src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJ_GeomW4P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomW4P1_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              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128i          ewitab;
 105     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     real             *ewtab;
 107     __m128d          dummy_mask,cutoff_mask;
 108     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 109     __m128d          one     = _mm_set1_pd(1.0);
 110     __m128d          two     = _mm_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_pd(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 136     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 137     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 164                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 165
 166         fix0             = _mm_setzero_pd();
 167         fiy0             = _mm_setzero_pd();
 168         fiz0             = _mm_setzero_pd();
 169         fix1             = _mm_setzero_pd();
 170         fiy1             = _mm_setzero_pd();
 171         fiz1             = _mm_setzero_pd();
 172         fix2             = _mm_setzero_pd();
 173         fiy2             = _mm_setzero_pd();
 174         fiz2             = _mm_setzero_pd();
 175         fix3             = _mm_setzero_pd();
 176         fiy3             = _mm_setzero_pd();
 177         fiz3             = _mm_setzero_pd();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_pd();
 181         vvdwsum          = _mm_setzero_pd();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             j_coord_offsetA  = DIM*jnrA;
 191             j_coord_offsetB  = DIM*jnrB;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_pd(ix0,jx0);
 199             dy00             = _mm_sub_pd(iy0,jy0);
 200             dz00             = _mm_sub_pd(iz0,jz0);
 201             dx10             = _mm_sub_pd(ix1,jx0);
 202             dy10             = _mm_sub_pd(iy1,jy0);
 203             dz10             = _mm_sub_pd(iz1,jz0);
 204             dx20             = _mm_sub_pd(ix2,jx0);
 205             dy20             = _mm_sub_pd(iy2,jy0);
 206             dz20             = _mm_sub_pd(iz2,jz0);
 207             dx30             = _mm_sub_pd(ix3,jx0);
 208             dy30             = _mm_sub_pd(iy3,jy0);
 209             dz30             = _mm_sub_pd(iz3,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 213             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 215             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 216
 217             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 218             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 219             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 220
 221             rinvsq00         = gmx_mm_inv_pd(rsq00);
 222             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 223             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 224             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230
 231             fjx0             = _mm_setzero_pd();
 232             fjy0             = _mm_setzero_pd();
 233             fjz0             = _mm_setzero_pd();
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 241                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 242
 243             /* LENNARD-JONES DISPERSION/REPULSION */
 244
 245             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 246             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 247             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 248             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 249             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 250
 251             /* Update potential sum for this i atom from the interaction with this j atom. */
 252             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 253
 254             fscal            = fvdw;
 255
 256             /* Update vectorial force */
 257             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 258             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 259             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 260
 261             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 262             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 263             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             r10              = _mm_mul_pd(rsq10,rinv10);
 270
 271             /* Compute parameters for interactions between i and j atoms */
 272             qq10             = _mm_mul_pd(iq1,jq0);
 273
 274             /* EWALD ELECTROSTATICS */
 275
 276             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 277             ewrt             = _mm_mul_pd(r10,ewtabscale);
 278             ewitab           = _mm_cvttpd_epi32(ewrt);
 279 #ifdef __XOP__
 280             eweps            = _mm_frcz_pd(ewrt);
 281 #else
 282             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 283 #endif
 284             twoeweps         = _mm_add_pd(eweps,eweps);
 285             ewitab           = _mm_slli_epi32(ewitab,2);
 286             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 287             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 288             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 289             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 290             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 291             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 292             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 293             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 294             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 295             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm_add_pd(velecsum,velec);
 299
 300             fscal            = felec;
 301
 302             /* Update vectorial force */
 303             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 304             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 305             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 306
 307             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 308             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 309             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             r20              = _mm_mul_pd(rsq20,rinv20);
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq20             = _mm_mul_pd(iq2,jq0);
 319
 320             /* EWALD ELECTROSTATICS */
 321
 322             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 323             ewrt             = _mm_mul_pd(r20,ewtabscale);
 324             ewitab           = _mm_cvttpd_epi32(ewrt);
 325 #ifdef __XOP__
 326             eweps            = _mm_frcz_pd(ewrt);
 327 #else
 328             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 329 #endif
 330             twoeweps         = _mm_add_pd(eweps,eweps);
 331             ewitab           = _mm_slli_epi32(ewitab,2);
 332             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 333             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 334             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 335             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 336             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 337             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 338             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 339             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 340             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 341             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm_add_pd(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             /* Update vectorial force */
 349             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 350             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 351             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 352
 353             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 354             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 355             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             r30              = _mm_mul_pd(rsq30,rinv30);
 362
 363             /* Compute parameters for interactions between i and j atoms */
 364             qq30             = _mm_mul_pd(iq3,jq0);
 365
 366             /* EWALD ELECTROSTATICS */
 367
 368             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 369             ewrt             = _mm_mul_pd(r30,ewtabscale);
 370             ewitab           = _mm_cvttpd_epi32(ewrt);
 371 #ifdef __XOP__
 372             eweps            = _mm_frcz_pd(ewrt);
