src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_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_ElecEw_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_ElecEw_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     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 158                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 159
 160         fix1             = _mm_setzero_pd();
 161         fiy1             = _mm_setzero_pd();
 162         fiz1             = _mm_setzero_pd();
 163         fix2             = _mm_setzero_pd();
 164         fiy2             = _mm_setzero_pd();
 165         fiz2             = _mm_setzero_pd();
 166         fix3             = _mm_setzero_pd();
 167         fiy3             = _mm_setzero_pd();
 168         fiz3             = _mm_setzero_pd();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_pd();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182
 183             /* load j atom coordinates */
 184             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 185                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 186
 187             /* Calculate displacement vector */
 188             dx11             = _mm_sub_pd(ix1,jx1);
 189             dy11             = _mm_sub_pd(iy1,jy1);
 190             dz11             = _mm_sub_pd(iz1,jz1);
 191             dx12             = _mm_sub_pd(ix1,jx2);
 192             dy12             = _mm_sub_pd(iy1,jy2);
 193             dz12             = _mm_sub_pd(iz1,jz2);
 194             dx13             = _mm_sub_pd(ix1,jx3);
 195             dy13             = _mm_sub_pd(iy1,jy3);
 196             dz13             = _mm_sub_pd(iz1,jz3);
 197             dx21             = _mm_sub_pd(ix2,jx1);
 198             dy21             = _mm_sub_pd(iy2,jy1);
 199             dz21             = _mm_sub_pd(iz2,jz1);
 200             dx22             = _mm_sub_pd(ix2,jx2);
 201             dy22             = _mm_sub_pd(iy2,jy2);
 202             dz22             = _mm_sub_pd(iz2,jz2);
 203             dx23             = _mm_sub_pd(ix2,jx3);
 204             dy23             = _mm_sub_pd(iy2,jy3);
 205             dz23             = _mm_sub_pd(iz2,jz3);
 206             dx31             = _mm_sub_pd(ix3,jx1);
 207             dy31             = _mm_sub_pd(iy3,jy1);
 208             dz31             = _mm_sub_pd(iz3,jz1);
 209             dx32             = _mm_sub_pd(ix3,jx2);
 210             dy32             = _mm_sub_pd(iy3,jy2);
 211             dz32             = _mm_sub_pd(iz3,jz2);
 212             dx33             = _mm_sub_pd(ix3,jx3);
 213             dy33             = _mm_sub_pd(iy3,jy3);
 214             dz33             = _mm_sub_pd(iz3,jz3);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 218             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 219             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 220             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 221             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 222             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 223             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 224             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 225             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 226
 227             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 228             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 229             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 230             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 231             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 232             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 233             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 234             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 235             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 236
 237             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 238             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 239             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 240             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 241             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 242             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 243             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 244             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 245             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 246
 247             fjx1             = _mm_setzero_pd();
 248             fjy1             = _mm_setzero_pd();
 249             fjz1             = _mm_setzero_pd();
 250             fjx2             = _mm_setzero_pd();
 251             fjy2             = _mm_setzero_pd();
 252             fjz2             = _mm_setzero_pd();
 253             fjx3             = _mm_setzero_pd();
 254             fjy3             = _mm_setzero_pd();
 255             fjz3             = _mm_setzero_pd();
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             r11              = _mm_mul_pd(rsq11,rinv11);
 262
 263             /* EWALD ELECTROSTATICS */
 264
 265             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 266             ewrt             = _mm_mul_pd(r11,ewtabscale);
 267             ewitab           = _mm_cvttpd_epi32(ewrt);
 268 #ifdef __XOP__
 269             eweps            = _mm_frcz_pd(ewrt);
 270 #else
 271             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 272 #endif
 273             twoeweps         = _mm_add_pd(eweps,eweps);
 274             ewitab           = _mm_slli_epi32(ewitab,2);
 275             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 276             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 277             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 278             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 279             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 280             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 281             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 282             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 283             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 284             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             velecsum         = _mm_add_pd(velecsum,velec);
 288
 289             fscal            = felec;
 290
 291             /* Update vectorial force */
 292             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 293             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 294             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 295
 296             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 297             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 298             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r12              = _mm_mul_pd(rsq12,rinv12);
 305
 306             /* EWALD ELECTROSTATICS */
 307
 308             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 309             ewrt             = _mm_mul_pd(r12,ewtabscale);
 310             ewitab           = _mm_cvttpd_epi32(ewrt);
 311 #ifdef __XOP__
 312             eweps            = _mm_frcz_pd(ewrt);
 313 #else
 314             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 315 #endif
 316             twoeweps         = _mm_add_pd(eweps,eweps);
 317             ewitab           = _mm_slli_epi32(ewitab,2);
 318             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 319             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 320             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 321             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 322             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 323             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 324             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 325             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 326             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 327             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velecsum         = _mm_add_pd(velecsum,velec);
 331
 332             fscal            = felec;
 333
 334             /* Update vectorial force */
 335             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 336             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 337             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 338
 339             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 340             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 341             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             r13              = _mm_mul_pd(rsq13,rinv13);
 348
 349             /* EWALD ELECTROSTATICS */
 350
 351             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 352             ewrt             = _mm_mul_pd(r13,ewtabscale);
 353             ewitab           = _mm_cvttpd_epi32(ewrt);
 354 #ifdef __XOP__
 355             eweps            = _mm_frcz_pd(ewrt);
 356 #else
 357             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 358 #endif
 359             twoeweps         = _mm_add_pd(eweps,eweps);
 360             ewitab           = _mm_slli_epi32(ewitab,2);
 361             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 362             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 363             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 364             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 365             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 366             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 367             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 368             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 369             velec            = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
 370             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velecsum         = _mm_add_pd(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377             /* Update vectorial force */
 378             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 379             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 380             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 381
 382             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 383             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 384             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 385
 386             /**************************
 387              * CALCULATE INTERACTIONS *
 388              **************************/
 389
 390             r21              = _mm_mul_pd(rsq21,rinv21);
 391
 392             /* EWALD ELECTROSTATICS */
 393
 394             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 395             ewrt             = _mm_mul_pd(r21,ewtabscale);
 396             ewitab           = _mm_cvttpd_epi32(ewrt);
 397 #ifdef __XOP__
 398             eweps            = _mm_frcz_pd(ewrt);
 399 #else
 400             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 401 #endif
 402             twoeweps         = _mm_add_pd(eweps,eweps);
 403             ewitab           = _mm_slli_epi32(ewitab,2);
 404             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 405             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 406             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 407             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 408             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 409             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 410             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 411             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 412             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 413             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 414
