src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwNone_GeomW3W3_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3W3_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Water3
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwNone_GeomW3W3_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     int              vdwjidx1A,vdwjidx1B;
  88     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  89     int              vdwjidx2A,vdwjidx2B;
  90     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  91     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  93     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  94     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  96     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  97     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  99     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 100     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     __m128i          ewitab;
 103     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     real             *ewtab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122
 123     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 124     ewtab            = fr->ic->tabq_coul_FDV0;
 125     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 126     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 131     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 132     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 133
 134     jq0              = _mm_set1_pd(charge[inr+0]);
 135     jq1              = _mm_set1_pd(charge[inr+1]);
 136     jq2              = _mm_set1_pd(charge[inr+2]);
 137     qq00             = _mm_mul_pd(iq0,jq0);
 138     qq01             = _mm_mul_pd(iq0,jq1);
 139     qq02             = _mm_mul_pd(iq0,jq2);
 140     qq10             = _mm_mul_pd(iq1,jq0);
 141     qq11             = _mm_mul_pd(iq1,jq1);
 142     qq12             = _mm_mul_pd(iq1,jq2);
 143     qq20             = _mm_mul_pd(iq2,jq0);
 144     qq21             = _mm_mul_pd(iq2,jq1);
 145     qq22             = _mm_mul_pd(iq2,jq2);
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 172
 173         fix0             = _mm_setzero_pd();
 174         fiy0             = _mm_setzero_pd();
 175         fiz0             = _mm_setzero_pd();
 176         fix1             = _mm_setzero_pd();
 177         fiy1             = _mm_setzero_pd();
 178         fiz1             = _mm_setzero_pd();
 179         fix2             = _mm_setzero_pd();
 180         fiy2             = _mm_setzero_pd();
 181         fiz2             = _mm_setzero_pd();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm_setzero_pd();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _mm_sub_pd(ix0,jx0);
 202             dy00             = _mm_sub_pd(iy0,jy0);
 203             dz00             = _mm_sub_pd(iz0,jz0);
 204             dx01             = _mm_sub_pd(ix0,jx1);
 205             dy01             = _mm_sub_pd(iy0,jy1);
 206             dz01             = _mm_sub_pd(iz0,jz1);
 207             dx02             = _mm_sub_pd(ix0,jx2);
 208             dy02             = _mm_sub_pd(iy0,jy2);
 209             dz02             = _mm_sub_pd(iz0,jz2);
 210             dx10             = _mm_sub_pd(ix1,jx0);
 211             dy10             = _mm_sub_pd(iy1,jy0);
 212             dz10             = _mm_sub_pd(iz1,jz0);
 213             dx11             = _mm_sub_pd(ix1,jx1);
 214             dy11             = _mm_sub_pd(iy1,jy1);
 215             dz11             = _mm_sub_pd(iz1,jz1);
 216             dx12             = _mm_sub_pd(ix1,jx2);
 217             dy12             = _mm_sub_pd(iy1,jy2);
 218             dz12             = _mm_sub_pd(iz1,jz2);
 219             dx20             = _mm_sub_pd(ix2,jx0);
 220             dy20             = _mm_sub_pd(iy2,jy0);
 221             dz20             = _mm_sub_pd(iz2,jz0);
 222             dx21             = _mm_sub_pd(ix2,jx1);
 223             dy21             = _mm_sub_pd(iy2,jy1);
 224             dz21             = _mm_sub_pd(iz2,jz1);
 225             dx22             = _mm_sub_pd(ix2,jx2);
 226             dy22             = _mm_sub_pd(iy2,jy2);
 227             dz22             = _mm_sub_pd(iz2,jz2);
 228
 229             /* Calculate squared distance and things based on it */
 230             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 231             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 232             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 233             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 234             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 235             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 236             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 237             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 238             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 239
 240             rinv00           = avx128fma_invsqrt_d(rsq00);
 241             rinv01           = avx128fma_invsqrt_d(rsq01);
 242             rinv02           = avx128fma_invsqrt_d(rsq02);
 243             rinv10           = avx128fma_invsqrt_d(rsq10);
 244             rinv11           = avx128fma_invsqrt_d(rsq11);
 245             rinv12           = avx128fma_invsqrt_d(rsq12);
 246             rinv20           = avx128fma_invsqrt_d(rsq20);
 247             rinv21           = avx128fma_invsqrt_d(rsq21);
 248             rinv22           = avx128fma_invsqrt_d(rsq22);
 249
 250             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 251             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 252             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 253             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 254             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 255             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 256             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 257             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 258             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 259
 260             fjx0             = _mm_setzero_pd();
 261             fjy0             = _mm_setzero_pd();
 262             fjz0             = _mm_setzero_pd();
 263             fjx1             = _mm_setzero_pd();
 264             fjy1             = _mm_setzero_pd();
 265             fjz1             = _mm_setzero_pd();
 266             fjx2             = _mm_setzero_pd();
 267             fjy2             = _mm_setzero_pd();
 268             fjz2             = _mm_setzero_pd();
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             r00              = _mm_mul_pd(rsq00,rinv00);
 275
 276             /* EWALD ELECTROSTATICS */
 277
 278             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 279             ewrt             = _mm_mul_pd(r00,ewtabscale);
 280             ewitab           = _mm_cvttpd_epi32(ewrt);
 281 #ifdef __XOP__
 282             eweps            = _mm_frcz_pd(ewrt);
 283 #else
 284             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 285 #endif
 286             twoeweps         = _mm_add_pd(eweps,eweps);
 287             ewitab           = _mm_slli_epi32(ewitab,2);
 288             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 289             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 290             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 291             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 292             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 293             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 294             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 295             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 296             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 297             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velecsum         = _mm_add_pd(velecsum,velec);
 301
 302             fscal            = felec;
 303
 304             /* Update vectorial force */
 305             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 306             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 307             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 308
 309             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 310             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 311             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             r01              = _mm_mul_pd(rsq01,rinv01);
 318
 319             /* EWALD ELECTROSTATICS */
 320
 321             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 322             ewrt             = _mm_mul_pd(r01,ewtabscale);
 323             ewitab           = _mm_cvttpd_epi32(ewrt);
 324 #ifdef __XOP__
 325             eweps            = _mm_frcz_pd(ewrt);
 326 #else
 327             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 328 #endif
 329             twoeweps         = _mm_add_pd(eweps,eweps);
 330             ewitab           = _mm_slli_epi32(ewitab,2);
 331             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 332             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 333             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 334             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 335             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 336             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 337             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 338             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 339             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
 340             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velecsum         = _mm_add_pd(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             /* Update vectorial force */
 348             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 349             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 350             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 351
 352             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 353             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 354             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 355
 356             /**************************
 357              * CALCULATE INTERACTIONS *
 358              **************************/
 359
 360             r02              = _mm_mul_pd(rsq02,rinv02);
 361
 362             /* EWALD ELECTROSTATICS */
 363
 364             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 365             ewrt             = _mm_mul_pd(r02,ewtabscale);
 366             ewitab           = _mm_cvttpd_epi32(ewrt);
 367 #ifdef __XOP__
 368             eweps            = _mm_frcz_pd(ewrt);
 369 #else
 370             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 371 #endif
 372             twoeweps         = _mm_add_pd(eweps,eweps);
 373             ewitab           = _mm_slli_epi32(ewitab,2);
 374             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 375             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 376             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 377             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 378             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 379             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 380             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 381             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 382             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
 383             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velecsum         = _mm_add_pd(velecsum,velec);
 387
 388             fscal            = felec;
 389
 390             /* Update vectorial force */
 391             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 392             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 393             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 394
 395             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 396             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 397             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 398
 399             /**************************
 400              * CALCULATE INTERACTIONS *
 401              **************************/
 402
 403             r10              = _mm_mul_pd(rsq10,rinv10);
 404
 405             /* EWALD ELECTROSTATICS */
 406
 407             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 408             ewrt             = _mm_mul_pd(r10,ewtabscale);
 409             ewitab           = _mm_cvttpd_epi32(ewrt);
