src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwNone_GeomW3W3_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3W3_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     int              vdwjidx1A,vdwjidx1B;
  91     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  92     int              vdwjidx2A,vdwjidx2B;
  93     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  94     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  96     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  97     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  99     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 100     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 102     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 103     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 104     real             *charge;
 105     __m128i          ewitab;
 106     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     real             *ewtab;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125
 126     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 134     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 135     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 136
 137     jq0              = _mm_set1_pd(charge[inr+0]);
 138     jq1              = _mm_set1_pd(charge[inr+1]);
 139     jq2              = _mm_set1_pd(charge[inr+2]);
 140     qq00             = _mm_mul_pd(iq0,jq0);
 141     qq01             = _mm_mul_pd(iq0,jq1);
 142     qq02             = _mm_mul_pd(iq0,jq2);
 143     qq10             = _mm_mul_pd(iq1,jq0);
 144     qq11             = _mm_mul_pd(iq1,jq1);
 145     qq12             = _mm_mul_pd(iq1,jq2);
 146     qq20             = _mm_mul_pd(iq2,jq0);
 147     qq21             = _mm_mul_pd(iq2,jq1);
 148     qq22             = _mm_mul_pd(iq2,jq2);
 149
 150     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 151     rcutoff_scalar   = fr->rcoulomb;
 152     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 153     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 154
 155     /* Avoid stupid compiler warnings */
 156     jnrA = jnrB = 0;
 157     j_coord_offsetA = 0;
 158     j_coord_offsetB = 0;
 159
 160     outeriter        = 0;
 161     inneriter        = 0;
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 180
 181         fix0             = _mm_setzero_pd();
 182         fiy0             = _mm_setzero_pd();
 183         fiz0             = _mm_setzero_pd();
 184         fix1             = _mm_setzero_pd();
 185         fiy1             = _mm_setzero_pd();
 186         fiz1             = _mm_setzero_pd();
 187         fix2             = _mm_setzero_pd();
 188         fiy2             = _mm_setzero_pd();
 189         fiz2             = _mm_setzero_pd();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm_setzero_pd();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm_sub_pd(ix0,jx0);
 210             dy00             = _mm_sub_pd(iy0,jy0);
 211             dz00             = _mm_sub_pd(iz0,jz0);
 212             dx01             = _mm_sub_pd(ix0,jx1);
 213             dy01             = _mm_sub_pd(iy0,jy1);
 214             dz01             = _mm_sub_pd(iz0,jz1);
 215             dx02             = _mm_sub_pd(ix0,jx2);
 216             dy02             = _mm_sub_pd(iy0,jy2);
 217             dz02             = _mm_sub_pd(iz0,jz2);
 218             dx10             = _mm_sub_pd(ix1,jx0);
 219             dy10             = _mm_sub_pd(iy1,jy0);
 220             dz10             = _mm_sub_pd(iz1,jz0);
 221             dx11             = _mm_sub_pd(ix1,jx1);
 222             dy11             = _mm_sub_pd(iy1,jy1);
 223             dz11             = _mm_sub_pd(iz1,jz1);
 224             dx12             = _mm_sub_pd(ix1,jx2);
 225             dy12             = _mm_sub_pd(iy1,jy2);
 226             dz12             = _mm_sub_pd(iz1,jz2);
 227             dx20             = _mm_sub_pd(ix2,jx0);
 228             dy20             = _mm_sub_pd(iy2,jy0);
 229             dz20             = _mm_sub_pd(iz2,jz0);
 230             dx21             = _mm_sub_pd(ix2,jx1);
 231             dy21             = _mm_sub_pd(iy2,jy1);
 232             dz21             = _mm_sub_pd(iz2,jz1);
 233             dx22             = _mm_sub_pd(ix2,jx2);
 234             dy22             = _mm_sub_pd(iy2,jy2);
 235             dz22             = _mm_sub_pd(iz2,jz2);
 236
 237             /* Calculate squared distance and things based on it */
 238             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 239             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 240             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 241             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 242             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 243             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 244             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 245             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 246             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 247
 248             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 249             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 250             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 251             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 252             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 253             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 254             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 255             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 256             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 257
 258             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 259             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 260             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 261             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 262             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 263             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 264             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 265             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 266             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 267
 268             fjx0             = _mm_setzero_pd();
 269             fjy0             = _mm_setzero_pd();
 270             fjz0             = _mm_setzero_pd();
 271             fjx1             = _mm_setzero_pd();
 272             fjy1             = _mm_setzero_pd();
 273             fjz1             = _mm_setzero_pd();
 274             fjx2             = _mm_setzero_pd();
 275             fjy2             = _mm_setzero_pd();
 276             fjz2             = _mm_setzero_pd();
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             if (gmx_mm_any_lt(rsq00,rcutoff2))
 283             {
 284
 285             r00              = _mm_mul_pd(rsq00,rinv00);
 286
 287             /* EWALD ELECTROSTATICS */
 288
 289             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 290             ewrt             = _mm_mul_pd(r00,ewtabscale);
 291             ewitab           = _mm_cvttpd_epi32(ewrt);
 292 #ifdef __XOP__
 293             eweps            = _mm_frcz_pd(ewrt);
 294 #else
 295             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 296 #endif
 297             twoeweps         = _mm_add_pd(eweps,eweps);
 298             ewitab           = _mm_slli_epi32(ewitab,2);
 299             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 300             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 301             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 302             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 303             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 304             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 305             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 306             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 307             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 308             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 309
 310             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velec            = _mm_and_pd(velec,cutoff_mask);
 314             velecsum         = _mm_add_pd(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             fscal            = _mm_and_pd(fscal,cutoff_mask);
 319
 320             /* Update vectorial force */
 321             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 322             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 323             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 324
 325             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 326             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 327             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 328
 329             }
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             if (gmx_mm_any_lt(rsq01,rcutoff2))
 336             {
 337
 338             r01              = _mm_mul_pd(rsq01,rinv01);
 339
 340             /* EWALD ELECTROSTATICS */
 341
 342             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 343             ewrt             = _mm_mul_pd(r01,ewtabscale);
 344             ewitab           = _mm_cvttpd_epi32(ewrt);
 345 #ifdef __XOP__
 346             eweps            = _mm_frcz_pd(ewrt);
 347 #else
 348             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 349 #endif
 350             twoeweps         = _mm_add_pd(eweps,eweps);
 351             ewitab           = _mm_slli_epi32(ewitab,2);
 352             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 353             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 354             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 355             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 356             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 357             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 358             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 359             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 360             velec            = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
 361             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 362
 363             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
 364
 365             /* Update potential sum for this i atom from the interaction with this j atom. */
 366             velec            = _mm_and_pd(velec,cutoff_mask);
 367             velecsum         = _mm_add_pd(velecsum,velec);
 368
 369             fscal            = felec;
 370
 371             fscal            = _mm_and_pd(fscal,cutoff_mask);
 372
 373             /* Update vectorial force */
 374             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 375             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 376             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 377
 378             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 379             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 380             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 381
 382             }
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             if (gmx_mm_any_lt(rsq02,rcutoff2))
 389             {
 390
 391             r02              = _mm_mul_pd(rsq02,rinv02);
 392
 393             /* EWALD ELECTROSTATICS */
 394
 395             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 396             ewrt             = _mm_mul_pd(r02,ewtabscale);
 397             ewitab           = _mm_cvttpd_epi32(ewrt);
 398 #ifdef __XOP__
 399             eweps            = _mm_frcz_pd(ewrt);
 400 #else
 401             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 402 #endif
 403             twoeweps         = _mm_add_pd(eweps,eweps);
 404             ewitab           = _mm_slli_epi32(ewitab,2);
 405             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 406             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 407             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 408             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 409             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 410             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 411             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 412             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 413             velec            = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
 414             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 415
 416             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
 417
 418             /* Update potential sum for this i atom from the interaction with this j atom. */
 419             velec            = _mm_and_pd(velec,cutoff_mask);
 420             velecsum         = _mm_add_pd(velecsum,velec);
 421
 422             fscal            = felec;
 423
 424             fscal            = _mm_and_pd(fscal,cutoff_mask);
 425
 426             /* Update vectorial force */
 427             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 428             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
 429             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
 430
 431             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
 432             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
 433             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
 434
 435             }
 436
 437             /**************************
 438              * CALCULATE INTERACTIONS *
 439              **************************/
 440
 441             if (gmx_mm_any_lt(rsq10,rcutoff2))
 442             {
 443
 444             r10              = _mm_mul_pd(rsq10,rinv10);
 445
 446             /* EWALD ELECTROSTATICS */
 447
 448             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 449             ewrt             = _mm_mul_pd(r10,ewtabscale);
 450             ewitab           = _mm_cvttpd_epi32(ewrt);
 451 #ifdef __XOP__
 452             eweps            = _mm_frcz_pd(ewrt);
 453 #else