 373 #else
 374             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 375 #endif
 376             twoeweps         = _mm_add_pd(eweps,eweps);
 377             ewitab           = _mm_slli_epi32(ewitab,2);
 378             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 379             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 380             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 381             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 382             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 383             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 384             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 385             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 386             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 387             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velecsum         = _mm_add_pd(velecsum,velec);
 391
 392             fscal            = felec;
 393
 394             /* Update vectorial force */
 395             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 396             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 397             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 398
 399             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 400             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 401             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 402
 403             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 404
 405             /* Inner loop uses 170 flops */
 406         }
 407
 408         if(jidx<j_index_end)
 409         {
 410
 411             jnrA             = jjnr[jidx];
 412             j_coord_offsetA  = DIM*jnrA;
 413
 414             /* load j atom coordinates */
 415             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 416                                               &jx0,&jy0,&jz0);
 417
 418             /* Calculate displacement vector */
 419             dx00             = _mm_sub_pd(ix0,jx0);
 420             dy00             = _mm_sub_pd(iy0,jy0);
 421             dz00             = _mm_sub_pd(iz0,jz0);
 422             dx10             = _mm_sub_pd(ix1,jx0);
 423             dy10             = _mm_sub_pd(iy1,jy0);
 424             dz10             = _mm_sub_pd(iz1,jz0);
 425             dx20             = _mm_sub_pd(ix2,jx0);
 426             dy20             = _mm_sub_pd(iy2,jy0);
 427             dz20             = _mm_sub_pd(iz2,jz0);
 428             dx30             = _mm_sub_pd(ix3,jx0);
 429             dy30             = _mm_sub_pd(iy3,jy0);
 430             dz30             = _mm_sub_pd(iz3,jz0);
 431
 432             /* Calculate squared distance and things based on it */
 433             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 434             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 435             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 436             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 437
 438             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 439             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 440             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 441
 442             rinvsq00         = gmx_mm_inv_pd(rsq00);
 443             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 444             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 445             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 446
 447             /* Load parameters for j particles */
 448             jq0              = _mm_load_sd(charge+jnrA+0);
 449             vdwjidx0A        = 2*vdwtype[jnrA+0];
 450
 451             fjx0             = _mm_setzero_pd();
 452             fjy0             = _mm_setzero_pd();
 453             fjz0             = _mm_setzero_pd();
 454
 455             /**************************
 456              * CALCULATE INTERACTIONS *
 457              **************************/
 458
 459             /* Compute parameters for interactions between i and j atoms */
 460             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 461
 462             /* LENNARD-JONES DISPERSION/REPULSION */
 463
 464             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 465             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 466             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 467             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 468             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 469
 470             /* Update potential sum for this i atom from the interaction with this j atom. */
 471             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 472             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 473
 474             fscal            = fvdw;
 475
 476             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 477
 478             /* Update vectorial force */
 479             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 480             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 481             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 482
 483             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 484             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 485             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             r10              = _mm_mul_pd(rsq10,rinv10);
 492
 493             /* Compute parameters for interactions between i and j atoms */
 494             qq10             = _mm_mul_pd(iq1,jq0);
 495
 496             /* EWALD ELECTROSTATICS */
 497
 498             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 499             ewrt             = _mm_mul_pd(r10,ewtabscale);
 500             ewitab           = _mm_cvttpd_epi32(ewrt);
 501 #ifdef __XOP__
 502             eweps            = _mm_frcz_pd(ewrt);
 503 #else
 504             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 505 #endif
 506             twoeweps         = _mm_add_pd(eweps,eweps);
 507             ewitab           = _mm_slli_epi32(ewitab,2);
 508             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 509             ewtabD           = _mm_setzero_pd();
 510             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 511             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 512             ewtabFn          = _mm_setzero_pd();
 513             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 514             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 515             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 516             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 517             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 518
 519             /* Update potential sum for this i atom from the interaction with this j atom. */
 520             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 521             velecsum         = _mm_add_pd(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 526
 527             /* Update vectorial force */
 528             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 529             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 530             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 531
 532             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 533             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 534             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 535
 536             /**************************
 537              * CALCULATE INTERACTIONS *
 538              **************************/
 539
 540             r20              = _mm_mul_pd(rsq20,rinv20);
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             qq20             = _mm_mul_pd(iq2,jq0);
 544
 545             /* EWALD ELECTROSTATICS */
 546
 547             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 548             ewrt             = _mm_mul_pd(r20,ewtabscale);
 549             ewitab           = _mm_cvttpd_epi32(ewrt);
 550 #ifdef __XOP__
 551             eweps            = _mm_frcz_pd(ewrt);
 552 #else
 553             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 554 #endif
 555             twoeweps         = _mm_add_pd(eweps,eweps);
 556             ewitab           = _mm_slli_epi32(ewitab,2);
 557             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 558             ewtabD           = _mm_setzero_pd();
 559             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 560             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 561             ewtabFn          = _mm_setzero_pd();
