src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwNone_GeomW4W4_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_ElecCSTab_VdwNone_GeomW4W4_VF_avx_128_fma_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Water4
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_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              vdwioffset1;
  81     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  82     int              vdwioffset2;
  83     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  84     int              vdwioffset3;
  85     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  86     int              vdwjidx1A,vdwjidx1B;
  87     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  88     int              vdwjidx2A,vdwjidx2B;
  89     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  90     int              vdwjidx3A,vdwjidx3B;
  91     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  92     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  93     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  94     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  95     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  96     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  97     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  98     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  99     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 100     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 101     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 102     real             *charge;
 103     __m128i          vfitab;
 104     __m128i          ifour       = _mm_set1_epi32(4);
 105     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 106     real             *vftab;
 107     __m128d          dummy_mask,cutoff_mask;
 108     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 109     __m128d          one     = _mm_set1_pd(1.0);
 110     __m128d          two     = _mm_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_pd(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124
 125     vftab            = kernel_data->table_elec->data;
 126     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 131     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 132     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 133
 134     jq1              = _mm_set1_pd(charge[inr+1]);
 135     jq2              = _mm_set1_pd(charge[inr+2]);
 136     jq3              = _mm_set1_pd(charge[inr+3]);
 137     qq11             = _mm_mul_pd(iq1,jq1);
 138     qq12             = _mm_mul_pd(iq1,jq2);
 139     qq13             = _mm_mul_pd(iq1,jq3);
 140     qq21             = _mm_mul_pd(iq2,jq1);
 141     qq22             = _mm_mul_pd(iq2,jq2);
 142     qq23             = _mm_mul_pd(iq2,jq3);
 143     qq31             = _mm_mul_pd(iq3,jq1);
 144     qq32             = _mm_mul_pd(iq3,jq2);
 145     qq33             = _mm_mul_pd(iq3,jq3);
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 171                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 172
 173         fix1             = _mm_setzero_pd();
 174         fiy1             = _mm_setzero_pd();
 175         fiz1             = _mm_setzero_pd();
 176         fix2             = _mm_setzero_pd();
 177         fiy2             = _mm_setzero_pd();
 178         fiz2             = _mm_setzero_pd();
 179         fix3             = _mm_setzero_pd();
 180         fiy3             = _mm_setzero_pd();
 181         fiz3             = _mm_setzero_pd();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm_setzero_pd();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 198                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 199
 200             /* Calculate displacement vector */
 201             dx11             = _mm_sub_pd(ix1,jx1);
 202             dy11             = _mm_sub_pd(iy1,jy1);
 203             dz11             = _mm_sub_pd(iz1,jz1);
 204             dx12             = _mm_sub_pd(ix1,jx2);
 205             dy12             = _mm_sub_pd(iy1,jy2);
 206             dz12             = _mm_sub_pd(iz1,jz2);
 207             dx13             = _mm_sub_pd(ix1,jx3);
 208             dy13             = _mm_sub_pd(iy1,jy3);
 209             dz13             = _mm_sub_pd(iz1,jz3);
 210             dx21             = _mm_sub_pd(ix2,jx1);
 211             dy21             = _mm_sub_pd(iy2,jy1);
 212             dz21             = _mm_sub_pd(iz2,jz1);
 213             dx22             = _mm_sub_pd(ix2,jx2);
 214             dy22             = _mm_sub_pd(iy2,jy2);
 215             dz22             = _mm_sub_pd(iz2,jz2);
 216             dx23             = _mm_sub_pd(ix2,jx3);
 217             dy23             = _mm_sub_pd(iy2,jy3);
 218             dz23             = _mm_sub_pd(iz2,jz3);
 219             dx31             = _mm_sub_pd(ix3,jx1);
 220             dy31             = _mm_sub_pd(iy3,jy1);
 221             dz31             = _mm_sub_pd(iz3,jz1);
 222             dx32             = _mm_sub_pd(ix3,jx2);
 223             dy32             = _mm_sub_pd(iy3,jy2);
 224             dz32             = _mm_sub_pd(iz3,jz2);
 225             dx33             = _mm_sub_pd(ix3,jx3);
 226             dy33             = _mm_sub_pd(iy3,jy3);
 227             dz33             = _mm_sub_pd(iz3,jz3);
 228
 229             /* Calculate squared distance and things based on it */
 230             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 231             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 232             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 233             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 234             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 235             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 236             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 237             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 238             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 239
 240             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 241             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 242             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 243             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 244             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 245             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 246             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 247             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 248             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 249
 250             fjx1             = _mm_setzero_pd();
 251             fjy1             = _mm_setzero_pd();
 252             fjz1             = _mm_setzero_pd();
 253             fjx2             = _mm_setzero_pd();
 254             fjy2             = _mm_setzero_pd();
 255             fjz2             = _mm_setzero_pd();
 256             fjx3             = _mm_setzero_pd();
 257             fjy3             = _mm_setzero_pd();
 258             fjz3             = _mm_setzero_pd();
 259
 260             /**************************
 261              * CALCULATE INTERACTIONS *
 262              **************************/
 263
 264             r11              = _mm_mul_pd(rsq11,rinv11);
 265
 266             /* Calculate table index by multiplying r with table scale and truncate to integer */
 267             rt               = _mm_mul_pd(r11,vftabscale);
 268             vfitab           = _mm_cvttpd_epi32(rt);
 269 #ifdef __XOP__
 270             vfeps            = _mm_frcz_pd(rt);
 271 #else
 272             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 273 #endif
 274             twovfeps         = _mm_add_pd(vfeps,vfeps);
 275             vfitab           = _mm_slli_epi32(vfitab,2);
 276
 277             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 278             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 279             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 280             GMX_MM_TRANSPOSE2_PD(Y,F);
 281             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 282             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 283             GMX_MM_TRANSPOSE2_PD(G,H);
 284             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 285             VV               = _mm_macc_pd(vfeps,Fp,Y);
 286             velec            = _mm_mul_pd(qq11,VV);
 287             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 288             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             velecsum         = _mm_add_pd(velecsum,velec);
 292
 293             fscal            = felec;
 294
 295             /* Update vectorial force */
 296             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 297             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 298             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 299
 300             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 301             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 302             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             r12              = _mm_mul_pd(rsq12,rinv12);
 309
 310             /* Calculate table index by multiplying r with table scale and truncate to integer */
 311             rt               = _mm_mul_pd(r12,vftabscale);
 312             vfitab           = _mm_cvttpd_epi32(rt);
 313 #ifdef __XOP__
 314             vfeps            = _mm_frcz_pd(rt);
 315 #else
 316             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 317 #endif
 318             twovfeps         = _mm_add_pd(vfeps,vfeps);
 319             vfitab           = _mm_slli_epi32(vfitab,2);
 320
 321             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 322             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 323             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 324             GMX_MM_TRANSPOSE2_PD(Y,F);
 325             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 326             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 327             GMX_MM_TRANSPOSE2_PD(G,H);
 328             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 329             VV               = _mm_macc_pd(vfeps,Fp,Y);
 330             velec            = _mm_mul_pd(qq12,VV);
 331             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 332             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velecsum         = _mm_add_pd(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             /* Update vectorial force */
 340             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 341             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 342             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 343
 344             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 345             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 346             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             r13              = _mm_mul_pd(rsq13,rinv13);
 353
 354             /* Calculate table index by multiplying r with table scale and truncate to integer */
 355             rt               = _mm_mul_pd(r13,vftabscale);
 356             vfitab           = _mm_cvttpd_epi32(rt);
 357 #ifdef __XOP__
 358             vfeps            = _mm_frcz_pd(rt);
 359 #else
 360             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 361 #endif
 362             twovfeps         = _mm_add_pd(vfeps,vfeps);
 363             vfitab           = _mm_slli_epi32(vfitab,2);
 364
 365             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 366             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 367             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 368             GMX_MM_TRANSPOSE2_PD(Y,F);
 369             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 370             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 371             GMX_MM_TRANSPOSE2_PD(G,H);
 372             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 373             VV               = _mm_macc_pd(vfeps,Fp,Y);
 374             velec            = _mm_mul_pd(qq13,VV);
 375             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 376             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq13,FF),_mm_mul_pd(vftabscale,rinv13)));
 377
 378             /* Update potential sum for this i atom from the interaction with this j atom. */
 379             velecsum         = _mm_add_pd(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             /* Update vectorial force */
 384             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 385             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 386             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 387
