src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwNone_GeomW4W4_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4W4_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Water4
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwNone_GeomW4W4_VF_avx_128_fma_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset1;
  67     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  68     int              vdwioffset2;
  69     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  70     int              vdwioffset3;
  71     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  72     int              vdwjidx1A,vdwjidx1B;
  73     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  74     int              vdwjidx2A,vdwjidx2B;
  75     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  76     int              vdwjidx3A,vdwjidx3B;
  77     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  78     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  79     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  80     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  81     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  82     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  83     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  84     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  85     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  86     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128i          ewitab;
  90     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  91     real             *ewtab;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109
 110     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 111     ewtab            = fr->ic->tabq_coul_FDV0;
 112     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 113     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 114
 115     /* Setup water-specific parameters */
 116     inr              = nlist->iinr[0];
 117     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 118     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 119     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 120
 121     jq1              = _mm_set1_pd(charge[inr+1]);
 122     jq2              = _mm_set1_pd(charge[inr+2]);
 123     jq3              = _mm_set1_pd(charge[inr+3]);
 124     qq11             = _mm_mul_pd(iq1,jq1);
 125     qq12             = _mm_mul_pd(iq1,jq2);
 126     qq13             = _mm_mul_pd(iq1,jq3);
 127     qq21             = _mm_mul_pd(iq2,jq1);
 128     qq22             = _mm_mul_pd(iq2,jq2);
 129     qq23             = _mm_mul_pd(iq2,jq3);
 130     qq31             = _mm_mul_pd(iq3,jq1);
 131     qq32             = _mm_mul_pd(iq3,jq2);
 132     qq33             = _mm_mul_pd(iq3,jq3);
 133
 134     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 135     rcutoff_scalar   = fr->rcoulomb;
 136     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 137     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 163                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 164
 165         fix1             = _mm_setzero_pd();
 166         fiy1             = _mm_setzero_pd();
 167         fiz1             = _mm_setzero_pd();
 168         fix2             = _mm_setzero_pd();
 169         fiy2             = _mm_setzero_pd();
 170         fiz2             = _mm_setzero_pd();
 171         fix3             = _mm_setzero_pd();
 172         fiy3             = _mm_setzero_pd();
 173         fiz3             = _mm_setzero_pd();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_pd();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187
 188             /* load j atom coordinates */
 189             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 190                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 191
 192             /* Calculate displacement vector */
 193             dx11             = _mm_sub_pd(ix1,jx1);
 194             dy11             = _mm_sub_pd(iy1,jy1);
 195             dz11             = _mm_sub_pd(iz1,jz1);
 196             dx12             = _mm_sub_pd(ix1,jx2);
 197             dy12             = _mm_sub_pd(iy1,jy2);
 198             dz12             = _mm_sub_pd(iz1,jz2);
 199             dx13             = _mm_sub_pd(ix1,jx3);
 200             dy13             = _mm_sub_pd(iy1,jy3);
 201             dz13             = _mm_sub_pd(iz1,jz3);
 202             dx21             = _mm_sub_pd(ix2,jx1);
 203             dy21             = _mm_sub_pd(iy2,jy1);
 204             dz21             = _mm_sub_pd(iz2,jz1);
 205             dx22             = _mm_sub_pd(ix2,jx2);
 206             dy22             = _mm_sub_pd(iy2,jy2);
 207             dz22             = _mm_sub_pd(iz2,jz2);
 208             dx23             = _mm_sub_pd(ix2,jx3);
 209             dy23             = _mm_sub_pd(iy2,jy3);
 210             dz23             = _mm_sub_pd(iz2,jz3);
 211             dx31             = _mm_sub_pd(ix3,jx1);
 212             dy31             = _mm_sub_pd(iy3,jy1);
 213             dz31             = _mm_sub_pd(iz3,jz1);
 214             dx32             = _mm_sub_pd(ix3,jx2);
 215             dy32             = _mm_sub_pd(iy3,jy2);
 216             dz32             = _mm_sub_pd(iz3,jz2);
 217             dx33             = _mm_sub_pd(ix3,jx3);
 218             dy33             = _mm_sub_pd(iy3,jy3);
 219             dz33             = _mm_sub_pd(iz3,jz3);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 223             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 224             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 225             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 226             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 227             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 228             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 229             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 230             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 231
 232             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 233             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 234             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 235             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 236             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 237             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 238             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 239             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 240             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 241
 242             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 243             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 244             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 245             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 246             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 247             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 248             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 249             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 250             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 251
 252             fjx1             = _mm_setzero_pd();
 253             fjy1             = _mm_setzero_pd();
 254             fjz1             = _mm_setzero_pd();
 255             fjx2             = _mm_setzero_pd();
 256             fjy2             = _mm_setzero_pd();
 257             fjz2             = _mm_setzero_pd();
 258             fjx3             = _mm_setzero_pd();
 259             fjy3             = _mm_setzero_pd();
 260             fjz3             = _mm_setzero_pd();
 261
 262             /**************************
 263              * CALCULATE INTERACTIONS *
 264              **************************/
 265
 266             if (gmx_mm_any_lt(rsq11,rcutoff2))
 267             {
 268
 269             r11              = _mm_mul_pd(rsq11,rinv11);
 270
 271             /* EWALD ELECTROSTATICS */
 272
 273             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 274             ewrt             = _mm_mul_pd(r11,ewtabscale);
 275             ewitab           = _mm_cvttpd_epi32(ewrt);
 276 #ifdef __XOP__
 277             eweps            = _mm_frcz_pd(ewrt);
 278 #else
 279             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 280 #endif
 281             twoeweps         = _mm_add_pd(eweps,eweps);
 282             ewitab           = _mm_slli_epi32(ewitab,2);
 283             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 284             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 285             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 286             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 287             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 288             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 289             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 290             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 291             velec            = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
 292             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 293
 294             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velec            = _mm_and_pd(velec,cutoff_mask);
 298             velecsum         = _mm_add_pd(velecsum,velec);
 299
 300             fscal            = felec;
 301
 302             fscal            = _mm_and_pd(fscal,cutoff_mask);
 303
 304             /* Update vectorial force */
 305             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 306             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 307             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 308
 309             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 310             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 311             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 312
 313             }
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             if (gmx_mm_any_lt(rsq12,rcutoff2))
 320             {
 321
 322             r12              = _mm_mul_pd(rsq12,rinv12);
 323
 324             /* EWALD ELECTROSTATICS */
 325
 326             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 327             ewrt             = _mm_mul_pd(r12,ewtabscale);
 328             ewitab           = _mm_cvttpd_epi32(ewrt);
 329 #ifdef __XOP__
 330             eweps            = _mm_frcz_pd(ewrt);
 331 #else
 332             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 333 #endif
 334             twoeweps         = _mm_add_pd(eweps,eweps);
 335             ewitab           = _mm_slli_epi32(ewitab,2);
 336             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 337             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 338             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 339             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 340             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 341             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 342             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 343             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 344             velec            = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
 345             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 346
 347             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velec            = _mm_and_pd(velec,cutoff_mask);
 351             velecsum         = _mm_add_pd(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355             fscal            = _mm_and_pd(fscal,cutoff_mask);
 356
 357             /* Update vectorial force */
 358             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 359             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 360             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 361
 362             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 363             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 364             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 365
 366             }
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             if (gmx_mm_any_lt(rsq13,rcutoff2))
 373             {
 374
 375             r13              = _mm_mul_pd(rsq13,rinv13);
 376
 377             /* EWALD ELECTROSTATICS */
 378
 379             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 380             ewrt             = _mm_mul_pd(r13,ewtabscale);
 381             ewitab           = _mm_cvttpd_epi32(ewrt);
 382 #ifdef __XOP__
 383             eweps            = _mm_frcz_pd(ewrt);
 384 #else
 385             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 386 #endif
 387             twoeweps         = _mm_add_pd(eweps,eweps);
 388             ewitab           = _mm_slli_epi32(ewitab,2);