 454             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 455 #endif
 456             twoeweps         = _mm_add_pd(eweps,eweps);
 457             ewitab           = _mm_slli_epi32(ewitab,2);
 458             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 459             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 460             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 461             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 462             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 463             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 464             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 465             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 466             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 467             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 468
 469             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 470
 471             /* Update potential sum for this i atom from the interaction with this j atom. */
 472             velec            = _mm_and_pd(velec,cutoff_mask);
 473             velecsum         = _mm_add_pd(velecsum,velec);
 474
 475             fscal            = felec;
 476
 477             fscal            = _mm_and_pd(fscal,cutoff_mask);
 478
 479             /* Update vectorial force */
 480             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 481             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 482             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 483
 484             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 485             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 486             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 487
 488             }
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             if (gmx_mm_any_lt(rsq11,rcutoff2))
 495             {
 496
 497             r11              = _mm_mul_pd(rsq11,rinv11);
 498
 499             /* EWALD ELECTROSTATICS */
 500
 501             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 502             ewrt             = _mm_mul_pd(r11,ewtabscale);
 503             ewitab           = _mm_cvttpd_epi32(ewrt);
 504 #ifdef __XOP__
 505             eweps            = _mm_frcz_pd(ewrt);
 506 #else
 507             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 508 #endif
 509             twoeweps         = _mm_add_pd(eweps,eweps);
 510             ewitab           = _mm_slli_epi32(ewitab,2);
 511             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 512             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 513             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 514             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 515             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 516             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 517             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 518             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 519             velec            = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
 520             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
 521
 522             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
 523
 524             /* Update potential sum for this i atom from the interaction with this j atom. */
 525             velec            = _mm_and_pd(velec,cutoff_mask);
 526             velecsum         = _mm_add_pd(velecsum,velec);
 527
 528             fscal            = felec;
 529
 530             fscal            = _mm_and_pd(fscal,cutoff_mask);
 531
 532             /* Update vectorial force */
 533             fix1             = _mm_macc_pd(dx11,fscal,fix1);
 534             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
 535             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
 536
 537             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
 538             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
 539             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
 540
 541             }
 542
 543             /**************************
 544              * CALCULATE INTERACTIONS *
 545              **************************/
 546
 547             if (gmx_mm_any_lt(rsq12,rcutoff2))
 548             {
 549
 550             r12              = _mm_mul_pd(rsq12,rinv12);
 551
 552             /* EWALD ELECTROSTATICS */
 553
 554             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 555             ewrt             = _mm_mul_pd(r12,ewtabscale);
 556             ewitab           = _mm_cvttpd_epi32(ewrt);
 557 #ifdef __XOP__
 558             eweps            = _mm_frcz_pd(ewrt);
 559 #else
 560             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 561 #endif
 562             twoeweps         = _mm_add_pd(eweps,eweps);
 563             ewitab           = _mm_slli_epi32(ewitab,2);
 564             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 565             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 566             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 567             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 568             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 569             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 570             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 571             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 572             velec            = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
 573             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
 574
 575             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm_and_pd(velec,cutoff_mask);
 579             velecsum         = _mm_add_pd(velecsum,velec);
 580
 581             fscal            = felec;
 582
 583             fscal            = _mm_and_pd(fscal,cutoff_mask);
 584
 585             /* Update vectorial force */
 586             fix1             = _mm_macc_pd(dx12,fscal,fix1);
 587             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
 588             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
 589
 590             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
 591             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
 592             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
 593
 594             }
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             if (gmx_mm_any_lt(rsq20,rcutoff2))
 601             {
 602
 603             r20              = _mm_mul_pd(rsq20,rinv20);
 604
 605             /* EWALD ELECTROSTATICS */
 606
 607             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 608             ewrt             = _mm_mul_pd(r20,ewtabscale);
 609             ewitab           = _mm_cvttpd_epi32(ewrt);
 610 #ifdef __XOP__
 611             eweps            = _mm_frcz_pd(ewrt);
 612 #else
 613             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 614 #endif
 615             twoeweps         = _mm_add_pd(eweps,eweps);
 616             ewitab           = _mm_slli_epi32(ewitab,2);
 617             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 618             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 619             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 620             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 621             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 622             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 623             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 624             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 625             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 626             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 627
 628             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 629
 630             /* Update potential sum for this i atom from the interaction with this j atom. */
 631             velec            = _mm_and_pd(velec,cutoff_mask);
 632             velecsum         = _mm_add_pd(velecsum,velec);
 633
 634             fscal            = felec;
 635
 636             fscal            = _mm_and_pd(fscal,cutoff_mask);
 637
 638             /* Update vectorial force */
 639             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 640             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 641             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 642
 643             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 644             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 645             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 646
 647             }
 648
 649             /**************************
 650              * CALCULATE INTERACTIONS *
 651              **************************/
 652
 653             if (gmx_mm_any_lt(rsq21,rcutoff2))
 654             {
 655
 656             r21              = _mm_mul_pd(rsq21,rinv21);
 657
 658             /* EWALD ELECTROSTATICS */
 659
 660             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 661             ewrt             = _mm_mul_pd(r21,ewtabscale);
 662             ewitab           = _mm_cvttpd_epi32(ewrt);
 663 #ifdef __XOP__
 664             eweps            = _mm_frcz_pd(ewrt);
 665 #else
 666             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 667 #endif
 668             twoeweps         = _mm_add_pd(eweps,eweps);
 669             ewitab           = _mm_slli_epi32(ewitab,2);
 670             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 671             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 672             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 673             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 674             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 675             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 676             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 677             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 678             velec            = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
 679             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
 680
 681             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
 682
 683             /* Update potential sum for this i atom from the interaction with this j atom. */
 684             velec            = _mm_and_pd(velec,cutoff_mask);
 685             velecsum         = _mm_add_pd(velecsum,velec);
 686
 687             fscal            = felec;
 688
 689             fscal            = _mm_and_pd(fscal,cutoff_mask);
 690
 691             /* Update vectorial force */
 692             fix2             = _mm_macc_pd(dx21,fscal,fix2);
 693             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
 694             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
 695
 696             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
 697             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
 698             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
 699
 700             }
 701
 702             /**************************
 703              * CALCULATE INTERACTIONS *
 704              **************************/
 705
 706             if (gmx_mm_any_lt(rsq22,rcutoff2))
 707             {
 708
 709             r22              = _mm_mul_pd(rsq22,rinv22);
 710
 711             /* EWALD ELECTROSTATICS */
 712
 713             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 714             ewrt             = _mm_mul_pd(r22,ewtabscale);
 715             ewitab           = _mm_cvttpd_epi32(ewrt);
 716 #ifdef __XOP__
 717             eweps            = _mm_frcz_pd(ewrt);
 718 #else
 719             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 720 #endif
 721             twoeweps         = _mm_add_pd(eweps,eweps);
 722             ewitab           = _mm_slli_epi32(ewitab,2);
 723             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 724             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 725             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 726             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 727             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 728             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 729             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 730             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 731             velec            = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
 732             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
 733
 734             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
 735
 736             /* Update potential sum for this i atom from the interaction with this j atom. */
 737             velec            = _mm_and_pd(velec,cutoff_mask);
 738             velecsum         = _mm_add_pd(velecsum,velec);
 739
 740             fscal            = felec;
 741
 742             fscal            = _mm_and_pd(fscal,cutoff_mask);
 743
 744             /* Update vectorial force */
 745             fix2             = _mm_macc_pd(dx22,fscal,fix2);
 746             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
 747             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
 748
 749             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
 750             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
 751             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
 752
 753             }
 754
 755             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 756
 757             /* Inner loop uses 441 flops */
 758         }
 759
 760         if(jidx<j_index_end)
 761         {
 762
 763             jnrA             = jjnr[jidx];
 764             j_coord_offsetA  = DIM*jnrA;
 765
 766             /* load j atom coordinates */
 767             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 768                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 769
 770             /* Calculate displacement vector */
 771             dx00             = _mm_sub_pd(ix0,jx0);
 772             dy00             = _mm_sub_pd(iy0,jy0);