 388             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 389             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 390             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             r21              = _mm_mul_pd(rsq21,rinv21);
 397
 398             /* Calculate table index by multiplying r with table scale and truncate to integer */
 399             rt               = _mm_mul_pd(r21,vftabscale);
 400             vfitab           = _mm_cvttpd_epi32(rt);
 401 #ifdef __XOP__
 402             vfeps            = _mm_frcz_pd(rt);
 403 #else
 404             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 405 #endif
 406             twovfeps         = _mm_add_pd(vfeps,vfeps);
 407             vfitab           = _mm_slli_epi32(vfitab,2);
 408
 409             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 410             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 411             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 412             GMX_MM_TRANSPOSE2_PD(Y,F);
 413             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 414             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 415             GMX_MM_TRANSPOSE2_PD(G,H);
 416             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 417             VV               = _mm_macc_pd(vfeps,Fp,Y);
 418             velec            = _mm_mul_pd(qq21,VV);
 419             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 420             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velecsum         = _mm_add_pd(velecsum,velec);
 424
 425             fscal            = felec;
 426
 427             /* Update vectorial force */
 428             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 429             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 430             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 431
 432             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 433             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 434             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             r22              = _mm_mul_pd(rsq22,rinv22);
 441
 442             /* Calculate table index by multiplying r with table scale and truncate to integer */
 443             rt               = _mm_mul_pd(r22,vftabscale);
 444             vfitab           = _mm_cvttpd_epi32(rt);
 445 #ifdef __XOP__
 446             vfeps            = _mm_frcz_pd(rt);
 447 #else
 448             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 449 #endif
 450             twovfeps         = _mm_add_pd(vfeps,vfeps);
 451             vfitab           = _mm_slli_epi32(vfitab,2);
 452
 453             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 454             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 455             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 456             GMX_MM_TRANSPOSE2_PD(Y,F);
 457             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 458             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 459             GMX_MM_TRANSPOSE2_PD(G,H);
 460             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 461             VV               = _mm_macc_pd(vfeps,Fp,Y);
 462             velec            = _mm_mul_pd(qq22,VV);
 463             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 464             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
 465
 466             /* Update potential sum for this i atom from the interaction with this j atom. */
 467             velecsum         = _mm_add_pd(velecsum,velec);
 468
 469             fscal            = felec;
 470
 471             /* Update vectorial force */
 472             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 473             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 474             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 475
 476             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 477             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 478             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r23              = _mm_mul_pd(rsq23,rinv23);
 485
 486             /* Calculate table index by multiplying r with table scale and truncate to integer */
 487             rt               = _mm_mul_pd(r23,vftabscale);
 488             vfitab           = _mm_cvttpd_epi32(rt);
 489 #ifdef __XOP__
 490             vfeps            = _mm_frcz_pd(rt);
 491 #else
 492             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 493 #endif
 494             twovfeps         = _mm_add_pd(vfeps,vfeps);
 495             vfitab           = _mm_slli_epi32(vfitab,2);
 496
 497             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 498             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 499             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 500             GMX_MM_TRANSPOSE2_PD(Y,F);
 501             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 502             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 503             GMX_MM_TRANSPOSE2_PD(G,H);
 504             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 505             VV               = _mm_macc_pd(vfeps,Fp,Y);
 506             velec            = _mm_mul_pd(qq23,VV);
 507             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 508             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq23,FF),_mm_mul_pd(vftabscale,rinv23)));
 509
 510             /* Update potential sum for this i atom from the interaction with this j atom. */
 511             velecsum         = _mm_add_pd(velecsum,velec);
 512
 513             fscal            = felec;
 514
 515             /* Update vectorial force */
 516             fix2             = _mm_macc_pd(dx23,fscal,fix2);
 517             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
 518             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
 519
 520             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
 521             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
 522             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             r31              = _mm_mul_pd(rsq31,rinv31);
 529
 530             /* Calculate table index by multiplying r with table scale and truncate to integer */
 531             rt               = _mm_mul_pd(r31,vftabscale);
 532             vfitab           = _mm_cvttpd_epi32(rt);
 533 #ifdef __XOP__
 534             vfeps            = _mm_frcz_pd(rt);
 535 #else
 536             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 537 #endif
 538             twovfeps         = _mm_add_pd(vfeps,vfeps);
 539             vfitab           = _mm_slli_epi32(vfitab,2);
 540
 541             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 542             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 543             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 544             GMX_MM_TRANSPOSE2_PD(Y,F);
 545             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 546             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 547             GMX_MM_TRANSPOSE2_PD(G,H);
 548             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 549             VV               = _mm_macc_pd(vfeps,Fp,Y);
 550             velec            = _mm_mul_pd(qq31,VV);
 551             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 552             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq31,FF),_mm_mul_pd(vftabscale,rinv31)));
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velecsum         = _mm_add_pd(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             /* Update vectorial force */
 560             fix3             = _mm_macc_pd(dx31,fscal,fix3);
 561             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
 562             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
 563
 564             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
 565             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
 566             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             r32              = _mm_mul_pd(rsq32,rinv32);
 573
 574             /* Calculate table index by multiplying r with table scale and truncate to integer */
 575             rt               = _mm_mul_pd(r32,vftabscale);
 576             vfitab           = _mm_cvttpd_epi32(rt);
 577 #ifdef __XOP__
 578             vfeps            = _mm_frcz_pd(rt);
 579 #else
 580             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 581 #endif
 582             twovfeps         = _mm_add_pd(vfeps,vfeps);
 583             vfitab           = _mm_slli_epi32(vfitab,2);
 584
 585             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 586             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 587             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 588             GMX_MM_TRANSPOSE2_PD(Y,F);
 589             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 590             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 591             GMX_MM_TRANSPOSE2_PD(G,H);
 592             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 593             VV               = _mm_macc_pd(vfeps,Fp,Y);
 594             velec            = _mm_mul_pd(qq32,VV);
 595             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 596             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq32,FF),_mm_mul_pd(vftabscale,rinv32)));
 597
 598             /* Update potential sum for this i atom from the interaction with this j atom. */
 599             velecsum         = _mm_add_pd(velecsum,velec);
 600
 601             fscal            = felec;
 602
 603             /* Update vectorial force */
 604             fix3             = _mm_macc_pd(dx32,fscal,fix3);
 605             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
 606             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
 607
 608             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
 609             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
 610             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             r33              = _mm_mul_pd(rsq33,rinv33);
 617
 618             /* Calculate table index by multiplying r with table scale and truncate to integer */
 619             rt               = _mm_mul_pd(r33,vftabscale);
 620             vfitab           = _mm_cvttpd_epi32(rt);
 621 #ifdef __XOP__
 622             vfeps            = _mm_frcz_pd(rt);
 623 #else
 624             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 625 #endif
 626             twovfeps         = _mm_add_pd(vfeps,vfeps);
 627             vfitab           = _mm_slli_epi32(vfitab,2);
 628
 629             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 630             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 631             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 632             GMX_MM_TRANSPOSE2_PD(Y,F);
 633             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 634             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 635             GMX_MM_TRANSPOSE2_PD(G,H);
 636             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 637             VV               = _mm_macc_pd(vfeps,Fp,Y);
 638             velec            = _mm_mul_pd(qq33,VV);
 639             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 640             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq33,FF),_mm_mul_pd(vftabscale,rinv33)));
 641
 642             /* Update potential sum for this i atom from the interaction with this j atom. */
 643             velecsum         = _mm_add_pd(velecsum,velec);
 644
 645             fscal            = felec;
 646
 647             /* Update vectorial force */
 648             fix3             = _mm_macc_pd(dx33,fscal,fix3);
 649             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
 650             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
 651
 652             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
 653             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
 654             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
 655
 656             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA+DIM,f+j_coord_offsetB+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 657
 658             /* Inner loop uses 414 flops */
 659         }
 660
 661         if(jidx<j_index_end)
 662         {
 663
 664             jnrA             = jjnr[jidx];
 665             j_coord_offsetA  = DIM*jnrA;
 666
 667             /* load j atom coordinates */
 668             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
 669                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 670
