fjz1             = _mm_add_pd(fjz1,tz);
1933
1934             }
1935
1936             /**************************
1937              * CALCULATE INTERACTIONS *
1938              **************************/
1939
1940             if (gmx_mm_any_lt(rsq02,rcutoff2))
1941             {
1942
1943             r02              = _mm_mul_pd(rsq02,rinv02);
1944
1945             /* EWALD ELECTROSTATICS */
1946
1947             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1948             ewrt             = _mm_mul_pd(r02,ewtabscale);
1949             ewitab           = _mm_cvttpd_epi32(ewrt);
1950             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1951             ewitab           = _mm_slli_epi32(ewitab,2);
1952             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1953             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1954             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1955             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
1956             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
1957             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1958             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
1959             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
1960             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
1961             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1962
1963             d                = _mm_sub_pd(r02,rswitch);
1964             d                = _mm_max_pd(d,_mm_setzero_pd());
1965             d2               = _mm_mul_pd(d,d);
1966             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1967
1968             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1969
1970             /* Evaluate switch function */
1971             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1972             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv02,_mm_mul_pd(velec,dsw)) );
1973             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
1974
1975             fscal            = felec;
1976
1977             fscal            = _mm_and_pd(fscal,cutoff_mask);
1978
1979             /* Calculate temporary vectorial force */
1980             tx               = _mm_mul_pd(fscal,dx02);
1981             ty               = _mm_mul_pd(fscal,dy02);
1982             tz               = _mm_mul_pd(fscal,dz02);
1983
1984             /* Update vectorial force */
1985             fix0             = _mm_add_pd(fix0,tx);
1986             fiy0             = _mm_add_pd(fiy0,ty);
1987             fiz0             = _mm_add_pd(fiz0,tz);
1988
1989             fjx2             = _mm_add_pd(fjx2,tx);
1990             fjy2             = _mm_add_pd(fjy2,ty);
1991             fjz2             = _mm_add_pd(fjz2,tz);
1992
1993             }
1994
1995             /**************************
1996              * CALCULATE INTERACTIONS *
1997              **************************/
1998
1999             if (gmx_mm_any_lt(rsq10,rcutoff2))
2000             {
2001
2002             r10              = _mm_mul_pd(rsq10,rinv10);
2003
2004             /* EWALD ELECTROSTATICS */
2005
2006             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2007             ewrt             = _mm_mul_pd(r10,ewtabscale);
2008             ewitab           = _mm_cvttpd_epi32(ewrt);
2009             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2010             ewitab           = _mm_slli_epi32(ewitab,2);
2011             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2012             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2013             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2014             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2015             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
2016             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2017             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2018             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2019             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
2020             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2021
2022             d                = _mm_sub_pd(r10,rswitch);
2023             d                = _mm_max_pd(d,_mm_setzero_pd());
2024             d2               = _mm_mul_pd(d,d);
2025             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2026
2027             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2028
2029             /* Evaluate switch function */
2030             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2031             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
2032             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
2033
2034             fscal            = felec;
2035
2036             fscal            = _mm_and_pd(fscal,cutoff_mask);
2037
2038             /* Calculate temporary vectorial force */
2039             tx               = _mm_mul_pd(fscal,dx10);
2040             ty               = _mm_mul_pd(fscal,dy10);
2041             tz               = _mm_mul_pd(fscal,dz10);
2042
2043             /* Update vectorial force */
2044             fix1             = _mm_add_pd(fix1,tx);
2045             fiy1             = _mm_add_pd(fiy1,ty);
2046             fiz1             = _mm_add_pd(fiz1,tz);
2047
2048             fjx0             = _mm_add_pd(fjx0,tx);
2049             fjy0             = _mm_add_pd(fjy0,ty);
2050             fjz0             = _mm_add_pd(fjz0,tz);
2051
2052             }
2053
2054             /**************************
2055              * CALCULATE INTERACTIONS *
2056              **************************/
2057
2058             if (gmx_mm_any_lt(rsq11,rcutoff2))
2059             {
2060
2061             r11              = _mm_mul_pd(rsq11,rinv11);
2062
2063             /* EWALD ELECTROSTATICS */
2064
2065             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2066             ewrt             = _mm_mul_pd(r11,ewtabscale);
2067             ewitab           = _mm_cvttpd_epi32(ewrt);
2068             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2069             ewitab           = _mm_slli_epi32(ewitab,2);