 773             dz00             = _mm_sub_pd(iz0,jz0);
 774             dx01             = _mm_sub_pd(ix0,jx1);
 775             dy01             = _mm_sub_pd(iy0,jy1);
 776             dz01             = _mm_sub_pd(iz0,jz1);
 777             dx02             = _mm_sub_pd(ix0,jx2);
 778             dy02             = _mm_sub_pd(iy0,jy2);
 779             dz02             = _mm_sub_pd(iz0,jz2);
 780             dx10             = _mm_sub_pd(ix1,jx0);
 781             dy10             = _mm_sub_pd(iy1,jy0);
 782             dz10             = _mm_sub_pd(iz1,jz0);
 783             dx11             = _mm_sub_pd(ix1,jx1);
 784             dy11             = _mm_sub_pd(iy1,jy1);
 785             dz11             = _mm_sub_pd(iz1,jz1);
 786             dx12             = _mm_sub_pd(ix1,jx2);
 787             dy12             = _mm_sub_pd(iy1,jy2);
 788             dz12             = _mm_sub_pd(iz1,jz2);
 789             dx20             = _mm_sub_pd(ix2,jx0);
 790             dy20             = _mm_sub_pd(iy2,jy0);
 791             dz20             = _mm_sub_pd(iz2,jz0);
 792             dx21             = _mm_sub_pd(ix2,jx1);
 793             dy21             = _mm_sub_pd(iy2,jy1);
 794             dz21             = _mm_sub_pd(iz2,jz1);
 795             dx22             = _mm_sub_pd(ix2,jx2);
 796             dy22             = _mm_sub_pd(iy2,jy2);
 797             dz22             = _mm_sub_pd(iz2,jz2);
 798
 799             /* Calculate squared distance and things based on it */
 800             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 801             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
 802             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
 803             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 804             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
 805             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
 806             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 807             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
 808             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
 809
 810             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 811             rinv01           = gmx_mm_invsqrt_pd(rsq01);
 812             rinv02           = gmx_mm_invsqrt_pd(rsq02);
 813             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 814             rinv11           = gmx_mm_invsqrt_pd(rsq11);
 815             rinv12           = gmx_mm_invsqrt_pd(rsq12);
 816             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 817             rinv21           = gmx_mm_invsqrt_pd(rsq21);
 818             rinv22           = gmx_mm_invsqrt_pd(rsq22);
 819
 820             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 821             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
 822             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
 823             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 824             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
 825             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
 826             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 827             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
 828             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
 829
 830             fjx0             = _mm_setzero_pd();
 831             fjy0             = _mm_setzero_pd();
 832             fjz0             = _mm_setzero_pd();
 833             fjx1             = _mm_setzero_pd();
 834             fjy1             = _mm_setzero_pd();
 835             fjz1             = _mm_setzero_pd();
 836             fjx2             = _mm_setzero_pd();
 837             fjy2             = _mm_setzero_pd();
 838             fjz2             = _mm_setzero_pd();
 839
 840             /**************************
 841              * CALCULATE INTERACTIONS *
 842              **************************/
 843
 844             if (gmx_mm_any_lt(rsq00,rcutoff2))
 845             {
 846
 847             r00              = _mm_mul_pd(rsq00,rinv00);
 848
 849             /* EWALD ELECTROSTATICS */
 850
 851             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 852             ewrt             = _mm_mul_pd(r00,ewtabscale);
 853             ewitab           = _mm_cvttpd_epi32(ewrt);
 854 #ifdef __XOP__
 855             eweps            = _mm_frcz_pd(ewrt);
 856 #else
 857             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 858 #endif
 859             twoeweps         = _mm_add_pd(eweps,eweps);
 860             ewitab           = _mm_slli_epi32(ewitab,2);
 861             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 862             ewtabD           = _mm_setzero_pd();
 863             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 864             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 865             ewtabFn          = _mm_setzero_pd();
 866             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 867             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 868             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 869             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 870             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 871
 872             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 873
 874             /* Update potential sum for this i atom from the interaction with this j atom. */
 875             velec            = _mm_and_pd(velec,cutoff_mask);
 876             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 877             velecsum         = _mm_add_pd(velecsum,velec);
 878
 879             fscal            = felec;
 880
 881             fscal            = _mm_and_pd(fscal,cutoff_mask);
 882
 883             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 884
 885             /* Update vectorial force */
 886             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 887             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 888             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 889
 890             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 891             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 892             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 893
 894             }
 895
 896             /**************************
 897              * CALCULATE INTERACTIONS *
 898              **************************/
 899
 900             if (gmx_mm_any_lt(rsq01,rcutoff2))
 901             {
 902
 903             r01              = _mm_mul_pd(rsq01,rinv01);
 904
 905             /* EWALD ELECTROSTATICS */
 906
 907             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 908             ewrt             = _mm_mul_pd(r01,ewtabscale);
 909             ewitab           = _mm_cvttpd_epi32(ewrt);
 910 #ifdef __XOP__
 911             eweps            = _mm_frcz_pd(ewrt);
 912 #else
 913             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 914 #endif
 915             twoeweps         = _mm_add_pd(eweps,eweps);
 916             ewitab           = _mm_slli_epi32(ewitab,2);
 917             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 918             ewtabD           = _mm_setzero_pd();
 919             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 920             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 921             ewtabFn          = _mm_setzero_pd();
 922             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 923             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 924             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 925             velec            = _mm_mul_pd(qq01,_mm_sub_pd(_mm_sub_pd(rinv01,sh_ewald),velec));
 926             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
 927
 928             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
 929
 930             /* Update potential sum for this i atom from the interaction with this j atom. */
 931             velec            = _mm_and_pd(velec,cutoff_mask);
 932             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 933             velecsum         = _mm_add_pd(velecsum,velec);
 934
 935             fscal            = felec;
 936
 937             fscal            = _mm_and_pd(fscal,cutoff_mask);
 938
 939             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 940
 941             /* Update vectorial force */
 942             fix0             = _mm_macc_pd(dx01,fscal,fix0);
 943             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
 944             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
 945
 946             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
 947             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
 948             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
 949
 950             }
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             if (gmx_mm_any_lt(rsq02,rcutoff2))
 957             {
 958
 959             r02              = _mm_mul_pd(rsq02,rinv02);
 960
 961             /* EWALD ELECTROSTATICS */
 962
 963             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 964             ewrt             = _mm_mul_pd(r02,ewtabscale);
 965             ewitab           = _mm_cvttpd_epi32(ewrt);
 966 #ifdef __XOP__
 967             eweps            = _mm_frcz_pd(ewrt);
 968 #else
 969             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 970 #endif
 971             twoeweps         = _mm_add_pd(eweps,eweps);
 972             ewitab           = _mm_slli_epi32(ewitab,2);
 973             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 974             ewtabD           = _mm_setzero_pd();
 975             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 976             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 977             ewtabFn          = _mm_setzero_pd();
 978             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 979             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 980             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 981             velec            = _mm_mul_pd(qq02,_mm_sub_pd(_mm_sub_pd(rinv02,sh_ewald),velec));
 982             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
 983
 984             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
 985
 986             /* Update potential sum for this i atom from the interaction with this j atom. */
 987             velec            = _mm_and_pd(velec,cutoff_mask);
 988             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 989             velecsum         = _mm_add_pd(velecsum,velec);
 990
 991             fscal            = felec;
 992
 993             fscal            = _mm_and_pd(fscal,cutoff_mask);
 994
 995             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 996
 997             /* Update vectorial force */
 998             fix0             = _mm_macc_pd(dx02,fscal,fix0);
 999             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1000             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1001
1002             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1003             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1004             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1005
1006             }
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             if (gmx_mm_any_lt(rsq10,rcutoff2))
1013             {
1014
1015             r10              = _mm_mul_pd(rsq10,rinv10);
1016
1017             /* EWALD ELECTROSTATICS */
1018
1019             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1020             ewrt             = _mm_mul_pd(r10,ewtabscale);
1021             ewitab           = _mm_cvttpd_epi32(ewrt);
1022 #ifdef __XOP__
1023             eweps            = _mm_frcz_pd(ewrt);
1024 #else
1025             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1026 #endif
1027             twoeweps         = _mm_add_pd(eweps,eweps);
1028             ewitab           = _mm_slli_epi32(ewitab,2);
1029             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1030             ewtabD           = _mm_setzero_pd();
1031             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1032             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1033             ewtabFn          = _mm_setzero_pd();
1034             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1035             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1036             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1037             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
1038             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1039
1040             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1041
1042             /* Update potential sum for this i atom from the interaction with this j atom. */
1043             velec            = _mm_and_pd(velec,cutoff_mask);
1044             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1045             velecsum         = _mm_add_pd(velecsum,velec);
1046
1047             fscal            = felec;
1048
1049             fscal            = _mm_and_pd(fscal,cutoff_mask);
1050
1051             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1052
1053             /* Update vectorial force */
1054             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1055             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1056             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1057
1058             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1059             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1060             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1061
1062             }
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             if (gmx_mm_any_lt(rsq11,rcutoff2))
1069             {
1070
1071             r11              = _mm_mul_pd(rsq11,rinv11);
1072
1073             /* EWALD ELECTROSTATICS */
1074
1075             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1076             ewrt             = _mm_mul_pd(r11,ewtabscale);
1077             ewitab           = _mm_cvttpd_epi32(ewrt);
1078 #ifdef __XOP__