 671             /* Calculate displacement vector */
 672             dx11             = _mm_sub_pd(ix1,jx1);
 673             dy11             = _mm_sub_pd(iy1,jy1);
 674             dz11             = _mm_sub_pd(iz1,jz1);
 675             dx12             = _mm_sub_pd(ix1,jx2);
 676             dy12             = _mm_sub_pd(iy1,jy2);
 677             dz12             = _mm_sub_pd(iz1,jz2);
 678             dx13             = _mm_sub_pd(ix1,jx3);
 679             dy13             = _mm_sub_pd(iy1,jy3);
 680             dz13             = _mm_sub_pd(iz1,jz3);
 681             dx21             = _mm_sub_pd(ix2,jx1);
 682             dy21             = _mm_sub_pd(iy2,jy1);
 683             dz21             = _mm_sub_pd(iz2,jz1);
 684             dx22             = _mm_sub_pd(ix2,jx2);
 685             dy22             = _mm_sub_pd(iy2,jy2);
 686             dz22             = _mm_sub_pd(iz2,jz2);
 687             dx23             = _mm_sub_pd(ix2,jx3);
 688             dy23             = _mm_sub_pd(iy2,jy3);
 689             dz23             = _mm_sub_pd(iz2,jz3);
 690             dx31             = _mm_sub_pd(ix3,jx1);
 691             dy31             = _mm_sub_pd(iy3,jy1);
 692             dz31             = _mm_sub_pd(iz3,jz1);
 693             dx32             = _mm_sub_pd(ix3,jx2);
 694             dy32             = _mm_sub_pd(iy3,jy2);
 695             dz32             = _mm_sub_pd(iz3,jz2);
 696             dx33             = _mm_sub_pd(ix3,jx3);
 697             dy33             = _mm_sub_pd(iy3,jy3);
 698             dz33             = _mm_sub_pd(iz3,jz3);
 699
 700             /* Calculate squared distance and things based on it */
 701             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 702             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 703             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 704             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 705             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 706             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 707             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 708             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 709             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 710
 711             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 712             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 713             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 714             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 715             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 716             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 717             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 718             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 719             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 720
 721             fjx1             = _mm_setzero_pd();
 722             fjy1             = _mm_setzero_pd();
 723             fjz1             = _mm_setzero_pd();
 724             fjx2             = _mm_setzero_pd();
 725             fjy2             = _mm_setzero_pd();
 726             fjz2             = _mm_setzero_pd();
 727             fjx3             = _mm_setzero_pd();
 728             fjy3             = _mm_setzero_pd();
 729             fjz3             = _mm_setzero_pd();
 730
 731             /**************************
 732              * CALCULATE INTERACTIONS *
 733              **************************/
 734
 735             r11              = _mm_mul_pd(rsq11,rinv11);
 736
 737             /* Calculate table index by multiplying r with table scale and truncate to integer */
 738             rt               = _mm_mul_pd(r11,vftabscale);
 739             vfitab           = _mm_cvttpd_epi32(rt);
 740 #ifdef __XOP__
 741             vfeps            = _mm_frcz_pd(rt);
 742 #else
 743             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 744 #endif
 745             twovfeps         = _mm_add_pd(vfeps,vfeps);
 746             vfitab           = _mm_slli_epi32(vfitab,2);
 747
 748             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 749             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 750             F                = _mm_setzero_pd();
 751             GMX_MM_TRANSPOSE2_PD(Y,F);
 752             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 753             H                = _mm_setzero_pd();
 754             GMX_MM_TRANSPOSE2_PD(G,H);
 755             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 756             VV               = _mm_macc_pd(vfeps,Fp,Y);
 757             velec            = _mm_mul_pd(qq11,VV);
 758             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 759             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
 760
 761             /* Update potential sum for this i atom from the interaction with this j atom. */
 762             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 763             velecsum         = _mm_add_pd(velecsum,velec);
 764
 765             fscal            = felec;
 766
 767             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 768
 769             /* Update vectorial force */
 770             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 771             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 772             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 773
 774             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 775             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 776             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 777
 778             /**************************
 779              * CALCULATE INTERACTIONS *
 780              **************************/
 781
 782             r12              = _mm_mul_pd(rsq12,rinv12);
 783
 784             /* Calculate table index by multiplying r with table scale and truncate to integer */
 785             rt               = _mm_mul_pd(r12,vftabscale);
 786             vfitab           = _mm_cvttpd_epi32(rt);
 787 #ifdef __XOP__
 788             vfeps            = _mm_frcz_pd(rt);
 789 #else
 790             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 791 #endif
 792             twovfeps         = _mm_add_pd(vfeps,vfeps);
 793             vfitab           = _mm_slli_epi32(vfitab,2);
 794
 795             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 796             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 797             F                = _mm_setzero_pd();
 798             GMX_MM_TRANSPOSE2_PD(Y,F);
 799             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 800             H                = _mm_setzero_pd();
 801             GMX_MM_TRANSPOSE2_PD(G,H);
 802             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 803             VV               = _mm_macc_pd(vfeps,Fp,Y);
 804             velec            = _mm_mul_pd(qq12,VV);
 805             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 806             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
 807
 808             /* Update potential sum for this i atom from the interaction with this j atom. */
 809             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 810             velecsum         = _mm_add_pd(velecsum,velec);
 811
 812             fscal            = felec;
 813
 814             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 815
 816             /* Update vectorial force */
 817             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 818             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 819             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 820
 821             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 822             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 823             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 824
 825             /**************************
 826              * CALCULATE INTERACTIONS *
 827              **************************/
 828
 829             r13              = _mm_mul_pd(rsq13,rinv13);
 830
 831             /* Calculate table index by multiplying r with table scale and truncate to integer */
 832             rt               = _mm_mul_pd(r13,vftabscale);
 833             vfitab           = _mm_cvttpd_epi32(rt);
 834 #ifdef __XOP__
 835             vfeps            = _mm_frcz_pd(rt);
 836 #else
 837             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 838 #endif
 839             twovfeps         = _mm_add_pd(vfeps,vfeps);
 840             vfitab           = _mm_slli_epi32(vfitab,2);
 841
 842             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 843             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 844             F                = _mm_setzero_pd();
 845             GMX_MM_TRANSPOSE2_PD(Y,F);
 846             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 847             H                = _mm_setzero_pd();
 848             GMX_MM_TRANSPOSE2_PD(G,H);
 849             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 850             VV               = _mm_macc_pd(vfeps,Fp,Y);
 851             velec            = _mm_mul_pd(qq13,VV);
 852             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 853             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq13,FF),_mm_mul_pd(vftabscale,rinv13)));
 854
 855             /* Update potential sum for this i atom from the interaction with this j atom. */
 856             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 857             velecsum         = _mm_add_pd(velecsum,velec);
 858
 859             fscal            = felec;
 860
 861             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 862
 863             /* Update vectorial force */
 864             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 865             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 866             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 867
 868             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 869             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 870             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 871
 872             /**************************
 873              * CALCULATE INTERACTIONS *
 874              **************************/
 875
 876             r21              = _mm_mul_pd(rsq21,rinv21);
 877
 878             /* Calculate table index by multiplying r with table scale and truncate to integer */
 879             rt               = _mm_mul_pd(r21,vftabscale);
 880             vfitab           = _mm_cvttpd_epi32(rt);
 881 #ifdef __XOP__
 882             vfeps            = _mm_frcz_pd(rt);
 883 #else
 884             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 885 #endif
 886             twovfeps         = _mm_add_pd(vfeps,vfeps);
 887             vfitab           = _mm_slli_epi32(vfitab,2);
 888
 889             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 890             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 891             F                = _mm_setzero_pd();
 892             GMX_MM_TRANSPOSE2_PD(Y,F);
 893             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 894             H                = _mm_setzero_pd();
 895             GMX_MM_TRANSPOSE2_PD(G,H);
 896             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 897             VV               = _mm_macc_pd(vfeps,Fp,Y);
 898             velec            = _mm_mul_pd(qq21,VV);
 899             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 900             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
 901
 902             /* Update potential sum for this i atom from the interaction with this j atom. */
 903             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 904             velecsum         = _mm_add_pd(velecsum,velec);
 905
 906             fscal            = felec;
 907
 908             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 909
 910             /* Update vectorial force */
 911             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 912             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 913             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 914
 915             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 916             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 917             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 918
 919             /**************************
 920              * CALCULATE INTERACTIONS *
 921              **************************/
 922
 923             r22              = _mm_mul_pd(rsq22,rinv22);
 924
 925             /* Calculate table index by multiplying r with table scale and truncate to integer */
