src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSw_VdwNone_GeomW3W3_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "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_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3W3_VF_avx_128_fma_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_avx_128_fma_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,twoeweps,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 #ifdef __XOP__
 305             eweps            = _mm_frcz_pd(ewrt);
 306 #else
 307             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 308 #endif
 309             twoeweps         = _mm_add_pd(eweps,eweps);
 310             ewitab           = _mm_slli_epi32(ewitab,2);
 311             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 312             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 313             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 314             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 315             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 316             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 317             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 318             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 319             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 320             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 321
 322             d                = _mm_sub_pd(r00,rswitch);
 323             d                = _mm_max_pd(d,_mm_setzero_pd());
 324             d2               = _mm_mul_pd(d,d);
 325             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 326
 327             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 328
 329             /* Evaluate switch function */
 330             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 331             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 332             velec            = _mm_mul_pd(velec,sw);
 333             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _mm_and_pd(velec,cutoff_mask);
 337             velecsum         = _mm_add_pd(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341             fscal            = _mm_and_pd(fscal,cutoff_mask);
 342
 343             /* Update vectorial force */
 344             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 345             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 346             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 347
 348             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 349             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 350             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 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 #ifdef __XOP__
 369             eweps            = _mm_frcz_pd(ewrt);
 370 #else
 371             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 372 #endif
 373             twoeweps         = _mm_add_pd(eweps,eweps);
 374             ewitab           = _mm_slli_epi32(ewitab,2);
 375             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 376             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 377             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 378             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 379             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 380             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 381             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 382             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 383             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
 384             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 385
 386             d                = _mm_sub_pd(r01,rswitch);
 387             d                = _mm_max_pd(d,_mm_setzero_pd());
 388             d2               = _mm_mul_pd(d,d);
 389             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 390
 391             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 392
 393             /* Evaluate switch function */
 394             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 395             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv01,_mm_mul_pd(velec,dsw)) );
 396             velec            = _mm_mul_pd(velec,sw);
 397             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
 398
 399             /* Update potential sum for this i atom from the interaction with this j atom. */
 400             velec            = _mm_and_pd(velec,cutoff_mask);
 401             velecsum         = _mm_add_pd(velecsum,velec);
 402
 403             fscal            = felec;
 404
 405             fscal            = _mm_and_pd(fscal,cutoff_mask);
 406
 407             /* Update vectorial force */
 408             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 409             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 410             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 411
 412             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 413             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 414             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 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 #ifdef __XOP__
 433             eweps            = _mm_frcz_pd(ewrt);
 434 #else
 435             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 436 #endif
 437             twoeweps         = _mm_add_pd(eweps,eweps);
 438             ewitab           = _mm_slli_epi32(ewitab,2);
 439             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 440             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 441             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 442             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 443             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 444             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 445             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 446             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 447             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
 448             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 449
 450             d                = _mm_sub_pd(r02,rswitch);
 451             d                = _mm_max_pd(d,_mm_setzero_pd());
 452             d2               = _mm_mul_pd(d,d);
 453             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 454
 455             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 456
 457             /* Evaluate switch function */
 458             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 459             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv02,_mm_mul_pd(velec,dsw)) );
 460             velec            = _mm_mul_pd(velec,sw);
 461             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
 462
 463             /* Update potential sum for this i atom from the interaction with this j atom. */
 464             velec            = _mm_and_pd(velec,cutoff_mask);
 465             velecsum         = _mm_add_pd(velecsum,velec);
 466
 467             fscal            = felec;
 468
 469             fscal            = _mm_and_pd(fscal,cutoff_mask);
 470
 471             /* Update vectorial force */
 472             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 473             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 474             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 475
 476             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 477             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 478             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 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 #ifdef __XOP__
 497             eweps            = _mm_frcz_pd(ewrt);
 498 #else
 499             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 500 #endif
 501             twoeweps         = _mm_add_pd(eweps,eweps);
 502             ewitab           = _mm_slli_epi32(ewitab,2);
 503             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 504             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 505             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 506             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 507             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 508             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 509             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 510             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 511             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 512             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 513
 514             d                = _mm_sub_pd(r10,rswitch);
 515             d                = _mm_max_pd(d,_mm_setzero_pd());
 516             d2               = _mm_mul_pd(d,d);
 517             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 518
 519             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 520
 521             /* Evaluate switch function */
 522             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 523             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 524             velec            = _mm_mul_pd(velec,sw);
 525             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 526
 527             /* Update potential sum for this i atom from the interaction with this j atom. */
 528             velec            = _mm_and_pd(velec,cutoff_mask);
 529             velecsum         = _mm_add_pd(velecsum,velec);
 530
 531             fscal            = felec;
 532
 533             fscal            = _mm_and_pd(fscal,cutoff_mask);
 534
 535             /* Update vectorial force */
 536             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 537             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 538             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 539
 540             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 541             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 542             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 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 #ifdef __XOP__
 561             eweps            = _mm_frcz_pd(ewrt);
 562 #else
 563             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 564 #endif
 565             twoeweps         = _mm_add_pd(eweps,eweps);
 566             ewitab           = _mm_slli_epi32(ewitab,2);