 410 #ifdef __XOP__
 411             eweps            = _mm_frcz_pd(ewrt);
 412 #else
 413             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 414 #endif
 415             twoeweps         = _mm_add_pd(eweps,eweps);
 416             ewitab           = _mm_slli_epi32(ewitab,2);
 417             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 418             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 419             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 420             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 421             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 422             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 423             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 424             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 425             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 426             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velecsum         = _mm_add_pd(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             /* Update vectorial force */
 434             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 435             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 436             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 437
 438             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 439             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 440             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 441
 442             /**************************
 443              * CALCULATE INTERACTIONS *
 444              **************************/
 445
 446             r11              = _mm_mul_pd(rsq11,rinv11);
 447
 448             /* EWALD ELECTROSTATICS */
 449
 450             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 451             ewrt             = _mm_mul_pd(r11,ewtabscale);
 452             ewitab           = _mm_cvttpd_epi32(ewrt);
 453 #ifdef __XOP__
 454             eweps            = _mm_frcz_pd(ewrt);
 455 #else
 456             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 457 #endif
 458             twoeweps         = _mm_add_pd(eweps,eweps);
 459             ewitab           = _mm_slli_epi32(ewitab,2);
 460             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 461             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 462             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 463             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 464             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 465             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 466             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 467             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 468             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 469             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 470
 471             /* Update potential sum for this i atom from the interaction with this j atom. */
 472             velecsum         = _mm_add_pd(velecsum,velec);
 473
 474             fscal            = felec;
 475
 476             /* Update vectorial force */
 477             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 478             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 479             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 480
 481             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 482             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 483             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 484
 485             /**************************
 486              * CALCULATE INTERACTIONS *
 487              **************************/
 488
 489             r12              = _mm_mul_pd(rsq12,rinv12);
 490
 491             /* EWALD ELECTROSTATICS */
 492
 493             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 494             ewrt             = _mm_mul_pd(r12,ewtabscale);
 495             ewitab           = _mm_cvttpd_epi32(ewrt);
 496 #ifdef __XOP__
 497             eweps            = _mm_frcz_pd(ewrt);
 498 #else
 499             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 500 #endif
 501             twoeweps         = _mm_add_pd(eweps,eweps);
 502             ewitab           = _mm_slli_epi32(ewitab,2);
 503             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 504             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 505             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 506             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 507             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 508             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 509             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 510             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 511             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 512             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 513
 514             /* Update potential sum for this i atom from the interaction with this j atom. */
 515             velecsum         = _mm_add_pd(velecsum,velec);
 516
 517             fscal            = felec;
 518
 519             /* Update vectorial force */
 520             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 521             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 522             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 523
 524             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 525             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 526             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             r20              = _mm_mul_pd(rsq20,rinv20);
 533
 534             /* EWALD ELECTROSTATICS */
 535
 536             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 537             ewrt             = _mm_mul_pd(r20,ewtabscale);
 538             ewitab           = _mm_cvttpd_epi32(ewrt);
 539 #ifdef __XOP__
 540             eweps            = _mm_frcz_pd(ewrt);
 541 #else
 542             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 543 #endif
 544             twoeweps         = _mm_add_pd(eweps,eweps);
 545             ewitab           = _mm_slli_epi32(ewitab,2);
 546             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 547             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 548             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 549             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 550             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 551             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 552             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 553             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 554             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 555             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             velecsum         = _mm_add_pd(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             /* Update vectorial force */
 563             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 564             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 565             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 566
 567             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 568             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 569             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 570
 571             /**************************
 572              * CALCULATE INTERACTIONS *
 573              **************************/
 574
 575             r21              = _mm_mul_pd(rsq21,rinv21);
 576
 577             /* EWALD ELECTROSTATICS */
 578
 579             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 580             ewrt             = _mm_mul_pd(r21,ewtabscale);
 581             ewitab           = _mm_cvttpd_epi32(ewrt);
 582 #ifdef __XOP__
 583             eweps            = _mm_frcz_pd(ewrt);
 584 #else
 585             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 586 #endif
 587             twoeweps         = _mm_add_pd(eweps,eweps);
 588             ewitab           = _mm_slli_epi32(ewitab,2);
 589             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 590             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 591             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 592             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 593             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 594             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 595             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 596             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 597             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
 598             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 599
 600             /* Update potential sum for this i atom from the interaction with this j atom. */
 601             velecsum         = _mm_add_pd(velecsum,velec);
 602
 603             fscal            = felec;
 604
 605             /* Update vectorial force */
 606             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 607             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 608             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 609
 610             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 611             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 612             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 613
 614             /**************************
 615              * CALCULATE INTERACTIONS *
 616              **************************/
 617
 618             r22              = _mm_mul_pd(rsq22,rinv22);
 619
 620             /* EWALD ELECTROSTATICS */
 621
 622             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 623             ewrt             = _mm_mul_pd(r22,ewtabscale);
 624             ewitab           = _mm_cvttpd_epi32(ewrt);
 625 #ifdef __XOP__
 626             eweps            = _mm_frcz_pd(ewrt);
 627 #else
 628             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 629 #endif
 630             twoeweps         = _mm_add_pd(eweps,eweps);
 631             ewitab           = _mm_slli_epi32(ewitab,2);
 632             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 633             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 634             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 635             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 636             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 637             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 638             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 639             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 640             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
 641             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 642
 643             /* Update potential sum for this i atom from the interaction with this j atom. */
 644             velecsum         = _mm_add_pd(velecsum,velec);
 645
 646             fscal            = felec;
 647
 648             /* Update vectorial force */
 649             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 650             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 651             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 652
 653             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 654             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 655             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 656
 657             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 658
 659             /* Inner loop uses 396 flops */
 660         }
 661
 662         if(jidx<j_index_end)
 663         {
 664
 665             jnrA             = jjnr[jidx];
 666             j_coord_offsetA  = DIM*jnrA;
 667
 668             /* load j atom coordinates */
 669             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 670                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 671
 672             /* Calculate displacement vector */
 673             dx00             = _mm_sub_pd(ix0,jx0);
 674             dy00             = _mm_sub_pd(iy0,jy0);
 675             dz00             = _mm_sub_pd(iz0,jz0);
 676             dx01             = _mm_sub_pd(ix0,jx1);
 677             dy01             = _mm_sub_pd(iy0,jy1);
 678             dz01             = _mm_sub_pd(iz0,jz1);