 926             rt               = _mm_mul_pd(r22,vftabscale);
 927             vfitab           = _mm_cvttpd_epi32(rt);
 928 #ifdef __XOP__
 929             vfeps            = _mm_frcz_pd(rt);
 930 #else
 931             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 932 #endif
 933             twovfeps         = _mm_add_pd(vfeps,vfeps);
 934             vfitab           = _mm_slli_epi32(vfitab,2);
 935
 936             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 937             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 938             F                = _mm_setzero_pd();
 939             GMX_MM_TRANSPOSE2_PD(Y,F);
 940             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 941             H                = _mm_setzero_pd();
 942             GMX_MM_TRANSPOSE2_PD(G,H);
 943             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 944             VV               = _mm_macc_pd(vfeps,Fp,Y);
 945             velec            = _mm_mul_pd(qq22,VV);
 946             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 947             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
 948
 949             /* Update potential sum for this i atom from the interaction with this j atom. */
 950             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 951             velecsum         = _mm_add_pd(velecsum,velec);
 952
 953             fscal            = felec;
 954
 955             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 956
 957             /* Update vectorial force */
 958             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 959             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 960             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 961
 962             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 963             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 964             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             r23              = _mm_mul_pd(rsq23,rinv23);
 971
 972             /* Calculate table index by multiplying r with table scale and truncate to integer */
 973             rt               = _mm_mul_pd(r23,vftabscale);
 974             vfitab           = _mm_cvttpd_epi32(rt);
 975 #ifdef __XOP__
 976             vfeps            = _mm_frcz_pd(rt);
 977 #else
 978             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 979 #endif
 980             twovfeps         = _mm_add_pd(vfeps,vfeps);
 981             vfitab           = _mm_slli_epi32(vfitab,2);
 982
 983             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 984             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 985             F                = _mm_setzero_pd();
 986             GMX_MM_TRANSPOSE2_PD(Y,F);
 987             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 988             H                = _mm_setzero_pd();
 989             GMX_MM_TRANSPOSE2_PD(G,H);
 990             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 991             VV               = _mm_macc_pd(vfeps,Fp,Y);
 992             velec            = _mm_mul_pd(qq23,VV);
 993             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 994             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq23,FF),_mm_mul_pd(vftabscale,rinv23)));
 995
 996             /* Update potential sum for this i atom from the interaction with this j atom. */
 997             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 998             velecsum         = _mm_add_pd(velecsum,velec);
 999
1000             fscal            = felec;
1001
1002             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1003
1004             /* Update vectorial force */
1005             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1006             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1007             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1008
1009             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1010             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1011             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             r31              = _mm_mul_pd(rsq31,rinv31);
1018
1019             /* Calculate table index by multiplying r with table scale and truncate to integer */
1020             rt               = _mm_mul_pd(r31,vftabscale);
1021             vfitab           = _mm_cvttpd_epi32(rt);
1022 #ifdef __XOP__
1023             vfeps            = _mm_frcz_pd(rt);
1024 #else
1025             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1026 #endif
1027             twovfeps         = _mm_add_pd(vfeps,vfeps);
1028             vfitab           = _mm_slli_epi32(vfitab,2);
1029
1030             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1031             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1032             F                = _mm_setzero_pd();
1033             GMX_MM_TRANSPOSE2_PD(Y,F);
1034             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1035             H                = _mm_setzero_pd();
1036             GMX_MM_TRANSPOSE2_PD(G,H);
1037             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1038             VV               = _mm_macc_pd(vfeps,Fp,Y);
1039             velec            = _mm_mul_pd(qq31,VV);
1040             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1041             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq31,FF),_mm_mul_pd(vftabscale,rinv31)));
1042
1043             /* Update potential sum for this i atom from the interaction with this j atom. */
1044             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1045             velecsum         = _mm_add_pd(velecsum,velec);
1046
1047             fscal            = felec;
1048
1049             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1050
1051             /* Update vectorial force */
1052             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1053             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1054             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1055
1056             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1057             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1058             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1059
1060             /**************************
1061              * CALCULATE INTERACTIONS *
1062              **************************/
1063
1064             r32              = _mm_mul_pd(rsq32,rinv32);
1065
1066             /* Calculate table index by multiplying r with table scale and truncate to integer */
1067             rt               = _mm_mul_pd(r32,vftabscale);
1068             vfitab           = _mm_cvttpd_epi32(rt);
1069 #ifdef __XOP__
1070             vfeps            = _mm_frcz_pd(rt);
1071 #else
1072             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1073 #endif
1074             twovfeps         = _mm_add_pd(vfeps,vfeps);
1075             vfitab           = _mm_slli_epi32(vfitab,2);
1076
1077             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1078             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1079             F                = _mm_setzero_pd();
1080             GMX_MM_TRANSPOSE2_PD(Y,F);
1081             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1082             H                = _mm_setzero_pd();
1083             GMX_MM_TRANSPOSE2_PD(G,H);
1084             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1085             VV               = _mm_macc_pd(vfeps,Fp,Y);
1086             velec            = _mm_mul_pd(qq32,VV);
1087             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1088             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq32,FF),_mm_mul_pd(vftabscale,rinv32)));
1089
1090             /* Update potential sum for this i atom from the interaction with this j atom. */
1091             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1092             velecsum         = _mm_add_pd(velecsum,velec);
1093
1094             fscal            = felec;
1095
1096             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1097
1098             /* Update vectorial force */
1099             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1100             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1101             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1102
1103             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1104             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1105             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1106
1107             /**************************
1108              * CALCULATE INTERACTIONS *
1109              **************************/
1110
1111             r33              = _mm_mul_pd(rsq33,rinv33);
1112
1113             /* Calculate table index by multiplying r with table scale and truncate to integer */
1114             rt               = _mm_mul_pd(r33,vftabscale);
1115             vfitab           = _mm_cvttpd_epi32(rt);
1116 #ifdef __XOP__
1117             vfeps            = _mm_frcz_pd(rt);
1118 #else
1119             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1120 #endif
1121             twovfeps         = _mm_add_pd(vfeps,vfeps);
1122             vfitab           = _mm_slli_epi32(vfitab,2);
1123
1124             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1125             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1126             F                = _mm_setzero_pd();
1127             GMX_MM_TRANSPOSE2_PD(Y,F);
1128             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1129             H                = _mm_setzero_pd();
1130             GMX_MM_TRANSPOSE2_PD(G,H);
1131             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1132             VV               = _mm_macc_pd(vfeps,Fp,Y);
1133             velec            = _mm_mul_pd(qq33,VV);
1134             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1135             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq33,FF),_mm_mul_pd(vftabscale,rinv33)));
1136
1137             /* Update potential sum for this i atom from the interaction with this j atom. */
1138             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1139             velecsum         = _mm_add_pd(velecsum,velec);
1140
1141             fscal            = felec;
1142
1143             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1144
1145             /* Update vectorial force */
1146             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1147             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1148             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1149
1150             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1151             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1152             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1153
1154             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1155
1156             /* Inner loop uses 414 flops */
1157         }
1158
1159         /* End of innermost loop */
1160
1161         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1162                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1163
1164         ggid                        = gid[iidx];
1165         /* Update potential energies */
1166         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1167
1168         /* Increment number of inner iterations */
1169         inneriter                  += j_index_end - j_index_start;
1170
1171         /* Outer loop uses 19 flops */
1172     }
1173
1174     /* Increment number of outer iterations */
1175     outeriter        += nri;
1176
1177     /* Update outer/inner flops */
1178
1179     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*414);
1180 }
1181 /*
1182  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_avx_128_fma_double
1183  * Electrostatics interaction: CubicSplineTable
1184  * VdW interaction:            None
1185  * Geometry:                   Water4-Water4
1186  * Calculate force/pot:        Force
1187  */
1188 void
1189 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_avx_128_fma_double
1190                     (t_nblist                    * gmx_restrict       nlist,
1191                      rvec                        * gmx_restrict          xx,
1192                      rvec                        * gmx_restrict          ff,
1193                      t_forcerec                  * gmx_restrict          fr,
1194                      t_mdatoms                   * gmx_restrict     mdatoms,
1195                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1196                      t_nrnb                      * gmx_restrict        nrnb)
1197 {
1198     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1199      * just 0 for non-waters.