 415             /* Update potential sum for this i atom from the interaction with this j atom. */
 416             velecsum         = _mm_add_pd(velecsum,velec);
 417
 418             fscal            = felec;
 419
 420             /* Update vectorial force */
 421             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 422             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 423             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 424
 425             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 426             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 427             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 428
 429             /**************************
 430              * CALCULATE INTERACTIONS *
 431              **************************/
 432
 433             r22              = _mm_mul_pd(rsq22,rinv22);
 434
 435             /* EWALD ELECTROSTATICS */
 436
 437             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 438             ewrt             = _mm_mul_pd(r22,ewtabscale);
 439             ewitab           = _mm_cvttpd_epi32(ewrt);
 440 #ifdef __XOP__
 441             eweps            = _mm_frcz_pd(ewrt);
 442 #else
 443             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 444 #endif
 445             twoeweps         = _mm_add_pd(eweps,eweps);
 446             ewitab           = _mm_slli_epi32(ewitab,2);
 447             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 448             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 449             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 450             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 451             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 452             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 453             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 454             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 455             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 456             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velecsum         = _mm_add_pd(velecsum,velec);
 460
 461             fscal            = felec;
 462
 463             /* Update vectorial force */
 464             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 465             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 466             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 467
 468             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 469             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 470             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             r23              = _mm_mul_pd(rsq23,rinv23);
 477
 478             /* EWALD ELECTROSTATICS */
 479
 480             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 481             ewrt             = _mm_mul_pd(r23,ewtabscale);
 482             ewitab           = _mm_cvttpd_epi32(ewrt);
 483 #ifdef __XOP__
 484             eweps            = _mm_frcz_pd(ewrt);
 485 #else
 486             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 487 #endif
 488             twoeweps         = _mm_add_pd(eweps,eweps);
 489             ewitab           = _mm_slli_epi32(ewitab,2);
 490             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 491             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 492             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 493             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 494             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 495             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 496             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 497             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 498             velec            = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
 499             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
 500
 501             /* Update potential sum for this i atom from the interaction with this j atom. */
 502             velecsum         = _mm_add_pd(velecsum,velec);
 503
 504             fscal            = felec;
 505
 506             /* Update vectorial force */
 507             fix2             = _mm_macc_pd(dx23,fscal,fix2);
 508             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
 509             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
 510
 511             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
 512             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
 513             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
 514
 515             /**************************
 516              * CALCULATE INTERACTIONS *
 517              **************************/
 518
 519             r31              = _mm_mul_pd(rsq31,rinv31);
 520
 521             /* EWALD ELECTROSTATICS */
 522
 523             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 524             ewrt             = _mm_mul_pd(r31,ewtabscale);
 525             ewitab           = _mm_cvttpd_epi32(ewrt);
 526 #ifdef __XOP__
 527             eweps            = _mm_frcz_pd(ewrt);
 528 #else
 529             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 530 #endif
 531             twoeweps         = _mm_add_pd(eweps,eweps);
 532             ewitab           = _mm_slli_epi32(ewitab,2);
 533             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 534             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 535             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 536             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 537             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 538             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 539             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 540             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 541             velec            = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
 542             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             velecsum         = _mm_add_pd(velecsum,velec);
 546
 547             fscal            = felec;
 548
 549             /* Update vectorial force */
 550             fix3             = _mm_macc_pd(dx31,fscal,fix3);
 551             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
 552             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
 553
 554             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
 555             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
 556             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             r32              = _mm_mul_pd(rsq32,rinv32);
 563
 564             /* EWALD ELECTROSTATICS */
 565
 566             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 567             ewrt             = _mm_mul_pd(r32,ewtabscale);
 568             ewitab           = _mm_cvttpd_epi32(ewrt);
 569 #ifdef __XOP__
 570             eweps            = _mm_frcz_pd(ewrt);
 571 #else
 572             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 573 #endif
 574             twoeweps         = _mm_add_pd(eweps,eweps);
 575             ewitab           = _mm_slli_epi32(ewitab,2);
 576             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 577             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 578             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 579             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 580             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 581             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 582             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 583             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 584             velec            = _mm_mul_pd(qq32,_mm_sub_pd(rinv32,velec));
 585             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
 586
 587             /* Update potential sum for this i atom from the interaction with this j atom. */
 588             velecsum         = _mm_add_pd(velecsum,velec);
 589
 590             fscal            = felec;
 591
 592             /* Update vectorial force */
 593             fix3             = _mm_macc_pd(dx32,fscal,fix3);
 594             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
 595             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
 596
 597             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
 598             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
 599             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             r33              = _mm_mul_pd(rsq33,rinv33);
 606
 607             /* EWALD ELECTROSTATICS */
 608
 609             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 610             ewrt             = _mm_mul_pd(r33,ewtabscale);
 611             ewitab           = _mm_cvttpd_epi32(ewrt);
 612 #ifdef __XOP__
 613             eweps            = _mm_frcz_pd(ewrt);
 614 #else
 615             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 616 #endif
 617             twoeweps         = _mm_add_pd(eweps,eweps);
 618             ewitab           = _mm_slli_epi32(ewitab,2);
 619             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 620             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 621             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 622             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 623             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 624             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 625             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 626             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 627             velec            = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
 628             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
 629
 630             /* Update potential sum for this i atom from the interaction with this j atom. */
 631             velecsum         = _mm_add_pd(velecsum,velec);
 632
 633             fscal            = felec;
 634
 635             /* Update vectorial force */
 636             fix3             = _mm_macc_pd(dx33,fscal,fix3);
 637             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
 638             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
 639
 640             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
 641             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
 642             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
 643
 644             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);
 645
 646             /* Inner loop uses 396 flops */
 647         }
 648
 649         if(jidx<j_index_end)
 650         {
 651
 652             jnrA             = jjnr[jidx];
 653             j_coord_offsetA  = DIM*jnrA;
 654
 655             /* load j atom coordinates */
 656             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
 657                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 658
 659             /* Calculate displacement vector */
 660             dx11             = _mm_sub_pd(ix1,jx1);
 661             dy11             = _mm_sub_pd(iy1,jy1);
 662             dz11             = _mm_sub_pd(iz1,jz1);
 663             dx12             = _mm_sub_pd(ix1,jx2);
 664             dy12             = _mm_sub_pd(iy1,jy2);
 665             dz12             = _mm_sub_pd(iz1,jz2);
 666             dx13             = _mm_sub_pd(ix1,jx3);
 667             dy13             = _mm_sub_pd(iy1,jy3);
 668             dz13             = _mm_sub_pd(iz1,jz3);
 669             dx21             = _mm_sub_pd(ix2,jx1);
 670             dy21             = _mm_sub_pd(iy2,jy1);
 671             dz21             = _mm_sub_pd(iz2,jz1);
 672             dx22             = _mm_sub_pd(ix2,jx2);