 389             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 390             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 391             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 392             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 393             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 394             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 395             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 396             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 397             velec            = _mm_mul_pd(qq13,_mm_sub_pd(_mm_sub_pd(rinv13,sh_ewald),velec));
 398             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 399
 400             cutoff_mask      = _mm_cmplt_pd(rsq13,rcutoff2);
 401
 402             /* Update potential sum for this i atom from the interaction with this j atom. */
 403             velec            = _mm_and_pd(velec,cutoff_mask);
 404             velecsum         = _mm_add_pd(velecsum,velec);
 405
 406             fscal            = felec;
 407
 408             fscal            = _mm_and_pd(fscal,cutoff_mask);
 409
 410             /* Update vectorial force */
 411             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 412             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 413             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 414
 415             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 416             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 417             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 418
 419             }
 420
 421             /**************************
 422              * CALCULATE INTERACTIONS *
 423              **************************/
 424
 425             if (gmx_mm_any_lt(rsq21,rcutoff2))
 426             {
 427
 428             r21              = _mm_mul_pd(rsq21,rinv21);
 429
 430             /* EWALD ELECTROSTATICS */
 431
 432             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 433             ewrt             = _mm_mul_pd(r21,ewtabscale);
 434             ewitab           = _mm_cvttpd_epi32(ewrt);
 435 #ifdef __XOP__
 436             eweps            = _mm_frcz_pd(ewrt);
 437 #else
 438             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 439 #endif
 440             twoeweps         = _mm_add_pd(eweps,eweps);
 441             ewitab           = _mm_slli_epi32(ewitab,2);
 442             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 443             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 444             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 445             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 446             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 447             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 448             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 449             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 450             velec            = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
 451             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 452
 453             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
 454
 455             /* Update potential sum for this i atom from the interaction with this j atom. */
 456             velec            = _mm_and_pd(velec,cutoff_mask);
 457             velecsum         = _mm_add_pd(velecsum,velec);
 458
 459             fscal            = felec;
 460
 461             fscal            = _mm_and_pd(fscal,cutoff_mask);
 462
 463             /* Update vectorial force */
 464             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 465             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 466             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 467
 468             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 469             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 470             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 471
 472             }
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             if (gmx_mm_any_lt(rsq22,rcutoff2))
 479             {
 480
 481             r22              = _mm_mul_pd(rsq22,rinv22);
 482
 483             /* EWALD ELECTROSTATICS */
 484
 485             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 486             ewrt             = _mm_mul_pd(r22,ewtabscale);
 487             ewitab           = _mm_cvttpd_epi32(ewrt);
 488 #ifdef __XOP__
 489             eweps            = _mm_frcz_pd(ewrt);
 490 #else
 491             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 492 #endif
 493             twoeweps         = _mm_add_pd(eweps,eweps);
 494             ewitab           = _mm_slli_epi32(ewitab,2);
 495             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 496             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 497             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 498             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 499             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 500             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 501             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 502             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 503             velec            = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
 504             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 505
 506             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             velec            = _mm_and_pd(velec,cutoff_mask);
 510             velecsum         = _mm_add_pd(velecsum,velec);
 511
 512             fscal            = felec;
 513
 514             fscal            = _mm_and_pd(fscal,cutoff_mask);
 515
 516             /* Update vectorial force */
 517             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 518             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 519             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 520
 521             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 522             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 523             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 524
 525             }
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             if (gmx_mm_any_lt(rsq23,rcutoff2))
 532             {
 533
 534             r23              = _mm_mul_pd(rsq23,rinv23);
 535
 536             /* EWALD ELECTROSTATICS */
 537
 538             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 539             ewrt             = _mm_mul_pd(r23,ewtabscale);
 540             ewitab           = _mm_cvttpd_epi32(ewrt);
 541 #ifdef __XOP__
 542             eweps            = _mm_frcz_pd(ewrt);
 543 #else
 544             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 545 #endif
 546             twoeweps         = _mm_add_pd(eweps,eweps);
 547             ewitab           = _mm_slli_epi32(ewitab,2);
 548             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 549             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 550             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 551             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 552             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 553             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 554             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 555             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 556             velec            = _mm_mul_pd(qq23,_mm_sub_pd(_mm_sub_pd(rinv23,sh_ewald),velec));
 557             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
 558
 559             cutoff_mask      = _mm_cmplt_pd(rsq23,rcutoff2);
 560
 561             /* Update potential sum for this i atom from the interaction with this j atom. */
 562             velec            = _mm_and_pd(velec,cutoff_mask);
 563             velecsum         = _mm_add_pd(velecsum,velec);
 564
 565             fscal            = felec;
 566
 567             fscal            = _mm_and_pd(fscal,cutoff_mask);
 568
 569             /* Update vectorial force */
 570             fix2             = _mm_macc_pd(dx23,fscal,fix2);
 571             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
 572             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
 573
 574             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
 575             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
 576             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
 577
 578             }
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             if (gmx_mm_any_lt(rsq31,rcutoff2))
 585             {
 586
 587             r31              = _mm_mul_pd(rsq31,rinv31);
 588
 589             /* EWALD ELECTROSTATICS */
 590
 591             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 592             ewrt             = _mm_mul_pd(r31,ewtabscale);
 593             ewitab           = _mm_cvttpd_epi32(ewrt);
 594 #ifdef __XOP__
 595             eweps            = _mm_frcz_pd(ewrt);
 596 #else
 597             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 598 #endif
 599             twoeweps         = _mm_add_pd(eweps,eweps);
 600             ewitab           = _mm_slli_epi32(ewitab,2);
 601             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 602             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 603             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 604             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 605             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 606             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 607             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 608             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 609             velec            = _mm_mul_pd(qq31,_mm_sub_pd(_mm_sub_pd(rinv31,sh_ewald),velec));
 610             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
 611
 612             cutoff_mask      = _mm_cmplt_pd(rsq31,rcutoff2);
 613
 614             /* Update potential sum for this i atom from the interaction with this j atom. */
 615             velec            = _mm_and_pd(velec,cutoff_mask);
 616             velecsum         = _mm_add_pd(velecsum,velec);
 617
 618             fscal            = felec;
 619
 620             fscal            = _mm_and_pd(fscal,cutoff_mask);
 621
 622             /* Update vectorial force */
 623             fix3             = _mm_macc_pd(dx31,fscal,fix3);
 624             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
 625             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
 626
 627             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
 628             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
 629             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
 630
 631             }
 632
 633             /**************************
 634              * CALCULATE INTERACTIONS *
 635              **************************/
 636
 637             if (gmx_mm_any_lt(rsq32,rcutoff2))
 638             {
 639
 640             r32              = _mm_mul_pd(rsq32,rinv32);
 641
 642             /* EWALD ELECTROSTATICS */
 643
 644             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 645             ewrt             = _mm_mul_pd(r32,ewtabscale);
 646             ewitab           = _mm_cvttpd_epi32(ewrt);
 647 #ifdef __XOP__
 648             eweps            = _mm_frcz_pd(ewrt);
 649 #else
 650             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 651 #endif
 652             twoeweps         = _mm_add_pd(eweps,eweps);
 653             ewitab           = _mm_slli_epi32(ewitab,2);
 654             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 655             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 656             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 657             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 658             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 659             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 660             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 661             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 662             velec            = _mm_mul_pd(qq32,_mm_sub_pd(_mm_sub_pd(rinv32,sh_ewald),velec));
 663             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
 664
 665             cutoff_mask      = _mm_cmplt_pd(rsq32,rcutoff2);
 666
 667             /* Update potential sum for this i atom from the interaction with this j atom. */
 668             velec            = _mm_and_pd(velec,cutoff_mask);
 669             velecsum         = _mm_add_pd(velecsum,velec);
 670
 671             fscal            = felec;
 672
 673             fscal            = _mm_and_pd(fscal,cutoff_mask);
 674
 675             /* Update vectorial force */