2070             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2071             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2072             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2073             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2074             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
2075             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2076             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2077             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2078             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
2079             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2080
2081             d                = _mm_sub_pd(r11,rswitch);
2082             d                = _mm_max_pd(d,_mm_setzero_pd());
2083             d2               = _mm_mul_pd(d,d);
2084             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2085
2086             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2087
2088             /* Evaluate switch function */
2089             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2090             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
2091             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
2092
2093             fscal            = felec;
2094
2095             fscal            = _mm_and_pd(fscal,cutoff_mask);
2096
2097             /* Calculate temporary vectorial force */
2098             tx               = _mm_mul_pd(fscal,dx11);
2099             ty               = _mm_mul_pd(fscal,dy11);
2100             tz               = _mm_mul_pd(fscal,dz11);
2101
2102             /* Update vectorial force */
2103             fix1             = _mm_add_pd(fix1,tx);
2104             fiy1             = _mm_add_pd(fiy1,ty);
2105             fiz1             = _mm_add_pd(fiz1,tz);
2106
2107             fjx1             = _mm_add_pd(fjx1,tx);
2108             fjy1             = _mm_add_pd(fjy1,ty);
2109             fjz1             = _mm_add_pd(fjz1,tz);
2110
2111             }
2112
2113             /**************************
2114              * CALCULATE INTERACTIONS *
2115              **************************/
2116
2117             if (gmx_mm_any_lt(rsq12,rcutoff2))
2118             {
2119
2120             r12              = _mm_mul_pd(rsq12,rinv12);
2121
2122             /* EWALD ELECTROSTATICS */
2123
2124             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2125             ewrt             = _mm_mul_pd(r12,ewtabscale);
2126             ewitab           = _mm_cvttpd_epi32(ewrt);
2127             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2128             ewitab           = _mm_slli_epi32(ewitab,2);
2129             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2130             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2131             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2132             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2133             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
2134             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2135             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2136             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2137             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
2138             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2139
2140             d                = _mm_sub_pd(r12,rswitch);
2141             d                = _mm_max_pd(d,_mm_setzero_pd());
2142             d2               = _mm_mul_pd(d,d);
2143             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2144
2145             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2146
2147             /* Evaluate switch function */
2148             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2149             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
2150             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
2151
2152             fscal            = felec;
2153
2154             fscal            = _mm_and_pd(fscal,cutoff_mask);
2155
2156             /* Calculate temporary vectorial force */
2157             tx               = _mm_mul_pd(fscal,dx12);
2158             ty               = _mm_mul_pd(fscal,dy12);
2159             tz               = _mm_mul_pd(fscal,dz12);
2160
2161             /* Update vectorial force */
2162             fix1             = _mm_add_pd(fix1,tx);
2163             fiy1             = _mm_add_pd(fiy1,ty);
2164             fiz1             = _mm_add_pd(fiz1,tz);
2165
2166             fjx2             = _mm_add_pd(fjx2,tx);
2167             fjy2             = _mm_add_pd(fjy2,ty);
2168             fjz2             = _mm_add_pd(fjz2,tz);
2169
2170             }
2171
2172             /**************************
2173              * CALCULATE INTERACTIONS *
2174              **************************/
2175
2176             if (gmx_mm_any_lt(rsq20,rcutoff2))
2177             {
2178
2179             r20              = _mm_mul_pd(rsq20,rinv20);
2180
2181             /* EWALD ELECTROSTATICS */
2182
2183             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2184             ewrt             = _mm_mul_pd(r20,ewtabscale);
2185             ewitab           = _mm_cvttpd_epi32(ewrt);
2186             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2187             ewitab           = _mm_slli_epi32(ewitab,2);
2188             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2189             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2190             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2191             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2192             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
2193             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2194             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2195             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2196             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
2197             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2198