 679             dx02             = _mm_sub_pd(ix0,jx2);
 680             dy02             = _mm_sub_pd(iy0,jy2);
 681             dz02             = _mm_sub_pd(iz0,jz2);
 682             dx10             = _mm_sub_pd(ix1,jx0);
 683             dy10             = _mm_sub_pd(iy1,jy0);
 684             dz10             = _mm_sub_pd(iz1,jz0);
 685             dx11             = _mm_sub_pd(ix1,jx1);
 686             dy11             = _mm_sub_pd(iy1,jy1);
 687             dz11             = _mm_sub_pd(iz1,jz1);
 688             dx12             = _mm_sub_pd(ix1,jx2);
 689             dy12             = _mm_sub_pd(iy1,jy2);
 690             dz12             = _mm_sub_pd(iz1,jz2);
 691             dx20             = _mm_sub_pd(ix2,jx0);
 692             dy20             = _mm_sub_pd(iy2,jy0);
 693             dz20             = _mm_sub_pd(iz2,jz0);
 694             dx21             = _mm_sub_pd(ix2,jx1);
 695             dy21             = _mm_sub_pd(iy2,jy1);
 696             dz21             = _mm_sub_pd(iz2,jz1);
 697             dx22             = _mm_sub_pd(ix2,jx2);
 698             dy22             = _mm_sub_pd(iy2,jy2);
 699             dz22             = _mm_sub_pd(iz2,jz2);
 700
 701             /* Calculate squared distance and things based on it */
 702             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 703             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 704             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 705             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 706             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 707             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 708             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 709             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 710             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 711
 712             rinv00           = avx128fma_invsqrt_d(rsq00);
 713             rinv01           = avx128fma_invsqrt_d(rsq01);
 714             rinv02           = avx128fma_invsqrt_d(rsq02);
 715             rinv10           = avx128fma_invsqrt_d(rsq10);
 716             rinv11           = avx128fma_invsqrt_d(rsq11);
 717             rinv12           = avx128fma_invsqrt_d(rsq12);
 718             rinv20           = avx128fma_invsqrt_d(rsq20);
 719             rinv21           = avx128fma_invsqrt_d(rsq21);
 720             rinv22           = avx128fma_invsqrt_d(rsq22);
 721
 722             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 723             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 724             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 725             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 726             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 727             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 728             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 729             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 730             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 731
 732             fjx0             = _mm_setzero_pd();
 733             fjy0             = _mm_setzero_pd();
 734             fjz0             = _mm_setzero_pd();
 735             fjx1             = _mm_setzero_pd();
 736             fjy1             = _mm_setzero_pd();
 737             fjz1             = _mm_setzero_pd();
 738             fjx2             = _mm_setzero_pd();
 739             fjy2             = _mm_setzero_pd();
 740             fjz2             = _mm_setzero_pd();
 741
 742             /**************************
 743              * CALCULATE INTERACTIONS *
 744              **************************/
 745
 746             r00              = _mm_mul_pd(rsq00,rinv00);
 747
 748             /* EWALD ELECTROSTATICS */
 749
 750             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 751             ewrt             = _mm_mul_pd(r00,ewtabscale);
 752             ewitab           = _mm_cvttpd_epi32(ewrt);
 753 #ifdef __XOP__
 754             eweps            = _mm_frcz_pd(ewrt);
 755 #else
 756             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 757 #endif
 758             twoeweps         = _mm_add_pd(eweps,eweps);
 759             ewitab           = _mm_slli_epi32(ewitab,2);
 760             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 761             ewtabD           = _mm_setzero_pd();
 762             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 763             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 764             ewtabFn          = _mm_setzero_pd();
 765             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 766             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 767             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 768             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 769             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 770
 771             /* Update potential sum for this i atom from the interaction with this j atom. */
 772             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 773             velecsum         = _mm_add_pd(velecsum,velec);
 774
 775             fscal            = felec;
 776
 777             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 778
 779             /* Update vectorial force */
 780             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 781             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 782             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 783
 784             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 785             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 786             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 787
 788             /**************************
 789              * CALCULATE INTERACTIONS *
 790              **************************/
 791
 792             r01              = _mm_mul_pd(rsq01,rinv01);
 793
 794             /* EWALD ELECTROSTATICS */
 795
 796             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 797             ewrt             = _mm_mul_pd(r01,ewtabscale);
 798             ewitab           = _mm_cvttpd_epi32(ewrt);
 799 #ifdef __XOP__
 800             eweps            = _mm_frcz_pd(ewrt);
 801 #else
 802             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 803 #endif
 804             twoeweps         = _mm_add_pd(eweps,eweps);
 805             ewitab           = _mm_slli_epi32(ewitab,2);
 806             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 807             ewtabD           = _mm_setzero_pd();
 808             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 809             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 810             ewtabFn          = _mm_setzero_pd();
 811             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 812             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 813             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 814             velec            = _mm_mul_pd(qq01,_mm_sub_pd(rinv01,velec));
 815             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 816
 817             /* Update potential sum for this i atom from the interaction with this j atom. */
 818             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 819             velecsum         = _mm_add_pd(velecsum,velec);
 820
 821             fscal            = felec;
 822
 823             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 824
 825             /* Update vectorial force */
 826             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 827             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 828             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 829
 830             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 831             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 832             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 833
 834             /**************************
 835              * CALCULATE INTERACTIONS *
 836              **************************/
 837
 838             r02              = _mm_mul_pd(rsq02,rinv02);
 839
 840             /* EWALD ELECTROSTATICS */
 841
 842             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 843             ewrt             = _mm_mul_pd(r02,ewtabscale);
 844             ewitab           = _mm_cvttpd_epi32(ewrt);
 845 #ifdef __XOP__
 846             eweps            = _mm_frcz_pd(ewrt);
 847 #else
 848             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 849 #endif
 850             twoeweps         = _mm_add_pd(eweps,eweps);
 851             ewitab           = _mm_slli_epi32(ewitab,2);
 852             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 853             ewtabD           = _mm_setzero_pd();
 854             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 855             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 856             ewtabFn          = _mm_setzero_pd();
 857             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 858             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 859             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 860             velec            = _mm_mul_pd(qq02,_mm_sub_pd(rinv02,velec));
 861             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 862
 863             /* Update potential sum for this i atom from the interaction with this j atom. */
 864             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 865             velecsum         = _mm_add_pd(velecsum,velec);
 866
 867             fscal            = felec;
 868
 869             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 870
 871             /* Update vectorial force */
 872             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 873             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 874             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 875
 876             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 877             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 878             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 879
 880             /**************************
 881              * CALCULATE INTERACTIONS *
 882              **************************/
 883
 884             r10              = _mm_mul_pd(rsq10,rinv10);
 885
 886             /* EWALD ELECTROSTATICS */
 887
 888             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 889             ewrt             = _mm_mul_pd(r10,ewtabscale);
 890             ewitab           = _mm_cvttpd_epi32(ewrt);
 891 #ifdef __XOP__
 892             eweps            = _mm_frcz_pd(ewrt);
 893 #else
 894             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 895 #endif
 896             twoeweps         = _mm_add_pd(eweps,eweps);
 897             ewitab           = _mm_slli_epi32(ewitab,2);
 898             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 899             ewtabD           = _mm_setzero_pd();
 900             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 901             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 902             ewtabFn          = _mm_setzero_pd();
 903             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 904             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 905             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 906             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 907             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 908
 909             /* Update potential sum for this i atom from the interaction with this j atom. */
 910             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 911             velecsum         = _mm_add_pd(velecsum,velec);
 912
 913             fscal            = felec;
 914
 915             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 916
 917             /* Update vectorial force */
 918             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 919             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 920             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 921
 922             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 923             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 924             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 925
 926             /**************************
 927              * CALCULATE INTERACTIONS *
 928              **************************/
 929
 930             r11              = _mm_mul_pd(rsq11,rinv11);
 931