 676             fix3             = _mm_macc_pd(dx32,fscal,fix3);
 677             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
 678             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
 679
 680             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
 681             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
 682             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
 683
 684             }
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             if (gmx_mm_any_lt(rsq33,rcutoff2))
 691             {
 692
 693             r33              = _mm_mul_pd(rsq33,rinv33);
 694
 695             /* EWALD ELECTROSTATICS */
 696
 697             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 698             ewrt             = _mm_mul_pd(r33,ewtabscale);
 699             ewitab           = _mm_cvttpd_epi32(ewrt);
 700 #ifdef __XOP__
 701             eweps            = _mm_frcz_pd(ewrt);
 702 #else
 703             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 704 #endif
 705             twoeweps         = _mm_add_pd(eweps,eweps);
 706             ewitab           = _mm_slli_epi32(ewitab,2);
 707             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 708             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 709             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 710             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 711             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 712             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 713             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 714             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 715             velec            = _mm_mul_pd(qq33,_mm_sub_pd(_mm_sub_pd(rinv33,sh_ewald),velec));
 716             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
 717
 718             cutoff_mask      = _mm_cmplt_pd(rsq33,rcutoff2);
 719
 720             /* Update potential sum for this i atom from the interaction with this j atom. */
 721             velec            = _mm_and_pd(velec,cutoff_mask);
 722             velecsum         = _mm_add_pd(velecsum,velec);
 723
 724             fscal            = felec;
 725
 726             fscal            = _mm_and_pd(fscal,cutoff_mask);
 727
 728             /* Update vectorial force */
 729             fix3             = _mm_macc_pd(dx33,fscal,fix3);
 730             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
 731             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
 732
 733             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
 734             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
 735             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
 736
 737             }
 738
 739             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);
 740
 741             /* Inner loop uses 441 flops */
 742         }
 743
 744         if(jidx<j_index_end)
 745         {
 746
 747             jnrA             = jjnr[jidx];
 748             j_coord_offsetA  = DIM*jnrA;
 749
 750             /* load j atom coordinates */
 751             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
 752                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 753
 754             /* Calculate displacement vector */
 755             dx11             = _mm_sub_pd(ix1,jx1);
 756             dy11             = _mm_sub_pd(iy1,jy1);
 757             dz11             = _mm_sub_pd(iz1,jz1);
 758             dx12             = _mm_sub_pd(ix1,jx2);
 759             dy12             = _mm_sub_pd(iy1,jy2);
 760             dz12             = _mm_sub_pd(iz1,jz2);
 761             dx13             = _mm_sub_pd(ix1,jx3);
 762             dy13             = _mm_sub_pd(iy1,jy3);
 763             dz13             = _mm_sub_pd(iz1,jz3);
 764             dx21             = _mm_sub_pd(ix2,jx1);
 765             dy21             = _mm_sub_pd(iy2,jy1);
 766             dz21             = _mm_sub_pd(iz2,jz1);
 767             dx22             = _mm_sub_pd(ix2,jx2);
 768             dy22             = _mm_sub_pd(iy2,jy2);
 769             dz22             = _mm_sub_pd(iz2,jz2);
 770             dx23             = _mm_sub_pd(ix2,jx3);
 771             dy23             = _mm_sub_pd(iy2,jy3);
 772             dz23             = _mm_sub_pd(iz2,jz3);
 773             dx31             = _mm_sub_pd(ix3,jx1);
 774             dy31             = _mm_sub_pd(iy3,jy1);
 775             dz31             = _mm_sub_pd(iz3,jz1);
 776             dx32             = _mm_sub_pd(ix3,jx2);
 777             dy32             = _mm_sub_pd(iy3,jy2);
 778             dz32             = _mm_sub_pd(iz3,jz2);
 779             dx33             = _mm_sub_pd(ix3,jx3);
 780             dy33             = _mm_sub_pd(iy3,jy3);
 781             dz33             = _mm_sub_pd(iz3,jz3);
 782
 783             /* Calculate squared distance and things based on it */
 784             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 785             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 786             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 787             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 788             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 789             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 790             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 791             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 792             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 793
 794             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 795             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 796             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 797             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 798             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 799             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 800             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 801             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 802             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 803
 804             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 805             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 806             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 807             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 808             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 809             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 810             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 811             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 812             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 813
 814             fjx1             = _mm_setzero_pd();
 815             fjy1             = _mm_setzero_pd();
 816             fjz1             = _mm_setzero_pd();
 817             fjx2             = _mm_setzero_pd();
 818             fjy2             = _mm_setzero_pd();
 819             fjz2             = _mm_setzero_pd();
 820             fjx3             = _mm_setzero_pd();
 821             fjy3             = _mm_setzero_pd();
 822             fjz3             = _mm_setzero_pd();
 823
 824             /**************************
 825              * CALCULATE INTERACTIONS *
 826              **************************/
 827
 828             if (gmx_mm_any_lt(rsq11,rcutoff2))
 829             {
 830
 831             r11              = _mm_mul_pd(rsq11,rinv11);
 832
 833             /* EWALD ELECTROSTATICS */
 834
 835             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 836             ewrt             = _mm_mul_pd(r11,ewtabscale);
 837             ewitab           = _mm_cvttpd_epi32(ewrt);
 838 #ifdef __XOP__
 839             eweps            = _mm_frcz_pd(ewrt);
 840 #else
 841             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 842 #endif
 843             twoeweps         = _mm_add_pd(eweps,eweps);
 844             ewitab           = _mm_slli_epi32(ewitab,2);
 845             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 846             ewtabD           = _mm_setzero_pd();
 847             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 848             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 849             ewtabFn          = _mm_setzero_pd();
 850             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 851             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 852             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 853             velec            = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
 854             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 855
 856             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
 857
 858             /* Update potential sum for this i atom from the interaction with this j atom. */
 859             velec            = _mm_and_pd(velec,cutoff_mask);
 860             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 861             velecsum         = _mm_add_pd(velecsum,velec);
 862
 863             fscal            = felec;
 864
 865             fscal            = _mm_and_pd(fscal,cutoff_mask);
 866
 867             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 868
 869             /* Update vectorial force */
 870             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 871             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 872             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 873
 874             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 875             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 876             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 877
 878             }
 879
 880             /**************************
 881              * CALCULATE INTERACTIONS *
 882              **************************/
 883
 884             if (gmx_mm_any_lt(rsq12,rcutoff2))
 885             {
 886
 887             r12              = _mm_mul_pd(rsq12,rinv12);
 888
 889             /* EWALD ELECTROSTATICS */
 890
 891             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 892             ewrt             = _mm_mul_pd(r12,ewtabscale);
 893             ewitab           = _mm_cvttpd_epi32(ewrt);
 894 #ifdef __XOP__
 895             eweps            = _mm_frcz_pd(ewrt);
 896 #else
 897             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 898 #endif
 899             twoeweps         = _mm_add_pd(eweps,eweps);
 900             ewitab           = _mm_slli_epi32(ewitab,2);
 901             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 902             ewtabD           = _mm_setzero_pd();
 903             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 904             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 905             ewtabFn          = _mm_setzero_pd();
 906             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 907             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 908             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 909             velec            = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
 910             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 911
 912             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
 913
 914             /* Update potential sum for this i atom from the interaction with this j atom. */
 915             velec            = _mm_and_pd(velec,cutoff_mask);
 916             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 917             velecsum         = _mm_add_pd(velecsum,velec);
 918
 919             fscal            = felec;
 920
 921             fscal            = _mm_and_pd(fscal,cutoff_mask);
 922
 923             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 924
 925             /* Update vectorial force */
 926             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 927             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 928             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 929
 930             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 931             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 932             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 933
 934             }
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             if (gmx_mm_any_lt(rsq13,rcutoff2))
 941             {
 942
 943             r13              = _mm_mul_pd(rsq13,rinv13);
 944
 945             /* EWALD ELECTROSTATICS */
 946
 947             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 948             ewrt             = _mm_mul_pd(r13,ewtabscale);
 949             ewitab           = _mm_cvttpd_epi32(ewrt);
 950 #ifdef __XOP__
 951             eweps            = _mm_frcz_pd(ewrt);
 952 #else
 953             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 954 #endif