1200      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1201      * jnr indices corresponding to data put in the four positions in the SIMD register.
1202      */
1203     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1204     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1205     int              jnrA,jnrB;
1206     int              j_coord_offsetA,j_coord_offsetB;
1207     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1208     real             rcutoff_scalar;
1209     real             *shiftvec,*fshift,*x,*f;
1210     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1211     int              vdwioffset1;
1212     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1213     int              vdwioffset2;
1214     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1215     int              vdwioffset3;
1216     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1217     int              vdwjidx1A,vdwjidx1B;
1218     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1219     int              vdwjidx2A,vdwjidx2B;
1220     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1221     int              vdwjidx3A,vdwjidx3B;
1222     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1223     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1224     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1225     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1226     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1227     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1228     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1229     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1230     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1231     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1232     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1233     real             *charge;
1234     __m128i          vfitab;
1235     __m128i          ifour       = _mm_set1_epi32(4);
1236     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
1237     real             *vftab;
1238     __m128d          dummy_mask,cutoff_mask;
1239     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1240     __m128d          one     = _mm_set1_pd(1.0);
1241     __m128d          two     = _mm_set1_pd(2.0);
1242     x                = xx[0];
1243     f                = ff[0];
1244
1245     nri              = nlist->nri;
1246     iinr             = nlist->iinr;
1247     jindex           = nlist->jindex;
1248     jjnr             = nlist->jjnr;
1249     shiftidx         = nlist->shift;
1250     gid              = nlist->gid;
1251     shiftvec         = fr->shift_vec[0];
1252     fshift           = fr->fshift[0];
1253     facel            = _mm_set1_pd(fr->epsfac);
1254     charge           = mdatoms->chargeA;
1255
1256     vftab            = kernel_data->table_elec->data;
1257     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
1258
1259     /* Setup water-specific parameters */
1260     inr              = nlist->iinr[0];
1261     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1262     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1263     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1264
1265     jq1              = _mm_set1_pd(charge[inr+1]);
1266     jq2              = _mm_set1_pd(charge[inr+2]);
1267     jq3              = _mm_set1_pd(charge[inr+3]);
1268     qq11             = _mm_mul_pd(iq1,jq1);
1269     qq12             = _mm_mul_pd(iq1,jq2);
1270     qq13             = _mm_mul_pd(iq1,jq3);
1271     qq21             = _mm_mul_pd(iq2,jq1);
1272     qq22             = _mm_mul_pd(iq2,jq2);
1273     qq23             = _mm_mul_pd(iq2,jq3);
1274     qq31             = _mm_mul_pd(iq3,jq1);
1275     qq32             = _mm_mul_pd(iq3,jq2);
1276     qq33             = _mm_mul_pd(iq3,jq3);
1277
1278     /* Avoid stupid compiler warnings */
1279     jnrA = jnrB = 0;
1280     j_coord_offsetA = 0;
1281     j_coord_offsetB = 0;
1282
1283     outeriter        = 0;
1284     inneriter        = 0;
1285
1286     /* Start outer loop over neighborlists */
1287     for(iidx=0; iidx<nri; iidx++)
1288     {
1289         /* Load shift vector for this list */
1290         i_shift_offset   = DIM*shiftidx[iidx];
1291
1292         /* Load limits for loop over neighbors */
1293         j_index_start    = jindex[iidx];
1294         j_index_end      = jindex[iidx+1];
1295
1296         /* Get outer coordinate index */
1297         inr              = iinr[iidx];
1298         i_coord_offset   = DIM*inr;
1299
1300         /* Load i particle coords and add shift vector */
1301         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1302                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1303
1304         fix1             = _mm_setzero_pd();
1305         fiy1             = _mm_setzero_pd();
1306         fiz1             = _mm_setzero_pd();
1307         fix2             = _mm_setzero_pd();
1308         fiy2             = _mm_setzero_pd();
1309         fiz2             = _mm_setzero_pd();
1310         fix3             = _mm_setzero_pd();
1311         fiy3             = _mm_setzero_pd();
1312         fiz3             = _mm_setzero_pd();
1313
1314         /* Start inner kernel loop */
1315         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1316         {
1317
1318             /* Get j neighbor index, and coordinate index */
1319             jnrA             = jjnr[jidx];
1320             jnrB             = jjnr[jidx+1];
1321             j_coord_offsetA  = DIM*jnrA;
1322             j_coord_offsetB  = DIM*jnrB;
1323
1324             /* load j atom coordinates */
1325             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1326                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1327
1328             /* Calculate displacement vector */
1329             dx11             = _mm_sub_pd(ix1,jx1);
1330             dy11             = _mm_sub_pd(iy1,jy1);
1331             dz11             = _mm_sub_pd(iz1,jz1);
1332             dx12             = _mm_sub_pd(ix1,jx2);
1333             dy12             = _mm_sub_pd(iy1,jy2);
1334             dz12             = _mm_sub_pd(iz1,jz2);
1335             dx13             = _mm_sub_pd(ix1,jx3);
1336             dy13             = _mm_sub_pd(iy1,jy3);
1337             dz13             = _mm_sub_pd(iz1,jz3);
1338             dx21             = _mm_sub_pd(ix2,jx1);
1339             dy21             = _mm_sub_pd(iy2,jy1);
1340             dz21             = _mm_sub_pd(iz2,jz1);
1341             dx22             = _mm_sub_pd(ix2,jx2);
1342             dy22             = _mm_sub_pd(iy2,jy2);
1343             dz22             = _mm_sub_pd(iz2,jz2);
1344             dx23             = _mm_sub_pd(ix2,jx3);
1345             dy23             = _mm_sub_pd(iy2,jy3);
1346             dz23             = _mm_sub_pd(iz2,jz3);
1347             dx31             = _mm_sub_pd(ix3,jx1);
1348             dy31             = _mm_sub_pd(iy3,jy1);
1349             dz31             = _mm_sub_pd(iz3,jz1);
1350             dx32             = _mm_sub_pd(ix3,jx2);
1351             dy32             = _mm_sub_pd(iy3,jy2);
1352             dz32             = _mm_sub_pd(iz3,jz2);
1353             dx33             = _mm_sub_pd(ix3,jx3);
1354             dy33             = _mm_sub_pd(iy3,jy3);
1355             dz33             = _mm_sub_pd(iz3,jz3);
1356
1357             /* Calculate squared distance and things based on it */
1358             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1359             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1360             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1361             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1362             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1363             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1364             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1365             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1366             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1367
1368             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1369             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1370             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1371             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1372             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1373             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1374             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1375             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1376             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1377
1378             fjx1             = _mm_setzero_pd();
1379             fjy1             = _mm_setzero_pd();
1380             fjz1             = _mm_setzero_pd();
1381             fjx2             = _mm_setzero_pd();
1382             fjy2             = _mm_setzero_pd();
1383             fjz2             = _mm_setzero_pd();
1384             fjx3             = _mm_setzero_pd();
1385             fjy3             = _mm_setzero_pd();
1386             fjz3             = _mm_setzero_pd();
1387
1388             /**************************
1389              * CALCULATE INTERACTIONS *
1390              **************************/
1391
1392             r11              = _mm_mul_pd(rsq11,rinv11);
1393
1394             /* Calculate table index by multiplying r with table scale and truncate to integer */
1395             rt               = _mm_mul_pd(r11,vftabscale);
1396             vfitab           = _mm_cvttpd_epi32(rt);
1397 #ifdef __XOP__
1398             vfeps            = _mm_frcz_pd(rt);
1399 #else
1400             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1401 #endif
1402             twovfeps         = _mm_add_pd(vfeps,vfeps);
1403             vfitab           = _mm_slli_epi32(vfitab,2);
1404
1405             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1406             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1407             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1408             GMX_MM_TRANSPOSE2_PD(Y,F);
1409             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1410             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1411             GMX_MM_TRANSPOSE2_PD(G,H);
1412             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1413             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1414             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
1415
1416             fscal            = felec;
1417
1418             /* Update vectorial force */
1419             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1420             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1421             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1422
1423             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1424             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1425             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1426
1427             /**************************
1428              * CALCULATE INTERACTIONS *
1429              **************************/
1430
1431             r12              = _mm_mul_pd(rsq12,rinv12);
1432
1433             /* Calculate table index by multiplying r with table scale and truncate to integer */
1434             rt               = _mm_mul_pd(r12,vftabscale);
1435             vfitab           = _mm_cvttpd_epi32(rt);
1436 #ifdef __XOP__
1437             vfeps            = _mm_frcz_pd(rt);
1438 #else
1439             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1440 #endif
1441             twovfeps         = _mm_add_pd(vfeps,vfeps);
1442             vfitab           = _mm_slli_epi32(vfitab,2);
1443
1444             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1445             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1446             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1447             GMX_MM_TRANSPOSE2_PD(Y,F);
1448             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1449             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1450             GMX_MM_TRANSPOSE2_PD(G,H);