 567             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 568             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 569             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 570             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 571             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 572             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 573             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 574             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 575             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 576             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 577
 578             d                = _mm_sub_pd(r11,rswitch);
 579             d                = _mm_max_pd(d,_mm_setzero_pd());
 580             d2               = _mm_mul_pd(d,d);
 581             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 582
 583             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 584
 585             /* Evaluate switch function */
 586             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 587             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
 588             velec            = _mm_mul_pd(velec,sw);
 589             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velec            = _mm_and_pd(velec,cutoff_mask);
 593             velecsum         = _mm_add_pd(velecsum,velec);
 594
 595             fscal            = felec;
 596
 597             fscal            = _mm_and_pd(fscal,cutoff_mask);
 598
 599             /* Update vectorial force */
 600             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 601             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 602             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 603
 604             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 605             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 606             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 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 #ifdef __XOP__
 625             eweps            = _mm_frcz_pd(ewrt);
 626 #else
 627             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 628 #endif
 629             twoeweps         = _mm_add_pd(eweps,eweps);
 630             ewitab           = _mm_slli_epi32(ewitab,2);
 631             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 632             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 633             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 634             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 635             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 636             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 637             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 638             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 639             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 640             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 641
 642             d                = _mm_sub_pd(r12,rswitch);
 643             d                = _mm_max_pd(d,_mm_setzero_pd());
 644             d2               = _mm_mul_pd(d,d);
 645             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 646
 647             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 648
 649             /* Evaluate switch function */
 650             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 651             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
 652             velec            = _mm_mul_pd(velec,sw);
 653             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
 654
 655             /* Update potential sum for this i atom from the interaction with this j atom. */
 656             velec            = _mm_and_pd(velec,cutoff_mask);
 657             velecsum         = _mm_add_pd(velecsum,velec);
 658
 659             fscal            = felec;
 660
 661             fscal            = _mm_and_pd(fscal,cutoff_mask);
 662
 663             /* Update vectorial force */
 664             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 665             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 666             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 667
 668             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 669             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 670             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 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 #ifdef __XOP__
 689             eweps            = _mm_frcz_pd(ewrt);
 690 #else
 691             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 692 #endif
 693             twoeweps         = _mm_add_pd(eweps,eweps);
 694             ewitab           = _mm_slli_epi32(ewitab,2);
 695             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 696             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 697             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 698             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 699             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 700             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 701             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 702             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 703             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 704             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 705
 706             d                = _mm_sub_pd(r20,rswitch);
 707             d                = _mm_max_pd(d,_mm_setzero_pd());
 708             d2               = _mm_mul_pd(d,d);
 709             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 710
 711             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 712
 713             /* Evaluate switch function */
 714             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 715             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 716             velec            = _mm_mul_pd(velec,sw);
 717             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 718
 719             /* Update potential sum for this i atom from the interaction with this j atom. */
 720             velec            = _mm_and_pd(velec,cutoff_mask);
 721             velecsum         = _mm_add_pd(velecsum,velec);
 722
 723             fscal            = felec;
 724
 725             fscal            = _mm_and_pd(fscal,cutoff_mask);
 726
 727             /* Update vectorial force */
 728             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 729             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 730             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 731
 732             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 733             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 734             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 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 #ifdef __XOP__
 753             eweps            = _mm_frcz_pd(ewrt);
 754 #else
 755             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 756 #endif
 757             twoeweps         = _mm_add_pd(eweps,eweps);
 758             ewitab           = _mm_slli_epi32(ewitab,2);
 759             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 760             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 761             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 762             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 763             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 764             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 765             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 766             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 767             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 768             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 769
 770             d                = _mm_sub_pd(r21,rswitch);
 771             d                = _mm_max_pd(d,_mm_setzero_pd());
 772             d2               = _mm_mul_pd(d,d);
 773             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 774
 775             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 776
 777             /* Evaluate switch function */
 778             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 779             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
 780             velec            = _mm_mul_pd(velec,sw);
 781             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
 782
 783             /* Update potential sum for this i atom from the interaction with this j atom. */
 784             velec            = _mm_and_pd(velec,cutoff_mask);
 785             velecsum         = _mm_add_pd(velecsum,velec);
 786
 787             fscal            = felec;
 788
 789             fscal            = _mm_and_pd(fscal,cutoff_mask);
 790
 791             /* Update vectorial force */
 792             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 793             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 794             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 795
 796             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 797             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 798             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 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 #ifdef __XOP__
 817             eweps            = _mm_frcz_pd(ewrt);
 818 #else
 819             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 820 #endif
 821             twoeweps         = _mm_add_pd(eweps,eweps);
 822             ewitab           = _mm_slli_epi32(ewitab,2);
 823             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 824             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 825             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 826             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 827             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 828             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 829             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 830             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 831             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 832             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 833
 834             d                = _mm_sub_pd(r22,rswitch);
 835             d                = _mm_max_pd(d,_mm_setzero_pd());
 836             d2               = _mm_mul_pd(d,d);
 837             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 838
 839             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 840
 841             /* Evaluate switch function */
 842             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 843             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
 844             velec            = _mm_mul_pd(velec,sw);
 845             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