 932             /* EWALD ELECTROSTATICS */
 933
 934             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 935             ewrt             = _mm_mul_pd(r11,ewtabscale);
 936             ewitab           = _mm_cvttpd_epi32(ewrt);
 937 #ifdef __XOP__
 938             eweps            = _mm_frcz_pd(ewrt);
 939 #else
 940             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 941 #endif
 942             twoeweps         = _mm_add_pd(eweps,eweps);
 943             ewitab           = _mm_slli_epi32(ewitab,2);
 944             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 945             ewtabD           = _mm_setzero_pd();
 946             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 947             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 948             ewtabFn          = _mm_setzero_pd();
 949             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 950             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 951             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 952             velec            = _mm_mul_pd(qq11,_mm_sub_pd(rinv11,velec));
 953             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 954
 955             /* Update potential sum for this i atom from the interaction with this j atom. */
 956             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 957             velecsum         = _mm_add_pd(velecsum,velec);
 958
 959             fscal            = felec;
 960
 961             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 962
 963             /* Update vectorial force */
 964             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 965             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 966             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 967
 968             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 969             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 970             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 971
 972             /**************************
 973              * CALCULATE INTERACTIONS *
 974              **************************/
 975
 976             r12              = _mm_mul_pd(rsq12,rinv12);
 977
 978             /* EWALD ELECTROSTATICS */
 979
 980             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 981             ewrt             = _mm_mul_pd(r12,ewtabscale);
 982             ewitab           = _mm_cvttpd_epi32(ewrt);
 983 #ifdef __XOP__
 984             eweps            = _mm_frcz_pd(ewrt);
 985 #else
 986             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 987 #endif
 988             twoeweps         = _mm_add_pd(eweps,eweps);
 989             ewitab           = _mm_slli_epi32(ewitab,2);
 990             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 991             ewtabD           = _mm_setzero_pd();
 992             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 993             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 994             ewtabFn          = _mm_setzero_pd();
 995             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 996             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 997             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 998             velec            = _mm_mul_pd(qq12,_mm_sub_pd(rinv12,velec));
 999             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1000
1001             /* Update potential sum for this i atom from the interaction with this j atom. */
1002             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1003             velecsum         = _mm_add_pd(velecsum,velec);
1004
1005             fscal            = felec;
1006
1007             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1008
1009             /* Update vectorial force */
1010             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1011             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1012             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1013
1014             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1015             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1016             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1017
1018             /**************************
1019              * CALCULATE INTERACTIONS *
1020              **************************/
1021
1022             r20              = _mm_mul_pd(rsq20,rinv20);
1023
1024             /* EWALD ELECTROSTATICS */
1025
1026             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1027             ewrt             = _mm_mul_pd(r20,ewtabscale);
1028             ewitab           = _mm_cvttpd_epi32(ewrt);
1029 #ifdef __XOP__
1030             eweps            = _mm_frcz_pd(ewrt);
1031 #else
1032             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1033 #endif
1034             twoeweps         = _mm_add_pd(eweps,eweps);
1035             ewitab           = _mm_slli_epi32(ewitab,2);
1036             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1037             ewtabD           = _mm_setzero_pd();
1038             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1039             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1040             ewtabFn          = _mm_setzero_pd();
1041             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1042             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1043             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1044             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1045             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1046
1047             /* Update potential sum for this i atom from the interaction with this j atom. */
1048             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1049             velecsum         = _mm_add_pd(velecsum,velec);
1050
1051             fscal            = felec;
1052
1053             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1054
1055             /* Update vectorial force */
1056             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1057             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1058             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1059
1060             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1061             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1062             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             r21              = _mm_mul_pd(rsq21,rinv21);
1069
1070             /* EWALD ELECTROSTATICS */
1071
1072             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1073             ewrt             = _mm_mul_pd(r21,ewtabscale);
1074             ewitab           = _mm_cvttpd_epi32(ewrt);
1075 #ifdef __XOP__
1076             eweps            = _mm_frcz_pd(ewrt);
1077 #else
1078             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1079 #endif
1080             twoeweps         = _mm_add_pd(eweps,eweps);
1081             ewitab           = _mm_slli_epi32(ewitab,2);
1082             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1083             ewtabD           = _mm_setzero_pd();
1084             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1085             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1086             ewtabFn          = _mm_setzero_pd();
1087             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1088             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1089             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1090             velec            = _mm_mul_pd(qq21,_mm_sub_pd(rinv21,velec));
1091             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1092
1093             /* Update potential sum for this i atom from the interaction with this j atom. */
1094             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1095             velecsum         = _mm_add_pd(velecsum,velec);
1096
1097             fscal            = felec;
1098
1099             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1100
1101             /* Update vectorial force */
1102             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1103             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1104             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1105
1106             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1107             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1108             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1109
1110             /**************************
1111              * CALCULATE INTERACTIONS *
1112              **************************/
1113
1114             r22              = _mm_mul_pd(rsq22,rinv22);
1115
1116             /* EWALD ELECTROSTATICS */
1117
1118             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1119             ewrt             = _mm_mul_pd(r22,ewtabscale);
1120             ewitab           = _mm_cvttpd_epi32(ewrt);
1121 #ifdef __XOP__
1122             eweps            = _mm_frcz_pd(ewrt);
1123 #else
1124             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1125 #endif
1126             twoeweps         = _mm_add_pd(eweps,eweps);
1127             ewitab           = _mm_slli_epi32(ewitab,2);
1128             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1129             ewtabD           = _mm_setzero_pd();
1130             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1131             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1132             ewtabFn          = _mm_setzero_pd();
1133             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1134             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1135             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1136             velec            = _mm_mul_pd(qq22,_mm_sub_pd(rinv22,velec));
1137             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1138
1139             /* Update potential sum for this i atom from the interaction with this j atom. */
1140             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1141             velecsum         = _mm_add_pd(velecsum,velec);
1142
1143             fscal            = felec;
1144
1145             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1146
1147             /* Update vectorial force */
1148             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1149             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1150             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1151
1152             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1153             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1154             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1155
1156             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1157
1158             /* Inner loop uses 396 flops */
1159         }
1160
1161         /* End of innermost loop */
1162
1163         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1164                                               f+i_coord_offset,fshift+i_shift_offset);
1165
1166         ggid                        = gid[iidx];
1167         /* Update potential energies */
1168         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1169
1170         /* Increment number of inner iterations */
1171         inneriter                  += j_index_end - j_index_start;
1172
1173         /* Outer loop uses 19 flops */
1174     }
1175
1176     /* Increment number of outer iterations */
1177     outeriter        += nri;
1178
1179     /* Update outer/inner flops */
1180
1181     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*396);
1182 }
1183 /*
1184  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3W3_F_avx_128_fma_double
1185  * Electrostatics interaction: Ewald
1186  * VdW interaction:            None
1187  * Geometry:                   Water3-Water3
1188  * Calculate force/pot:        Force
1189  */
1190 void
1191 nb_kernel_ElecEw_VdwNone_GeomW3W3_F_avx_128_fma_double
1192                     (t_nblist                    * gmx_restrict       nlist,
1193                      rvec                        * gmx_restrict          xx,
1194                      rvec                        * gmx_restrict          ff,
1195                      struct t_forcerec           * gmx_restrict          fr,
1196                      t_mdatoms                   * gmx_restrict     mdatoms,
1197                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1198                      t_nrnb                      * gmx_restrict        nrnb)
1199 {
1200     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1201      * just 0 for non-waters.