1451             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1452             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1453             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
1454
1455             fscal            = felec;
1456
1457             /* Update vectorial force */
1458             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1459             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1460             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1461
1462             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1463             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1464             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1465
1466             /**************************
1467              * CALCULATE INTERACTIONS *
1468              **************************/
1469
1470             r13              = _mm_mul_pd(rsq13,rinv13);
1471
1472             /* Calculate table index by multiplying r with table scale and truncate to integer */
1473             rt               = _mm_mul_pd(r13,vftabscale);
1474             vfitab           = _mm_cvttpd_epi32(rt);
1475 #ifdef __XOP__
1476             vfeps            = _mm_frcz_pd(rt);
1477 #else
1478             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1479 #endif
1480             twovfeps         = _mm_add_pd(vfeps,vfeps);
1481             vfitab           = _mm_slli_epi32(vfitab,2);
1482
1483             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1484             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1485             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1486             GMX_MM_TRANSPOSE2_PD(Y,F);
1487             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1488             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1489             GMX_MM_TRANSPOSE2_PD(G,H);
1490             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1491             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1492             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq13,FF),_mm_mul_pd(vftabscale,rinv13)));
1493
1494             fscal            = felec;
1495
1496             /* Update vectorial force */
1497             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1498             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1499             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1500
1501             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1502             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1503             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1504
1505             /**************************
1506              * CALCULATE INTERACTIONS *
1507              **************************/
1508
1509             r21              = _mm_mul_pd(rsq21,rinv21);
1510
1511             /* Calculate table index by multiplying r with table scale and truncate to integer */
1512             rt               = _mm_mul_pd(r21,vftabscale);
1513             vfitab           = _mm_cvttpd_epi32(rt);
1514 #ifdef __XOP__
1515             vfeps            = _mm_frcz_pd(rt);
1516 #else
1517             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1518 #endif
1519             twovfeps         = _mm_add_pd(vfeps,vfeps);
1520             vfitab           = _mm_slli_epi32(vfitab,2);
1521
1522             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1523             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1524             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1525             GMX_MM_TRANSPOSE2_PD(Y,F);
1526             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1527             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1528             GMX_MM_TRANSPOSE2_PD(G,H);
1529             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1530             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1531             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
1532
1533             fscal            = felec;
1534
1535             /* Update vectorial force */
1536             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1537             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1538             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1539
1540             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1541             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1542             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1543
1544             /**************************
1545              * CALCULATE INTERACTIONS *
1546              **************************/
1547
1548             r22              = _mm_mul_pd(rsq22,rinv22);
1549
1550             /* Calculate table index by multiplying r with table scale and truncate to integer */
1551             rt               = _mm_mul_pd(r22,vftabscale);
1552             vfitab           = _mm_cvttpd_epi32(rt);
1553 #ifdef __XOP__
1554             vfeps            = _mm_frcz_pd(rt);
1555 #else
1556             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1557 #endif
1558             twovfeps         = _mm_add_pd(vfeps,vfeps);
1559             vfitab           = _mm_slli_epi32(vfitab,2);
1560
1561             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1562             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1563             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1564             GMX_MM_TRANSPOSE2_PD(Y,F);
1565             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1566             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1567             GMX_MM_TRANSPOSE2_PD(G,H);
1568             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1569             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1570             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
1571
1572             fscal            = felec;
1573
1574             /* Update vectorial force */
1575             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1576             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1577             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1578
1579             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1580             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1581             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1582
1583             /**************************
1584              * CALCULATE INTERACTIONS *
1585              **************************/
1586
1587             r23              = _mm_mul_pd(rsq23,rinv23);
1588
1589             /* Calculate table index by multiplying r with table scale and truncate to integer */
1590             rt               = _mm_mul_pd(r23,vftabscale);
1591             vfitab           = _mm_cvttpd_epi32(rt);
1592 #ifdef __XOP__
1593             vfeps            = _mm_frcz_pd(rt);
1594 #else
1595             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1596 #endif
1597             twovfeps         = _mm_add_pd(vfeps,vfeps);
1598             vfitab           = _mm_slli_epi32(vfitab,2);
1599
1600             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1601             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1602             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1603             GMX_MM_TRANSPOSE2_PD(Y,F);
1604             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1605             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1606             GMX_MM_TRANSPOSE2_PD(G,H);
1607             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1608             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1609             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq23,FF),_mm_mul_pd(vftabscale,rinv23)));
1610
1611             fscal            = felec;
1612
1613             /* Update vectorial force */
1614             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1615             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1616             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1617
1618             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1619             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1620             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1621
1622             /**************************
1623              * CALCULATE INTERACTIONS *
1624              **************************/
1625
1626             r31              = _mm_mul_pd(rsq31,rinv31);
1627
1628             /* Calculate table index by multiplying r with table scale and truncate to integer */
1629             rt               = _mm_mul_pd(r31,vftabscale);
1630             vfitab           = _mm_cvttpd_epi32(rt);
1631 #ifdef __XOP__
1632             vfeps            = _mm_frcz_pd(rt);
1633 #else
1634             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1635 #endif
1636             twovfeps         = _mm_add_pd(vfeps,vfeps);
1637             vfitab           = _mm_slli_epi32(vfitab,2);
1638
1639             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1640             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1641             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1642             GMX_MM_TRANSPOSE2_PD(Y,F);
1643             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1644             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1645             GMX_MM_TRANSPOSE2_PD(G,H);
1646             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1647             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1648             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq31,FF),_mm_mul_pd(vftabscale,rinv31)));
1649
1650             fscal            = felec;
1651
1652             /* Update vectorial force */
1653             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1654             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1655             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1656
1657             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1658             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1659             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1660
1661             /**************************
1662              * CALCULATE INTERACTIONS *
1663              **************************/
1664
1665             r32              = _mm_mul_pd(rsq32,rinv32);
1666
1667             /* Calculate table index by multiplying r with table scale and truncate to integer */
1668             rt               = _mm_mul_pd(r32,vftabscale);
1669             vfitab           = _mm_cvttpd_epi32(rt);
1670 #ifdef __XOP__
1671             vfeps            = _mm_frcz_pd(rt);
1672 #else
1673             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1674 #endif
1675             twovfeps         = _mm_add_pd(vfeps,vfeps);
1676             vfitab           = _mm_slli_epi32(vfitab,2);
1677
1678             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1679             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1680             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1681             GMX_MM_TRANSPOSE2_PD(Y,F);
1682             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1683             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1684             GMX_MM_TRANSPOSE2_PD(G,H);
1685             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1686             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1687             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq32,FF),_mm_mul_pd(vftabscale,rinv32)));
1688
1689             fscal            = felec;
1690
1691             /* Update vectorial force */
1692             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1693             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1694             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1695
1696             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1697             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1698             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1699