 955             twoeweps         = _mm_add_pd(eweps,eweps);
 956             ewitab           = _mm_slli_epi32(ewitab,2);
 957             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 958             ewtabD           = _mm_setzero_pd();
 959             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 960             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 961             ewtabFn          = _mm_setzero_pd();
 962             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 963             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 964             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 965             velec            = _mm_mul_pd(qq13,_mm_sub_pd(_mm_sub_pd(rinv13,sh_ewald),velec));
 966             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 967
 968             cutoff_mask      = _mm_cmplt_pd(rsq13,rcutoff2);
 969
 970             /* Update potential sum for this i atom from the interaction with this j atom. */
 971             velec            = _mm_and_pd(velec,cutoff_mask);
 972             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 973             velecsum         = _mm_add_pd(velecsum,velec);
 974
 975             fscal            = felec;
 976
 977             fscal            = _mm_and_pd(fscal,cutoff_mask);
 978
 979             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 980
 981             /* Update vectorial force */
 982             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 983             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 984             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 985
 986             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 987             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 988             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 989
 990             }
 991
 992             /**************************
 993              * CALCULATE INTERACTIONS *
 994              **************************/
 995
 996             if (gmx_mm_any_lt(rsq21,rcutoff2))
 997             {
 998
 999             r21              = _mm_mul_pd(rsq21,rinv21);
1000
1001             /* EWALD ELECTROSTATICS */
1002
1003             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1004             ewrt             = _mm_mul_pd(r21,ewtabscale);
1005             ewitab           = _mm_cvttpd_epi32(ewrt);
1006 #ifdef __XOP__
1007             eweps            = _mm_frcz_pd(ewrt);
1008 #else
1009             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1010 #endif
1011             twoeweps         = _mm_add_pd(eweps,eweps);
1012             ewitab           = _mm_slli_epi32(ewitab,2);
1013             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1014             ewtabD           = _mm_setzero_pd();
1015             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1016             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1017             ewtabFn          = _mm_setzero_pd();
1018             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1019             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1020             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1021             velec            = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
1022             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1023
1024             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
1025
1026             /* Update potential sum for this i atom from the interaction with this j atom. */
1027             velec            = _mm_and_pd(velec,cutoff_mask);
1028             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1029             velecsum         = _mm_add_pd(velecsum,velec);
1030
1031             fscal            = felec;
1032
1033             fscal            = _mm_and_pd(fscal,cutoff_mask);
1034
1035             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1036
1037             /* Update vectorial force */
1038             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1039             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1040             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1041
1042             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1043             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1044             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1045
1046             }
1047
1048             /**************************
1049              * CALCULATE INTERACTIONS *
1050              **************************/
1051
1052             if (gmx_mm_any_lt(rsq22,rcutoff2))
1053             {
1054
1055             r22              = _mm_mul_pd(rsq22,rinv22);
1056
1057             /* EWALD ELECTROSTATICS */
1058
1059             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1060             ewrt             = _mm_mul_pd(r22,ewtabscale);
1061             ewitab           = _mm_cvttpd_epi32(ewrt);
1062 #ifdef __XOP__
1063             eweps            = _mm_frcz_pd(ewrt);
1064 #else
1065             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1066 #endif
1067             twoeweps         = _mm_add_pd(eweps,eweps);
1068             ewitab           = _mm_slli_epi32(ewitab,2);
1069             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1070             ewtabD           = _mm_setzero_pd();
1071             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1072             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1073             ewtabFn          = _mm_setzero_pd();
1074             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1075             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1076             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1077             velec            = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
1078             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1079
1080             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
1081
1082             /* Update potential sum for this i atom from the interaction with this j atom. */
1083             velec            = _mm_and_pd(velec,cutoff_mask);
1084             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1085             velecsum         = _mm_add_pd(velecsum,velec);
1086
1087             fscal            = felec;
1088
1089             fscal            = _mm_and_pd(fscal,cutoff_mask);
1090
1091             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1092
1093             /* Update vectorial force */
1094             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1095             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1096             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1097
1098             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1099             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1100             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1101
1102             }
1103
1104             /**************************
1105              * CALCULATE INTERACTIONS *
1106              **************************/
1107
1108             if (gmx_mm_any_lt(rsq23,rcutoff2))
1109             {
1110
1111             r23              = _mm_mul_pd(rsq23,rinv23);
1112
1113             /* EWALD ELECTROSTATICS */
1114
1115             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1116             ewrt             = _mm_mul_pd(r23,ewtabscale);
1117             ewitab           = _mm_cvttpd_epi32(ewrt);
1118 #ifdef __XOP__
1119             eweps            = _mm_frcz_pd(ewrt);
1120 #else
1121             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1122 #endif
1123             twoeweps         = _mm_add_pd(eweps,eweps);
1124             ewitab           = _mm_slli_epi32(ewitab,2);
1125             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1126             ewtabD           = _mm_setzero_pd();
1127             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1128             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1129             ewtabFn          = _mm_setzero_pd();
1130             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1131             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1132             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1133             velec            = _mm_mul_pd(qq23,_mm_sub_pd(_mm_sub_pd(rinv23,sh_ewald),velec));
1134             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1135
1136             cutoff_mask      = _mm_cmplt_pd(rsq23,rcutoff2);
1137
1138             /* Update potential sum for this i atom from the interaction with this j atom. */
1139             velec            = _mm_and_pd(velec,cutoff_mask);
1140             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1141             velecsum         = _mm_add_pd(velecsum,velec);
1142
1143             fscal            = felec;
1144
1145             fscal            = _mm_and_pd(fscal,cutoff_mask);
1146
1147             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1148
1149             /* Update vectorial force */
1150             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1151             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1152             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1153
1154             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1155             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1156             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1157
1158             }
1159
1160             /**************************
1161              * CALCULATE INTERACTIONS *
1162              **************************/
1163
1164             if (gmx_mm_any_lt(rsq31,rcutoff2))
1165             {
1166
1167             r31              = _mm_mul_pd(rsq31,rinv31);
1168
1169             /* EWALD ELECTROSTATICS */
1170
1171             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1172             ewrt             = _mm_mul_pd(r31,ewtabscale);
1173             ewitab           = _mm_cvttpd_epi32(ewrt);
1174 #ifdef __XOP__
1175             eweps            = _mm_frcz_pd(ewrt);
1176 #else
1177             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1178 #endif
1179             twoeweps         = _mm_add_pd(eweps,eweps);
1180             ewitab           = _mm_slli_epi32(ewitab,2);
1181             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1182             ewtabD           = _mm_setzero_pd();
1183             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1184             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1185             ewtabFn          = _mm_setzero_pd();
1186             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1187             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1188             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1189             velec            = _mm_mul_pd(qq31,_mm_sub_pd(_mm_sub_pd(rinv31,sh_ewald),velec));
1190             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1191
1192             cutoff_mask      = _mm_cmplt_pd(rsq31,rcutoff2);
1193
1194             /* Update potential sum for this i atom from the interaction with this j atom. */
1195             velec            = _mm_and_pd(velec,cutoff_mask);
1196             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1197             velecsum         = _mm_add_pd(velecsum,velec);
1198
1199             fscal            = felec;
1200
1201             fscal            = _mm_and_pd(fscal,cutoff_mask);
1202
1203             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1204
1205             /* Update vectorial force */
1206             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1207             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1208             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1209
1210             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1211             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1212             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1213
1214             }
1215
1216             /**************************
1217              * CALCULATE INTERACTIONS *
1218              **************************/
1219
1220             if (gmx_mm_any_lt(rsq32,rcutoff2))
1221             {
1222
1223             r32              = _mm_mul_pd(rsq32,rinv32);
1224
1225             /* EWALD ELECTROSTATICS */
1226
1227             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1228             ewrt             = _mm_mul_pd(r32,ewtabscale);
1229             ewitab           = _mm_cvttpd_epi32(ewrt);
1230 #ifdef __XOP__
1231             eweps            = _mm_frcz_pd(ewrt);
1232 #else
1233             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1234 #endif
1235             twoeweps         = _mm_add_pd(eweps,eweps);
1236             ewitab           = _mm_slli_epi32(ewitab,2);
1237             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1238             ewtabD           = _mm_setzero_pd();