 673             dy22             = _mm_sub_pd(iy2,jy2);
 674             dz22             = _mm_sub_pd(iz2,jz2);
 675             dx23             = _mm_sub_pd(ix2,jx3);
 676             dy23             = _mm_sub_pd(iy2,jy3);
 677             dz23             = _mm_sub_pd(iz2,jz3);
 678             dx31             = _mm_sub_pd(ix3,jx1);
 679             dy31             = _mm_sub_pd(iy3,jy1);
 680             dz31             = _mm_sub_pd(iz3,jz1);
 681             dx32             = _mm_sub_pd(ix3,jx2);
 682             dy32             = _mm_sub_pd(iy3,jy2);
 683             dz32             = _mm_sub_pd(iz3,jz2);
 684             dx33             = _mm_sub_pd(ix3,jx3);
 685             dy33             = _mm_sub_pd(iy3,jy3);
 686             dz33             = _mm_sub_pd(iz3,jz3);
 687
 688             /* Calculate squared distance and things based on it */
 689             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 690             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 691             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
 692             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 693             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 694             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
 695             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
 696             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
 697             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
 698
 699             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 700             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 701             rinv13           = gmx_mm_invsqrt_pd(rsq13);
 702             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 703             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 704             rinv23           = gmx_mm_invsqrt_pd(rsq23);
 705             rinv31           = gmx_mm_invsqrt_pd(rsq31);
 706             rinv32           = gmx_mm_invsqrt_pd(rsq32);
 707             rinv33           = gmx_mm_invsqrt_pd(rsq33);
 708
 709             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 710             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 711             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
 712             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 713             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 714             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
 715             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
 716             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
 717             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
 718
 719             fjx1             = _mm_setzero_pd();
 720             fjy1             = _mm_setzero_pd();
 721             fjz1             = _mm_setzero_pd();
 722             fjx2             = _mm_setzero_pd();
 723             fjy2             = _mm_setzero_pd();
 724             fjz2             = _mm_setzero_pd();
 725             fjx3             = _mm_setzero_pd();
 726             fjy3             = _mm_setzero_pd();
 727             fjz3             = _mm_setzero_pd();
 728
 729             /**************************
 730              * CALCULATE INTERACTIONS *
 731              **************************/
 732
 733             r11              = _mm_mul_pd(rsq11,rinv11);
 734
 735             /* EWALD ELECTROSTATICS */
 736
 737             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 738             ewrt             = _mm_mul_pd(r11,ewtabscale);
 739             ewitab           = _mm_cvttpd_epi32(ewrt);
 740 #ifdef __XOP__
 741             eweps            = _mm_frcz_pd(ewrt);
 742 #else
 743             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 744 #endif
 745             twoeweps         = _mm_add_pd(eweps,eweps);
 746             ewitab           = _mm_slli_epi32(ewitab,2);
 747             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 748             ewtabD           = _mm_setzero_pd();
 749             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 750             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 751             ewtabFn          = _mm_setzero_pd();
 752             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 753             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 754             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 755             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 756             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 757
 758             /* Update potential sum for this i atom from the interaction with this j atom. */
 759             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 760             velecsum         = _mm_add_pd(velecsum,velec);
 761
 762             fscal            = felec;
 763
 764             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 765
 766             /* Update vectorial force */
 767             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 768             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 769             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 770
 771             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 772             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 773             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 774
 775             /**************************
 776              * CALCULATE INTERACTIONS *
 777              **************************/
 778
 779             r12              = _mm_mul_pd(rsq12,rinv12);
 780
 781             /* EWALD ELECTROSTATICS */
 782
 783             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 784             ewrt             = _mm_mul_pd(r12,ewtabscale);
 785             ewitab           = _mm_cvttpd_epi32(ewrt);
 786 #ifdef __XOP__
 787             eweps            = _mm_frcz_pd(ewrt);
 788 #else
 789             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 790 #endif
 791             twoeweps         = _mm_add_pd(eweps,eweps);
 792             ewitab           = _mm_slli_epi32(ewitab,2);
 793             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 794             ewtabD           = _mm_setzero_pd();
 795             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 796             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 797             ewtabFn          = _mm_setzero_pd();
 798             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 799             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 800             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 801             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 802             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 803
 804             /* Update potential sum for this i atom from the interaction with this j atom. */
 805             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 806             velecsum         = _mm_add_pd(velecsum,velec);
 807
 808             fscal            = felec;
 809
 810             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 811
 812             /* Update vectorial force */
 813             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 814             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 815             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 816
 817             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 818             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 819             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 820
 821             /**************************
 822              * CALCULATE INTERACTIONS *
 823              **************************/
 824
 825             r13              = _mm_mul_pd(rsq13,rinv13);
 826
 827             /* EWALD ELECTROSTATICS */
 828
 829             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 830             ewrt             = _mm_mul_pd(r13,ewtabscale);
 831             ewitab           = _mm_cvttpd_epi32(ewrt);
 832 #ifdef __XOP__
 833             eweps            = _mm_frcz_pd(ewrt);
 834 #else
 835             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 836 #endif
 837             twoeweps         = _mm_add_pd(eweps,eweps);
 838             ewitab           = _mm_slli_epi32(ewitab,2);
 839             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 840             ewtabD           = _mm_setzero_pd();
 841             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 842             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 843             ewtabFn          = _mm_setzero_pd();
 844             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 845             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 846             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 847             velec            = _mm_mul_pd(qq13,_mm_sub_pd(rinv13,velec));
 848             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
 849
 850             /* Update potential sum for this i atom from the interaction with this j atom. */
 851             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 852             velecsum         = _mm_add_pd(velecsum,velec);
 853
 854             fscal            = felec;
 855
 856             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 857
 858             /* Update vectorial force */
 859             fix1             = _mm_macc_pd(dx13,fscal,fix1);
 860             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
 861             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
 862
 863             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
 864             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
 865             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
 866
 867             /**************************
 868              * CALCULATE INTERACTIONS *
 869              **************************/
 870
 871             r21              = _mm_mul_pd(rsq21,rinv21);
 872
 873             /* EWALD ELECTROSTATICS */
 874
 875             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 876             ewrt             = _mm_mul_pd(r21,ewtabscale);
 877             ewitab           = _mm_cvttpd_epi32(ewrt);
 878 #ifdef __XOP__
 879             eweps            = _mm_frcz_pd(ewrt);
 880 #else
 881             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 882 #endif
 883             twoeweps         = _mm_add_pd(eweps,eweps);
 884             ewitab           = _mm_slli_epi32(ewitab,2);
 885             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 886             ewtabD           = _mm_setzero_pd();
 887             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 888             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 889             ewtabFn          = _mm_setzero_pd();
 890             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 891             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 892             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 893             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 894             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 895
 896             /* Update potential sum for this i atom from the interaction with this j atom. */
 897             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 898             velecsum         = _mm_add_pd(velecsum,velec);
 899
 900             fscal            = felec;
 901
 902             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 903
 904             /* Update vectorial force */
 905             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 906             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 907             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 908