2199             d                = _mm_sub_pd(r20,rswitch);
2200             d                = _mm_max_pd(d,_mm_setzero_pd());
2201             d2               = _mm_mul_pd(d,d);
2202             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2203
2204             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2205
2206             /* Evaluate switch function */
2207             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2208             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
2209             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
2210
2211             fscal            = felec;
2212
2213             fscal            = _mm_and_pd(fscal,cutoff_mask);
2214
2215             /* Calculate temporary vectorial force */
2216             tx               = _mm_mul_pd(fscal,dx20);
2217             ty               = _mm_mul_pd(fscal,dy20);
2218             tz               = _mm_mul_pd(fscal,dz20);
2219
2220             /* Update vectorial force */
2221             fix2             = _mm_add_pd(fix2,tx);
2222             fiy2             = _mm_add_pd(fiy2,ty);
2223             fiz2             = _mm_add_pd(fiz2,tz);
2224
2225             fjx0             = _mm_add_pd(fjx0,tx);
2226             fjy0             = _mm_add_pd(fjy0,ty);
2227             fjz0             = _mm_add_pd(fjz0,tz);
2228
2229             }
2230
2231             /**************************
2232              * CALCULATE INTERACTIONS *
2233              **************************/
2234
2235             if (gmx_mm_any_lt(rsq21,rcutoff2))
2236             {
2237
2238             r21              = _mm_mul_pd(rsq21,rinv21);
2239
2240             /* EWALD ELECTROSTATICS */
2241
2242             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2243             ewrt             = _mm_mul_pd(r21,ewtabscale);
2244             ewitab           = _mm_cvttpd_epi32(ewrt);
2245             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2246             ewitab           = _mm_slli_epi32(ewitab,2);
2247             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2248             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2249             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2250             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2251             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
2252             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2253             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2254             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2255             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
2256             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2257
2258             d                = _mm_sub_pd(r21,rswitch);
2259             d                = _mm_max_pd(d,_mm_setzero_pd());
2260             d2               = _mm_mul_pd(d,d);
2261             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2262
2263             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2264
2265             /* Evaluate switch function */
2266             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2267             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
2268             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
2269
2270             fscal            = felec;
2271
2272             fscal            = _mm_and_pd(fscal,cutoff_mask);
2273
2274             /* Calculate temporary vectorial force */
2275             tx               = _mm_mul_pd(fscal,dx21);
2276             ty               = _mm_mul_pd(fscal,dy21);
2277             tz               = _mm_mul_pd(fscal,dz21);
2278
2279             /* Update vectorial force */
2280             fix2             = _mm_add_pd(fix2,tx);
2281             fiy2             = _mm_add_pd(fiy2,ty);
2282             fiz2             = _mm_add_pd(fiz2,tz);
2283
2284             fjx1             = _mm_add_pd(fjx1,tx);
2285             fjy1             = _mm_add_pd(fjy1,ty);
2286             fjz1             = _mm_add_pd(fjz1,tz);
2287
2288             }
2289
2290             /**************************
2291              * CALCULATE INTERACTIONS *
2292              **************************/
2293
2294             if (gmx_mm_any_lt(rsq22,rcutoff2))
2295             {
2296
2297             r22              = _mm_mul_pd(rsq22,rinv22);
2298
2299             /* EWALD ELECTROSTATICS */
2300
2301             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2302             ewrt             = _mm_mul_pd(r22,ewtabscale);
2303             ewitab           = _mm_cvttpd_epi32(ewrt);
2304             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2305             ewitab           = _mm_slli_epi32(ewitab,2);
2306             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2307             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2308             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2309             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2310             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
2311             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2312             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2313             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2314             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
2315             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2316
2317             d                = _mm_sub_pd(r22,rswitch);
2318             d                = _mm_max_pd(d,_mm_setzero_pd());
2319             d2               = _mm_mul_pd(d,d);
2320             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2321
2322             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2323
2324             /* Evaluate switch function */
2325             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2326             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
2327             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
2328
2329             fscal            = felec;
2330
2331             fscal            = _mm_and_pd(fscal,cutoff_mask);