1239             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1240             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1241             ewtabFn          = _mm_setzero_pd();
1242             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1243             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1244             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1245             velec            = _mm_mul_pd(qq32,_mm_sub_pd(_mm_sub_pd(rinv32,sh_ewald),velec));
1246             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1247
1248             cutoff_mask      = _mm_cmplt_pd(rsq32,rcutoff2);
1249
1250             /* Update potential sum for this i atom from the interaction with this j atom. */
1251             velec            = _mm_and_pd(velec,cutoff_mask);
1252             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1253             velecsum         = _mm_add_pd(velecsum,velec);
1254
1255             fscal            = felec;
1256
1257             fscal            = _mm_and_pd(fscal,cutoff_mask);
1258
1259             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1260
1261             /* Update vectorial force */
1262             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1263             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1264             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1265
1266             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1267             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1268             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1269
1270             }
1271
1272             /**************************
1273              * CALCULATE INTERACTIONS *
1274              **************************/
1275
1276             if (gmx_mm_any_lt(rsq33,rcutoff2))
1277             {
1278
1279             r33              = _mm_mul_pd(rsq33,rinv33);
1280
1281             /* EWALD ELECTROSTATICS */
1282
1283             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1284             ewrt             = _mm_mul_pd(r33,ewtabscale);
1285             ewitab           = _mm_cvttpd_epi32(ewrt);
1286 #ifdef __XOP__
1287             eweps            = _mm_frcz_pd(ewrt);
1288 #else
1289             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1290 #endif
1291             twoeweps         = _mm_add_pd(eweps,eweps);
1292             ewitab           = _mm_slli_epi32(ewitab,2);
1293             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1294             ewtabD           = _mm_setzero_pd();
1295             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1296             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1297             ewtabFn          = _mm_setzero_pd();
1298             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1299             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1300             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1301             velec            = _mm_mul_pd(qq33,_mm_sub_pd(_mm_sub_pd(rinv33,sh_ewald),velec));
1302             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1303
1304             cutoff_mask      = _mm_cmplt_pd(rsq33,rcutoff2);
1305
1306             /* Update potential sum for this i atom from the interaction with this j atom. */
1307             velec            = _mm_and_pd(velec,cutoff_mask);
1308             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1309             velecsum         = _mm_add_pd(velecsum,velec);
1310
1311             fscal            = felec;
1312
1313             fscal            = _mm_and_pd(fscal,cutoff_mask);
1314
1315             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1316
1317             /* Update vectorial force */
1318             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1319             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1320             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1321
1322             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1323             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1324             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1325
1326             }
1327
1328             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1329
1330             /* Inner loop uses 441 flops */
1331         }
1332
1333         /* End of innermost loop */
1334
1335         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1336                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1337
1338         ggid                        = gid[iidx];
1339         /* Update potential energies */
1340         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1341
1342         /* Increment number of inner iterations */
1343         inneriter                  += j_index_end - j_index_start;
1344
1345         /* Outer loop uses 19 flops */
1346     }
1347
1348     /* Increment number of outer iterations */
1349     outeriter        += nri;
1350
1351     /* Update outer/inner flops */
1352
1353     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*441);
1354 }
1355 /*
1356  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4W4_F_avx_128_fma_double
1357  * Electrostatics interaction: Ewald
1358  * VdW interaction:            None
1359  * Geometry:                   Water4-Water4
1360  * Calculate force/pot:        Force
1361  */
1362 void
1363 nb_kernel_ElecEwSh_VdwNone_GeomW4W4_F_avx_128_fma_double
1364                     (t_nblist * gmx_restrict                nlist,
1365                      rvec * gmx_restrict                    xx,
1366                      rvec * gmx_restrict                    ff,
1367                      t_forcerec * gmx_restrict              fr,
1368                      t_mdatoms * gmx_restrict               mdatoms,
1369                      nb_kernel_data_t * gmx_restrict        kernel_data,
1370                      t_nrnb * gmx_restrict                  nrnb)
1371 {
1372     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1373      * just 0 for non-waters.
1374      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1375      * jnr indices corresponding to data put in the four positions in the SIMD register.
1376      */
1377     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1378     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1379     int              jnrA,jnrB;
1380     int              j_coord_offsetA,j_coord_offsetB;
1381     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1382     real             rcutoff_scalar;
1383     real             *shiftvec,*fshift,*x,*f;
1384     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1385     int              vdwioffset1;
1386     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1387     int              vdwioffset2;
1388     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1389     int              vdwioffset3;
1390     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1391     int              vdwjidx1A,vdwjidx1B;
1392     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1393     int              vdwjidx2A,vdwjidx2B;
1394     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1395     int              vdwjidx3A,vdwjidx3B;
1396     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1397     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1398     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1399     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1400     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1401     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1402     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1403     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1404     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1405     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1406     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1407     real             *charge;
1408     __m128i          ewitab;
1409     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1410     real             *ewtab;
1411     __m128d          dummy_mask,cutoff_mask;
1412     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1413     __m128d          one     = _mm_set1_pd(1.0);
1414     __m128d          two     = _mm_set1_pd(2.0);
1415     x                = xx[0];
1416     f                = ff[0];
1417
1418     nri              = nlist->nri;
1419     iinr             = nlist->iinr;
1420     jindex           = nlist->jindex;
1421     jjnr             = nlist->jjnr;
1422     shiftidx         = nlist->shift;
1423     gid              = nlist->gid;
1424     shiftvec         = fr->shift_vec[0];
1425     fshift           = fr->fshift[0];
1426     facel            = _mm_set1_pd(fr->epsfac);
1427     charge           = mdatoms->chargeA;
1428
1429     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1430     ewtab            = fr->ic->tabq_coul_F;
1431     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1432     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1433
1434     /* Setup water-specific parameters */
1435     inr              = nlist->iinr[0];
1436     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1437     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1438     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1439
1440     jq1              = _mm_set1_pd(charge[inr+1]);
1441     jq2              = _mm_set1_pd(charge[inr+2]);
1442     jq3              = _mm_set1_pd(charge[inr+3]);
1443     qq11             = _mm_mul_pd(iq1,jq1);
1444     qq12             = _mm_mul_pd(iq1,jq2);
1445     qq13             = _mm_mul_pd(iq1,jq3);
1446     qq21             = _mm_mul_pd(iq2,jq1);
1447     qq22             = _mm_mul_pd(iq2,jq2);
1448     qq23             = _mm_mul_pd(iq2,jq3);
1449     qq31             = _mm_mul_pd(iq3,jq1);
1450     qq32             = _mm_mul_pd(iq3,jq2);
1451     qq33             = _mm_mul_pd(iq3,jq3);
1452
1453     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1454     rcutoff_scalar   = fr->rcoulomb;
1455     rcutoff          = _mm_set1_pd(rcutoff_scalar);
1456     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
1457
1458     /* Avoid stupid compiler warnings */
1459     jnrA = jnrB = 0;
1460     j_coord_offsetA = 0;
1461     j_coord_offsetB = 0;
1462
1463     outeriter        = 0;
1464     inneriter        = 0;
1465
1466     /* Start outer loop over neighborlists */
1467     for(iidx=0; iidx<nri; iidx++)
1468     {
1469         /* Load shift vector for this list */
1470         i_shift_offset   = DIM*shiftidx[iidx];
1471
1472         /* Load limits for loop over neighbors */
1473         j_index_start    = jindex[iidx];
1474         j_index_end      = jindex[iidx+1];
1475
1476         /* Get outer coordinate index */
1477         inr              = iinr[iidx];
1478         i_coord_offset   = DIM*inr;
1479
1480         /* Load i particle coords and add shift vector */
1481         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1482                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1483
1484         fix1             = _mm_setzero_pd();
1485         fiy1             = _mm_setzero_pd();
1486         fiz1             = _mm_setzero_pd();
1487         fix2             = _mm_setzero_pd();
1488         fiy2             = _mm_setzero_pd();
1489         fiz2             = _mm_setzero_pd();
1490         fix3             = _mm_setzero_pd();
1491         fiy3             = _mm_setzero_pd();
1492         fiz3             = _mm_setzero_pd();
1493
1494         /* Start inner kernel loop */
1495         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1496         {
1497
1498             /* Get j neighbor index, and coordinate index */
1499             jnrA             = jjnr[jidx];
1500             jnrB             = jjnr[jidx+1];
1501             j_coord_offsetA  = DIM*jnrA;
1502             j_coord_offsetB  = DIM*jnrB;
1503
1504             /* load j atom coordinates */
1505             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1506                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1507
1508             /* Calculate displacement vector */
1509             dx11             = _mm_sub_pd(ix1,jx1);
1510             dy11             = _mm_sub_pd(iy1,jy1);
1511             dz11             = _mm_sub_pd(iz1,jz1);
1512             dx12             = _mm_sub_pd(ix1,jx2);
1513             dy12             = _mm_sub_pd(iy1,jy2);
1514             dz12             = _mm_sub_pd(iz1,jz2);
1515             dx13             = _mm_sub_pd(ix1,jx3);
1516             dy13             = _mm_sub_pd(iy1,jy3);
1517             dz13             = _mm_sub_pd(iz1,jz3);
1518             dx21             = _mm_sub_pd(ix2,jx1);
1519             dy21             = _mm_sub_pd(iy2,jy1);
1520             dz21             = _mm_sub_pd(iz2,jz1);
1521             dx22             = _mm_sub_pd(ix2,jx2);
1522             dy22             = _mm_sub_pd(iy2,jy2);
1523             dz22             = _mm_sub_pd(iz2,jz2);
1524             dx23             = _mm_sub_pd(ix2,jx3);
1525             dy23             = _mm_sub_pd(iy2,jy3);