 909             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 910             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 911             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             r22              = _mm_mul_pd(rsq22,rinv22);
 918
 919             /* EWALD ELECTROSTATICS */
 920
 921             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 922             ewrt             = _mm_mul_pd(r22,ewtabscale);
 923             ewitab           = _mm_cvttpd_epi32(ewrt);
 924 #ifdef __XOP__
 925             eweps            = _mm_frcz_pd(ewrt);
 926 #else
 927             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 928 #endif
 929             twoeweps         = _mm_add_pd(eweps,eweps);
 930             ewitab           = _mm_slli_epi32(ewitab,2);
 931             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 932             ewtabD           = _mm_setzero_pd();
 933             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 934             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 935             ewtabFn          = _mm_setzero_pd();
 936             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 937             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 938             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 939             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 940             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 941
 942             /* Update potential sum for this i atom from the interaction with this j atom. */
 943             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 944             velecsum         = _mm_add_pd(velecsum,velec);
 945
 946             fscal            = felec;
 947
 948             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 949
 950             /* Update vectorial force */
 951             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 952             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 953             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 954
 955             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 956             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 957             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 958
 959             /**************************
 960              * CALCULATE INTERACTIONS *
 961              **************************/
 962
 963             r23              = _mm_mul_pd(rsq23,rinv23);
 964
 965             /* EWALD ELECTROSTATICS */
 966
 967             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 968             ewrt             = _mm_mul_pd(r23,ewtabscale);
 969             ewitab           = _mm_cvttpd_epi32(ewrt);
 970 #ifdef __XOP__
 971             eweps            = _mm_frcz_pd(ewrt);
 972 #else
 973             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 974 #endif
 975             twoeweps         = _mm_add_pd(eweps,eweps);
 976             ewitab           = _mm_slli_epi32(ewitab,2);
 977             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 978             ewtabD           = _mm_setzero_pd();
 979             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 980             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 981             ewtabFn          = _mm_setzero_pd();
 982             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 983             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 984             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 985             velec            = _mm_mul_pd(qq23,_mm_sub_pd(rinv23,velec));
 986             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
 987
 988             /* Update potential sum for this i atom from the interaction with this j atom. */
 989             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 990             velecsum         = _mm_add_pd(velecsum,velec);
 991
 992             fscal            = felec;
 993
 994             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 995
 996             /* Update vectorial force */
 997             fix2             = _mm_macc_pd(dx23,fscal,fix2);
 998             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
 999             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1000
1001             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1002             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1003             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             r31              = _mm_mul_pd(rsq31,rinv31);
1010
1011             /* EWALD ELECTROSTATICS */
1012
1013             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1014             ewrt             = _mm_mul_pd(r31,ewtabscale);
1015             ewitab           = _mm_cvttpd_epi32(ewrt);
1016 #ifdef __XOP__
1017             eweps            = _mm_frcz_pd(ewrt);
1018 #else
1019             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1020 #endif
1021             twoeweps         = _mm_add_pd(eweps,eweps);
1022             ewitab           = _mm_slli_epi32(ewitab,2);
1023             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1024             ewtabD           = _mm_setzero_pd();
1025             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1026             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1027             ewtabFn          = _mm_setzero_pd();
1028             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1029             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1030             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1031             velec            = _mm_mul_pd(qq31,_mm_sub_pd(rinv31,velec));
1032             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1033
1034             /* Update potential sum for this i atom from the interaction with this j atom. */
1035             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1036             velecsum         = _mm_add_pd(velecsum,velec);
1037
1038             fscal            = felec;
1039
1040             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1041
1042             /* Update vectorial force */
1043             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1044             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1045             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1046
1047             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1048             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1049             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1050
1051             /**************************
1052              * CALCULATE INTERACTIONS *
1053              **************************/
1054
1055             r32              = _mm_mul_pd(rsq32,rinv32);
1056
1057             /* EWALD ELECTROSTATICS */
1058
1059             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1060             ewrt             = _mm_mul_pd(r32,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(qq32,_mm_sub_pd(rinv32,velec));
1078             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1079
1080             /* Update potential sum for this i atom from the interaction with this j atom. */
1081             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1082             velecsum         = _mm_add_pd(velecsum,velec);
1083
1084             fscal            = felec;
1085
1086             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1087
1088             /* Update vectorial force */
1089             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1090             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1091             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1092
1093             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1094             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1095             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             r33              = _mm_mul_pd(rsq33,rinv33);
1102
1103             /* EWALD ELECTROSTATICS */
1104
1105             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1106             ewrt             = _mm_mul_pd(r33,ewtabscale);
1107             ewitab           = _mm_cvttpd_epi32(ewrt);
1108 #ifdef __XOP__
1109             eweps            = _mm_frcz_pd(ewrt);
1110 #else
1111             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1112 #endif
1113             twoeweps         = _mm_add_pd(eweps,eweps);
1114             ewitab           = _mm_slli_epi32(ewitab,2);
1115             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1116             ewtabD           = _mm_setzero_pd();
1117             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1118             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1119             ewtabFn          = _mm_setzero_pd();
1120             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1121             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1122             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1123             velec            = _mm_mul_pd(qq33,_mm_sub_pd(rinv33,velec));
1124             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1125
1126             /* Update potential sum for this i atom from the interaction with this j atom. */
1127             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1128             velecsum         = _mm_add_pd(velecsum,velec);
1129
1130             fscal            = felec;
1131
1132             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1133
1134             /* Update vectorial force */
1135             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1136             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1137             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1138
1139             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1140             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1141             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1142
1143             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1144
1145             /* Inner loop uses 396 flops */
1146         }
1147
1148         /* End of innermost loop */
1149
1150         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1151                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1152
1153         ggid                        = gid[iidx];
1154         /* Update potential energies */
1155         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1156
1157         /* Increment number of inner iterations */
1158         inneriter                  += j_index_end - j_index_start;
1159
1160         /* Outer loop uses 19 flops */
1161     }
1162
1163     /* Increment number of outer iterations */
1164     outeriter        += nri;
1165
1166     /* Update outer/inner flops */
1167
1168     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*396);
1169 }
1170 /*
1171  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_128_fma_double
1172  * Electrostatics interaction: Ewald
1173  * VdW interaction:            None
1174  * Geometry:                   Water4-Water4
1175  * Calculate force/pot:        Force
1176  */
1177 void
1178 nb_kernel_ElecEw_VdwNone_GeomW4W4_F_avx_128_fma_double
1179                     (t_nblist * gmx_restrict                nlist,
1180                      rvec * gmx_restrict                    xx,
1181                      rvec * gmx_restrict                    ff,
1182                      t_forcerec * gmx_restrict              fr,
1183                      t_mdatoms * gmx_restrict               mdatoms,
1184                      nb_kernel_data_t * gmx_restrict        kernel_data,
1185                      t_nrnb * gmx_restrict                  nrnb)
1186 {
1187     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1188      * just 0 for non-waters.
1189      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1190      * jnr indices corresponding to data put in the four positions in the SIMD register.