 846
 847             /* Update potential sum for this i atom from the interaction with this j atom. */
 848             velec            = _mm_and_pd(velec,cutoff_mask);
 849             velecsum         = _mm_add_pd(velecsum,velec);
 850
 851             fscal            = felec;
 852
 853             fscal            = _mm_and_pd(fscal,cutoff_mask);
 854
 855             /* Update vectorial force */
 856             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 857             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 858             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 859
 860             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 861             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 862             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 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 612 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 #ifdef __XOP__
 966             eweps            = _mm_frcz_pd(ewrt);
 967 #else
 968             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 969 #endif
 970             twoeweps         = _mm_add_pd(eweps,eweps);
 971             ewitab           = _mm_slli_epi32(ewitab,2);
 972             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 973             ewtabD           = _mm_setzero_pd();
 974             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 975             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 976             ewtabFn          = _mm_setzero_pd();
 977             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 978             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 979             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 980             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 981             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 982
 983             d                = _mm_sub_pd(r00,rswitch);
 984             d                = _mm_max_pd(d,_mm_setzero_pd());
 985             d2               = _mm_mul_pd(d,d);
 986             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 987
 988             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 989
 990             /* Evaluate switch function */
 991             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 992             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 993             velec            = _mm_mul_pd(velec,sw);
 994             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 995
 996             /* Update potential sum for this i atom from the interaction with this j atom. */
 997             velec            = _mm_and_pd(velec,cutoff_mask);
 998             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 999             velecsum         = _mm_add_pd(velecsum,velec);
1000
1001             fscal            = felec;
1002
1003             fscal            = _mm_and_pd(fscal,cutoff_mask);
1004
1005             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1006
1007             /* Update vectorial force */
1008             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1009             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1010             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1011
1012             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1013             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1014             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
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 #ifdef __XOP__
1033             eweps            = _mm_frcz_pd(ewrt);
1034 #else
1035             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1036 #endif
1037             twoeweps         = _mm_add_pd(eweps,eweps);
1038             ewitab           = _mm_slli_epi32(ewitab,2);
1039             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1040             ewtabD           = _mm_setzero_pd();
1041             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1042             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1043             ewtabFn          = _mm_setzero_pd();
1044             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1045             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1046             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1047             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
1048             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1049
1050             d                = _mm_sub_pd(r01,rswitch);
1051             d                = _mm_max_pd(d,_mm_setzero_pd());
1052             d2               = _mm_mul_pd(d,d);
1053             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1054
1055             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1056
1057             /* Evaluate switch function */
1058             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1059             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv01,_mm_mul_pd(velec,dsw)) );
1060             velec            = _mm_mul_pd(velec,sw);
1061             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
1062
1063             /* Update potential sum for this i atom from the interaction with this j atom. */
1064             velec            = _mm_and_pd(velec,cutoff_mask);
1065             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1066             velecsum         = _mm_add_pd(velecsum,velec);
1067
1068             fscal            = felec;
1069
1070             fscal            = _mm_and_pd(fscal,cutoff_mask);
1071
1072             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1073
1074             /* Update vectorial force */
1075             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1076             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1077             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1078
1079             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1080             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1081             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
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 #ifdef __XOP__
1100             eweps            = _mm_frcz_pd(ewrt);
1101 #else
1102             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1103 #endif
1104             twoeweps         = _mm_add_pd(eweps,eweps);
1105             ewitab           = _mm_slli_epi32(ewitab,2);
1106             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1107             ewtabD           = _mm_setzero_pd();
1108             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1109             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1110             ewtabFn          = _mm_setzero_pd();
1111             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1112             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1113             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1114             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
1115             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1116
1117             d                = _mm_sub_pd(r02,rswitch);
1118             d                = _mm_max_pd(d,_mm_setzero_pd());
1119             d2               = _mm_mul_pd(d,d);
1120             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1121
1122             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1123
1124             /* Evaluate switch function */
1125             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1126             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv02,_mm_mul_pd(velec,dsw)) );
1127             velec            = _mm_mul_pd(velec,sw);
1128             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
1129
1130             /* Update potential sum for this i atom from the interaction with this j atom. */
1131             velec            = _mm_and_pd(velec,cutoff_mask);
1132             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1133             velecsum         = _mm_add_pd(velecsum,velec);
1134
1135             fscal            = felec;
1136
1137             fscal            = _mm_and_pd(fscal,cutoff_mask);
1138
1139             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1140
1141             /* Update vectorial force */
1142             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1143             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1144             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1145
1146             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1147             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1148             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
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 #ifdef __XOP__
1167             eweps            = _mm_frcz_pd(ewrt);
1168 #else
1169             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1170 #endif
1171             twoeweps         = _mm_add_pd(eweps,eweps);
1172             ewitab           = _mm_slli_epi32(ewitab,2);
1173             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1174             ewtabD           = _mm_setzero_pd();
1175             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1176             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1177             ewtabFn          = _mm_setzero_pd();
1178             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1179             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1180             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1181             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1182             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1183
1184             d                = _mm_sub_pd(r10,rswitch);
1185             d                = _mm_max_pd(d,_mm_setzero_pd());
1186             d2               = _mm_mul_pd(d,d);
1187             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1188
1189             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1190
1191             /* Evaluate switch function */
1192             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1193             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1194             velec            = _mm_mul_pd(velec,sw);
1195             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1196
1197             /* Update potential sum for this i atom from the interaction with this j atom. */
1198             velec            = _mm_and_pd(velec,cutoff_mask);
1199             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1200             velecsum         = _mm_add_pd(velecsum,velec);
1201
1202             fscal            = felec;
1203
1204             fscal            = _mm_and_pd(fscal,cutoff_mask);
1205