1202      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1203      * jnr indices corresponding to data put in the four positions in the SIMD register.
1204      */
1205     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1206     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1207     int              jnrA,jnrB;
1208     int              j_coord_offsetA,j_coord_offsetB;
1209     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1210     real             rcutoff_scalar;
1211     real             *shiftvec,*fshift,*x,*f;
1212     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1213     int              vdwioffset0;
1214     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1215     int              vdwioffset1;
1216     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1217     int              vdwioffset2;
1218     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1219     int              vdwjidx0A,vdwjidx0B;
1220     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1221     int              vdwjidx1A,vdwjidx1B;
1222     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1223     int              vdwjidx2A,vdwjidx2B;
1224     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1225     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1226     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1227     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1228     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1229     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1230     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1231     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1232     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1233     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1234     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1235     real             *charge;
1236     __m128i          ewitab;
1237     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1238     real             *ewtab;
1239     __m128d          dummy_mask,cutoff_mask;
1240     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1241     __m128d          one     = _mm_set1_pd(1.0);
1242     __m128d          two     = _mm_set1_pd(2.0);
1243     x                = xx[0];
1244     f                = ff[0];
1245
1246     nri              = nlist->nri;
1247     iinr             = nlist->iinr;
1248     jindex           = nlist->jindex;
1249     jjnr             = nlist->jjnr;
1250     shiftidx         = nlist->shift;
1251     gid              = nlist->gid;
1252     shiftvec         = fr->shift_vec[0];
1253     fshift           = fr->fshift[0];
1254     facel            = _mm_set1_pd(fr->ic->epsfac);
1255     charge           = mdatoms->chargeA;
1256
1257     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1258     ewtab            = fr->ic->tabq_coul_F;
1259     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1260     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1261
1262     /* Setup water-specific parameters */
1263     inr              = nlist->iinr[0];
1264     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1265     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1266     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1267
1268     jq0              = _mm_set1_pd(charge[inr+0]);
1269     jq1              = _mm_set1_pd(charge[inr+1]);
1270     jq2              = _mm_set1_pd(charge[inr+2]);
1271     qq00             = _mm_mul_pd(iq0,jq0);
1272     qq01             = _mm_mul_pd(iq0,jq1);
1273     qq02             = _mm_mul_pd(iq0,jq2);
1274     qq10             = _mm_mul_pd(iq1,jq0);
1275     qq11             = _mm_mul_pd(iq1,jq1);
1276     qq12             = _mm_mul_pd(iq1,jq2);
1277     qq20             = _mm_mul_pd(iq2,jq0);
1278     qq21             = _mm_mul_pd(iq2,jq1);
1279     qq22             = _mm_mul_pd(iq2,jq2);
1280
1281     /* Avoid stupid compiler warnings */
1282     jnrA = jnrB = 0;
1283     j_coord_offsetA = 0;
1284     j_coord_offsetB = 0;
1285
1286     outeriter        = 0;
1287     inneriter        = 0;
1288
1289     /* Start outer loop over neighborlists */
1290     for(iidx=0; iidx<nri; iidx++)
1291     {
1292         /* Load shift vector for this list */
1293         i_shift_offset   = DIM*shiftidx[iidx];
1294
1295         /* Load limits for loop over neighbors */
1296         j_index_start    = jindex[iidx];
1297         j_index_end      = jindex[iidx+1];
1298
1299         /* Get outer coordinate index */
1300         inr              = iinr[iidx];
1301         i_coord_offset   = DIM*inr;
1302
1303         /* Load i particle coords and add shift vector */
1304         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1305                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1306
1307         fix0             = _mm_setzero_pd();
1308         fiy0             = _mm_setzero_pd();
1309         fiz0             = _mm_setzero_pd();
1310         fix1             = _mm_setzero_pd();
1311         fiy1             = _mm_setzero_pd();
1312         fiz1             = _mm_setzero_pd();
1313         fix2             = _mm_setzero_pd();
1314         fiy2             = _mm_setzero_pd();
1315         fiz2             = _mm_setzero_pd();
1316
1317         /* Start inner kernel loop */
1318         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1319         {
1320
1321             /* Get j neighbor index, and coordinate index */
1322             jnrA             = jjnr[jidx];
1323             jnrB             = jjnr[jidx+1];
1324             j_coord_offsetA  = DIM*jnrA;
1325             j_coord_offsetB  = DIM*jnrB;
1326
1327             /* load j atom coordinates */
1328             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1329                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1330
1331             /* Calculate displacement vector */
1332             dx00             = _mm_sub_pd(ix0,jx0);
1333             dy00             = _mm_sub_pd(iy0,jy0);
1334             dz00             = _mm_sub_pd(iz0,jz0);
1335             dx01             = _mm_sub_pd(ix0,jx1);
1336             dy01             = _mm_sub_pd(iy0,jy1);
1337             dz01             = _mm_sub_pd(iz0,jz1);
1338             dx02             = _mm_sub_pd(ix0,jx2);
1339             dy02             = _mm_sub_pd(iy0,jy2);
1340             dz02             = _mm_sub_pd(iz0,jz2);
1341             dx10             = _mm_sub_pd(ix1,jx0);
1342             dy10             = _mm_sub_pd(iy1,jy0);
1343             dz10             = _mm_sub_pd(iz1,jz0);
1344             dx11             = _mm_sub_pd(ix1,jx1);
1345             dy11             = _mm_sub_pd(iy1,jy1);
1346             dz11             = _mm_sub_pd(iz1,jz1);
1347             dx12             = _mm_sub_pd(ix1,jx2);
1348             dy12             = _mm_sub_pd(iy1,jy2);
1349             dz12             = _mm_sub_pd(iz1,jz2);
1350             dx20             = _mm_sub_pd(ix2,jx0);
1351             dy20             = _mm_sub_pd(iy2,jy0);
1352             dz20             = _mm_sub_pd(iz2,jz0);
1353             dx21             = _mm_sub_pd(ix2,jx1);
1354             dy21             = _mm_sub_pd(iy2,jy1);
1355             dz21             = _mm_sub_pd(iz2,jz1);
1356             dx22             = _mm_sub_pd(ix2,jx2);
1357             dy22             = _mm_sub_pd(iy2,jy2);
1358             dz22             = _mm_sub_pd(iz2,jz2);
1359
1360             /* Calculate squared distance and things based on it */
1361             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1362             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1363             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1364             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1365             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1366             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1367             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1368             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1369             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1370
1371             rinv00           = avx128fma_invsqrt_d(rsq00);
1372             rinv01           = avx128fma_invsqrt_d(rsq01);
1373             rinv02           = avx128fma_invsqrt_d(rsq02);
1374             rinv10           = avx128fma_invsqrt_d(rsq10);
1375             rinv11           = avx128fma_invsqrt_d(rsq11);
1376             rinv12           = avx128fma_invsqrt_d(rsq12);
1377             rinv20           = avx128fma_invsqrt_d(rsq20);
1378             rinv21           = avx128fma_invsqrt_d(rsq21);
1379             rinv22           = avx128fma_invsqrt_d(rsq22);
1380
1381             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1382             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1383             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1384             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1385             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1386             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1387             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1388             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1389             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1390
1391             fjx0             = _mm_setzero_pd();
1392             fjy0             = _mm_setzero_pd();
1393             fjz0             = _mm_setzero_pd();
1394             fjx1             = _mm_setzero_pd();
1395             fjy1             = _mm_setzero_pd();
1396             fjz1             = _mm_setzero_pd();
1397             fjx2             = _mm_setzero_pd();
1398             fjy2             = _mm_setzero_pd();
1399             fjz2             = _mm_setzero_pd();
1400
1401             /**************************
1402              * CALCULATE INTERACTIONS *
1403              **************************/
1404
1405             r00              = _mm_mul_pd(rsq00,rinv00);
1406
1407             /* EWALD ELECTROSTATICS */
1408
1409             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1410             ewrt             = _mm_mul_pd(r00,ewtabscale);
1411             ewitab           = _mm_cvttpd_epi32(ewrt);
1412 #ifdef __XOP__
1413             eweps            = _mm_frcz_pd(ewrt);
1414 #else
1415             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1416 #endif