1191      */
1192     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1193     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1194     int              jnrA,jnrB;
1195     int              j_coord_offsetA,j_coord_offsetB;
1196     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1197     real             rcutoff_scalar;
1198     real             *shiftvec,*fshift,*x,*f;
1199     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1200     int              vdwioffset1;
1201     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1202     int              vdwioffset2;
1203     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1204     int              vdwioffset3;
1205     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1206     int              vdwjidx1A,vdwjidx1B;
1207     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1208     int              vdwjidx2A,vdwjidx2B;
1209     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1210     int              vdwjidx3A,vdwjidx3B;
1211     __m128d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1212     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1213     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1214     __m128d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1215     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1216     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1217     __m128d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1218     __m128d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1219     __m128d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1220     __m128d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1221     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1222     real             *charge;
1223     __m128i          ewitab;
1224     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1225     real             *ewtab;
1226     __m128d          dummy_mask,cutoff_mask;
1227     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1228     __m128d          one     = _mm_set1_pd(1.0);
1229     __m128d          two     = _mm_set1_pd(2.0);
1230     x                = xx[0];
1231     f                = ff[0];
1232
1233     nri              = nlist->nri;
1234     iinr             = nlist->iinr;
1235     jindex           = nlist->jindex;
1236     jjnr             = nlist->jjnr;
1237     shiftidx         = nlist->shift;
1238     gid              = nlist->gid;
1239     shiftvec         = fr->shift_vec[0];
1240     fshift           = fr->fshift[0];
1241     facel            = _mm_set1_pd(fr->epsfac);
1242     charge           = mdatoms->chargeA;
1243
1244     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1245     ewtab            = fr->ic->tabq_coul_F;
1246     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1247     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1248
1249     /* Setup water-specific parameters */
1250     inr              = nlist->iinr[0];
1251     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1252     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1253     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
1254
1255     jq1              = _mm_set1_pd(charge[inr+1]);
1256     jq2              = _mm_set1_pd(charge[inr+2]);
1257     jq3              = _mm_set1_pd(charge[inr+3]);
1258     qq11             = _mm_mul_pd(iq1,jq1);
1259     qq12             = _mm_mul_pd(iq1,jq2);
1260     qq13             = _mm_mul_pd(iq1,jq3);
1261     qq21             = _mm_mul_pd(iq2,jq1);
1262     qq22             = _mm_mul_pd(iq2,jq2);
1263     qq23             = _mm_mul_pd(iq2,jq3);
1264     qq31             = _mm_mul_pd(iq3,jq1);
1265     qq32             = _mm_mul_pd(iq3,jq2);
1266     qq33             = _mm_mul_pd(iq3,jq3);
1267
1268     /* Avoid stupid compiler warnings */
1269     jnrA = jnrB = 0;
1270     j_coord_offsetA = 0;
1271     j_coord_offsetB = 0;
1272
1273     outeriter        = 0;
1274     inneriter        = 0;
1275
1276     /* Start outer loop over neighborlists */
1277     for(iidx=0; iidx<nri; iidx++)
1278     {
1279         /* Load shift vector for this list */
1280         i_shift_offset   = DIM*shiftidx[iidx];
1281
1282         /* Load limits for loop over neighbors */
1283         j_index_start    = jindex[iidx];
1284         j_index_end      = jindex[iidx+1];
1285
1286         /* Get outer coordinate index */
1287         inr              = iinr[iidx];
1288         i_coord_offset   = DIM*inr;
1289
1290         /* Load i particle coords and add shift vector */
1291         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1292                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1293
1294         fix1             = _mm_setzero_pd();
1295         fiy1             = _mm_setzero_pd();
1296         fiz1             = _mm_setzero_pd();
1297         fix2             = _mm_setzero_pd();
1298         fiy2             = _mm_setzero_pd();
1299         fiz2             = _mm_setzero_pd();
1300         fix3             = _mm_setzero_pd();
1301         fiy3             = _mm_setzero_pd();
1302         fiz3             = _mm_setzero_pd();
1303
1304         /* Start inner kernel loop */
1305         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1306         {
1307
1308             /* Get j neighbor index, and coordinate index */
1309             jnrA             = jjnr[jidx];
1310             jnrB             = jjnr[jidx+1];
1311             j_coord_offsetA  = DIM*jnrA;
1312             j_coord_offsetB  = DIM*jnrB;
1313
1314             /* load j atom coordinates */
1315             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1316                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1317
1318             /* Calculate displacement vector */
1319             dx11             = _mm_sub_pd(ix1,jx1);
1320             dy11             = _mm_sub_pd(iy1,jy1);
1321             dz11             = _mm_sub_pd(iz1,jz1);
1322             dx12             = _mm_sub_pd(ix1,jx2);
1323             dy12             = _mm_sub_pd(iy1,jy2);
1324             dz12             = _mm_sub_pd(iz1,jz2);
1325             dx13             = _mm_sub_pd(ix1,jx3);
1326             dy13             = _mm_sub_pd(iy1,jy3);
1327             dz13             = _mm_sub_pd(iz1,jz3);
1328             dx21             = _mm_sub_pd(ix2,jx1);
1329             dy21             = _mm_sub_pd(iy2,jy1);
1330             dz21             = _mm_sub_pd(iz2,jz1);
1331             dx22             = _mm_sub_pd(ix2,jx2);
1332             dy22             = _mm_sub_pd(iy2,jy2);
1333             dz22             = _mm_sub_pd(iz2,jz2);
1334             dx23             = _mm_sub_pd(ix2,jx3);
1335             dy23             = _mm_sub_pd(iy2,jy3);
1336             dz23             = _mm_sub_pd(iz2,jz3);
1337             dx31             = _mm_sub_pd(ix3,jx1);
1338             dy31             = _mm_sub_pd(iy3,jy1);
1339             dz31             = _mm_sub_pd(iz3,jz1);
1340             dx32             = _mm_sub_pd(ix3,jx2);
1341             dy32             = _mm_sub_pd(iy3,jy2);
1342             dz32             = _mm_sub_pd(iz3,jz2);
1343             dx33             = _mm_sub_pd(ix3,jx3);
1344             dy33             = _mm_sub_pd(iy3,jy3);
1345             dz33             = _mm_sub_pd(iz3,jz3);
1346
1347             /* Calculate squared distance and things based on it */
1348             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1349             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1350             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1351             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1352             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1353             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1354             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1355             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1356             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1357
1358             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1359             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1360             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1361             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1362             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1363             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1364             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1365             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1366             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1367
1368             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1369             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1370             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1371             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1372             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1373             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1374             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1375             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1376             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1377
1378             fjx1             = _mm_setzero_pd();
1379             fjy1             = _mm_setzero_pd();
1380             fjz1             = _mm_setzero_pd();
1381             fjx2             = _mm_setzero_pd();
1382             fjy2             = _mm_setzero_pd();
1383             fjz2             = _mm_setzero_pd();
1384             fjx3             = _mm_setzero_pd();
1385             fjy3             = _mm_setzero_pd();
1386             fjz3             = _mm_setzero_pd();
1387
1388             /**************************
1389              * CALCULATE INTERACTIONS *
1390              **************************/
1391
1392             r11              = _mm_mul_pd(rsq11,rinv11);
1393
1394             /* EWALD ELECTROSTATICS */
1395
1396             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1397             ewrt             = _mm_mul_pd(r11,ewtabscale);
1398             ewitab           = _mm_cvttpd_epi32(ewrt);
1399 #ifdef __XOP__
1400             eweps            = _mm_frcz_pd(ewrt);
1401 #else
1402             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1403 #endif
1404             twoeweps         = _mm_add_pd(eweps,eweps);