1079             eweps            = _mm_frcz_pd(ewrt);
1080 #else
1081             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1082 #endif
1083             twoeweps         = _mm_add_pd(eweps,eweps);
1084             ewitab           = _mm_slli_epi32(ewitab,2);
1085             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1086             ewtabD           = _mm_setzero_pd();
1087             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1088             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1089             ewtabFn          = _mm_setzero_pd();
1090             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1091             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1092             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1093             velec            = _mm_mul_pd(qq11,_mm_sub_pd(_mm_sub_pd(rinv11,sh_ewald),velec));
1094             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1095
1096             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
1097
1098             /* Update potential sum for this i atom from the interaction with this j atom. */
1099             velec            = _mm_and_pd(velec,cutoff_mask);
1100             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1101             velecsum         = _mm_add_pd(velecsum,velec);
1102
1103             fscal            = felec;
1104
1105             fscal            = _mm_and_pd(fscal,cutoff_mask);
1106
1107             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1108
1109             /* Update vectorial force */
1110             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1111             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1112             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1113
1114             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1115             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1116             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1117
1118             }
1119
1120             /**************************
1121              * CALCULATE INTERACTIONS *
1122              **************************/
1123
1124             if (gmx_mm_any_lt(rsq12,rcutoff2))
1125             {
1126
1127             r12              = _mm_mul_pd(rsq12,rinv12);
1128
1129             /* EWALD ELECTROSTATICS */
1130
1131             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1132             ewrt             = _mm_mul_pd(r12,ewtabscale);
1133             ewitab           = _mm_cvttpd_epi32(ewrt);
1134 #ifdef __XOP__
1135             eweps            = _mm_frcz_pd(ewrt);
1136 #else
1137             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1138 #endif
1139             twoeweps         = _mm_add_pd(eweps,eweps);
1140             ewitab           = _mm_slli_epi32(ewitab,2);
1141             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1142             ewtabD           = _mm_setzero_pd();
1143             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1144             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1145             ewtabFn          = _mm_setzero_pd();
1146             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1147             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1148             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1149             velec            = _mm_mul_pd(qq12,_mm_sub_pd(_mm_sub_pd(rinv12,sh_ewald),velec));
1150             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1151
1152             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
1153
1154             /* Update potential sum for this i atom from the interaction with this j atom. */
1155             velec            = _mm_and_pd(velec,cutoff_mask);
1156             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1157             velecsum         = _mm_add_pd(velecsum,velec);
1158
1159             fscal            = felec;
1160
1161             fscal            = _mm_and_pd(fscal,cutoff_mask);
1162
1163             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1164
1165             /* Update vectorial force */
1166             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1167             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1168             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1169
1170             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1171             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1172             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1173
1174             }
1175
1176             /**************************
1177              * CALCULATE INTERACTIONS *
1178              **************************/
1179
1180             if (gmx_mm_any_lt(rsq20,rcutoff2))
1181             {
1182
1183             r20              = _mm_mul_pd(rsq20,rinv20);
1184
1185             /* EWALD ELECTROSTATICS */
1186
1187             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1188             ewrt             = _mm_mul_pd(r20,ewtabscale);
1189             ewitab           = _mm_cvttpd_epi32(ewrt);
1190 #ifdef __XOP__
1191             eweps            = _mm_frcz_pd(ewrt);
1192 #else
1193             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1194 #endif
1195             twoeweps         = _mm_add_pd(eweps,eweps);
1196             ewitab           = _mm_slli_epi32(ewitab,2);
1197             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1198             ewtabD           = _mm_setzero_pd();
1199             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1200             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1201             ewtabFn          = _mm_setzero_pd();
1202             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1203             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1204             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1205             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
1206             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1207
1208             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1209
1210             /* Update potential sum for this i atom from the interaction with this j atom. */
1211             velec            = _mm_and_pd(velec,cutoff_mask);
1212             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1213             velecsum         = _mm_add_pd(velecsum,velec);
1214
1215             fscal            = felec;
1216
1217             fscal            = _mm_and_pd(fscal,cutoff_mask);
1218
1219             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1220
1221             /* Update vectorial force */
1222             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1223             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1224             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1225
1226             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1227             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1228             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1229
1230             }
1231
1232             /**************************
1233              * CALCULATE INTERACTIONS *
1234              **************************/
1235
1236             if (gmx_mm_any_lt(rsq21,rcutoff2))
1237             {
1238
1239             r21              = _mm_mul_pd(rsq21,rinv21);
1240
1241             /* EWALD ELECTROSTATICS */
1242
1243             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1244             ewrt             = _mm_mul_pd(r21,ewtabscale);
1245             ewitab           = _mm_cvttpd_epi32(ewrt);
1246 #ifdef __XOP__
1247             eweps            = _mm_frcz_pd(ewrt);
1248 #else
1249             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1250 #endif
1251             twoeweps         = _mm_add_pd(eweps,eweps);
1252             ewitab           = _mm_slli_epi32(ewitab,2);
1253             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1254             ewtabD           = _mm_setzero_pd();
1255             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1256             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1257             ewtabFn          = _mm_setzero_pd();
1258             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1259             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1260             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1261             velec            = _mm_mul_pd(qq21,_mm_sub_pd(_mm_sub_pd(rinv21,sh_ewald),velec));
1262             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1263
1264             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
1265
1266             /* Update potential sum for this i atom from the interaction with this j atom. */
1267             velec            = _mm_and_pd(velec,cutoff_mask);
1268             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1269             velecsum         = _mm_add_pd(velecsum,velec);
1270
1271             fscal            = felec;
1272
1273             fscal            = _mm_and_pd(fscal,cutoff_mask);
1274
1275             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1276
1277             /* Update vectorial force */
1278             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1279             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1280             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1281
1282             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1283             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1284             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1285
1286             }
1287
1288             /**************************
1289              * CALCULATE INTERACTIONS *
1290              **************************/
1291
1292             if (gmx_mm_any_lt(rsq22,rcutoff2))
1293             {
1294
1295             r22              = _mm_mul_pd(rsq22,rinv22);
1296
1297             /* EWALD ELECTROSTATICS */
1298
1299             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1300             ewrt             = _mm_mul_pd(r22,ewtabscale);
1301             ewitab           = _mm_cvttpd_epi32(ewrt);
1302 #ifdef __XOP__
1303             eweps            = _mm_frcz_pd(ewrt);
1304 #else
1305             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1306 #endif
1307             twoeweps         = _mm_add_pd(eweps,eweps);
1308             ewitab           = _mm_slli_epi32(ewitab,2);
1309             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1310             ewtabD           = _mm_setzero_pd();
1311             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1312             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1313             ewtabFn          = _mm_setzero_pd();
1314             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1315             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1316             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1317             velec            = _mm_mul_pd(qq22,_mm_sub_pd(_mm_sub_pd(rinv22,sh_ewald),velec));
1318             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1319
1320             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
1321
1322             /* Update potential sum for this i atom from the interaction with this j atom. */
1323             velec            = _mm_and_pd(velec,cutoff_mask);
1324             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
1325             velecsum         = _mm_add_pd(velecsum,velec);
1326
1327             fscal            = felec;
1328
1329             fscal            = _mm_and_pd(fscal,cutoff_mask);
1330
1331             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1332
1333             /* Update vectorial force */
1334             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1335             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1336             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1337
1338             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1339             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1340             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1341
1342             }
1343
1344             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1345
1346             /* Inner loop uses 441 flops */
1347         }
1348
1349         /* End of innermost loop */
1350
1351         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1352                                               f+i_coord_offset,fshift+i_shift_offset);
1353
1354         ggid                        = gid[iidx];
1355         /* Update potential energies */
1356         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1357
1358         /* Increment number of inner iterations */
1359         inneriter                  += j_index_end - j_index_start;
1360
1361         /* Outer loop uses 19 flops */
1362     }
1363
1364     /* Increment number of outer iterations */
1365     outeriter        += nri;
1366
1367     /* Update outer/inner flops */
1368
1369     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*441);
1370 }
1371 /*
1372  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_avx_128_fma_double
1373  * Electrostatics interaction: Ewald
1374  * VdW interaction:            None
1375  * Geometry:                   Water3-Water3
1376  * Calculate force/pot:        Force
1377  */
1378 void
1379 nb_kernel_ElecEwSh_VdwNone_GeomW3W3_F_avx_128_fma_double
1380                     (t_nblist                    * gmx_restrict       nlist,
1381                      rvec                        * gmx_restrict          xx,
1382                      rvec                        * gmx_restrict          ff,
1383                      t_forcerec                  * gmx_restrict          fr,
1384                      t_mdatoms                   * gmx_restrict     mdatoms,
1385                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1386                      t_nrnb                      * gmx_restrict        nrnb)
1387 {
1388     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1389      * just 0 for non-waters.
1390      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
1391      * jnr indices corresponding to data put in the four positions in the SIMD register.