1417             twoeweps         = _mm_add_pd(eweps,eweps);
1418             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1419                                          &ewtabF,&ewtabFn);
1420             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1421             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1422
1423             fscal            = felec;
1424
1425             /* Update vectorial force */
1426             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1427             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1428             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1429
1430             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1431             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1432             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1433
1434             /**************************
1435              * CALCULATE INTERACTIONS *
1436              **************************/
1437
1438             r01              = _mm_mul_pd(rsq01,rinv01);
1439
1440             /* EWALD ELECTROSTATICS */
1441
1442             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1443             ewrt             = _mm_mul_pd(r01,ewtabscale);
1444             ewitab           = _mm_cvttpd_epi32(ewrt);
1445 #ifdef __XOP__
1446             eweps            = _mm_frcz_pd(ewrt);
1447 #else
1448             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1449 #endif
1450             twoeweps         = _mm_add_pd(eweps,eweps);
1451             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1452                                          &ewtabF,&ewtabFn);
1453             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1454             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1455
1456             fscal            = felec;
1457
1458             /* Update vectorial force */
1459             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1460             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1461             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1462
1463             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1464             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1465             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1466
1467             /**************************
1468              * CALCULATE INTERACTIONS *
1469              **************************/
1470
1471             r02              = _mm_mul_pd(rsq02,rinv02);
1472
1473             /* EWALD ELECTROSTATICS */
1474
1475             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1476             ewrt             = _mm_mul_pd(r02,ewtabscale);
1477             ewitab           = _mm_cvttpd_epi32(ewrt);
1478 #ifdef __XOP__
1479             eweps            = _mm_frcz_pd(ewrt);
1480 #else
1481             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1482 #endif
1483             twoeweps         = _mm_add_pd(eweps,eweps);
1484             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1485                                          &ewtabF,&ewtabFn);
1486             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1487             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1488
1489             fscal            = felec;
1490
1491             /* Update vectorial force */
1492             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1493             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1494             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1495
1496             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1497             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1498             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1499
1500             /**************************
1501              * CALCULATE INTERACTIONS *
1502              **************************/
1503
1504             r10              = _mm_mul_pd(rsq10,rinv10);
1505
1506             /* EWALD ELECTROSTATICS */
1507
1508             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1509             ewrt             = _mm_mul_pd(r10,ewtabscale);
1510             ewitab           = _mm_cvttpd_epi32(ewrt);
1511 #ifdef __XOP__
1512             eweps            = _mm_frcz_pd(ewrt);
1513 #else
1514             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1515 #endif
1516             twoeweps         = _mm_add_pd(eweps,eweps);
1517             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1518                                          &ewtabF,&ewtabFn);
1519             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1520             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1521
1522             fscal            = felec;
1523
1524             /* Update vectorial force */
1525             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1526             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1527             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1528
1529             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1530             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1531             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1532
1533             /**************************
1534              * CALCULATE INTERACTIONS *
1535              **************************/
1536
1537             r11              = _mm_mul_pd(rsq11,rinv11);
1538
1539             /* EWALD ELECTROSTATICS */
1540
1541             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1542             ewrt             = _mm_mul_pd(r11,ewtabscale);
1543             ewitab           = _mm_cvttpd_epi32(ewrt);
1544 #ifdef __XOP__
1545             eweps            = _mm_frcz_pd(ewrt);
1546 #else
1547             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1548 #endif
1549             twoeweps         = _mm_add_pd(eweps,eweps);
1550             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1551                                          &ewtabF,&ewtabFn);
1552             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1553             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1554
1555             fscal            = felec;
1556
1557             /* Update vectorial force */
1558             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1559             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1560             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1561
1562             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1563             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1564             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1565
1566             /**************************
1567              * CALCULATE INTERACTIONS *
1568              **************************/
1569
1570             r12              = _mm_mul_pd(rsq12,rinv12);
1571
1572             /* EWALD ELECTROSTATICS */
1573
1574             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1575             ewrt             = _mm_mul_pd(r12,ewtabscale);
1576             ewitab           = _mm_cvttpd_epi32(ewrt);
1577 #ifdef __XOP__
1578             eweps            = _mm_frcz_pd(ewrt);
1579 #else
1580             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1581 #endif
1582             twoeweps         = _mm_add_pd(eweps,eweps);
1583             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1584                                          &ewtabF,&ewtabFn);
1585             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1586             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1587
1588             fscal            = felec;
1589
1590             /* Update vectorial force */
1591             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1592             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1593             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1594
1595             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1596             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1597             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1598
1599             /**************************
1600              * CALCULATE INTERACTIONS *
1601              **************************/
1602
1603             r20              = _mm_mul_pd(rsq20,rinv20);
1604
1605             /* EWALD ELECTROSTATICS */
1606
1607             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1608             ewrt             = _mm_mul_pd(r20,ewtabscale);
1609             ewitab           = _mm_cvttpd_epi32(ewrt);
1610 #ifdef __XOP__
1611             eweps            = _mm_frcz_pd(ewrt);
1612 #else
1613             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1614 #endif
1615             twoeweps         = _mm_add_pd(eweps,eweps);
1616             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1617                                          &ewtabF,&ewtabFn);
1618             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1619             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1620
1621             fscal            = felec;
1622
1623             /* Update vectorial force */
1624             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1625             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1626             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1627
1628             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1629             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1630             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1631
1632             /**************************
1633              * CALCULATE INTERACTIONS *
1634              **************************/
1635
1636             r21              = _mm_mul_pd(rsq21,rinv21);
1637
1638             /* EWALD ELECTROSTATICS */
1639
1640             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1641             ewrt             = _mm_mul_pd(r21,ewtabscale);
1642             ewitab           = _mm_cvttpd_epi32(ewrt);
1643 #ifdef __XOP__
1644             eweps            = _mm_frcz_pd(ewrt);
1645 #else
1646             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1647 #endif
1648             twoeweps         = _mm_add_pd(eweps,eweps);