2332
2333             /* Calculate temporary vectorial force */
2334             tx               = _mm_mul_pd(fscal,dx22);
2335             ty               = _mm_mul_pd(fscal,dy22);
2336             tz               = _mm_mul_pd(fscal,dz22);
2337
2338             /* Update vectorial force */
2339             fix2             = _mm_add_pd(fix2,tx);
2340             fiy2             = _mm_add_pd(fiy2,ty);
2341             fiz2             = _mm_add_pd(fiz2,tz);
2342
2343             fjx2             = _mm_add_pd(fjx2,tx);
2344             fjy2             = _mm_add_pd(fjy2,ty);
2345             fjz2             = _mm_add_pd(fjz2,tz);
2346
2347             }
2348
2349             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2350
2351             /* Inner loop uses 558 flops */
2352         }
2353
2354         if(jidx<j_index_end)
2355         {
2356
2357             jnrA             = jjnr[jidx];
2358             j_coord_offsetA  = DIM*jnrA;
2359
2360             /* load j atom coordinates */
2361             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
2362                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2363
2364             /* Calculate displacement vector */
2365             dx00             = _mm_sub_pd(ix0,jx0);
2366             dy00             = _mm_sub_pd(iy0,jy0);
2367             dz00             = _mm_sub_pd(iz0,jz0);
2368             dx01             = _mm_sub_pd(ix0,jx1);
2369             dy01             = _mm_sub_pd(iy0,jy1);
2370             dz01             = _mm_sub_pd(iz0,jz1);
2371             dx02             = _mm_sub_pd(ix0,jx2);
2372             dy02             = _mm_sub_pd(iy0,jy2);
2373             dz02             = _mm_sub_pd(iz0,jz2);
2374             dx10             = _mm_sub_pd(ix1,jx0);
2375             dy10             = _mm_sub_pd(iy1,jy0);
2376             dz10             = _mm_sub_pd(iz1,jz0);
2377             dx11             = _mm_sub_pd(ix1,jx1);
2378             dy11             = _mm_sub_pd(iy1,jy1);
2379             dz11             = _mm_sub_pd(iz1,jz1);
2380             dx12             = _mm_sub_pd(ix1,jx2);
2381             dy12             = _mm_sub_pd(iy1,jy2);
2382             dz12             = _mm_sub_pd(iz1,jz2);
2383             dx20             = _mm_sub_pd(ix2,jx0);
2384             dy20             = _mm_sub_pd(iy2,jy0);
2385             dz20             = _mm_sub_pd(iz2,jz0);
2386             dx21             = _mm_sub_pd(ix2,jx1);
2387             dy21             = _mm_sub_pd(iy2,jy1);
2388             dz21             = _mm_sub_pd(iz2,jz1);
2389             dx22             = _mm_sub_pd(ix2,jx2);
2390             dy22             = _mm_sub_pd(iy2,jy2);
2391             dz22             = _mm_sub_pd(iz2,jz2);
2392
2393             /* Calculate squared distance and things based on it */
2394             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2395             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
2396             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
2397             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
2398             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2399             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2400             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
2401             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2402             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2403
2404             rinv00           = gmx_mm_invsqrt_pd(rsq00);
2405             rinv01           = gmx_mm_invsqrt_pd(rsq01);
2406             rinv02           = gmx_mm_invsqrt_pd(rsq02);
2407             rinv10           = gmx_mm_invsqrt_pd(rsq10);
2408             rinv11           = gmx_mm_invsqrt_pd(rsq11);
2409             rinv12           = gmx_mm_invsqrt_pd(rsq12);
2410             rinv20           = gmx_mm_invsqrt_pd(rsq20);
2411             rinv21           = gmx_mm_invsqrt_pd(rsq21);
2412             rinv22           = gmx_mm_invsqrt_pd(rsq22);
2413
2414             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
2415             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
2416             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
2417             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
2418             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
2419             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
2420             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
2421             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
2422             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
2423
2424             fjx0             = _mm_setzero_pd();
2425             fjy0             = _mm_setzero_pd();
2426             fjz0             = _mm_setzero_pd();
2427             fjx1             = _mm_setzero_pd();
2428             fjy1             = _mm_setzero_pd();
2429             fjz1             = _mm_setzero_pd();
2430             fjx2             = _mm_setzero_pd();
2431             fjy2             = _mm_setzero_pd();
2432             fjz2             = _mm_setzero_pd();
2433
2434             /**************************
2435              * CALCULATE INTERACTIONS *
2436              **************************/
2437
2438             if (gmx_mm_any_lt(rsq00,rcutoff2))
2439             {
2440
2441             r00              = _mm_mul_pd(rsq00,rinv00);
2442
2443             /* EWALD ELECTROSTATICS */
2444
2445             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2446             ewrt             = _mm_mul_pd(r00,ewtabscale);
2447             ewitab           = _mm_cvttpd_epi32(ewrt);
2448             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2449             ewitab           = _mm_slli_epi32(ewitab,2);
2450             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2451             ewtabD           = _mm_setzero_pd();
2452             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2453             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2454             ewtabFn          = _mm_setzero_pd();