1392      */
1393     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1394     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1395     int              jnrA,jnrB;
1396     int              j_coord_offsetA,j_coord_offsetB;
1397     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1398     real             rcutoff_scalar;
1399     real             *shiftvec,*fshift,*x,*f;
1400     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1401     int              vdwioffset0;
1402     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1403     int              vdwioffset1;
1404     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1405     int              vdwioffset2;
1406     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1407     int              vdwjidx0A,vdwjidx0B;
1408     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1409     int              vdwjidx1A,vdwjidx1B;
1410     __m128d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1411     int              vdwjidx2A,vdwjidx2B;
1412     __m128d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1413     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1414     __m128d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1415     __m128d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1416     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1417     __m128d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1418     __m128d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1419     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1420     __m128d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1421     __m128d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1422     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
1423     real             *charge;
1424     __m128i          ewitab;
1425     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1426     real             *ewtab;
1427     __m128d          dummy_mask,cutoff_mask;
1428     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
1429     __m128d          one     = _mm_set1_pd(1.0);
1430     __m128d          two     = _mm_set1_pd(2.0);
1431     x                = xx[0];
1432     f                = ff[0];
1433
1434     nri              = nlist->nri;
1435     iinr             = nlist->iinr;
1436     jindex           = nlist->jindex;
1437     jjnr             = nlist->jjnr;
1438     shiftidx         = nlist->shift;
1439     gid              = nlist->gid;
1440     shiftvec         = fr->shift_vec[0];
1441     fshift           = fr->fshift[0];
1442     facel            = _mm_set1_pd(fr->epsfac);
1443     charge           = mdatoms->chargeA;
1444
1445     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
1446     ewtab            = fr->ic->tabq_coul_F;
1447     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
1448     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
1449
1450     /* Setup water-specific parameters */
1451     inr              = nlist->iinr[0];
1452     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
1453     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
1454     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
1455
1456     jq0              = _mm_set1_pd(charge[inr+0]);
1457     jq1              = _mm_set1_pd(charge[inr+1]);
1458     jq2              = _mm_set1_pd(charge[inr+2]);
1459     qq00             = _mm_mul_pd(iq0,jq0);
1460     qq01             = _mm_mul_pd(iq0,jq1);
1461     qq02             = _mm_mul_pd(iq0,jq2);
1462     qq10             = _mm_mul_pd(iq1,jq0);
1463     qq11             = _mm_mul_pd(iq1,jq1);
1464     qq12             = _mm_mul_pd(iq1,jq2);
1465     qq20             = _mm_mul_pd(iq2,jq0);
1466     qq21             = _mm_mul_pd(iq2,jq1);
1467     qq22             = _mm_mul_pd(iq2,jq2);
1468
1469     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1470     rcutoff_scalar   = fr->rcoulomb;
1471     rcutoff          = _mm_set1_pd(rcutoff_scalar);
1472     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
1473
1474     /* Avoid stupid compiler warnings */
1475     jnrA = jnrB = 0;
1476     j_coord_offsetA = 0;
1477     j_coord_offsetB = 0;
1478
1479     outeriter        = 0;
1480     inneriter        = 0;
1481
1482     /* Start outer loop over neighborlists */
1483     for(iidx=0; iidx<nri; iidx++)
1484     {
1485         /* Load shift vector for this list */
1486         i_shift_offset   = DIM*shiftidx[iidx];
1487
1488         /* Load limits for loop over neighbors */
1489         j_index_start    = jindex[iidx];
1490         j_index_end      = jindex[iidx+1];
1491
1492         /* Get outer coordinate index */
1493         inr              = iinr[iidx];
1494         i_coord_offset   = DIM*inr;
1495
1496         /* Load i particle coords and add shift vector */
1497         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1498                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1499
1500         fix0             = _mm_setzero_pd();
1501         fiy0             = _mm_setzero_pd();
1502         fiz0             = _mm_setzero_pd();
1503         fix1             = _mm_setzero_pd();
1504         fiy1             = _mm_setzero_pd();
1505         fiz1             = _mm_setzero_pd();
1506         fix2             = _mm_setzero_pd();
1507         fiy2             = _mm_setzero_pd();
1508         fiz2             = _mm_setzero_pd();
1509
1510         /* Start inner kernel loop */
1511         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1512         {
1513
1514             /* Get j neighbor index, and coordinate index */
1515             jnrA             = jjnr[jidx];
1516             jnrB             = jjnr[jidx+1];
1517             j_coord_offsetA  = DIM*jnrA;
1518             j_coord_offsetB  = DIM*jnrB;
1519
1520             /* load j atom coordinates */
1521             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1522                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1523
1524             /* Calculate displacement vector */
1525             dx00             = _mm_sub_pd(ix0,jx0);
1526             dy00             = _mm_sub_pd(iy0,jy0);
1527             dz00             = _mm_sub_pd(iz0,jz0);
1528             dx01             = _mm_sub_pd(ix0,jx1);
1529             dy01             = _mm_sub_pd(iy0,jy1);
1530             dz01             = _mm_sub_pd(iz0,jz1);
1531             dx02             = _mm_sub_pd(ix0,jx2);
1532             dy02             = _mm_sub_pd(iy0,jy2);
1533             dz02             = _mm_sub_pd(iz0,jz2);
1534             dx10             = _mm_sub_pd(ix1,jx0);
1535             dy10             = _mm_sub_pd(iy1,jy0);
1536             dz10             = _mm_sub_pd(iz1,jz0);
1537             dx11             = _mm_sub_pd(ix1,jx1);
1538             dy11             = _mm_sub_pd(iy1,jy1);
1539             dz11             = _mm_sub_pd(iz1,jz1);
1540             dx12             = _mm_sub_pd(ix1,jx2);
1541             dy12             = _mm_sub_pd(iy1,jy2);
1542             dz12             = _mm_sub_pd(iz1,jz2);
1543             dx20             = _mm_sub_pd(ix2,jx0);
1544             dy20             = _mm_sub_pd(iy2,jy0);
1545             dz20             = _mm_sub_pd(iz2,jz0);
1546             dx21             = _mm_sub_pd(ix2,jx1);
1547             dy21             = _mm_sub_pd(iy2,jy1);
1548             dz21             = _mm_sub_pd(iz2,jz1);
1549             dx22             = _mm_sub_pd(ix2,jx2);
1550             dy22             = _mm_sub_pd(iy2,jy2);
1551             dz22             = _mm_sub_pd(iz2,jz2);
1552
1553             /* Calculate squared distance and things based on it */
1554             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1555             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
1556             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
1557             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1558             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
1559             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
1560             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1561             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
1562             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
1563
1564             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1565             rinv01           = gmx_mm_invsqrt_pd(rsq01);
1566             rinv02           = gmx_mm_invsqrt_pd(rsq02);
1567             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1568             rinv11           = gmx_mm_invsqrt_pd(rsq11);
1569             rinv12           = gmx_mm_invsqrt_pd(rsq12);
1570             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1571             rinv21           = gmx_mm_invsqrt_pd(rsq21);
1572             rinv22           = gmx_mm_invsqrt_pd(rsq22);
1573
1574             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1575             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
1576             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
1577             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1578             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
1579             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
1580             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1581             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
1582             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
1583
1584             fjx0             = _mm_setzero_pd();
1585             fjy0             = _mm_setzero_pd();
1586             fjz0             = _mm_setzero_pd();
1587             fjx1             = _mm_setzero_pd();
1588             fjy1             = _mm_setzero_pd();
1589             fjz1             = _mm_setzero_pd();
1590             fjx2             = _mm_setzero_pd();
1591             fjy2             = _mm_setzero_pd();
1592             fjz2             = _mm_setzero_pd();
1593
1594             /**************************
1595              * CALCULATE INTERACTIONS *
1596              **************************/
1597
1598             if (gmx_mm_any_lt(rsq00,rcutoff2))
1599             {
1600
1601             r00              = _mm_mul_pd(rsq00,rinv00);
1602
1603             /* EWALD ELECTROSTATICS */
1604
1605             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1606             ewrt             = _mm_mul_pd(r00,ewtabscale);
1607             ewitab           = _mm_cvttpd_epi32(ewrt);
1608 #ifdef __XOP__
1609             eweps            = _mm_frcz_pd(ewrt);
1610 #else
1611             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1612 #endif
1613             twoeweps         = _mm_add_pd(eweps,eweps);
1614             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1615                                          &ewtabF,&ewtabFn);
1616             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1617             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1618
1619             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1620
1621             fscal            = felec;
1622
1623             fscal            = _mm_and_pd(fscal,cutoff_mask);
1624
1625             /* Update vectorial force */
1626             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1627             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1628             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1629
1630             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1631             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1632             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1633