1206             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1207
1208             /* Update vectorial force */
1209             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1210             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1211             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1212
1213             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1214             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1215             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
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 #ifdef __XOP__
1234             eweps            = _mm_frcz_pd(ewrt);
1235 #else
1236             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1237 #endif
1238             twoeweps         = _mm_add_pd(eweps,eweps);
1239             ewitab           = _mm_slli_epi32(ewitab,2);
1240             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1241             ewtabD           = _mm_setzero_pd();
1242             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1243             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1244             ewtabFn          = _mm_setzero_pd();
1245             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1246             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1247             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1248             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
1249             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1250
1251             d                = _mm_sub_pd(r11,rswitch);
1252             d                = _mm_max_pd(d,_mm_setzero_pd());
1253             d2               = _mm_mul_pd(d,d);
1254             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1255
1256             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1257
1258             /* Evaluate switch function */
1259             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1260             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
1261             velec            = _mm_mul_pd(velec,sw);
1262             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
1263
1264             /* Update potential sum for this i atom from the interaction with this j atom. */
1265             velec            = _mm_and_pd(velec,cutoff_mask);
1266             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1267             velecsum         = _mm_add_pd(velecsum,velec);
1268
1269             fscal            = felec;
1270
1271             fscal            = _mm_and_pd(fscal,cutoff_mask);
1272
1273             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1274
1275             /* Update vectorial force */
1276             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1277             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1278             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1279
1280             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1281             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1282             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
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 #ifdef __XOP__
1301             eweps            = _mm_frcz_pd(ewrt);
1302 #else
1303             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1304 #endif
1305             twoeweps         = _mm_add_pd(eweps,eweps);
1306             ewitab           = _mm_slli_epi32(ewitab,2);
1307             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1308             ewtabD           = _mm_setzero_pd();
1309             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1310             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1311             ewtabFn          = _mm_setzero_pd();
1312             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1313             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1314             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1315             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
1316             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1317
1318             d                = _mm_sub_pd(r12,rswitch);
1319             d                = _mm_max_pd(d,_mm_setzero_pd());
1320             d2               = _mm_mul_pd(d,d);
1321             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1322
1323             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1324
1325             /* Evaluate switch function */
1326             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1327             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
1328             velec            = _mm_mul_pd(velec,sw);
1329             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
1330
1331             /* Update potential sum for this i atom from the interaction with this j atom. */
1332             velec            = _mm_and_pd(velec,cutoff_mask);
1333             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1334             velecsum         = _mm_add_pd(velecsum,velec);
1335
1336             fscal            = felec;
1337
1338             fscal            = _mm_and_pd(fscal,cutoff_mask);
1339
1340             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1341
1342             /* Update vectorial force */
1343             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1344             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1345             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1346
1347             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1348             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1349             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
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 #ifdef __XOP__
1368             eweps            = _mm_frcz_pd(ewrt);
1369 #else
1370             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1371 #endif
1372             twoeweps         = _mm_add_pd(eweps,eweps);
1373             ewitab           = _mm_slli_epi32(ewitab,2);
1374             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1375             ewtabD           = _mm_setzero_pd();
1376             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1377             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1378             ewtabFn          = _mm_setzero_pd();
1379             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1380             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1381             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1382             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1383             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1384
1385             d                = _mm_sub_pd(r20,rswitch);
1386             d                = _mm_max_pd(d,_mm_setzero_pd());
1387             d2               = _mm_mul_pd(d,d);
1388             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1389
1390             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1391
1392             /* Evaluate switch function */
1393             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1394             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1395             velec            = _mm_mul_pd(velec,sw);
1396             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1397
1398             /* Update potential sum for this i atom from the interaction with this j atom. */
1399             velec            = _mm_and_pd(velec,cutoff_mask);
1400             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1401             velecsum         = _mm_add_pd(velecsum,velec);
1402
1403             fscal            = felec;
1404
1405             fscal            = _mm_and_pd(fscal,cutoff_mask);
1406
1407             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1408
1409             /* Update vectorial force */
1410             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1411             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1412             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1413
1414             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1415             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1416             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
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 #ifdef __XOP__
1435             eweps            = _mm_frcz_pd(ewrt);
1436 #else
1437             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1438 #endif
1439             twoeweps         = _mm_add_pd(eweps,eweps);
1440             ewitab           = _mm_slli_epi32(ewitab,2);
1441             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1442             ewtabD           = _mm_setzero_pd();
1443             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1444             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1445             ewtabFn          = _mm_setzero_pd();
1446             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1447             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1448             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1449             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1450             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1451
1452             d                = _mm_sub_pd(r21,rswitch);
1453             d                = _mm_max_pd(d,_mm_setzero_pd());
1454             d2               = _mm_mul_pd(d,d);
1455             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1456
1457             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1458
1459             /* Evaluate switch function */
1460             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1461             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
1462             velec            = _mm_mul_pd(velec,sw);
1463             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
1464
1465             /* Update potential sum for this i atom from the interaction with this j atom. */
1466             velec            = _mm_and_pd(velec,cutoff_mask);
1467             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1468             velecsum         = _mm_add_pd(velecsum,velec);
1469
1470             fscal            = felec;
1471
1472             fscal            = _mm_and_pd(fscal,cutoff_mask);
1473
1474             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1475
1476             /* Update vectorial force */
1477             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1478             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1479             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1480
1481             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1482             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1483             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