2455             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2456             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2457             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2458             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
2459             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
2460
2461             d                = _mm_sub_pd(r00,rswitch);
2462             d                = _mm_max_pd(d,_mm_setzero_pd());
2463             d2               = _mm_mul_pd(d,d);
2464             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2465
2466             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2467
2468             /* Evaluate switch function */
2469             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2470             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
2471             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
2472
2473             fscal            = felec;
2474
2475             fscal            = _mm_and_pd(fscal,cutoff_mask);
2476
2477             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2478
2479             /* Calculate temporary vectorial force */
2480             tx               = _mm_mul_pd(fscal,dx00);
2481             ty               = _mm_mul_pd(fscal,dy00);
2482             tz               = _mm_mul_pd(fscal,dz00);
2483
2484             /* Update vectorial force */
2485             fix0             = _mm_add_pd(fix0,tx);
2486             fiy0             = _mm_add_pd(fiy0,ty);
2487             fiz0             = _mm_add_pd(fiz0,tz);
2488
2489             fjx0             = _mm_add_pd(fjx0,tx);
2490             fjy0             = _mm_add_pd(fjy0,ty);
2491             fjz0             = _mm_add_pd(fjz0,tz);
2492
2493             }
2494
2495             /**************************
2496              * CALCULATE INTERACTIONS *
2497              **************************/
2498
2499             if (gmx_mm_any_lt(rsq01,rcutoff2))
2500             {
2501
2502             r01              = _mm_mul_pd(rsq01,rinv01);
2503
2504             /* EWALD ELECTROSTATICS */
2505
2506             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2507             ewrt             = _mm_mul_pd(r01,ewtabscale);
2508             ewitab           = _mm_cvttpd_epi32(ewrt);
2509             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2510             ewitab           = _mm_slli_epi32(ewitab,2);
2511             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2512             ewtabD           = _mm_setzero_pd();
2513             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2514             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2515             ewtabFn          = _mm_setzero_pd();
2516             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2517             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2518             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2519             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
2520             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
2521
2522             d                = _mm_sub_pd(r01,rswitch);
2523             d                = _mm_max_pd(d,_mm_setzero_pd());
2524             d2               = _mm_mul_pd(d,d);
2525             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2526
2527             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2528
2529             /* Evaluate switch function */
2530             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2531             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv01,_mm_mul_pd(velec,dsw)) );
2532             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
2533
2534             fscal            = felec;
2535
2536             fscal            = _mm_and_pd(fscal,cutoff_mask);
2537
2538             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2539
2540             /* Calculate temporary vectorial force */
2541             tx               = _mm_mul_pd(fscal,dx01);
2542             ty               = _mm_mul_pd(fscal,dy01);
2543             tz               = _mm_mul_pd(fscal,dz01);
2544
2545             /* Update vectorial force */
2546             fix0             = _mm_add_pd(fix0,tx);
2547             fiy0             = _mm_add_pd(fiy0,ty);
2548             fiz0             = _mm_add_pd(fiz0,tz);
2549
2550             fjx1             = _mm_add_pd(fjx1,tx);
2551             fjy1             = _mm_add_pd(fjy1,ty);
2552             fjz1             = _mm_add_pd(fjz1,tz);
2553
2554             }
2555
2556             /**************************
2557              * CALCULATE INTERACTIONS *
2558              **************************/
2559
2560             if (gmx_mm_any_lt(rsq02,rcutoff2))
2561             {
2562
2563             r02              = _mm_mul_pd(rsq02,rinv02);
2564
2565             /* EWALD ELECTROSTATICS */
2566
2567             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2568             ewrt             = _mm_mul_pd(r02,ewtabscale);
2569             ewitab           = _mm_cvttpd_epi32(ewrt);
2570             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2571             ewitab           = _mm_slli_epi32(ewitab,2);
2572             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2573             ewtabD           = _mm_setzero_pd();
2574             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2575             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2576             ewtabFn          = _mm_setzero_pd();
2577             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2578             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2579             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2580             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
2581             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
2582
2583             d                = _mm_sub_pd(r02,rswitch);
2584             d                = _mm_max_pd(d,_mm_setzero_pd());
2585             d2               = _mm_mul_pd(d,d);
2586             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2587
2588             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2589