1634             }
1635
1636             /**************************
1637              * CALCULATE INTERACTIONS *
1638              **************************/
1639
1640             if (gmx_mm_any_lt(rsq01,rcutoff2))
1641             {
1642
1643             r01              = _mm_mul_pd(rsq01,rinv01);
1644
1645             /* EWALD ELECTROSTATICS */
1646
1647             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1648             ewrt             = _mm_mul_pd(r01,ewtabscale);
1649             ewitab           = _mm_cvttpd_epi32(ewrt);
1650 #ifdef __XOP__
1651             eweps            = _mm_frcz_pd(ewrt);
1652 #else
1653             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1654 #endif
1655             twoeweps         = _mm_add_pd(eweps,eweps);
1656             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1657                                          &ewtabF,&ewtabFn);
1658             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1659             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
1660
1661             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
1662
1663             fscal            = felec;
1664
1665             fscal            = _mm_and_pd(fscal,cutoff_mask);
1666
1667             /* Update vectorial force */
1668             fix0             = _mm_macc_pd(dx01,fscal,fix0);
1669             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
1670             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
1671
1672             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
1673             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
1674             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
1675
1676             }
1677
1678             /**************************
1679              * CALCULATE INTERACTIONS *
1680              **************************/
1681
1682             if (gmx_mm_any_lt(rsq02,rcutoff2))
1683             {
1684
1685             r02              = _mm_mul_pd(rsq02,rinv02);
1686
1687             /* EWALD ELECTROSTATICS */
1688
1689             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1690             ewrt             = _mm_mul_pd(r02,ewtabscale);
1691             ewitab           = _mm_cvttpd_epi32(ewrt);
1692 #ifdef __XOP__
1693             eweps            = _mm_frcz_pd(ewrt);
1694 #else
1695             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1696 #endif
1697             twoeweps         = _mm_add_pd(eweps,eweps);
1698             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1699                                          &ewtabF,&ewtabFn);
1700             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1701             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
1702
1703             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
1704
1705             fscal            = felec;
1706
1707             fscal            = _mm_and_pd(fscal,cutoff_mask);
1708
1709             /* Update vectorial force */
1710             fix0             = _mm_macc_pd(dx02,fscal,fix0);
1711             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
1712             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
1713
1714             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
1715             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
1716             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
1717
1718             }
1719
1720             /**************************
1721              * CALCULATE INTERACTIONS *
1722              **************************/
1723
1724             if (gmx_mm_any_lt(rsq10,rcutoff2))
1725             {
1726
1727             r10              = _mm_mul_pd(rsq10,rinv10);
1728
1729             /* EWALD ELECTROSTATICS */
1730
1731             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1732             ewrt             = _mm_mul_pd(r10,ewtabscale);
1733             ewitab           = _mm_cvttpd_epi32(ewrt);
1734 #ifdef __XOP__
1735             eweps            = _mm_frcz_pd(ewrt);
1736 #else
1737             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1738 #endif
1739             twoeweps         = _mm_add_pd(eweps,eweps);
1740             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1741                                          &ewtabF,&ewtabFn);
1742             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1743             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1744
1745             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1746
1747             fscal            = felec;
1748
1749             fscal            = _mm_and_pd(fscal,cutoff_mask);
1750
1751             /* Update vectorial force */
1752             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1753             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1754             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1755
1756             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1757             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1758             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1759
1760             }
1761
1762             /**************************
1763              * CALCULATE INTERACTIONS *
1764              **************************/
1765
1766             if (gmx_mm_any_lt(rsq11,rcutoff2))
1767             {
1768
1769             r11              = _mm_mul_pd(rsq11,rinv11);
1770
1771             /* EWALD ELECTROSTATICS */
1772
1773             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1774             ewrt             = _mm_mul_pd(r11,ewtabscale);
1775             ewitab           = _mm_cvttpd_epi32(ewrt);
1776 #ifdef __XOP__
1777             eweps            = _mm_frcz_pd(ewrt);
1778 #else
1779             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1780 #endif
1781             twoeweps         = _mm_add_pd(eweps,eweps);
1782             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1783                                          &ewtabF,&ewtabFn);
1784             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1785             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
1786
1787             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
1788
1789             fscal            = felec;
1790
1791             fscal            = _mm_and_pd(fscal,cutoff_mask);
1792
1793             /* Update vectorial force */
1794             fix1             = _mm_macc_pd(dx11,fscal,fix1);
1795             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
1796             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
1797
1798             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
1799             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
1800             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
1801
1802             }
1803
1804             /**************************
1805              * CALCULATE INTERACTIONS *
1806              **************************/
1807
1808             if (gmx_mm_any_lt(rsq12,rcutoff2))
1809             {
1810
1811             r12              = _mm_mul_pd(rsq12,rinv12);
1812
1813             /* EWALD ELECTROSTATICS */
1814
1815             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1816             ewrt             = _mm_mul_pd(r12,ewtabscale);
1817             ewitab           = _mm_cvttpd_epi32(ewrt);
1818 #ifdef __XOP__
1819             eweps            = _mm_frcz_pd(ewrt);
1820 #else
1821             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1822 #endif
1823             twoeweps         = _mm_add_pd(eweps,eweps);
1824             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1825                                          &ewtabF,&ewtabFn);
1826             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1827             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
1828
1829             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
1830
1831             fscal            = felec;
1832
1833             fscal            = _mm_and_pd(fscal,cutoff_mask);
1834
1835             /* Update vectorial force */
1836             fix1             = _mm_macc_pd(dx12,fscal,fix1);
1837             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
1838             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
1839
1840             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
1841             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
1842             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
1843
1844             }
1845
1846             /**************************
1847              * CALCULATE INTERACTIONS *
1848              **************************/
1849
1850             if (gmx_mm_any_lt(rsq20,rcutoff2))
1851             {
1852
1853             r20              = _mm_mul_pd(rsq20,rinv20);
1854
1855             /* EWALD ELECTROSTATICS */
1856
1857             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1858             ewrt             = _mm_mul_pd(r20,ewtabscale);
1859             ewitab           = _mm_cvttpd_epi32(ewrt);
1860 #ifdef __XOP__
1861             eweps            = _mm_frcz_pd(ewrt);
1862 #else
1863             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1864 #endif
1865             twoeweps         = _mm_add_pd(eweps,eweps);
1866             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1867                                          &ewtabF,&ewtabFn);
1868             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1869             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1870
1871             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1872
1873             fscal            = felec;
1874
1875             fscal            = _mm_and_pd(fscal,cutoff_mask);
1876
1877             /* Update vectorial force */
1878             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1879             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1880             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1881
1882             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1883             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1884             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1885
1886             }
1887
1888             /**************************
1889              * CALCULATE INTERACTIONS *
1890              **************************/
1891
1892             if (gmx_mm_any_lt(rsq21,rcutoff2))
1893             {
1894
1895             r21              = _mm_mul_pd(rsq21,rinv21);
1896
1897             /* EWALD ELECTROSTATICS */
1898
1899             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1900             ewrt             = _mm_mul_pd(r21,ewtabscale);
1901             ewitab           = _mm_cvttpd_epi32(ewrt);
1902 #ifdef __XOP__
1903             eweps            = _mm_frcz_pd(ewrt);
1904 #else
1905             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1906 #endif
1907             twoeweps         = _mm_add_pd(eweps,eweps);
1908             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1909                                          &ewtabF,&ewtabFn);
1910             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1911             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
1912
1913             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
1914