1405             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1406                                          &ewtabF,&ewtabFn);
1407             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1408             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1409
1410             fscal            = felec;
1411
1412             /* Update vectorial force */
1413             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1414             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1415             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1416
1417             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1418             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1419             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1420
1421             /**************************
1422              * CALCULATE INTERACTIONS *
1423              **************************/
1424
1425             r12              = _mm_mul_pd(rsq12,rinv12);
1426
1427             /* EWALD ELECTROSTATICS */
1428
1429             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1430             ewrt             = _mm_mul_pd(r12,ewtabscale);
1431             ewitab           = _mm_cvttpd_epi32(ewrt);
1432 #ifdef __XOP__
1433             eweps            = _mm_frcz_pd(ewrt);
1434 #else
1435             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1436 #endif
1437             twoeweps         = _mm_add_pd(eweps,eweps);
1438             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1439                                          &ewtabF,&ewtabFn);
1440             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1441             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1442
1443             fscal            = felec;
1444
1445             /* Update vectorial force */
1446             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1447             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1448             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1449
1450             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1451             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1452             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1453
1454             /**************************
1455              * CALCULATE INTERACTIONS *
1456              **************************/
1457
1458             r13              = _mm_mul_pd(rsq13,rinv13);
1459
1460             /* EWALD ELECTROSTATICS */
1461
1462             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1463             ewrt             = _mm_mul_pd(r13,ewtabscale);
1464             ewitab           = _mm_cvttpd_epi32(ewrt);
1465 #ifdef __XOP__
1466             eweps            = _mm_frcz_pd(ewrt);
1467 #else
1468             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1469 #endif
1470             twoeweps         = _mm_add_pd(eweps,eweps);
1471             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1472                                          &ewtabF,&ewtabFn);
1473             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1474             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1475
1476             fscal            = felec;
1477
1478             /* Update vectorial force */
1479             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1480             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1481             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1482
1483             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1484             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1485             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1486
1487             /**************************
1488              * CALCULATE INTERACTIONS *
1489              **************************/
1490
1491             r21              = _mm_mul_pd(rsq21,rinv21);
1492
1493             /* EWALD ELECTROSTATICS */
1494
1495             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1496             ewrt             = _mm_mul_pd(r21,ewtabscale);
1497             ewitab           = _mm_cvttpd_epi32(ewrt);
1498 #ifdef __XOP__
1499             eweps            = _mm_frcz_pd(ewrt);
1500 #else
1501             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1502 #endif
1503             twoeweps         = _mm_add_pd(eweps,eweps);
1504             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1505                                          &ewtabF,&ewtabFn);
1506             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1507             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1508
1509             fscal            = felec;
1510
1511             /* Update vectorial force */
1512             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1513             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1514             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1515
1516             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1517             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1518             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1519
1520             /**************************
1521              * CALCULATE INTERACTIONS *
1522              **************************/
1523
1524             r22              = _mm_mul_pd(rsq22,rinv22);
1525
1526             /* EWALD ELECTROSTATICS */
1527
1528             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1529             ewrt             = _mm_mul_pd(r22,ewtabscale);
1530             ewitab           = _mm_cvttpd_epi32(ewrt);
1531 #ifdef __XOP__
1532             eweps            = _mm_frcz_pd(ewrt);
1533 #else
1534             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1535 #endif
1536             twoeweps         = _mm_add_pd(eweps,eweps);
1537             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1538                                          &ewtabF,&ewtabFn);
1539             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1540             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1541
1542             fscal            = felec;
1543
1544             /* Update vectorial force */
1545             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1546             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1547             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1548
1549             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1550             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1551             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1552
1553             /**************************
1554              * CALCULATE INTERACTIONS *
1555              **************************/
1556
1557             r23              = _mm_mul_pd(rsq23,rinv23);
1558
1559             /* EWALD ELECTROSTATICS */
1560
1561             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1562             ewrt             = _mm_mul_pd(r23,ewtabscale);
1563             ewitab           = _mm_cvttpd_epi32(ewrt);
1564 #ifdef __XOP__
1565             eweps            = _mm_frcz_pd(ewrt);
1566 #else
1567             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1568 #endif
1569             twoeweps         = _mm_add_pd(eweps,eweps);
1570             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1571                                          &ewtabF,&ewtabFn);
1572             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1573             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1574
1575             fscal            = felec;
1576
1577             /* Update vectorial force */
1578             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1579             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1580             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1581
1582             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1583             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1584             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1585
1586             /**************************
1587              * CALCULATE INTERACTIONS *
1588              **************************/
1589
1590             r31              = _mm_mul_pd(rsq31,rinv31);
1591
1592             /* EWALD ELECTROSTATICS */
1593
1594             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1595             ewrt             = _mm_mul_pd(r31,ewtabscale);
1596             ewitab           = _mm_cvttpd_epi32(ewrt);
1597 #ifdef __XOP__
1598             eweps            = _mm_frcz_pd(ewrt);
1599 #else
1600             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1601 #endif
1602             twoeweps         = _mm_add_pd(eweps,eweps);
1603             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1604                                          &ewtabF,&ewtabFn);
1605             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1606             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1607
1608             fscal            = felec;
1609
1610             /* Update vectorial force */
1611             fix3             = _mm_macc_pd(dx31,fscal,fix3);
1612             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
1613             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
1614
1615             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
1616             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
1617             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
1618
1619             /**************************
1620              * CALCULATE INTERACTIONS *
1621              **************************/
1622
1623             r32              = _mm_mul_pd(rsq32,rinv32);
1624
1625             /* EWALD ELECTROSTATICS */
1626
1627             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1628             ewrt             = _mm_mul_pd(r32,ewtabscale);
1629             ewitab           = _mm_cvttpd_epi32(ewrt);
1630 #ifdef __XOP__
1631             eweps            = _mm_frcz_pd(ewrt);
1632 #else
1633             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1634 #endif
1635             twoeweps         = _mm_add_pd(eweps,eweps);
1636             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1637                                          &ewtabF,&ewtabFn);