2590             /* Evaluate switch function */
2591             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2592             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv02,_mm_mul_pd(velec,dsw)) );
2593             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
2594
2595             fscal            = felec;
2596
2597             fscal            = _mm_and_pd(fscal,cutoff_mask);
2598
2599             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2600
2601             /* Calculate temporary vectorial force */
2602             tx               = _mm_mul_pd(fscal,dx02);
2603             ty               = _mm_mul_pd(fscal,dy02);
2604             tz               = _mm_mul_pd(fscal,dz02);
2605
2606             /* Update vectorial force */
2607             fix0             = _mm_add_pd(fix0,tx);
2608             fiy0             = _mm_add_pd(fiy0,ty);
2609             fiz0             = _mm_add_pd(fiz0,tz);
2610
2611             fjx2             = _mm_add_pd(fjx2,tx);
2612             fjy2             = _mm_add_pd(fjy2,ty);
2613             fjz2             = _mm_add_pd(fjz2,tz);
2614
2615             }
2616
2617             /**************************
2618              * CALCULATE INTERACTIONS *
2619              **************************/
2620
2621             if (gmx_mm_any_lt(rsq10,rcutoff2))
2622             {
2623
2624             r10              = _mm_mul_pd(rsq10,rinv10);
2625
2626             /* EWALD ELECTROSTATICS */
2627
2628             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2629             ewrt             = _mm_mul_pd(r10,ewtabscale);
2630             ewitab           = _mm_cvttpd_epi32(ewrt);
2631             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2632             ewitab           = _mm_slli_epi32(ewitab,2);
2633             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2634             ewtabD           = _mm_setzero_pd();
2635             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2636             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2637             ewtabFn          = _mm_setzero_pd();
2638             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2639             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2640             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2641             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
2642             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2643
2644             d                = _mm_sub_pd(r10,rswitch);
2645             d                = _mm_max_pd(d,_mm_setzero_pd());
2646             d2               = _mm_mul_pd(d,d);
2647             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2648
2649             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2650
2651             /* Evaluate switch function */
2652             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2653             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
2654             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
2655
2656             fscal            = felec;
2657
2658             fscal            = _mm_and_pd(fscal,cutoff_mask);
2659
2660             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2661
2662             /* Calculate temporary vectorial force */
2663             tx               = _mm_mul_pd(fscal,dx10);
2664             ty               = _mm_mul_pd(fscal,dy10);
2665             tz               = _mm_mul_pd(fscal,dz10);
2666
2667             /* Update vectorial force */
2668             fix1             = _mm_add_pd(fix1,tx);
2669             fiy1             = _mm_add_pd(fiy1,ty);
2670             fiz1             = _mm_add_pd(fiz1,tz);
2671
2672             fjx0             = _mm_add_pd(fjx0,tx);
2673             fjy0             = _mm_add_pd(fjy0,ty);
2674             fjz0             = _mm_add_pd(fjz0,tz);
2675
2676             }
2677
2678             /**************************
2679              * CALCULATE INTERACTIONS *
2680              **************************/
2681
2682             if (gmx_mm_any_lt(rsq11,rcutoff2))
2683             {
2684
2685             r11              = _mm_mul_pd(rsq11,rinv11);
2686
2687             /* EWALD ELECTROSTATICS */
2688
2689             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2690             ewrt             = _mm_mul_pd(r11,ewtabscale);
2691             ewitab           = _mm_cvttpd_epi32(ewrt);
2692             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2693             ewitab           = _mm_slli_epi32(ewitab,2);
2694             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2695             ewtabD           = _mm_setzero_pd();
2696             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2697             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2698             ewtabFn          = _mm_setzero_pd();
2699             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2700             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2701             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2702             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
2703             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2704
2705             d                = _mm_sub_pd(r11,rswitch);
2706             d                = _mm_max_pd(d,_mm_setzero_pd());
2707             d2               = _mm_mul_pd(d,d);
2708             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2709
2710             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2711
2712             /* Evaluate switch function */
2713             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2714             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
2715             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
2716
2717             fscal            = felec;
2718
2719             fscal            = _mm_and_pd(fscal,cutoff_mask);
2720
2721             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2722
2723             /* Calculate temporary vectorial force */
2724             tx               = _mm_mul_pd(fscal,dx11);