1915             fscal            = felec;
1916
1917             fscal            = _mm_and_pd(fscal,cutoff_mask);
1918
1919             /* Update vectorial force */
1920             fix2             = _mm_macc_pd(dx21,fscal,fix2);
1921             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
1922             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
1923
1924             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
1925             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
1926             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
1927
1928             }
1929
1930             /**************************
1931              * CALCULATE INTERACTIONS *
1932              **************************/
1933
1934             if (gmx_mm_any_lt(rsq22,rcutoff2))
1935             {
1936
1937             r22              = _mm_mul_pd(rsq22,rinv22);
1938
1939             /* EWALD ELECTROSTATICS */
1940
1941             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1942             ewrt             = _mm_mul_pd(r22,ewtabscale);
1943             ewitab           = _mm_cvttpd_epi32(ewrt);
1944 #ifdef __XOP__
1945             eweps            = _mm_frcz_pd(ewrt);
1946 #else
1947             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1948 #endif
1949             twoeweps         = _mm_add_pd(eweps,eweps);
1950             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1951                                          &ewtabF,&ewtabFn);
1952             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1953             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
1954
1955             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
1956
1957             fscal            = felec;
1958
1959             fscal            = _mm_and_pd(fscal,cutoff_mask);
1960
1961             /* Update vectorial force */
1962             fix2             = _mm_macc_pd(dx22,fscal,fix2);
1963             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
1964             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
1965
1966             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
1967             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
1968             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
1969
1970             }
1971
1972             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1973
1974             /* Inner loop uses 378 flops */
1975         }
1976
1977         if(jidx<j_index_end)
1978         {
1979
1980             jnrA             = jjnr[jidx];
1981             j_coord_offsetA  = DIM*jnrA;
1982
1983             /* load j atom coordinates */
1984             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1985                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1986
1987             /* Calculate displacement vector */
1988             dx00             = _mm_sub_pd(ix0,jx0);
1989             dy00             = _mm_sub_pd(iy0,jy0);
1990             dz00             = _mm_sub_pd(iz0,jz0);
1991             dx01             = _mm_sub_pd(ix0,jx1);
1992             dy01             = _mm_sub_pd(iy0,jy1);
1993             dz01             = _mm_sub_pd(iz0,jz1);
1994             dx02             = _mm_sub_pd(ix0,jx2);
1995             dy02             = _mm_sub_pd(iy0,jy2);
1996             dz02             = _mm_sub_pd(iz0,jz2);
1997             dx10             = _mm_sub_pd(ix1,jx0);
1998             dy10             = _mm_sub_pd(iy1,jy0);
1999             dz10             = _mm_sub_pd(iz1,jz0);
2000             dx11             = _mm_sub_pd(ix1,jx1);
2001             dy11             = _mm_sub_pd(iy1,jy1);
2002             dz11             = _mm_sub_pd(iz1,jz1);
2003             dx12             = _mm_sub_pd(ix1,jx2);
2004             dy12             = _mm_sub_pd(iy1,jy2);
2005             dz12             = _mm_sub_pd(iz1,jz2);
2006             dx20             = _mm_sub_pd(ix2,jx0);
2007             dy20             = _mm_sub_pd(iy2,jy0);
2008             dz20             = _mm_sub_pd(iz2,jz0);
2009             dx21             = _mm_sub_pd(ix2,jx1);
2010             dy21             = _mm_sub_pd(iy2,jy1);
2011             dz21             = _mm_sub_pd(iz2,jz1);
2012             dx22             = _mm_sub_pd(ix2,jx2);
2013             dy22             = _mm_sub_pd(iy2,jy2);
2014             dz22             = _mm_sub_pd(iz2,jz2);
2015
2016             /* Calculate squared distance and things based on it */
2017             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
2018             rsq01            = gmx_mm_calc_rsq_pd(dx01,dy01,dz01);
2019             rsq02            = gmx_mm_calc_rsq_pd(dx02,dy02,dz02);
2020             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
2021             rsq11            = gmx_mm_calc_rsq_pd(dx11,dy11,dz11);
2022             rsq12            = gmx_mm_calc_rsq_pd(dx12,dy12,dz12);
2023             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
2024             rsq21            = gmx_mm_calc_rsq_pd(dx21,dy21,dz21);
2025             rsq22            = gmx_mm_calc_rsq_pd(dx22,dy22,dz22);
2026
2027             rinv00           = gmx_mm_invsqrt_pd(rsq00);
2028             rinv01           = gmx_mm_invsqrt_pd(rsq01);
2029             rinv02           = gmx_mm_invsqrt_pd(rsq02);
2030             rinv10           = gmx_mm_invsqrt_pd(rsq10);
2031             rinv11           = gmx_mm_invsqrt_pd(rsq11);
2032             rinv12           = gmx_mm_invsqrt_pd(rsq12);
2033             rinv20           = gmx_mm_invsqrt_pd(rsq20);
2034             rinv21           = gmx_mm_invsqrt_pd(rsq21);
2035             rinv22           = gmx_mm_invsqrt_pd(rsq22);
2036
2037             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
2038             rinvsq01         = _mm_mul_pd(rinv01,rinv01);
2039             rinvsq02         = _mm_mul_pd(rinv02,rinv02);
2040             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
2041             rinvsq11         = _mm_mul_pd(rinv11,rinv11);
2042             rinvsq12         = _mm_mul_pd(rinv12,rinv12);
2043             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
2044             rinvsq21         = _mm_mul_pd(rinv21,rinv21);
2045             rinvsq22         = _mm_mul_pd(rinv22,rinv22);
2046
2047             fjx0             = _mm_setzero_pd();
2048             fjy0             = _mm_setzero_pd();
2049             fjz0             = _mm_setzero_pd();
2050             fjx1             = _mm_setzero_pd();
2051             fjy1             = _mm_setzero_pd();
2052             fjz1             = _mm_setzero_pd();
2053             fjx2             = _mm_setzero_pd();
2054             fjy2             = _mm_setzero_pd();
2055             fjz2             = _mm_setzero_pd();
2056
2057             /**************************
2058              * CALCULATE INTERACTIONS *
2059              **************************/
2060
2061             if (gmx_mm_any_lt(rsq00,rcutoff2))
2062             {
2063
2064             r00              = _mm_mul_pd(rsq00,rinv00);
2065
2066             /* EWALD ELECTROSTATICS */
2067
2068             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2069             ewrt             = _mm_mul_pd(r00,ewtabscale);
2070             ewitab           = _mm_cvttpd_epi32(ewrt);
2071 #ifdef __XOP__
2072             eweps            = _mm_frcz_pd(ewrt);
2073 #else
2074             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2075 #endif
2076             twoeweps         = _mm_add_pd(eweps,eweps);
2077             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2078             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2079             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
2080
2081             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
2082
2083             fscal            = felec;
2084
2085             fscal            = _mm_and_pd(fscal,cutoff_mask);
2086
2087             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2088
2089             /* Update vectorial force */
2090             fix0             = _mm_macc_pd(dx00,fscal,fix0);
2091             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
2092             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
2093
2094             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
2095             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
2096             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
2097
2098             }
2099
2100             /**************************
2101              * CALCULATE INTERACTIONS *
2102              **************************/
2103
2104             if (gmx_mm_any_lt(rsq01,rcutoff2))
2105             {
2106
2107             r01              = _mm_mul_pd(rsq01,rinv01);
2108
2109             /* EWALD ELECTROSTATICS */
2110
2111             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2112             ewrt             = _mm_mul_pd(r01,ewtabscale);
2113             ewitab           = _mm_cvttpd_epi32(ewrt);
2114 #ifdef __XOP__
2115             eweps            = _mm_frcz_pd(ewrt);
2116 #else
2117             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2118 #endif
2119             twoeweps         = _mm_add_pd(eweps,eweps);
2120             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2121             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2122             felec            = _mm_mul_pd(_mm_mul_pd(qq01,rinv01),_mm_sub_pd(rinvsq01,felec));
2123
2124             cutoff_mask      = _mm_cmplt_pd(rsq01,rcutoff2);
2125
2126             fscal            = felec;
2127
2128             fscal            = _mm_and_pd(fscal,cutoff_mask);
2129
2130             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2131
2132             /* Update vectorial force */
2133             fix0             = _mm_macc_pd(dx01,fscal,fix0);
2134             fiy0             = _mm_macc_pd(dy01,fscal,fiy0);
2135             fiz0             = _mm_macc_pd(dz01,fscal,fiz0);
2136
2137             fjx1             = _mm_macc_pd(dx01,fscal,fjx1);
2138             fjy1             = _mm_macc_pd(dy01,fscal,fjy1);
2139             fjz1             = _mm_macc_pd(dz01,fscal,fjz1);
2140
2141             }
2142
2143             /**************************
2144              * CALCULATE INTERACTIONS *
2145              **************************/
2146
2147             if (gmx_mm_any_lt(rsq02,rcutoff2))
2148             {
2149
2150             r02              = _mm_mul_pd(rsq02,rinv02);
2151
2152             /* EWALD ELECTROSTATICS */
2153
2154             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2155             ewrt             = _mm_mul_pd(r02,ewtabscale);
2156             ewitab           = _mm_cvttpd_epi32(ewrt);
2157 #ifdef __XOP__
2158             eweps            = _mm_frcz_pd(ewrt);
2159 #else
2160             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2161 #endif
2162             twoeweps         = _mm_add_pd(eweps,eweps);
2163             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2164             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2165             felec            = _mm_mul_pd(_mm_mul_pd(qq02,rinv02),_mm_sub_pd(rinvsq02,felec));
2166
2167             cutoff_mask      = _mm_cmplt_pd(rsq02,rcutoff2);
2168
2169             fscal            = felec;
2170
2171             fscal            = _mm_and_pd(fscal,cutoff_mask);
2172
2173             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2174
2175             /* Update vectorial force */