1526             dz23             = _mm_sub_pd(iz2,jz3);
1527             dx31             = _mm_sub_pd(ix3,jx1);
1528             dy31             = _mm_sub_pd(iy3,jy1);
1529             dz31             = _mm_sub_pd(iz3,jz1);
1530             dx32             = _mm_sub_pd(ix3,jx2);
1531             dy32             = _mm_sub_pd(iy3,jy2);
1532             dz32             = _mm_sub_pd(iz3,jz2);
1533             dx33             = _mm_sub_pd(ix3,jx3);
1534             dy33             = _mm_sub_pd(iy3,jy3);
1535             dz33             = _mm_sub_pd(iz3,jz3);
1536
1537             /* Calculate squared distance and things based on it */
1538             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1539             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1540             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1541             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1542             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1543             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1544             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1545             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1546             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1547
1548             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1549             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1550             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1551             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1552             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1553             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1554             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1555             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1556             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1557
1558             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1559             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1560             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1561             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1562             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1563             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1564             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1565             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1566             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1567
1568             fjx1             = _mm_setzero_pd();
1569             fjy1             = _mm_setzero_pd();
1570             fjz1             = _mm_setzero_pd();
1571             fjx2             = _mm_setzero_pd();
1572             fjy2             = _mm_setzero_pd();
1573             fjz2             = _mm_setzero_pd();
1574             fjx3             = _mm_setzero_pd();
1575             fjy3             = _mm_setzero_pd();
1576             fjz3             = _mm_setzero_pd();
1577
1578             /**************************
1579              * CALCULATE INTERACTIONS *
1580              **************************/
1581
1582             if (gmx_mm_any_lt(rsq11,rcutoff2))
1583             {
1584
1585             r11              = _mm_mul_pd(rsq11,rinv11);
1586
1587             /* EWALD ELECTROSTATICS */
1588
1589             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1590             ewrt             = _mm_mul_pd(r11,ewtabscale);
1591             ewitab           = _mm_cvttpd_epi32(ewrt);
1592 #ifdef __XOP__
1593             eweps            = _mm_frcz_pd(ewrt);
1594 #else
1595             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1596 #endif
1597             twoeweps         = _mm_add_pd(eweps,eweps);
1598             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1599                                          &ewtabF,&ewtabFn);
1600             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1601             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1602
1603             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
1604
1605             fscal            = felec;
1606
1607             fscal            = _mm_and_pd(fscal,cutoff_mask);
1608
1609             /* Update vectorial force */
1610             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1611             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1612             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1613
1614             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1615             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1616             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1617
1618             }
1619
1620             /**************************
1621              * CALCULATE INTERACTIONS *
1622              **************************/
1623
1624             if (gmx_mm_any_lt(rsq12,rcutoff2))
1625             {
1626
1627             r12              = _mm_mul_pd(rsq12,rinv12);
1628
1629             /* EWALD ELECTROSTATICS */
1630
1631             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1632             ewrt             = _mm_mul_pd(r12,ewtabscale);
1633             ewitab           = _mm_cvttpd_epi32(ewrt);
1634 #ifdef __XOP__
1635             eweps            = _mm_frcz_pd(ewrt);
1636 #else
1637             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1638 #endif
1639             twoeweps         = _mm_add_pd(eweps,eweps);
1640             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1641                                          &ewtabF,&ewtabFn);
1642             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1643             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1644
1645             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
1646
1647             fscal            = felec;
1648
1649             fscal            = _mm_and_pd(fscal,cutoff_mask);
1650
1651             /* Update vectorial force */
1652             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1653             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1654             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1655
1656             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1657             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1658             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1659
1660             }
1661
1662             /**************************
1663              * CALCULATE INTERACTIONS *
1664              **************************/
1665
1666             if (gmx_mm_any_lt(rsq13,rcutoff2))
1667             {
1668
1669             r13              = _mm_mul_pd(rsq13,rinv13);
1670
1671             /* EWALD ELECTROSTATICS */
1672
1673             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1674             ewrt             = _mm_mul_pd(r13,ewtabscale);
1675             ewitab           = _mm_cvttpd_epi32(ewrt);
1676 #ifdef __XOP__
1677             eweps            = _mm_frcz_pd(ewrt);
1678 #else
1679             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1680 #endif
1681             twoeweps         = _mm_add_pd(eweps,eweps);
1682             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1683                                          &ewtabF,&ewtabFn);
1684             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1685             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1686
1687             cutoff_mask      = _mm_cmplt_pd(rsq13,rcutoff2);
1688
1689             fscal            = felec;
1690
1691             fscal            = _mm_and_pd(fscal,cutoff_mask);
1692
1693             /* Update vectorial force */
1694             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1695             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1696             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1697
1698             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1699             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1700             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1701
1702             }
1703
1704             /**************************
1705              * CALCULATE INTERACTIONS *
1706              **************************/
1707
1708             if (gmx_mm_any_lt(rsq21,rcutoff2))
1709             {
1710
1711             r21              = _mm_mul_pd(rsq21,rinv21);
1712
1713             /* EWALD ELECTROSTATICS */
1714
1715             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1716             ewrt             = _mm_mul_pd(r21,ewtabscale);
1717             ewitab           = _mm_cvttpd_epi32(ewrt);
1718 #ifdef __XOP__
1719             eweps            = _mm_frcz_pd(ewrt);
1720 #else
1721             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1722 #endif
1723             twoeweps         = _mm_add_pd(eweps,eweps);
1724             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1725                                          &ewtabF,&ewtabFn);
1726             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1727             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1728
1729             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
1730
1731             fscal            = felec;
1732
1733             fscal            = _mm_and_pd(fscal,cutoff_mask);
1734
1735             /* Update vectorial force */
1736             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1737             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1738             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1739
1740             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1741             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1742             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1743
1744             }
1745
1746             /**************************
1747              * CALCULATE INTERACTIONS *
1748              **************************/
1749
1750             if (gmx_mm_any_lt(rsq22,rcutoff2))
1751             {
1752
1753             r22              = _mm_mul_pd(rsq22,rinv22);
1754
1755             /* EWALD ELECTROSTATICS */
1756
1757             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1758             ewrt             = _mm_mul_pd(r22,ewtabscale);
1759             ewitab           = _mm_cvttpd_epi32(ewrt);
1760 #ifdef __XOP__
1761             eweps            = _mm_frcz_pd(ewrt);
1762 #else
1763             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1764 #endif
1765             twoeweps         = _mm_add_pd(eweps,eweps);
1766             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1767                                          &ewtabF,&ewtabFn);
1768             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1769             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1770
1771             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
1772
1773             fscal            = felec;
1774
1775             fscal            = _mm_and_pd(fscal,cutoff_mask);
1776
1777             /* Update vectorial force */
1778             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1779             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1780             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1781
1782             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1783             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1784             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1785
1786             }
1787
1788             /**************************
1789              * CALCULATE INTERACTIONS *
1790              **************************/
1791
1792             if (gmx_mm_any_lt(rsq23,rcutoff2))
1793             {
1794
1795             r23              = _mm_mul_pd(rsq23,rinv23);
1796
1797             /* EWALD ELECTROSTATICS */
1798
1799             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1800             ewrt             = _mm_mul_pd(r23,ewtabscale);
1801             ewitab           = _mm_cvttpd_epi32(ewrt);
1802 #ifdef __XOP__
1803             eweps            = _mm_frcz_pd(ewrt);
1804 #else
1805             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1806 #endif
1807             twoeweps         = _mm_add_pd(eweps,eweps);
1808             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1809                                          &ewtabF,&ewtabFn);
1810             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1811             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1812
1813             cutoff_mask      = _mm_cmplt_pd(rsq23,rcutoff2);
1814
1815             fscal            = felec;
1816
1817             fscal            = _mm_and_pd(fscal,cutoff_mask);
1818
1819             /* Update vectorial force */
1820             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1821             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1822             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1823