2725             ty               = _mm_mul_pd(fscal,dy11);
2726             tz               = _mm_mul_pd(fscal,dz11);
2727
2728             /* Update vectorial force */
2729             fix1             = _mm_add_pd(fix1,tx);
2730             fiy1             = _mm_add_pd(fiy1,ty);
2731             fiz1             = _mm_add_pd(fiz1,tz);
2732
2733             fjx1             = _mm_add_pd(fjx1,tx);
2734             fjy1             = _mm_add_pd(fjy1,ty);
2735             fjz1             = _mm_add_pd(fjz1,tz);
2736
2737             }
2738
2739             /**************************
2740              * CALCULATE INTERACTIONS *
2741              **************************/
2742
2743             if (gmx_mm_any_lt(rsq12,rcutoff2))
2744             {
2745
2746             r12              = _mm_mul_pd(rsq12,rinv12);
2747
2748             /* EWALD ELECTROSTATICS */
2749
2750             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2751             ewrt             = _mm_mul_pd(r12,ewtabscale);
2752             ewitab           = _mm_cvttpd_epi32(ewrt);
2753             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2754             ewitab           = _mm_slli_epi32(ewitab,2);
2755             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2756             ewtabD           = _mm_setzero_pd();
2757             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2758             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2759             ewtabFn          = _mm_setzero_pd();
2760             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2761             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2762             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2763             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
2764             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2765
2766             d                = _mm_sub_pd(r12,rswitch);
2767             d                = _mm_max_pd(d,_mm_setzero_pd());
2768             d2               = _mm_mul_pd(d,d);
2769             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2770
2771             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2772
2773             /* Evaluate switch function */
2774             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2775             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
2776             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
2777
2778             fscal            = felec;
2779
2780             fscal            = _mm_and_pd(fscal,cutoff_mask);
2781
2782             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2783
2784             /* Calculate temporary vectorial force */
2785             tx               = _mm_mul_pd(fscal,dx12);
2786             ty               = _mm_mul_pd(fscal,dy12);
2787             tz               = _mm_mul_pd(fscal,dz12);
2788
2789             /* Update vectorial force */
2790             fix1             = _mm_add_pd(fix1,tx);
2791             fiy1             = _mm_add_pd(fiy1,ty);
2792             fiz1             = _mm_add_pd(fiz1,tz);
2793
2794             fjx2             = _mm_add_pd(fjx2,tx);
2795             fjy2             = _mm_add_pd(fjy2,ty);
2796             fjz2             = _mm_add_pd(fjz2,tz);
2797
2798             }
2799
2800             /**************************
2801              * CALCULATE INTERACTIONS *
2802              **************************/
2803
2804             if (gmx_mm_any_lt(rsq20,rcutoff2))
2805             {
2806
2807             r20              = _mm_mul_pd(rsq20,rinv20);
2808
2809             /* EWALD ELECTROSTATICS */
2810
2811             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2812             ewrt             = _mm_mul_pd(r20,ewtabscale);
2813             ewitab           = _mm_cvttpd_epi32(ewrt);
2814             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2815             ewitab           = _mm_slli_epi32(ewitab,2);
2816             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2817             ewtabD           = _mm_setzero_pd();
2818             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2819             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2820             ewtabFn          = _mm_setzero_pd();
2821             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2822             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2823             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2824             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
2825             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2826
2827             d                = _mm_sub_pd(r20,rswitch);
2828             d                = _mm_max_pd(d,_mm_setzero_pd());
2829             d2               = _mm_mul_pd(d,d);
2830             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2831
2832             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2833
2834             /* Evaluate switch function */
2835             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2836             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
2837             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
2838
2839             fscal            = felec;
2840
2841             fscal            = _mm_and_pd(fscal,cutoff_mask);
2842
2843             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2844
2845             /* Calculate temporary vectorial force */
2846             tx               = _mm_mul_pd(fscal,dx20);
2847             ty               = _mm_mul_pd(fscal,dy20);
2848             tz               = _mm_mul_pd(fscal,dz20);
2849
2850             /* Update vectorial force */
2851             fix2             = _mm_add_pd(fix2,tx);
2852             fiy2             = _mm_add_pd(fiy2,ty);
2853             fiz2             = _mm_add_pd(fiz2,tz);
2854
2855             fjx0             = _mm_add_pd(fjx0,tx);
2856             fjy0             = _mm_add_pd(fjy0,ty);
2857             fjz0             = _mm_add_pd(fjz0,tz);
2858
2859             }
2860
2861             /**************************
2862              * CALCULATE INTERACTIONS *
2863              **************************/