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 #ifdef __XOP__
1502             eweps            = _mm_frcz_pd(ewrt);
1503 #else
1504             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1505 #endif
1506             twoeweps         = _mm_add_pd(eweps,eweps);
1507             ewitab           = _mm_slli_epi32(ewitab,2);
1508             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1509             ewtabD           = _mm_setzero_pd();
1510             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1511             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1512             ewtabFn          = _mm_setzero_pd();
1513             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1514             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1515             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1516             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1517             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1518
1519             d                = _mm_sub_pd(r22,rswitch);
1520             d                = _mm_max_pd(d,_mm_setzero_pd());
1521             d2               = _mm_mul_pd(d,d);
1522             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1523
1524             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1525
1526             /* Evaluate switch function */
1527             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1528             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
1529             velec            = _mm_mul_pd(velec,sw);
1530             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
1531
1532             /* Update potential sum for this i atom from the interaction with this j atom. */
1533             velec            = _mm_and_pd(velec,cutoff_mask);
1534             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1535             velecsum         = _mm_add_pd(velecsum,velec);
1536
1537             fscal            = felec;
1538
1539             fscal            = _mm_and_pd(fscal,cutoff_mask);
1540
1541             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1542
1543             /* Update vectorial force */
1544             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1545             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1546             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1547
1548             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1549             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1550             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
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 612 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*612);
1580 }
1581 /*
1582  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3W3_F_avx_128_fma_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_avx_128_fma_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,twoeweps,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 #ifdef __XOP__
1833             eweps            = _mm_frcz_pd(ewrt);
1834 #else
1835             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1836 #endif
1837             twoeweps         = _mm_add_pd(eweps,eweps);
1838             ewitab           = _mm_slli_epi32(ewitab,2);
1839             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1840             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1841             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1842             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1843             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1844             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1845             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1846             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1847             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
1848             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1849
1850             d                = _mm_sub_pd(r00,rswitch);
1851             d                = _mm_max_pd(d,_mm_setzero_pd());
1852             d2               = _mm_mul_pd(d,d);
1853             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1854
1855             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1856
1857             /* Evaluate switch function */
1858             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1859             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
1860             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1861
1862             fscal            = felec;
1863
1864             fscal            = _mm_and_pd(fscal,cutoff_mask);
1865
1866             /* Update vectorial force */
1867             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1868             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1869             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1870
1871             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1872             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1873             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
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 #ifdef __XOP__
1892             eweps            = _mm_frcz_pd(ewrt);
1893 #else
1894             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1895 #endif
1896             twoeweps         = _mm_add_pd(eweps,eweps);
1897             ewitab           = _mm_slli_epi32(ewitab,2);
1898             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1899             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1900             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1901             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1902             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1903             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1904             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1905             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1906             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
1907             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1908
1909             d                = _mm_sub_pd(r01,rswitch);
1910             d                = _mm_max_pd(d,_mm_setzero_pd());
1911             d2               = _mm_mul_pd(d,d);
1912             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1913
1914             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1915
1916             /* Evaluate switch function */
1917             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1918             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv01,_mm_mul_pd(velec,dsw)) );
1919             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
1920
1921             fscal            = felec;
1922
1923             fscal            = _mm_and_pd(fscal,cutoff_mask);
1924
1925             /* Update vectorial force */
1926             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1927             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1928             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1929
1930             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1931             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1932             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
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 #ifdef __XOP__
1951             eweps            = _mm_frcz_pd(ewrt);
1952 #else
1953             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1954 #endif
1955             twoeweps         = _mm_add_pd(eweps,eweps);
1956             ewitab           = _mm_slli_epi32(ewitab,2);
1957             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1958             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1959             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1960             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1961             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1962             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1963             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1964             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1965             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
1966             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1967
1968             d                = _mm_sub_pd(r02,rswitch);
1969             d                = _mm_max_pd(d,_mm_setzero_pd());
1970             d2               = _mm_mul_pd(d,d);
1971             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1972
1973             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1974
1975             /* Evaluate switch function */
1976             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1977             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv02,_mm_mul_pd(velec,dsw)) );
1978             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
1979
1980             fscal            = felec;
1981
1982             fscal            = _mm_and_pd(fscal,cutoff_mask);
1983
1984             /* Update vectorial force */
1985             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1986             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1987             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1988
1989             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1990             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1991             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
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 #ifdef __XOP__
2010             eweps            = _mm_frcz_pd(ewrt);
2011 #else
2012             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2013 #endif
2014             twoeweps         = _mm_add_pd(eweps,eweps);
2015             ewitab           = _mm_slli_epi32(ewitab,2);
2016             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2017             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2018             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2019             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2020             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2021             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2022             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2023             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2024             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
2025             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2026
2027             d                = _mm_sub_pd(r10,rswitch);
2028             d                = _mm_max_pd(d,_mm_setzero_pd());
2029             d2               = _mm_mul_pd(d,d);