1824             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1825             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1826             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1827
1828             }
1829
1830             /**************************
1831              * CALCULATE INTERACTIONS *
1832              **************************/
1833
1834             if (gmx_mm_any_lt(rsq31,rcutoff2))
1835             {
1836
1837             r31              = _mm_mul_pd(rsq31,rinv31);
1838
1839             /* EWALD ELECTROSTATICS */
1840
1841             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1842             ewrt             = _mm_mul_pd(r31,ewtabscale);
1843             ewitab           = _mm_cvttpd_epi32(ewrt);
1844 #ifdef __XOP__
1845             eweps            = _mm_frcz_pd(ewrt);
1846 #else
1847             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1848 #endif
1849             twoeweps         = _mm_add_pd(eweps,eweps);
1850             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1851                                          &ewtabF,&ewtabFn);
1852             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1853             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1854
1855             cutoff_mask      = _mm_cmplt_pd(rsq31,rcutoff2);
1856
1857             fscal            = felec;
1858
1859             fscal            = _mm_and_pd(fscal,cutoff_mask);
1860
1861             /* Update vectorial force */
1862             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1863             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1864             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1865
1866             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1867             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1868             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1869
1870             }
1871
1872             /**************************
1873              * CALCULATE INTERACTIONS *
1874              **************************/
1875
1876             if (gmx_mm_any_lt(rsq32,rcutoff2))
1877             {
1878
1879             r32              = _mm_mul_pd(rsq32,rinv32);
1880
1881             /* EWALD ELECTROSTATICS */
1882
1883             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1884             ewrt             = _mm_mul_pd(r32,ewtabscale);
1885             ewitab           = _mm_cvttpd_epi32(ewrt);
1886 #ifdef __XOP__
1887             eweps            = _mm_frcz_pd(ewrt);
1888 #else
1889             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1890 #endif
1891             twoeweps         = _mm_add_pd(eweps,eweps);
1892             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1893                                          &ewtabF,&ewtabFn);
1894             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1895             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1896
1897             cutoff_mask      = _mm_cmplt_pd(rsq32,rcutoff2);
1898
1899             fscal            = felec;
1900
1901             fscal            = _mm_and_pd(fscal,cutoff_mask);
1902
1903             /* Update vectorial force */
1904             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1905             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1906             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1907
1908             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1909             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1910             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1911
1912             }
1913
1914             /**************************
1915              * CALCULATE INTERACTIONS *
1916              **************************/
1917
1918             if (gmx_mm_any_lt(rsq33,rcutoff2))
1919             {
1920
1921             r33              = _mm_mul_pd(rsq33,rinv33);
1922
1923             /* EWALD ELECTROSTATICS */
1924
1925             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1926             ewrt             = _mm_mul_pd(r33,ewtabscale);
1927             ewitab           = _mm_cvttpd_epi32(ewrt);
1928 #ifdef __XOP__
1929             eweps            = _mm_frcz_pd(ewrt);
1930 #else
1931             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1932 #endif
1933             twoeweps         = _mm_add_pd(eweps,eweps);
1934             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1935                                          &ewtabF,&ewtabFn);
1936             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1937             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1938
1939             cutoff_mask      = _mm_cmplt_pd(rsq33,rcutoff2);
1940
1941             fscal            = felec;
1942
1943             fscal            = _mm_and_pd(fscal,cutoff_mask);
1944
1945             /* Update vectorial force */
1946             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1947             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1948             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1949
1950             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1951             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1952             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1953
1954             }
1955
1956             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);
1957
1958             /* Inner loop uses 378 flops */
1959         }
1960
1961         if(jidx<j_index_end)
1962         {
1963
1964             jnrA             = jjnr[jidx];
1965             j_coord_offsetA  = DIM*jnrA;
1966
1967             /* load j atom coordinates */
1968             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
1969                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1970
1971             /* Calculate displacement vector */
1972             dx11             = _mm_sub_pd(ix1,jx1);
1973             dy11             = _mm_sub_pd(iy1,jy1);
1974             dz11             = _mm_sub_pd(iz1,jz1);
1975             dx12             = _mm_sub_pd(ix1,jx2);
1976             dy12             = _mm_sub_pd(iy1,jy2);
1977             dz12             = _mm_sub_pd(iz1,jz2);
1978             dx13             = _mm_sub_pd(ix1,jx3);
1979             dy13             = _mm_sub_pd(iy1,jy3);
1980             dz13             = _mm_sub_pd(iz1,jz3);
1981             dx21             = _mm_sub_pd(ix2,jx1);
1982             dy21             = _mm_sub_pd(iy2,jy1);
1983             dz21             = _mm_sub_pd(iz2,jz1);
1984             dx22             = _mm_sub_pd(ix2,jx2);
1985             dy22             = _mm_sub_pd(iy2,jy2);
1986             dz22             = _mm_sub_pd(iz2,jz2);
1987             dx23             = _mm_sub_pd(ix2,jx3);
1988             dy23             = _mm_sub_pd(iy2,jy3);
1989             dz23             = _mm_sub_pd(iz2,jz3);
1990             dx31             = _mm_sub_pd(ix3,jx1);
1991             dy31             = _mm_sub_pd(iy3,jy1);
1992             dz31             = _mm_sub_pd(iz3,jz1);
1993             dx32             = _mm_sub_pd(ix3,jx2);
1994             dy32             = _mm_sub_pd(iy3,jy2);
1995             dz32             = _mm_sub_pd(iz3,jz2);
1996             dx33             = _mm_sub_pd(ix3,jx3);
1997             dy33             = _mm_sub_pd(iy3,jy3);
1998             dz33             = _mm_sub_pd(iz3,jz3);
1999
2000             /* Calculate squared distance and things based on it */
2001             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2002             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2003             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
2004             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2005             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2006             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
2007             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
2008             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
2009             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
2010
2011             rinv11           = gmx_mm_invsqrt_pd(rsq11);
2012             rinv12           = gmx_mm_invsqrt_pd(rsq12);
2013             rinv13           = gmx_mm_invsqrt_pd(rsq13);
2014             rinv21           = gmx_mm_invsqrt_pd(rsq21);
2015             rinv22           = gmx_mm_invsqrt_pd(rsq22);
2016             rinv23           = gmx_mm_invsqrt_pd(rsq23);
2017             rinv31           = gmx_mm_invsqrt_pd(rsq31);
2018             rinv32           = gmx_mm_invsqrt_pd(rsq32);
2019             rinv33           = gmx_mm_invsqrt_pd(rsq33);
2020
2021             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
2022             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
2023             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
2024             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
2025             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
2026             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
2027             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
2028             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
2029             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
2030
2031             fjx1             = _mm_setzero_pd();
2032             fjy1             = _mm_setzero_pd();
2033             fjz1             = _mm_setzero_pd();
2034             fjx2             = _mm_setzero_pd();
2035             fjy2             = _mm_setzero_pd();
2036             fjz2             = _mm_setzero_pd();
2037             fjx3             = _mm_setzero_pd();
2038             fjy3             = _mm_setzero_pd();
2039             fjz3             = _mm_setzero_pd();
2040
2041             /**************************
2042              * CALCULATE INTERACTIONS *
2043              **************************/
2044
2045             if (gmx_mm_any_lt(rsq11,rcutoff2))
2046             {
2047
2048             r11              = _mm_mul_pd(rsq11,rinv11);
2049
2050             /* EWALD ELECTROSTATICS */
2051
2052             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2053             ewrt             = _mm_mul_pd(r11,ewtabscale);
2054             ewitab           = _mm_cvttpd_epi32(ewrt);
2055 #ifdef __XOP__
2056             eweps            = _mm_frcz_pd(ewrt);
2057 #else
2058             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2059 #endif
2060             twoeweps         = _mm_add_pd(eweps,eweps);
2061             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2062             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2063             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2064
2065             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
2066
2067             fscal            = felec;
2068
2069             fscal            = _mm_and_pd(fscal,cutoff_mask);
2070
2071             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2072
2073             /* Update vectorial force */
2074             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2075             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2076             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2077
2078             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2079             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2080             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2081
2082             }
2083
2084             /**************************
2085              * CALCULATE INTERACTIONS *
2086              **************************/
2087
2088             if (gmx_mm_any_lt(rsq12,rcutoff2))
2089             {
2090
2091             r12              = _mm_mul_pd(rsq12,rinv12);
2092
2093             /* EWALD ELECTROSTATICS */
2094
2095             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2096             ewrt             = _mm_mul_pd(r12,ewtabscale);
2097             ewitab           = _mm_cvttpd_epi32(ewrt);
2098 #ifdef __XOP__
2099             eweps            = _mm_frcz_pd(ewrt);
2100 #else
2101             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2102 #endif
2103             twoeweps         = _mm_add_pd(eweps,eweps);
2104             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2105             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2106             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2107
2108             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
2109
2110             fscal            = felec;
2111
2112             fscal            = _mm_and_pd(fscal,cutoff_mask);
2113
2114             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2115
2116             /* Update vectorial force */
2117             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2118             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2119             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2120