1649             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1650                                          &ewtabF,&ewtabFn);
1651             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1652             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1653
1654             fscal            = felec;
1655
1656             /* Update vectorial force */
1657             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1658             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1659             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1660
1661             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1662             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1663             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1664
1665             /**************************
1666              * CALCULATE INTERACTIONS *
1667              **************************/
1668
1669             r22              = _mm_mul_pd(rsq22,rinv22);
1670
1671             /* EWALD ELECTROSTATICS */
1672
1673             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1674             ewrt             = _mm_mul_pd(r22,ewtabscale);
1675             ewitab           = _mm_cvttpd_epi32(ewrt);
1676 #ifdef __XOP__
1677             eweps            = _mm_frcz_pd(ewrt);
1678 #else
1679             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1680 #endif
1681             twoeweps         = _mm_add_pd(eweps,eweps);
1682             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1683                                          &ewtabF,&ewtabFn);
1684             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1685             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1686
1687             fscal            = felec;
1688
1689             /* Update vectorial force */
1690             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1691             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1692             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1693
1694             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1695             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1696             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1697
1698             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1699
1700             /* Inner loop uses 351 flops */
1701         }
1702
1703         if(jidx<j_index_end)
1704         {
1705
1706             jnrA             = jjnr[jidx];
1707             j_coord_offsetA  = DIM*jnrA;
1708
1709             /* load j atom coordinates */
1710             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1711                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1712
1713             /* Calculate displacement vector */
1714             dx00             = _mm_sub_pd(ix0,jx0);
1715             dy00             = _mm_sub_pd(iy0,jy0);
1716             dz00             = _mm_sub_pd(iz0,jz0);
1717             dx01             = _mm_sub_pd(ix0,jx1);
1718             dy01             = _mm_sub_pd(iy0,jy1);
1719             dz01             = _mm_sub_pd(iz0,jz1);
1720             dx02             = _mm_sub_pd(ix0,jx2);
1721             dy02             = _mm_sub_pd(iy0,jy2);
1722             dz02             = _mm_sub_pd(iz0,jz2);
1723             dx10             = _mm_sub_pd(ix1,jx0);
1724             dy10             = _mm_sub_pd(iy1,jy0);
1725             dz10             = _mm_sub_pd(iz1,jz0);
1726             dx11             = _mm_sub_pd(ix1,jx1);
1727             dy11             = _mm_sub_pd(iy1,jy1);
1728             dz11             = _mm_sub_pd(iz1,jz1);
1729             dx12             = _mm_sub_pd(ix1,jx2);
1730             dy12             = _mm_sub_pd(iy1,jy2);
1731             dz12             = _mm_sub_pd(iz1,jz2);
1732             dx20             = _mm_sub_pd(ix2,jx0);
1733             dy20             = _mm_sub_pd(iy2,jy0);
1734             dz20             = _mm_sub_pd(iz2,jz0);
1735             dx21             = _mm_sub_pd(ix2,jx1);
1736             dy21             = _mm_sub_pd(iy2,jy1);
1737             dz21             = _mm_sub_pd(iz2,jz1);
1738             dx22             = _mm_sub_pd(ix2,jx2);
1739             dy22             = _mm_sub_pd(iy2,jy2);
1740             dz22             = _mm_sub_pd(iz2,jz2);
1741
1742             /* Calculate squared distance and things based on it */
1743             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1744             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1745             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1746             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1747             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1748             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1749             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1750             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1751             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1752
1753             rinv00           = avx128fma_invsqrt_d(rsq00);
1754             rinv01           = avx128fma_invsqrt_d(rsq01);
1755             rinv02           = avx128fma_invsqrt_d(rsq02);
1756             rinv10           = avx128fma_invsqrt_d(rsq10);
1757             rinv11           = avx128fma_invsqrt_d(rsq11);
1758             rinv12           = avx128fma_invsqrt_d(rsq12);
1759             rinv20           = avx128fma_invsqrt_d(rsq20);
1760             rinv21           = avx128fma_invsqrt_d(rsq21);
1761             rinv22           = avx128fma_invsqrt_d(rsq22);
1762
1763             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1764             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1765             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1766             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1767             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1768             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1769             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1770             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1771             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1772
1773             fjx0             = _mm_setzero_pd();
1774             fjy0             = _mm_setzero_pd();
1775             fjz0             = _mm_setzero_pd();
1776             fjx1             = _mm_setzero_pd();
1777             fjy1             = _mm_setzero_pd();
1778             fjz1             = _mm_setzero_pd();
1779             fjx2             = _mm_setzero_pd();
1780             fjy2             = _mm_setzero_pd();
1781             fjz2             = _mm_setzero_pd();
1782
1783             /**************************
1784              * CALCULATE INTERACTIONS *
1785              **************************/
1786
1787             r00              = _mm_mul_pd(rsq00,rinv00);
1788
1789             /* EWALD ELECTROSTATICS */
1790
1791             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1792             ewrt             = _mm_mul_pd(r00,ewtabscale);
1793             ewitab           = _mm_cvttpd_epi32(ewrt);
1794 #ifdef __XOP__
1795             eweps            = _mm_frcz_pd(ewrt);
1796 #else
1797             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1798 #endif
1799             twoeweps         = _mm_add_pd(eweps,eweps);
1800             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1801             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1802             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1803
1804             fscal            = felec;
1805
1806             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1807
1808             /* Update vectorial force */
1809             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1810             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1811             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1812
1813             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1814             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1815             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1816
1817             /**************************
1818              * CALCULATE INTERACTIONS *
1819              **************************/
1820
1821             r01              = _mm_mul_pd(rsq01,rinv01);
1822
1823             /* EWALD ELECTROSTATICS */
1824
1825             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1826             ewrt             = _mm_mul_pd(r01,ewtabscale);
1827             ewitab           = _mm_cvttpd_epi32(ewrt);
1828 #ifdef __XOP__
1829             eweps            = _mm_frcz_pd(ewrt);
1830 #else
1831             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1832 #endif
1833             twoeweps         = _mm_add_pd(eweps,eweps);
1834             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1835             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1836             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1837
1838             fscal            = felec;
1839
1840             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1841
1842             /* Update vectorial force */
1843             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1844             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1845             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1846
1847             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1848             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1849             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1850
1851             /**************************
1852              * CALCULATE INTERACTIONS *
1853              **************************/
1854
1855             r02              = _mm_mul_pd(rsq02,rinv02);
1856
1857             /* EWALD ELECTROSTATICS */
1858
1859             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1860             ewrt             = _mm_mul_pd(r02,ewtabscale);
1861             ewitab           = _mm_cvttpd_epi32(ewrt);
1862 #ifdef __XOP__
1863             eweps            = _mm_frcz_pd(ewrt);
1864 #else
1865             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1866 #endif
1867             twoeweps         = _mm_add_pd(eweps,eweps);