2030             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2031
2032             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2033
2034             /* Evaluate switch function */
2035             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2036             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
2037             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
2038
2039             fscal            = felec;
2040
2041             fscal            = _mm_and_pd(fscal,cutoff_mask);
2042
2043             /* Update vectorial force */
2044             fix1             = _mm_macc_pd(dx10,fscal,fix1);
2045             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
2046             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
2047
2048             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
2049             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
2050             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
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 #ifdef __XOP__
2069             eweps            = _mm_frcz_pd(ewrt);
2070 #else
2071             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2072 #endif
2073             twoeweps         = _mm_add_pd(eweps,eweps);
2074             ewitab           = _mm_slli_epi32(ewitab,2);
2075             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2076             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2077             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2078             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2079             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2080             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2081             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2082             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2083             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
2084             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2085
2086             d                = _mm_sub_pd(r11,rswitch);
2087             d                = _mm_max_pd(d,_mm_setzero_pd());
2088             d2               = _mm_mul_pd(d,d);
2089             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2090
2091             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2092
2093             /* Evaluate switch function */
2094             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2095             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
2096             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
2097
2098             fscal            = felec;
2099
2100             fscal            = _mm_and_pd(fscal,cutoff_mask);
2101
2102             /* Update vectorial force */
2103             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2104             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2105             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2106
2107             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2108             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2109             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
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 #ifdef __XOP__
2128             eweps            = _mm_frcz_pd(ewrt);
2129 #else
2130             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2131 #endif
2132             twoeweps         = _mm_add_pd(eweps,eweps);
2133             ewitab           = _mm_slli_epi32(ewitab,2);
2134             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2135             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2136             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2137             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2138             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2139             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2140             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2141             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2142             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
2143             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2144
2145             d                = _mm_sub_pd(r12,rswitch);
2146             d                = _mm_max_pd(d,_mm_setzero_pd());
2147             d2               = _mm_mul_pd(d,d);
2148             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2149
2150             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2151
2152             /* Evaluate switch function */
2153             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2154             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
2155             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
2156
2157             fscal            = felec;
2158
2159             fscal            = _mm_and_pd(fscal,cutoff_mask);
2160
2161             /* Update vectorial force */
2162             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2163             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2164             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2165
2166             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2167             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2168             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
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 #ifdef __XOP__
2187             eweps            = _mm_frcz_pd(ewrt);
2188 #else
2189             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2190 #endif
2191             twoeweps         = _mm_add_pd(eweps,eweps);
2192             ewitab           = _mm_slli_epi32(ewitab,2);
2193             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2194             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2195             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2196             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2197             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2198             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2199             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2200             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2201             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
2202             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2203
2204             d                = _mm_sub_pd(r20,rswitch);
2205             d                = _mm_max_pd(d,_mm_setzero_pd());
2206             d2               = _mm_mul_pd(d,d);
2207             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2208
2209             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2210
2211             /* Evaluate switch function */
2212             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2213             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
2214             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
2215
2216             fscal            = felec;
2217
2218             fscal            = _mm_and_pd(fscal,cutoff_mask);
2219
2220             /* Update vectorial force */
2221             fix2             = _mm_macc_pd(dx20,fscal,fix2);
2222             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
2223             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
2224
2225             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2226             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2227             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
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 #ifdef __XOP__
2246             eweps            = _mm_frcz_pd(ewrt);
2247 #else
2248             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2249 #endif
2250             twoeweps         = _mm_add_pd(eweps,eweps);
2251             ewitab           = _mm_slli_epi32(ewitab,2);
2252             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2253             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2254             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2255             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2256             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2257             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2258             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2259             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2260             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
2261             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2262
2263             d                = _mm_sub_pd(r21,rswitch);
2264             d                = _mm_max_pd(d,_mm_setzero_pd());
2265             d2               = _mm_mul_pd(d,d);
2266             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2267
2268             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2269
2270             /* Evaluate switch function */
2271             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2272             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
2273             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
2274
2275             fscal            = felec;
2276
2277             fscal            = _mm_and_pd(fscal,cutoff_mask);
2278
2279             /* Update vectorial force */
2280             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2281             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2282             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2283
2284             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2285             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2286             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
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 #ifdef __XOP__
2305             eweps            = _mm_frcz_pd(ewrt);
2306 #else
2307             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2308 #endif
2309             twoeweps         = _mm_add_pd(eweps,eweps);
2310             ewitab           = _mm_slli_epi32(ewitab,2);
2311             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2312             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
2313             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2314             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2315             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
2316             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2317             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2318             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2319             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
2320             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2321