2121             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2122             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2123             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2124
2125             }
2126
2127             /**************************
2128              * CALCULATE INTERACTIONS *
2129              **************************/
2130
2131             if (gmx_mm_any_lt(rsq13,rcutoff2))
2132             {
2133
2134             r13              = _mm_mul_pd(rsq13,rinv13);
2135
2136             /* EWALD ELECTROSTATICS */
2137
2138             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2139             ewrt             = _mm_mul_pd(r13,ewtabscale);
2140             ewitab           = _mm_cvttpd_epi32(ewrt);
2141 #ifdef __XOP__
2142             eweps            = _mm_frcz_pd(ewrt);
2143 #else
2144             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2145 #endif
2146             twoeweps         = _mm_add_pd(eweps,eweps);
2147             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2148             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2149             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
2150
2151             cutoff_mask      = _mm_cmplt_pd(rsq13,rcutoff2);
2152
2153             fscal            = felec;
2154
2155             fscal            = _mm_and_pd(fscal,cutoff_mask);
2156
2157             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2158
2159             /* Update vectorial force */
2160             fix1             = _mm_macc_pd(dx13,fscal,fix1);
2161             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
2162             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
2163
2164             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
2165             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
2166             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
2167
2168             }
2169
2170             /**************************
2171              * CALCULATE INTERACTIONS *
2172              **************************/
2173
2174             if (gmx_mm_any_lt(rsq21,rcutoff2))
2175             {
2176
2177             r21              = _mm_mul_pd(rsq21,rinv21);
2178
2179             /* EWALD ELECTROSTATICS */
2180
2181             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2182             ewrt             = _mm_mul_pd(r21,ewtabscale);
2183             ewitab           = _mm_cvttpd_epi32(ewrt);
2184 #ifdef __XOP__
2185             eweps            = _mm_frcz_pd(ewrt);
2186 #else
2187             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2188 #endif
2189             twoeweps         = _mm_add_pd(eweps,eweps);
2190             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2191             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2192             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2193
2194             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
2195
2196             fscal            = felec;
2197
2198             fscal            = _mm_and_pd(fscal,cutoff_mask);
2199
2200             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2201
2202             /* Update vectorial force */
2203             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2204             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2205             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2206
2207             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2208             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2209             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2210
2211             }
2212
2213             /**************************
2214              * CALCULATE INTERACTIONS *
2215              **************************/
2216
2217             if (gmx_mm_any_lt(rsq22,rcutoff2))
2218             {
2219
2220             r22              = _mm_mul_pd(rsq22,rinv22);
2221
2222             /* EWALD ELECTROSTATICS */
2223
2224             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2225             ewrt             = _mm_mul_pd(r22,ewtabscale);
2226             ewitab           = _mm_cvttpd_epi32(ewrt);
2227 #ifdef __XOP__
2228             eweps            = _mm_frcz_pd(ewrt);
2229 #else
2230             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2231 #endif
2232             twoeweps         = _mm_add_pd(eweps,eweps);
2233             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2234             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2235             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2236
2237             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
2238
2239             fscal            = felec;
2240
2241             fscal            = _mm_and_pd(fscal,cutoff_mask);
2242
2243             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2244
2245             /* Update vectorial force */
2246             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2247             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2248             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2249
2250             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2251             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2252             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2253
2254             }
2255
2256             /**************************
2257              * CALCULATE INTERACTIONS *
2258              **************************/
2259
2260             if (gmx_mm_any_lt(rsq23,rcutoff2))
2261             {
2262
2263             r23              = _mm_mul_pd(rsq23,rinv23);
2264
2265             /* EWALD ELECTROSTATICS */
2266
2267             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2268             ewrt             = _mm_mul_pd(r23,ewtabscale);
2269             ewitab           = _mm_cvttpd_epi32(ewrt);
2270 #ifdef __XOP__
2271             eweps            = _mm_frcz_pd(ewrt);
2272 #else
2273             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2274 #endif
2275             twoeweps         = _mm_add_pd(eweps,eweps);
2276             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2277             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2278             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
2279
2280             cutoff_mask      = _mm_cmplt_pd(rsq23,rcutoff2);
2281
2282             fscal            = felec;
2283
2284             fscal            = _mm_and_pd(fscal,cutoff_mask);
2285
2286             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2287
2288             /* Update vectorial force */
2289             fix2             = _mm_macc_pd(dx23,fscal,fix2);
2290             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
2291             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
2292
2293             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
2294             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
2295             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
2296
2297             }
2298
2299             /**************************
2300              * CALCULATE INTERACTIONS *
2301              **************************/
2302
2303             if (gmx_mm_any_lt(rsq31,rcutoff2))
2304             {
2305
2306             r31              = _mm_mul_pd(rsq31,rinv31);
2307
2308             /* EWALD ELECTROSTATICS */
2309
2310             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2311             ewrt             = _mm_mul_pd(r31,ewtabscale);
2312             ewitab           = _mm_cvttpd_epi32(ewrt);
2313 #ifdef __XOP__
2314             eweps            = _mm_frcz_pd(ewrt);
2315 #else
2316             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2317 #endif
2318             twoeweps         = _mm_add_pd(eweps,eweps);
2319             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2320             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2321             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
2322
2323             cutoff_mask      = _mm_cmplt_pd(rsq31,rcutoff2);
2324
2325             fscal            = felec;
2326
2327             fscal            = _mm_and_pd(fscal,cutoff_mask);
2328
2329             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2330
2331             /* Update vectorial force */
2332             fix3             = _mm_macc_pd(dx31,fscal,fix3);
2333             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
2334             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
2335
2336             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
2337             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
2338             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
2339
2340             }
2341
2342             /**************************
2343              * CALCULATE INTERACTIONS *
2344              **************************/
2345
2346             if (gmx_mm_any_lt(rsq32,rcutoff2))
2347             {
2348
2349             r32              = _mm_mul_pd(rsq32,rinv32);
2350
2351             /* EWALD ELECTROSTATICS */
2352
2353             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2354             ewrt             = _mm_mul_pd(r32,ewtabscale);
2355             ewitab           = _mm_cvttpd_epi32(ewrt);
2356 #ifdef __XOP__
2357             eweps            = _mm_frcz_pd(ewrt);
2358 #else
2359             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2360 #endif
2361             twoeweps         = _mm_add_pd(eweps,eweps);
2362             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2363             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2364             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2365
2366             cutoff_mask      = _mm_cmplt_pd(rsq32,rcutoff2);
2367
2368             fscal            = felec;
2369
2370             fscal            = _mm_and_pd(fscal,cutoff_mask);
2371
2372             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2373
2374             /* Update vectorial force */
2375             fix3             = _mm_macc_pd(dx32,fscal,fix3);
2376             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
2377             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
2378
2379             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
2380             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
2381             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
2382
2383             }
2384
2385             /**************************
2386              * CALCULATE INTERACTIONS *
2387              **************************/
2388
2389             if (gmx_mm_any_lt(rsq33,rcutoff2))
2390             {
2391
2392             r33              = _mm_mul_pd(rsq33,rinv33);
2393
2394             /* EWALD ELECTROSTATICS */
2395
2396             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2397             ewrt             = _mm_mul_pd(r33,ewtabscale);
2398             ewitab           = _mm_cvttpd_epi32(ewrt);
2399 #ifdef __XOP__
2400             eweps            = _mm_frcz_pd(ewrt);
2401 #else
2402             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2403 #endif
2404             twoeweps         = _mm_add_pd(eweps,eweps);
2405             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2406             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2407             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2408
2409             cutoff_mask      = _mm_cmplt_pd(rsq33,rcutoff2);
2410
2411             fscal            = felec;
2412
2413             fscal            = _mm_and_pd(fscal,cutoff_mask);
2414
2415             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2416
2417             /* Update vectorial force */
2418             fix3             = _mm_macc_pd(dx33,fscal,fix3);
2419             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
2420             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
2421
2422             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
2423             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
2424             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
2425
2426             }
2427
2428             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2429
2430             /* Inner loop uses 378 flops */
2431         }
2432
2433         /* End of innermost loop */
2434
2435         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2436                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
2437
2438         /* Increment number of inner iterations */
2439         inneriter                  += j_index_end - j_index_start;
2440
2441         /* Outer loop uses 18 flops */
2442     }
2443
2444     /* Increment number of outer iterations */
2445     outeriter        += nri;
2446
2447     /* Update outer/inner flops */
2448
2449     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*378);
2450 }