2864
2865             if (gmx_mm_any_lt(rsq21,rcutoff2))
2866             {
2867
2868             r21              = _mm_mul_pd(rsq21,rinv21);
2869
2870             /* EWALD ELECTROSTATICS */
2871
2872             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2873             ewrt             = _mm_mul_pd(r21,ewtabscale);
2874             ewitab           = _mm_cvttpd_epi32(ewrt);
2875             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2876             ewitab           = _mm_slli_epi32(ewitab,2);
2877             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2878             ewtabD           = _mm_setzero_pd();
2879             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2880             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2881             ewtabFn          = _mm_setzero_pd();
2882             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2883             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2884             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2885             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
2886             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2887
2888             d                = _mm_sub_pd(r21,rswitch);
2889             d                = _mm_max_pd(d,_mm_setzero_pd());
2890             d2               = _mm_mul_pd(d,d);
2891             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2892
2893             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2894
2895             /* Evaluate switch function */
2896             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2897             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
2898             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
2899
2900             fscal            = felec;
2901
2902             fscal            = _mm_and_pd(fscal,cutoff_mask);
2903
2904             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2905
2906             /* Calculate temporary vectorial force */
2907             tx               = _mm_mul_pd(fscal,dx21);
2908             ty               = _mm_mul_pd(fscal,dy21);
2909             tz               = _mm_mul_pd(fscal,dz21);
2910
2911             /* Update vectorial force */
2912             fix2             = _mm_add_pd(fix2,tx);
2913             fiy2             = _mm_add_pd(fiy2,ty);
2914             fiz2             = _mm_add_pd(fiz2,tz);
2915
2916             fjx1             = _mm_add_pd(fjx1,tx);
2917             fjy1             = _mm_add_pd(fjy1,ty);
2918             fjz1             = _mm_add_pd(fjz1,tz);
2919
2920             }
2921
2922             /**************************
2923              * CALCULATE INTERACTIONS *
2924              **************************/
2925
2926             if (gmx_mm_any_lt(rsq22,rcutoff2))
2927             {
2928
2929             r22              = _mm_mul_pd(rsq22,rinv22);
2930
2931             /* EWALD ELECTROSTATICS */
2932
2933             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2934             ewrt             = _mm_mul_pd(r22,ewtabscale);
2935             ewitab           = _mm_cvttpd_epi32(ewrt);
2936             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
2937             ewitab           = _mm_slli_epi32(ewitab,2);
2938             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2939             ewtabD           = _mm_setzero_pd();
2940             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2941             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
2942             ewtabFn          = _mm_setzero_pd();
2943             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2944             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
2945             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
2946             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
2947             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2948
2949             d                = _mm_sub_pd(r22,rswitch);
2950             d                = _mm_max_pd(d,_mm_setzero_pd());
2951             d2               = _mm_mul_pd(d,d);
2952             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
2953
2954             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
2955
2956             /* Evaluate switch function */
2957             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2958             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
2959             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
2960
2961             fscal            = felec;
2962
2963             fscal            = _mm_and_pd(fscal,cutoff_mask);
2964
2965             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2966
2967             /* Calculate temporary vectorial force */
2968             tx               = _mm_mul_pd(fscal,dx22);
2969             ty               = _mm_mul_pd(fscal,dy22);
2970             tz               = _mm_mul_pd(fscal,dz22);
2971
2972             /* Update vectorial force */
2973             fix2             = _mm_add_pd(fix2,tx);
2974             fiy2             = _mm_add_pd(fiy2,ty);
2975             fiz2             = _mm_add_pd(fiz2,tz);
2976
2977             fjx2             = _mm_add_pd(fjx2,tx);
2978             fjy2             = _mm_add_pd(fjy2,ty);
2979             fjz2             = _mm_add_pd(fjz2,tz);
2980
2981             }
2982
2983             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2984
2985             /* Inner loop uses 558 flops */
2986         }
2987
2988         /* End of innermost loop */
2989
2990         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2991                                               f+i_coord_offset,fshift+i_shift_offset);
2992
2993         /* Increment number of inner iterations */
2994         inneriter                  += j_index_end - j_index_start;
2995
2996         /* Outer loop uses 18 flops */
2997     }
2998
2999     /* Increment number of outer iterations */
3000     outeriter        += nri;
3001
3002     /* Update outer/inner flops */
3003
3004     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*558);
3005 }