2322             d                = _mm_sub_pd(r22,rswitch);
2323             d                = _mm_max_pd(d,_mm_setzero_pd());
2324             d2               = _mm_mul_pd(d,d);
2325             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2326
2327             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2328
2329             /* Evaluate switch function */
2330             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2331             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
2332             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
2333
2334             fscal            = felec;
2335
2336             fscal            = _mm_and_pd(fscal,cutoff_mask);
2337
2338             /* Update vectorial force */
2339             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2340             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2341             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2342
2343             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2344             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2345             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
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 585 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 #ifdef __XOP__
2449             eweps            = _mm_frcz_pd(ewrt);
2450 #else
2451             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2452 #endif
2453             twoeweps         = _mm_add_pd(eweps,eweps);
2454             ewitab           = _mm_slli_epi32(ewitab,2);
2455             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2456             ewtabD           = _mm_setzero_pd();
2457             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2458             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2459             ewtabFn          = _mm_setzero_pd();
2460             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2461             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2462             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2463             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
2464             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
2465
2466             d                = _mm_sub_pd(r00,rswitch);
2467             d                = _mm_max_pd(d,_mm_setzero_pd());
2468             d2               = _mm_mul_pd(d,d);
2469             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2470
2471             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2472
2473             /* Evaluate switch function */
2474             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2475             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
2476             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
2477
2478             fscal            = felec;
2479
2480             fscal            = _mm_and_pd(fscal,cutoff_mask);
2481
2482             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2483
2484             /* Update vectorial force */
2485             fix0             = _mm_macc_pd(dx00,fscal,fix0);
2486             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
2487             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
2488
2489             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
2490             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
2491             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
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 #ifdef __XOP__
2510             eweps            = _mm_frcz_pd(ewrt);
2511 #else
2512             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2513 #endif
2514             twoeweps         = _mm_add_pd(eweps,eweps);
2515             ewitab           = _mm_slli_epi32(ewitab,2);
2516             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2517             ewtabD           = _mm_setzero_pd();
2518             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2519             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2520             ewtabFn          = _mm_setzero_pd();
2521             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2522             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2523             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2524             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
2525             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
2526
2527             d                = _mm_sub_pd(r01,rswitch);
2528             d                = _mm_max_pd(d,_mm_setzero_pd());
2529             d2               = _mm_mul_pd(d,d);
2530             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2531
2532             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2533
2534             /* Evaluate switch function */
2535             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2536             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv01,_mm_mul_pd(velec,dsw)) );
2537             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
2538
2539             fscal            = felec;
2540
2541             fscal            = _mm_and_pd(fscal,cutoff_mask);
2542
2543             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2544
2545             /* Update vectorial force */
2546             fix0             = _mm_macc_pd(dx01,fscal,fix0);
2547             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
2548             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
2549
2550             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
2551             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
2552             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
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 #ifdef __XOP__
2571             eweps            = _mm_frcz_pd(ewrt);
2572 #else
2573             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2574 #endif
2575             twoeweps         = _mm_add_pd(eweps,eweps);
2576             ewitab           = _mm_slli_epi32(ewitab,2);
2577             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2578             ewtabD           = _mm_setzero_pd();
2579             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2580             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2581             ewtabFn          = _mm_setzero_pd();
2582             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2583             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2584             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2585             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
2586             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
2587
2588             d                = _mm_sub_pd(r02,rswitch);
2589             d                = _mm_max_pd(d,_mm_setzero_pd());
2590             d2               = _mm_mul_pd(d,d);
2591             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2592
2593             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2594
2595             /* Evaluate switch function */
2596             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2597             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv02,_mm_mul_pd(velec,dsw)) );
2598             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
2599
2600             fscal            = felec;
2601
2602             fscal            = _mm_and_pd(fscal,cutoff_mask);
2603
2604             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2605
2606             /* Update vectorial force */
2607             fix0             = _mm_macc_pd(dx02,fscal,fix0);
2608             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
2609             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
2610
2611             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
2612             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
2613             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
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 #ifdef __XOP__
2632             eweps            = _mm_frcz_pd(ewrt);
2633 #else
2634             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2635 #endif
2636             twoeweps         = _mm_add_pd(eweps,eweps);
2637             ewitab           = _mm_slli_epi32(ewitab,2);
2638             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2639             ewtabD           = _mm_setzero_pd();
2640             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2641             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2642             ewtabFn          = _mm_setzero_pd();
2643             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2644             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2645             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2646             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
2647             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2648
2649             d                = _mm_sub_pd(r10,rswitch);
2650             d                = _mm_max_pd(d,_mm_setzero_pd());
2651             d2               = _mm_mul_pd(d,d);
2652             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2653
2654             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2655
2656             /* Evaluate switch function */
2657             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2658             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
2659             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
2660
2661             fscal            = felec;
2662
2663             fscal            = _mm_and_pd(fscal,cutoff_mask);
2664
2665             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2666
2667             /* Update vectorial force */
2668             fix1             = _mm_macc_pd(dx10,fscal,fix1);
2669             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
2670             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
2671
2672             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
2673             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
2674             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
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 #ifdef __XOP__
2693             eweps            = _mm_frcz_pd(ewrt);
2694 #else
2695             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2696 #endif
2697             twoeweps         = _mm_add_pd(eweps,eweps);
2698             ewitab           = _mm_slli_epi32(ewitab,2);
2699             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2700             ewtabD           = _mm_setzero_pd();
2701             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2702             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2703             ewtabFn          = _mm_setzero_pd();