2176             fix0             = _mm_macc_pd(dx02,fscal,fix0);
2177             fiy0             = _mm_macc_pd(dy02,fscal,fiy0);
2178             fiz0             = _mm_macc_pd(dz02,fscal,fiz0);
2179
2180             fjx2             = _mm_macc_pd(dx02,fscal,fjx2);
2181             fjy2             = _mm_macc_pd(dy02,fscal,fjy2);
2182             fjz2             = _mm_macc_pd(dz02,fscal,fjz2);
2183
2184             }
2185
2186             /**************************
2187              * CALCULATE INTERACTIONS *
2188              **************************/
2189
2190             if (gmx_mm_any_lt(rsq10,rcutoff2))
2191             {
2192
2193             r10              = _mm_mul_pd(rsq10,rinv10);
2194
2195             /* EWALD ELECTROSTATICS */
2196
2197             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2198             ewrt             = _mm_mul_pd(r10,ewtabscale);
2199             ewitab           = _mm_cvttpd_epi32(ewrt);
2200 #ifdef __XOP__
2201             eweps            = _mm_frcz_pd(ewrt);
2202 #else
2203             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2204 #endif
2205             twoeweps         = _mm_add_pd(eweps,eweps);
2206             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2207             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2208             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
2209
2210             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
2211
2212             fscal            = felec;
2213
2214             fscal            = _mm_and_pd(fscal,cutoff_mask);
2215
2216             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2217
2218             /* Update vectorial force */
2219             fix1             = _mm_macc_pd(dx10,fscal,fix1);
2220             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
2221             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
2222
2223             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
2224             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
2225             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
2226
2227             }
2228
2229             /**************************
2230              * CALCULATE INTERACTIONS *
2231              **************************/
2232
2233             if (gmx_mm_any_lt(rsq11,rcutoff2))
2234             {
2235
2236             r11              = _mm_mul_pd(rsq11,rinv11);
2237
2238             /* EWALD ELECTROSTATICS */
2239
2240             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2241             ewrt             = _mm_mul_pd(r11,ewtabscale);
2242             ewitab           = _mm_cvttpd_epi32(ewrt);
2243 #ifdef __XOP__
2244             eweps            = _mm_frcz_pd(ewrt);
2245 #else
2246             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2247 #endif
2248             twoeweps         = _mm_add_pd(eweps,eweps);
2249             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2250             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2251             felec            = _mm_mul_pd(_mm_mul_pd(qq11,rinv11),_mm_sub_pd(rinvsq11,felec));
2252
2253             cutoff_mask      = _mm_cmplt_pd(rsq11,rcutoff2);
2254
2255             fscal            = felec;
2256
2257             fscal            = _mm_and_pd(fscal,cutoff_mask);
2258
2259             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2260
2261             /* Update vectorial force */
2262             fix1             = _mm_macc_pd(dx11,fscal,fix1);
2263             fiy1             = _mm_macc_pd(dy11,fscal,fiy1);
2264             fiz1             = _mm_macc_pd(dz11,fscal,fiz1);
2265
2266             fjx1             = _mm_macc_pd(dx11,fscal,fjx1);
2267             fjy1             = _mm_macc_pd(dy11,fscal,fjy1);
2268             fjz1             = _mm_macc_pd(dz11,fscal,fjz1);
2269
2270             }
2271
2272             /**************************
2273              * CALCULATE INTERACTIONS *
2274              **************************/
2275
2276             if (gmx_mm_any_lt(rsq12,rcutoff2))
2277             {
2278
2279             r12              = _mm_mul_pd(rsq12,rinv12);
2280
2281             /* EWALD ELECTROSTATICS */
2282
2283             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2284             ewrt             = _mm_mul_pd(r12,ewtabscale);
2285             ewitab           = _mm_cvttpd_epi32(ewrt);
2286 #ifdef __XOP__
2287             eweps            = _mm_frcz_pd(ewrt);
2288 #else
2289             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2290 #endif
2291             twoeweps         = _mm_add_pd(eweps,eweps);
2292             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2293             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2294             felec            = _mm_mul_pd(_mm_mul_pd(qq12,rinv12),_mm_sub_pd(rinvsq12,felec));
2295
2296             cutoff_mask      = _mm_cmplt_pd(rsq12,rcutoff2);
2297
2298             fscal            = felec;
2299
2300             fscal            = _mm_and_pd(fscal,cutoff_mask);
2301
2302             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2303
2304             /* Update vectorial force */
2305             fix1             = _mm_macc_pd(dx12,fscal,fix1);
2306             fiy1             = _mm_macc_pd(dy12,fscal,fiy1);
2307             fiz1             = _mm_macc_pd(dz12,fscal,fiz1);
2308
2309             fjx2             = _mm_macc_pd(dx12,fscal,fjx2);
2310             fjy2             = _mm_macc_pd(dy12,fscal,fjy2);
2311             fjz2             = _mm_macc_pd(dz12,fscal,fjz2);
2312
2313             }
2314
2315             /**************************
2316              * CALCULATE INTERACTIONS *
2317              **************************/
2318
2319             if (gmx_mm_any_lt(rsq20,rcutoff2))
2320             {
2321
2322             r20              = _mm_mul_pd(rsq20,rinv20);
2323
2324             /* EWALD ELECTROSTATICS */
2325
2326             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2327             ewrt             = _mm_mul_pd(r20,ewtabscale);
2328             ewitab           = _mm_cvttpd_epi32(ewrt);
2329 #ifdef __XOP__
2330             eweps            = _mm_frcz_pd(ewrt);
2331 #else
2332             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2333 #endif
2334             twoeweps         = _mm_add_pd(eweps,eweps);
2335             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2336             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2337             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
2338
2339             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
2340
2341             fscal            = felec;
2342
2343             fscal            = _mm_and_pd(fscal,cutoff_mask);
2344
2345             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2346
2347             /* Update vectorial force */
2348             fix2             = _mm_macc_pd(dx20,fscal,fix2);
2349             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
2350             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
2351
2352             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
2353             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
2354             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
2355
2356             }
2357
2358             /**************************
2359              * CALCULATE INTERACTIONS *
2360              **************************/
2361
2362             if (gmx_mm_any_lt(rsq21,rcutoff2))
2363             {
2364
2365             r21              = _mm_mul_pd(rsq21,rinv21);
2366
2367             /* EWALD ELECTROSTATICS */
2368
2369             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2370             ewrt             = _mm_mul_pd(r21,ewtabscale);
2371             ewitab           = _mm_cvttpd_epi32(ewrt);
2372 #ifdef __XOP__
2373             eweps            = _mm_frcz_pd(ewrt);
2374 #else
2375             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2376 #endif
2377             twoeweps         = _mm_add_pd(eweps,eweps);
2378             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2379             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2380             felec            = _mm_mul_pd(_mm_mul_pd(qq21,rinv21),_mm_sub_pd(rinvsq21,felec));
2381
2382             cutoff_mask      = _mm_cmplt_pd(rsq21,rcutoff2);
2383
2384             fscal            = felec;
2385
2386             fscal            = _mm_and_pd(fscal,cutoff_mask);
2387
2388             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2389
2390             /* Update vectorial force */
2391             fix2             = _mm_macc_pd(dx21,fscal,fix2);
2392             fiy2             = _mm_macc_pd(dy21,fscal,fiy2);
2393             fiz2             = _mm_macc_pd(dz21,fscal,fiz2);
2394
2395             fjx1             = _mm_macc_pd(dx21,fscal,fjx1);
2396             fjy1             = _mm_macc_pd(dy21,fscal,fjy1);
2397             fjz1             = _mm_macc_pd(dz21,fscal,fjz1);
2398
2399             }
2400
2401             /**************************
2402              * CALCULATE INTERACTIONS *
2403              **************************/
2404
2405             if (gmx_mm_any_lt(rsq22,rcutoff2))
2406             {
2407
2408             r22              = _mm_mul_pd(rsq22,rinv22);
2409
2410             /* EWALD ELECTROSTATICS */
2411
2412             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2413             ewrt             = _mm_mul_pd(r22,ewtabscale);
2414             ewitab           = _mm_cvttpd_epi32(ewrt);
2415 #ifdef __XOP__
2416             eweps            = _mm_frcz_pd(ewrt);
2417 #else
2418             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
2419 #endif
2420             twoeweps         = _mm_add_pd(eweps,eweps);
2421             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
2422             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
2423             felec            = _mm_mul_pd(_mm_mul_pd(qq22,rinv22),_mm_sub_pd(rinvsq22,felec));
2424
2425             cutoff_mask      = _mm_cmplt_pd(rsq22,rcutoff2);
2426
2427             fscal            = felec;
2428
2429             fscal            = _mm_and_pd(fscal,cutoff_mask);
2430
2431             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
2432
2433             /* Update vectorial force */
2434             fix2             = _mm_macc_pd(dx22,fscal,fix2);
2435             fiy2             = _mm_macc_pd(dy22,fscal,fiy2);
2436             fiz2             = _mm_macc_pd(dz22,fscal,fiz2);
2437
2438             fjx2             = _mm_macc_pd(dx22,fscal,fjx2);
2439             fjy2             = _mm_macc_pd(dy22,fscal,fjy2);
2440             fjz2             = _mm_macc_pd(dz22,fscal,fjz2);
2441
2442             }
2443
2444             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2445
2446             /* Inner loop uses 378 flops */
2447         }
2448
2449         /* End of innermost loop */
2450
2451         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2452                                               f+i_coord_offset,fshift+i_shift_offset);
2453
2454         /* Increment number of inner iterations */
2455         inneriter                  += j_index_end - j_index_start;
2456
2457         /* Outer loop uses 18 flops */
2458     }
2459
2460     /* Increment number of outer iterations */
2461     outeriter        += nri;
2462
2463     /* Update outer/inner flops */
2464
2465     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*378);
2466 }