1700             /**************************
1701              * CALCULATE INTERACTIONS *
1702              **************************/
1703
1704             r33              = _mm_mul_pd(rsq33,rinv33);
1705
1706             /* Calculate table index by multiplying r with table scale and truncate to integer */
1707             rt               = _mm_mul_pd(r33,vftabscale);
1708             vfitab           = _mm_cvttpd_epi32(rt);
1709 #ifdef __XOP__
1710             vfeps            = _mm_frcz_pd(rt);
1711 #else
1712             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1713 #endif
1714             twovfeps         = _mm_add_pd(vfeps,vfeps);
1715             vfitab           = _mm_slli_epi32(vfitab,2);
1716
1717             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1718             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1719             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1720             GMX_MM_TRANSPOSE2_PD(Y,F);
1721             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1722             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
1723             GMX_MM_TRANSPOSE2_PD(G,H);
1724             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1725             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1726             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq33,FF),_mm_mul_pd(vftabscale,rinv33)));
1727
1728             fscal            = felec;
1729
1730             /* Update vectorial force */
1731             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1732             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1733             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1734
1735             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1736             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1737             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1738
1739             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA+DIM,f+j_coord_offsetB+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1740
1741             /* Inner loop uses 378 flops */
1742         }
1743
1744         if(jidx<j_index_end)
1745         {
1746
1747             jnrA             = jjnr[jidx];
1748             j_coord_offsetA  = DIM*jnrA;
1749
1750             /* load j atom coordinates */
1751             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
1752                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1753
1754             /* Calculate displacement vector */
1755             dx11             = _mm_sub_pd(ix1,jx1);
1756             dy11             = _mm_sub_pd(iy1,jy1);
1757             dz11             = _mm_sub_pd(iz1,jz1);
1758             dx12             = _mm_sub_pd(ix1,jx2);
1759             dy12             = _mm_sub_pd(iy1,jy2);
1760             dz12             = _mm_sub_pd(iz1,jz2);
1761             dx13             = _mm_sub_pd(ix1,jx3);
1762             dy13             = _mm_sub_pd(iy1,jy3);
1763             dz13             = _mm_sub_pd(iz1,jz3);
1764             dx21             = _mm_sub_pd(ix2,jx1);
1765             dy21             = _mm_sub_pd(iy2,jy1);
1766             dz21             = _mm_sub_pd(iz2,jz1);
1767             dx22             = _mm_sub_pd(ix2,jx2);
1768             dy22             = _mm_sub_pd(iy2,jy2);
1769             dz22             = _mm_sub_pd(iz2,jz2);
1770             dx23             = _mm_sub_pd(ix2,jx3);
1771             dy23             = _mm_sub_pd(iy2,jy3);
1772             dz23             = _mm_sub_pd(iz2,jz3);
1773             dx31             = _mm_sub_pd(ix3,jx1);
1774             dy31             = _mm_sub_pd(iy3,jy1);
1775             dz31             = _mm_sub_pd(iz3,jz1);
1776             dx32             = _mm_sub_pd(ix3,jx2);
1777             dy32             = _mm_sub_pd(iy3,jy2);
1778             dz32             = _mm_sub_pd(iz3,jz2);
1779             dx33             = _mm_sub_pd(ix3,jx3);
1780             dy33             = _mm_sub_pd(iy3,jy3);
1781             dz33             = _mm_sub_pd(iz3,jz3);
1782
1783             /* Calculate squared distance and things based on it */
1784             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1785             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1786             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1787             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1788             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1789             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1790             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1791             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1792             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1793
1794             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1795             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1796             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1797             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1798             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1799             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1800             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1801             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1802             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1803
1804             fjx1             = _mm_setzero_pd();
1805             fjy1             = _mm_setzero_pd();
1806             fjz1             = _mm_setzero_pd();
1807             fjx2             = _mm_setzero_pd();
1808             fjy2             = _mm_setzero_pd();
1809             fjz2             = _mm_setzero_pd();
1810             fjx3             = _mm_setzero_pd();
1811             fjy3             = _mm_setzero_pd();
1812             fjz3             = _mm_setzero_pd();
1813
1814             /**************************
1815              * CALCULATE INTERACTIONS *
1816              **************************/
1817
1818             r11              = _mm_mul_pd(rsq11,rinv11);
1819
1820             /* Calculate table index by multiplying r with table scale and truncate to integer */
1821             rt               = _mm_mul_pd(r11,vftabscale);
1822             vfitab           = _mm_cvttpd_epi32(rt);
1823 #ifdef __XOP__
1824             vfeps            = _mm_frcz_pd(rt);
1825 #else
1826             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1827 #endif
1828             twovfeps         = _mm_add_pd(vfeps,vfeps);
1829             vfitab           = _mm_slli_epi32(vfitab,2);
1830
1831             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1832             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1833             F                = _mm_setzero_pd();
1834             GMX_MM_TRANSPOSE2_PD(Y,F);
1835             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1836             H                = _mm_setzero_pd();
1837             GMX_MM_TRANSPOSE2_PD(G,H);
1838             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1839             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1840             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq11,FF),_mm_mul_pd(vftabscale,rinv11)));
1841
1842             fscal            = felec;
1843
1844             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1845
1846             /* Update vectorial force */
1847             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1848             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1849             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1850
1851             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1852             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1853             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1854
1855             /**************************
1856              * CALCULATE INTERACTIONS *
1857              **************************/
1858
1859             r12              = _mm_mul_pd(rsq12,rinv12);
1860
1861             /* Calculate table index by multiplying r with table scale and truncate to integer */
1862             rt               = _mm_mul_pd(r12,vftabscale);
1863             vfitab           = _mm_cvttpd_epi32(rt);
1864 #ifdef __XOP__
1865             vfeps            = _mm_frcz_pd(rt);
1866 #else
1867             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1868 #endif
1869             twovfeps         = _mm_add_pd(vfeps,vfeps);
1870             vfitab           = _mm_slli_epi32(vfitab,2);
1871
1872             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1873             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1874             F                = _mm_setzero_pd();
1875             GMX_MM_TRANSPOSE2_PD(Y,F);
1876             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1877             H                = _mm_setzero_pd();
1878             GMX_MM_TRANSPOSE2_PD(G,H);
1879             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1880             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1881             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq12,FF),_mm_mul_pd(vftabscale,rinv12)));
1882
1883             fscal            = felec;
1884
1885             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1886
1887             /* Update vectorial force */
1888             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1889             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1890             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1891
1892             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1893             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1894             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1895
1896             /**************************
1897              * CALCULATE INTERACTIONS *
1898              **************************/
1899
1900             r13              = _mm_mul_pd(rsq13,rinv13);
1901
1902             /* Calculate table index by multiplying r with table scale and truncate to integer */
1903             rt               = _mm_mul_pd(r13,vftabscale);
1904             vfitab           = _mm_cvttpd_epi32(rt);
1905 #ifdef __XOP__
1906             vfeps            = _mm_frcz_pd(rt);
1907 #else
1908             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1909 #endif
1910             twovfeps         = _mm_add_pd(vfeps,vfeps);
1911             vfitab           = _mm_slli_epi32(vfitab,2);
1912
1913             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1914             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1915             F                = _mm_setzero_pd();
1916             GMX_MM_TRANSPOSE2_PD(Y,F);
1917             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1918             H                = _mm_setzero_pd();
1919             GMX_MM_TRANSPOSE2_PD(G,H);
1920             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1921             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1922             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq13,FF),_mm_mul_pd(vftabscale,rinv13)));
1923
1924             fscal            = felec;
1925
1926             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1927
1928             /* Update vectorial force */
1929             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1930             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1931             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1932
1933             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1934             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1935             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1936
1937             /**************************
1938              * CALCULATE INTERACTIONS *
1939              **************************/
1940
1941             r21              = _mm_mul_pd(rsq21,rinv21);
1942
1943             /* Calculate table index by multiplying r with table scale and truncate to integer */
1944             rt               = _mm_mul_pd(r21,vftabscale);