1638             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1639             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
1640
1641             fscal            = felec;
1642
1643             /* Update vectorial force */
1644             fix3             = _mm_macc_pd(dx32,fscal,fix3);
1645             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
1646             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
1647
1648             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
1649             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
1650             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
1651
1652             /**************************
1653              * CALCULATE INTERACTIONS *
1654              **************************/
1655
1656             r33              = _mm_mul_pd(rsq33,rinv33);
1657
1658             /* EWALD ELECTROSTATICS */
1659
1660             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1661             ewrt             = _mm_mul_pd(r33,ewtabscale);
1662             ewitab           = _mm_cvttpd_epi32(ewrt);
1663 #ifdef __XOP__
1664             eweps            = _mm_frcz_pd(ewrt);
1665 #else
1666             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1667 #endif
1668             twoeweps         = _mm_add_pd(eweps,eweps);
1669             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1670                                          &ewtabF,&ewtabFn);
1671             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1672             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
1673
1674             fscal            = felec;
1675
1676             /* Update vectorial force */
1677             fix3             = _mm_macc_pd(dx33,fscal,fix3);
1678             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
1679             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
1680
1681             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
1682             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
1683             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
1684
1685             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);
1686
1687             /* Inner loop uses 351 flops */
1688         }
1689
1690         if(jidx<j_index_end)
1691         {
1692
1693             jnrA             = jjnr[jidx];
1694             j_coord_offsetA  = DIM*jnrA;
1695
1696             /* load j atom coordinates */
1697             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
1698                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1699
1700             /* Calculate displacement vector */
1701             dx11             = _mm_sub_pd(ix1,jx1);
1702             dy11             = _mm_sub_pd(iy1,jy1);
1703             dz11             = _mm_sub_pd(iz1,jz1);
1704             dx12             = _mm_sub_pd(ix1,jx2);
1705             dy12             = _mm_sub_pd(iy1,jy2);
1706             dz12             = _mm_sub_pd(iz1,jz2);
1707             dx13             = _mm_sub_pd(ix1,jx3);
1708             dy13             = _mm_sub_pd(iy1,jy3);
1709             dz13             = _mm_sub_pd(iz1,jz3);
1710             dx21             = _mm_sub_pd(ix2,jx1);
1711             dy21             = _mm_sub_pd(iy2,jy1);
1712             dz21             = _mm_sub_pd(iz2,jz1);
1713             dx22             = _mm_sub_pd(ix2,jx2);
1714             dy22             = _mm_sub_pd(iy2,jy2);
1715             dz22             = _mm_sub_pd(iz2,jz2);
1716             dx23             = _mm_sub_pd(ix2,jx3);
1717             dy23             = _mm_sub_pd(iy2,jy3);
1718             dz23             = _mm_sub_pd(iz2,jz3);
1719             dx31             = _mm_sub_pd(ix3,jx1);
1720             dy31             = _mm_sub_pd(iy3,jy1);
1721             dz31             = _mm_sub_pd(iz3,jz1);
1722             dx32             = _mm_sub_pd(ix3,jx2);
1723             dy32             = _mm_sub_pd(iy3,jy2);
1724             dz32             = _mm_sub_pd(iz3,jz2);
1725             dx33             = _mm_sub_pd(ix3,jx3);
1726             dy33             = _mm_sub_pd(iy3,jy3);
1727             dz33             = _mm_sub_pd(iz3,jz3);
1728
1729             /* Calculate squared distance and things based on it */
1730             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1731             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1732             rsq13            = gmx_mm_calc_rsq_pd(dx13,dy13,dz13);
1733             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1734             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1735             rsq23            = gmx_mm_calc_rsq_pd(dx23,dy23,dz23);
1736             rsq31            = gmx_mm_calc_rsq_pd(dx31,dy31,dz31);
1737             rsq32            = gmx_mm_calc_rsq_pd(dx32,dy32,dz32);
1738             rsq33            = gmx_mm_calc_rsq_pd(dx33,dy33,dz33);
1739
1740             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1741             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1742             rinv13           = gmx_mm_invsqrt_pd(rsq13);
1743             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1744             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1745             rinv23           = gmx_mm_invsqrt_pd(rsq23);
1746             rinv31           = gmx_mm_invsqrt_pd(rsq31);
1747             rinv32           = gmx_mm_invsqrt_pd(rsq32);
1748             rinv33           = gmx_mm_invsqrt_pd(rsq33);
1749
1750             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1751             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1752             rinvsq13         = _mm_mul_pd(rinv13,rinv13);
1753             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1754             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1755             rinvsq23         = _mm_mul_pd(rinv23,rinv23);
1756             rinvsq31         = _mm_mul_pd(rinv31,rinv31);
1757             rinvsq32         = _mm_mul_pd(rinv32,rinv32);
1758             rinvsq33         = _mm_mul_pd(rinv33,rinv33);
1759
1760             fjx1             = _mm_setzero_pd();
1761             fjy1             = _mm_setzero_pd();
1762             fjz1             = _mm_setzero_pd();
1763             fjx2             = _mm_setzero_pd();
1764             fjy2             = _mm_setzero_pd();
1765             fjz2             = _mm_setzero_pd();
1766             fjx3             = _mm_setzero_pd();
1767             fjy3             = _mm_setzero_pd();
1768             fjz3             = _mm_setzero_pd();
1769
1770             /**************************
1771              * CALCULATE INTERACTIONS *
1772              **************************/
1773
1774             r11              = _mm_mul_pd(rsq11,rinv11);
1775
1776             /* EWALD ELECTROSTATICS */
1777
1778             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1779             ewrt             = _mm_mul_pd(r11,ewtabscale);
1780             ewitab           = _mm_cvttpd_epi32(ewrt);
1781 #ifdef __XOP__
1782             eweps            = _mm_frcz_pd(ewrt);
1783 #else
1784             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1785 #endif
1786             twoeweps         = _mm_add_pd(eweps,eweps);
1787             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1788             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1789             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1790
1791             fscal            = felec;
1792
1793             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1794
1795             /* Update vectorial force */
1796             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1797             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1798             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1799
1800             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1801             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1802             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1803
1804             /**************************
1805              * CALCULATE INTERACTIONS *
1806              **************************/
1807
1808             r12              = _mm_mul_pd(rsq12,rinv12);
1809
1810             /* EWALD ELECTROSTATICS */
1811
1812             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1813             ewrt             = _mm_mul_pd(r12,ewtabscale);
1814             ewitab           = _mm_cvttpd_epi32(ewrt);
1815 #ifdef __XOP__
1816             eweps            = _mm_frcz_pd(ewrt);
1817 #else
1818             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1819 #endif
1820             twoeweps         = _mm_add_pd(eweps,eweps);
1821             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1822             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1823             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1824
1825             fscal            = felec;
1826
1827             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1828
1829             /* Update vectorial force */
1830             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1831             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1832             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1833
1834             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1835             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1836             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1837
1838             /**************************
1839              * CALCULATE INTERACTIONS *
1840              **************************/
1841
1842             r13              = _mm_mul_pd(rsq13,rinv13);
1843
1844             /* EWALD ELECTROSTATICS */
1845
1846             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1847             ewrt             = _mm_mul_pd(r13,ewtabscale);
1848             ewitab           = _mm_cvttpd_epi32(ewrt);
1849 #ifdef __XOP__
1850             eweps            = _mm_frcz_pd(ewrt);
1851 #else
1852             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1853 #endif
1854             twoeweps         = _mm_add_pd(eweps,eweps);