2704             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2705             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2706             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2707             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
2708             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2709
2710             d                = _mm_sub_pd(r11,rswitch);
2711             d                = _mm_max_pd(d,_mm_setzero_pd());
2712             d2               = _mm_mul_pd(d,d);
2713             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2714
2715             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2716
2717             /* Evaluate switch function */
2718             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2719             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv11,_mm_mul_pd(velec,dsw)) );
2720             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
2721
2722             fscal            = felec;
2723
2724             fscal            = _mm_and_pd(fscal,cutoff_mask);
2725
2726             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2727
2728             /* Update vectorial force */
2729             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2730             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2731             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2732
2733             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2734             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2735             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
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 #ifdef __XOP__
2754             eweps            = _mm_frcz_pd(ewrt);
2755 #else
2756             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2757 #endif
2758             twoeweps         = _mm_add_pd(eweps,eweps);
2759             ewitab           = _mm_slli_epi32(ewitab,2);
2760             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2761             ewtabD           = _mm_setzero_pd();
2762             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2763             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2764             ewtabFn          = _mm_setzero_pd();
2765             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2766             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2767             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2768             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
2769             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2770
2771             d                = _mm_sub_pd(r12,rswitch);
2772             d                = _mm_max_pd(d,_mm_setzero_pd());
2773             d2               = _mm_mul_pd(d,d);
2774             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2775
2776             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2777
2778             /* Evaluate switch function */
2779             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2780             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv12,_mm_mul_pd(velec,dsw)) );
2781             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
2782
2783             fscal            = felec;
2784
2785             fscal            = _mm_and_pd(fscal,cutoff_mask);
2786
2787             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2788
2789             /* Update vectorial force */
2790             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2791             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2792             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2793
2794             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2795             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2796             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
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 #ifdef __XOP__
2815             eweps            = _mm_frcz_pd(ewrt);
2816 #else
2817             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2818 #endif
2819             twoeweps         = _mm_add_pd(eweps,eweps);
2820             ewitab           = _mm_slli_epi32(ewitab,2);
2821             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2822             ewtabD           = _mm_setzero_pd();
2823             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2824             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2825             ewtabFn          = _mm_setzero_pd();
2826             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2827             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2828             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2829             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
2830             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2831
2832             d                = _mm_sub_pd(r20,rswitch);
2833             d                = _mm_max_pd(d,_mm_setzero_pd());
2834             d2               = _mm_mul_pd(d,d);
2835             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2836
2837             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2838
2839             /* Evaluate switch function */
2840             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2841             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
2842             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
2843
2844             fscal            = felec;
2845
2846             fscal            = _mm_and_pd(fscal,cutoff_mask);
2847
2848             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2849
2850             /* Update vectorial force */
2851             fix2             = _mm_macc_pd(dx20,fscal,fix2);
2852             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
2853             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
2854
2855             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2856             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2857             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
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 #ifdef __XOP__
2876             eweps            = _mm_frcz_pd(ewrt);
2877 #else
2878             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2879 #endif
2880             twoeweps         = _mm_add_pd(eweps,eweps);
2881             ewitab           = _mm_slli_epi32(ewitab,2);
2882             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2883             ewtabD           = _mm_setzero_pd();
2884             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2885             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2886             ewtabFn          = _mm_setzero_pd();
2887             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2888             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2889             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2890             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
2891             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2892
2893             d                = _mm_sub_pd(r21,rswitch);
2894             d                = _mm_max_pd(d,_mm_setzero_pd());
2895             d2               = _mm_mul_pd(d,d);
2896             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2897
2898             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2899
2900             /* Evaluate switch function */
2901             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2902             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv21,_mm_mul_pd(velec,dsw)) );
2903             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
2904
2905             fscal            = felec;
2906
2907             fscal            = _mm_and_pd(fscal,cutoff_mask);
2908
2909             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2910
2911             /* Update vectorial force */
2912             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2913             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2914             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2915
2916             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2917             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2918             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
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 #ifdef __XOP__
2937             eweps            = _mm_frcz_pd(ewrt);
2938 #else
2939             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2940 #endif
2941             twoeweps         = _mm_add_pd(eweps,eweps);
2942             ewitab           = _mm_slli_epi32(ewitab,2);
2943             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
2944             ewtabD           = _mm_setzero_pd();
2945             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
2946             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
2947             ewtabFn          = _mm_setzero_pd();
2948             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
2949             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
2950             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
2951             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
2952             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2953
2954             d                = _mm_sub_pd(r22,rswitch);
2955             d                = _mm_max_pd(d,_mm_setzero_pd());
2956             d2               = _mm_mul_pd(d,d);
2957             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
2958
2959             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
2960
2961             /* Evaluate switch function */
2962             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2963             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv22,_mm_mul_pd(velec,dsw)) );
2964             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
2965
2966             fscal            = felec;
2967
2968             fscal            = _mm_and_pd(fscal,cutoff_mask);
2969
2970             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2971
2972             /* Update vectorial force */
2973             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2974             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2975             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2976
2977             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2978             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2979             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
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 585 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*585);
3005 }