1868             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1869             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1870             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1871
1872             fscal            = felec;
1873
1874             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1875
1876             /* Update vectorial force */
1877             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1878             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1879             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1880
1881             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1882             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1883             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1884
1885             /**************************
1886              * CALCULATE INTERACTIONS *
1887              **************************/
1888
1889             r10              = _mm_mul_pd(rsq10,rinv10);
1890
1891             /* EWALD ELECTROSTATICS */
1892
1893             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1894             ewrt             = _mm_mul_pd(r10,ewtabscale);
1895             ewitab           = _mm_cvttpd_epi32(ewrt);
1896 #ifdef __XOP__
1897             eweps            = _mm_frcz_pd(ewrt);
1898 #else
1899             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1900 #endif
1901             twoeweps         = _mm_add_pd(eweps,eweps);
1902             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1903             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1904             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1905
1906             fscal            = felec;
1907
1908             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1909
1910             /* Update vectorial force */
1911             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1912             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1913             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1914
1915             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1916             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1917             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1918
1919             /**************************
1920              * CALCULATE INTERACTIONS *
1921              **************************/
1922
1923             r11              = _mm_mul_pd(rsq11,rinv11);
1924
1925             /* EWALD ELECTROSTATICS */
1926
1927             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1928             ewrt             = _mm_mul_pd(r11,ewtabscale);
1929             ewitab           = _mm_cvttpd_epi32(ewrt);
1930 #ifdef __XOP__
1931             eweps            = _mm_frcz_pd(ewrt);
1932 #else
1933             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1934 #endif
1935             twoeweps         = _mm_add_pd(eweps,eweps);
1936             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1937             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1938             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1939
1940             fscal            = felec;
1941
1942             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1943
1944             /* Update vectorial force */
1945             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1946             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1947             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1948
1949             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1950             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1951             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1952
1953             /**************************
1954              * CALCULATE INTERACTIONS *
1955              **************************/
1956
1957             r12              = _mm_mul_pd(rsq12,rinv12);
1958
1959             /* EWALD ELECTROSTATICS */
1960
1961             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1962             ewrt             = _mm_mul_pd(r12,ewtabscale);
1963             ewitab           = _mm_cvttpd_epi32(ewrt);
1964 #ifdef __XOP__
1965             eweps            = _mm_frcz_pd(ewrt);
1966 #else
1967             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1968 #endif
1969             twoeweps         = _mm_add_pd(eweps,eweps);
1970             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1971             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1972             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1973
1974             fscal            = felec;
1975
1976             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1977
1978             /* Update vectorial force */
1979             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1980             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1981             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1982
1983             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1984             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1985             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1986
1987             /**************************
1988              * CALCULATE INTERACTIONS *
1989              **************************/
1990
1991             r20              = _mm_mul_pd(rsq20,rinv20);
1992
1993             /* EWALD ELECTROSTATICS */
1994
1995             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1996             ewrt             = _mm_mul_pd(r20,ewtabscale);
1997             ewitab           = _mm_cvttpd_epi32(ewrt);
1998 #ifdef __XOP__
1999             eweps            = _mm_frcz_pd(ewrt);
2000 #else
2001             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2002 #endif
2003             twoeweps         = _mm_add_pd(eweps,eweps);
2004             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2005             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2006             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2007
2008             fscal            = felec;
2009
2010             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2011
2012             /* Update vectorial force */
2013             fix2             = _mm_macc_pd(dx20,fscal,fix2);
2014             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
2015             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
2016
2017             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2018             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2019             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
2020
2021             /**************************
2022              * CALCULATE INTERACTIONS *
2023              **************************/
2024
2025             r21              = _mm_mul_pd(rsq21,rinv21);
2026
2027             /* EWALD ELECTROSTATICS */
2028
2029             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2030             ewrt             = _mm_mul_pd(r21,ewtabscale);
2031             ewitab           = _mm_cvttpd_epi32(ewrt);
2032 #ifdef __XOP__
2033             eweps            = _mm_frcz_pd(ewrt);
2034 #else
2035             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2036 #endif
2037             twoeweps         = _mm_add_pd(eweps,eweps);
2038             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2039             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2040             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2041
2042             fscal            = felec;
2043
2044             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2045
2046             /* Update vectorial force */
2047             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2048             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2049             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2050
2051             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2052             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2053             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2054
2055             /**************************
2056              * CALCULATE INTERACTIONS *
2057              **************************/
2058
2059             r22              = _mm_mul_pd(rsq22,rinv22);
2060
2061             /* EWALD ELECTROSTATICS */
2062
2063             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2064             ewrt             = _mm_mul_pd(r22,ewtabscale);
2065             ewitab           = _mm_cvttpd_epi32(ewrt);
2066 #ifdef __XOP__
2067             eweps            = _mm_frcz_pd(ewrt);
2068 #else
2069             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2070 #endif
2071             twoeweps         = _mm_add_pd(eweps,eweps);
2072             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2073             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2074             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2075
2076             fscal            = felec;
2077
2078             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2079
2080             /* Update vectorial force */
2081             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2082             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2083             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2084
2085             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2086             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2087             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2088
2089             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2090
2091             /* Inner loop uses 351 flops */
2092         }
2093
2094         /* End of innermost loop */
2095
2096         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2097                                               f+i_coord_offset,fshift+i_shift_offset);
2098
2099         /* Increment number of inner iterations */
2100         inneriter                  += j_index_end - j_index_start;
2101
2102         /* Outer loop uses 18 flops */
2103     }
2104
2105     /* Increment number of outer iterations */
2106     outeriter        += nri;
2107
2108     /* Update outer/inner flops */
2109
2110     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*351);
2111 }