1855             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1856             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1857             felec            = _mm_mul_pd(_mm_mul_pd(qq13,rinv13),_mm_sub_pd(rinvsq13,felec));
1858
1859             fscal            = felec;
1860
1861             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1862
1863             /* Update vectorial force */
1864             fix1             = _mm_macc_pd(dx13,fscal,fix1);
1865             fiy1             = _mm_macc_pd(dy13,fscal,fiy1);
1866             fiz1             = _mm_macc_pd(dz13,fscal,fiz1);
1867
1868             fjx3             = _mm_macc_pd(dx13,fscal,fjx3);
1869             fjy3             = _mm_macc_pd(dy13,fscal,fjy3);
1870             fjz3             = _mm_macc_pd(dz13,fscal,fjz3);
1871
1872             /**************************
1873              * CALCULATE INTERACTIONS *
1874              **************************/
1875
1876             r21              = _mm_mul_pd(rsq21,rinv21);
1877
1878             /* EWALD ELECTROSTATICS */
1879
1880             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1881             ewrt             = _mm_mul_pd(r21,ewtabscale);
1882             ewitab           = _mm_cvttpd_epi32(ewrt);
1883 #ifdef __XOP__
1884             eweps            = _mm_frcz_pd(ewrt);
1885 #else
1886             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1887 #endif
1888             twoeweps         = _mm_add_pd(eweps,eweps);
1889             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1890             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1891             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1892
1893             fscal            = felec;
1894
1895             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1896
1897             /* Update vectorial force */
1898             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1899             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1900             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1901
1902             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1903             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1904             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1905
1906             /**************************
1907              * CALCULATE INTERACTIONS *
1908              **************************/
1909
1910             r22              = _mm_mul_pd(rsq22,rinv22);
1911
1912             /* EWALD ELECTROSTATICS */
1913
1914             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1915             ewrt             = _mm_mul_pd(r22,ewtabscale);
1916             ewitab           = _mm_cvttpd_epi32(ewrt);
1917 #ifdef __XOP__
1918             eweps            = _mm_frcz_pd(ewrt);
1919 #else
1920             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1921 #endif
1922             twoeweps         = _mm_add_pd(eweps,eweps);
1923             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1924             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1925             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1926
1927             fscal            = felec;
1928
1929             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1930
1931             /* Update vectorial force */
1932             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1933             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1934             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1935
1936             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1937             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1938             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1939
1940             /**************************
1941              * CALCULATE INTERACTIONS *
1942              **************************/
1943
1944             r23              = _mm_mul_pd(rsq23,rinv23);
1945
1946             /* EWALD ELECTROSTATICS */
1947
1948             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1949             ewrt             = _mm_mul_pd(r23,ewtabscale);
1950             ewitab           = _mm_cvttpd_epi32(ewrt);
1951 #ifdef __XOP__
1952             eweps            = _mm_frcz_pd(ewrt);
1953 #else
1954             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1955 #endif
1956             twoeweps         = _mm_add_pd(eweps,eweps);
1957             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1958             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1959             felec            = _mm_mul_pd(_mm_mul_pd(qq23,rinv23),_mm_sub_pd(rinvsq23,felec));
1960
1961             fscal            = felec;
1962
1963             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1964
1965             /* Update vectorial force */
1966             fix2             = _mm_macc_pd(dx23,fscal,fix2);
1967             fiy2             = _mm_macc_pd(dy23,fscal,fiy2);
1968             fiz2             = _mm_macc_pd(dz23,fscal,fiz2);
1969
1970             fjx3             = _mm_macc_pd(dx23,fscal,fjx3);
1971             fjy3             = _mm_macc_pd(dy23,fscal,fjy3);
1972             fjz3             = _mm_macc_pd(dz23,fscal,fjz3);
1973
1974             /**************************
1975              * CALCULATE INTERACTIONS *
1976              **************************/
1977
1978             r31              = _mm_mul_pd(rsq31,rinv31);
1979
1980             /* EWALD ELECTROSTATICS */
1981
1982             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1983             ewrt             = _mm_mul_pd(r31,ewtabscale);
1984             ewitab           = _mm_cvttpd_epi32(ewrt);
1985 #ifdef __XOP__
1986             eweps            = _mm_frcz_pd(ewrt);
1987 #else
1988             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1989 #endif
1990             twoeweps         = _mm_add_pd(eweps,eweps);
1991             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1992             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1993             felec            = _mm_mul_pd(_mm_mul_pd(qq31,rinv31),_mm_sub_pd(rinvsq31,felec));
1994
1995             fscal            = felec;
1996
1997             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1998
1999             /* Update vectorial force */
2000             fix3             = _mm_macc_pd(dx31,fscal,fix3);
2001             fiy3             = _mm_macc_pd(dy31,fscal,fiy3);
2002             fiz3             = _mm_macc_pd(dz31,fscal,fiz3);
2003
2004             fjx1             = _mm_macc_pd(dx31,fscal,fjx1);
2005             fjy1             = _mm_macc_pd(dy31,fscal,fjy1);
2006             fjz1             = _mm_macc_pd(dz31,fscal,fjz1);
2007
2008             /**************************
2009              * CALCULATE INTERACTIONS *
2010              **************************/
2011
2012             r32              = _mm_mul_pd(rsq32,rinv32);
2013
2014             /* EWALD ELECTROSTATICS */
2015
2016             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2017             ewrt             = _mm_mul_pd(r32,ewtabscale);
2018             ewitab           = _mm_cvttpd_epi32(ewrt);
2019 #ifdef __XOP__
2020             eweps            = _mm_frcz_pd(ewrt);
2021 #else
2022             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2023 #endif
2024             twoeweps         = _mm_add_pd(eweps,eweps);
2025             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2026             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2027             felec            = _mm_mul_pd(_mm_mul_pd(qq32,rinv32),_mm_sub_pd(rinvsq32,felec));
2028
2029             fscal            = felec;
2030
2031             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2032
2033             /* Update vectorial force */
2034             fix3             = _mm_macc_pd(dx32,fscal,fix3);
2035             fiy3             = _mm_macc_pd(dy32,fscal,fiy3);
2036             fiz3             = _mm_macc_pd(dz32,fscal,fiz3);
2037
2038             fjx2             = _mm_macc_pd(dx32,fscal,fjx2);
2039             fjy2             = _mm_macc_pd(dy32,fscal,fjy2);
2040             fjz2             = _mm_macc_pd(dz32,fscal,fjz2);
2041
2042             /**************************
2043              * CALCULATE INTERACTIONS *
2044              **************************/
2045
2046             r33              = _mm_mul_pd(rsq33,rinv33);
2047
2048             /* EWALD ELECTROSTATICS */
2049
2050             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2051             ewrt             = _mm_mul_pd(r33,ewtabscale);
2052             ewitab           = _mm_cvttpd_epi32(ewrt);
2053 #ifdef __XOP__
2054             eweps            = _mm_frcz_pd(ewrt);
2055 #else
2056             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2057 #endif
2058             twoeweps         = _mm_add_pd(eweps,eweps);
2059             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2060             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2061             felec            = _mm_mul_pd(_mm_mul_pd(qq33,rinv33),_mm_sub_pd(rinvsq33,felec));
2062
2063             fscal            = felec;
2064
2065             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2066
2067             /* Update vectorial force */
2068             fix3             = _mm_macc_pd(dx33,fscal,fix3);
2069             fiy3             = _mm_macc_pd(dy33,fscal,fiy3);
2070             fiz3             = _mm_macc_pd(dz33,fscal,fiz3);
2071
2072             fjx3             = _mm_macc_pd(dx33,fscal,fjx3);
2073             fjy3             = _mm_macc_pd(dy33,fscal,fjy3);
2074             fjz3             = _mm_macc_pd(dz33,fscal,fjz3);
2075
2076             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2077
2078             /* Inner loop uses 351 flops */
2079         }
2080
2081         /* End of innermost loop */
2082
2083         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2084                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
2085
2086         /* Increment number of inner iterations */
2087         inneriter                  += j_index_end - j_index_start;
2088
2089         /* Outer loop uses 18 flops */
2090     }
2091
2092     /* Increment number of outer iterations */
2093     outeriter        += nri;
2094
2095     /* Update outer/inner flops */
2096
2097     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*351);
2098 }