1945             vfitab           = _mm_cvttpd_epi32(rt);
1946 #ifdef __XOP__
1947             vfeps            = _mm_frcz_pd(rt);
1948 #else
1949             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1950 #endif
1951             twovfeps         = _mm_add_pd(vfeps,vfeps);
1952             vfitab           = _mm_slli_epi32(vfitab,2);
1953
1954             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1955             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1956             F                = _mm_setzero_pd();
1957             GMX_MM_TRANSPOSE2_PD(Y,F);
1958             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1959             H                = _mm_setzero_pd();
1960             GMX_MM_TRANSPOSE2_PD(G,H);
1961             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1962             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1963             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq21,FF),_mm_mul_pd(vftabscale,rinv21)));
1964
1965             fscal            = felec;
1966
1967             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1968
1969             /* Update vectorial force */
1970             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1971             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1972             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1973
1974             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1975             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1976             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1977
1978             /**************************
1979              * CALCULATE INTERACTIONS *
1980              **************************/
1981
1982             r22              = _mm_mul_pd(rsq22,rinv22);
1983
1984             /* Calculate table index by multiplying r with table scale and truncate to integer */
1985             rt               = _mm_mul_pd(r22,vftabscale);
1986             vfitab           = _mm_cvttpd_epi32(rt);
1987 #ifdef __XOP__
1988             vfeps            = _mm_frcz_pd(rt);
1989 #else
1990             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1991 #endif
1992             twovfeps         = _mm_add_pd(vfeps,vfeps);
1993             vfitab           = _mm_slli_epi32(vfitab,2);
1994
1995             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1996             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1997             F                = _mm_setzero_pd();
1998             GMX_MM_TRANSPOSE2_PD(Y,F);
1999             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2000             H                = _mm_setzero_pd();
2001             GMX_MM_TRANSPOSE2_PD(G,H);
2002             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2003             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2004             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq22,FF),_mm_mul_pd(vftabscale,rinv22)));
2005
2006             fscal            = felec;
2007
2008             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2009
2010             /* Update vectorial force */
2011             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2012             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2013             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2014
2015             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2016             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2017             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2018
2019             /**************************
2020              * CALCULATE INTERACTIONS *
2021              **************************/
2022
2023             r23              = _mm_mul_pd(rsq23,rinv23);
2024
2025             /* Calculate table index by multiplying r with table scale and truncate to integer */
2026             rt               = _mm_mul_pd(r23,vftabscale);
2027             vfitab           = _mm_cvttpd_epi32(rt);
2028 #ifdef __XOP__
2029             vfeps            = _mm_frcz_pd(rt);
2030 #else
2031             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2032 #endif
2033             twovfeps         = _mm_add_pd(vfeps,vfeps);
2034             vfitab           = _mm_slli_epi32(vfitab,2);
2035
2036             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2037             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2038             F                = _mm_setzero_pd();
2039             GMX_MM_TRANSPOSE2_PD(Y,F);
2040             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2041             H                = _mm_setzero_pd();
2042             GMX_MM_TRANSPOSE2_PD(G,H);
2043             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2044             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2045             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq23,FF),_mm_mul_pd(vftabscale,rinv23)));
2046
2047             fscal            = felec;
2048
2049             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2050
2051             /* Update vectorial force */
2052             fix2             = _mm_macc_pd(dx23,fscal,fix2);
2053             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
2054             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
2055
2056             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
2057             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
2058             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
2059
2060             /**************************
2061              * CALCULATE INTERACTIONS *
2062              **************************/
2063
2064             r31              = _mm_mul_pd(rsq31,rinv31);
2065
2066             /* Calculate table index by multiplying r with table scale and truncate to integer */
2067             rt               = _mm_mul_pd(r31,vftabscale);
2068             vfitab           = _mm_cvttpd_epi32(rt);
2069 #ifdef __XOP__
2070             vfeps            = _mm_frcz_pd(rt);
2071 #else
2072             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2073 #endif
2074             twovfeps         = _mm_add_pd(vfeps,vfeps);
2075             vfitab           = _mm_slli_epi32(vfitab,2);
2076
2077             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2078             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2079             F                = _mm_setzero_pd();
2080             GMX_MM_TRANSPOSE2_PD(Y,F);
2081             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2082             H                = _mm_setzero_pd();
2083             GMX_MM_TRANSPOSE2_PD(G,H);
2084             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2085             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2086             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq31,FF),_mm_mul_pd(vftabscale,rinv31)));
2087
2088             fscal            = felec;
2089
2090             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2091
2092             /* Update vectorial force */
2093             fix3             = _mm_macc_pd(dx31,fscal,fix3);
2094             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
2095             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
2096
2097             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
2098             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
2099             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
2100
2101             /**************************
2102              * CALCULATE INTERACTIONS *
2103              **************************/
2104
2105             r32              = _mm_mul_pd(rsq32,rinv32);
2106
2107             /* Calculate table index by multiplying r with table scale and truncate to integer */
2108             rt               = _mm_mul_pd(r32,vftabscale);
2109             vfitab           = _mm_cvttpd_epi32(rt);
2110 #ifdef __XOP__
2111             vfeps            = _mm_frcz_pd(rt);
2112 #else
2113             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2114 #endif
2115             twovfeps         = _mm_add_pd(vfeps,vfeps);
2116             vfitab           = _mm_slli_epi32(vfitab,2);
2117
2118             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2119             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2120             F                = _mm_setzero_pd();
2121             GMX_MM_TRANSPOSE2_PD(Y,F);
2122             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2123             H                = _mm_setzero_pd();
2124             GMX_MM_TRANSPOSE2_PD(G,H);
2125             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2126             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2127             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq32,FF),_mm_mul_pd(vftabscale,rinv32)));
2128
2129             fscal            = felec;
2130
2131             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2132
2133             /* Update vectorial force */
2134             fix3             = _mm_macc_pd(dx32,fscal,fix3);
2135             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
2136             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
2137
2138             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
2139             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
2140             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
2141
2142             /**************************
2143              * CALCULATE INTERACTIONS *
2144              **************************/
2145
2146             r33              = _mm_mul_pd(rsq33,rinv33);
2147
2148             /* Calculate table index by multiplying r with table scale and truncate to integer */
2149             rt               = _mm_mul_pd(r33,vftabscale);
2150             vfitab           = _mm_cvttpd_epi32(rt);
2151 #ifdef __XOP__
2152             vfeps            = _mm_frcz_pd(rt);
2153 #else
2154             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
2155 #endif
2156             twovfeps         = _mm_add_pd(vfeps,vfeps);
2157             vfitab           = _mm_slli_epi32(vfitab,2);
2158
2159             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2160             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2161             F                = _mm_setzero_pd();
2162             GMX_MM_TRANSPOSE2_PD(Y,F);
2163             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
2164             H                = _mm_setzero_pd();
2165             GMX_MM_TRANSPOSE2_PD(G,H);
2166             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
2167             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
2168             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq33,FF),_mm_mul_pd(vftabscale,rinv33)));
2169
2170             fscal            = felec;
2171
2172             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2173
2174             /* Update vectorial force */
2175             fix3             = _mm_macc_pd(dx33,fscal,fix3);
2176             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
2177             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
2178
2179             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
2180             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
2181             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
2182
2183             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2184
2185             /* Inner loop uses 378 flops */
2186         }
2187
2188         /* End of innermost loop */
2189
2190         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2191                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
2192
2193         /* Increment number of inner iterations */
2194         inneriter                  += j_index_end - j_index_start;
2195
2196         /* Outer loop uses 18 flops */
2197     }
2198
2199     /* Increment number of outer iterations */
2200     outeriter        += nri;
2201
2202     /* Update outer/inner flops */
2203
2204     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*378);
2205 }