 562             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 563             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 564             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 565             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 566             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 570             velecsum         = _mm_add_pd(velecsum,velec);
 571
 572             fscal            = felec;
 573
 574             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 575
 576             /* Update vectorial force */
 577             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 578             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 579             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 580
 581             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 582             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 583             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 584
 585             /**************************
 586              * CALCULATE INTERACTIONS *
 587              **************************/
 588
 589             r30              = _mm_mul_pd(rsq30,rinv30);
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq30             = _mm_mul_pd(iq3,jq0);
 593
 594             /* EWALD ELECTROSTATICS */
 595
 596             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 597             ewrt             = _mm_mul_pd(r30,ewtabscale);
 598             ewitab           = _mm_cvttpd_epi32(ewrt);
 599 #ifdef __XOP__
 600             eweps            = _mm_frcz_pd(ewrt);
 601 #else
 602             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 603 #endif
 604             twoeweps         = _mm_add_pd(eweps,eweps);
 605             ewitab           = _mm_slli_epi32(ewitab,2);
 606             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 607             ewtabD           = _mm_setzero_pd();
 608             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 609             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 610             ewtabFn          = _mm_setzero_pd();
 611             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 612             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 613             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 614             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 615             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 616
 617             /* Update potential sum for this i atom from the interaction with this j atom. */
 618             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 619             velecsum         = _mm_add_pd(velecsum,velec);
 620
 621             fscal            = felec;
 622
 623             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 624
 625             /* Update vectorial force */
 626             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 627             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 628             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 629
 630             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 631             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 632             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 633
 634             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 635
 636             /* Inner loop uses 170 flops */
 637         }
 638
 639         /* End of innermost loop */
 640
 641         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 642                                               f+i_coord_offset,fshift+i_shift_offset);
 643
 644         ggid                        = gid[iidx];
 645         /* Update potential energies */
 646         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 647         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 648
 649         /* Increment number of inner iterations */
 650         inneriter                  += j_index_end - j_index_start;
 651
 652         /* Outer loop uses 26 flops */
 653     }
 654
 655     /* Increment number of outer iterations */
 656     outeriter        += nri;
 657
 658     /* Update outer/inner flops */
 659
 660     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*170);
 661 }
 662 /*
 663  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_128_fma_double
 664  * Electrostatics interaction: Ewald
 665  * VdW interaction:            LennardJones
 666  * Geometry:                   Water4-Particle
 667  * Calculate force/pot:        Force
 668  */
 669 void
 670 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_128_fma_double
 671                     (t_nblist                    * gmx_restrict       nlist,
 672                      rvec                        * gmx_restrict          xx,
 673                      rvec                        * gmx_restrict          ff,
 674                      t_forcerec                  * gmx_restrict          fr,
 675                      t_mdatoms                   * gmx_restrict     mdatoms,
 676                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 677                      t_nrnb                      * gmx_restrict        nrnb)
 678 {
 679     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 680      * just 0 for non-waters.
 681      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 682      * jnr indices corresponding to data put in the four positions in the SIMD register.
 683      */
 684     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 685     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 686     int              jnrA,jnrB;
 687     int              j_coord_offsetA,j_coord_offsetB;
 688     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 689     real             rcutoff_scalar;
 690     real             *shiftvec,*fshift,*x,*f;
 691     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 692     int              vdwioffset0;
 693     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 694     int              vdwioffset1;
 695     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 696     int              vdwioffset2;
 697     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 698     int              vdwioffset3;
 699     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 700     int              vdwjidx0A,vdwjidx0B;
 701     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 702     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 703     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 704     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 705     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 706     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 707     real             *charge;
 708     int              nvdwtype;
 709     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 710     int              *vdwtype;
 711     real             *vdwparam;
 712     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 713     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 714     __m128i          ewitab;
 715     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 716     real             *ewtab;
 717     __m128d          dummy_mask,cutoff_mask;
 718     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 719     __m128d          one     = _mm_set1_pd(1.0);
 720     __m128d          two     = _mm_set1_pd(2.0);
 721     x                = xx[0];
 722     f                = ff[0];
 723
 724     nri              = nlist->nri;
 725     iinr             = nlist->iinr;
 726     jindex           = nlist->jindex;
 727     jjnr             = nlist->jjnr;
 728     shiftidx         = nlist->shift;
 729     gid              = nlist->gid;
 730     shiftvec         = fr->shift_vec[0];
 731     fshift           = fr->fshift[0];
 732     facel            = _mm_set1_pd(fr->epsfac);
 733     charge           = mdatoms->chargeA;
 734     nvdwtype         = fr->ntype;
 735     vdwparam         = fr->nbfp;
 736     vdwtype          = mdatoms->typeA;
 737
 738     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 739     ewtab            = fr->ic->tabq_coul_F;
 740     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 741     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 742
 743     /* Setup water-specific parameters */
 744     inr              = nlist->iinr[0];
 745     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 746     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 747     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 748     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 749
 750     /* Avoid stupid compiler warnings */
 751     jnrA = jnrB = 0;
 752     j_coord_offsetA = 0;
 753     j_coord_offsetB = 0;
 754
 755     outeriter        = 0;
 756     inneriter        = 0;
 757
 758     /* Start outer loop over neighborlists */
 759     for(iidx=0; iidx<nri; iidx++)
 760     {
 761         /* Load shift vector for this list */
 762         i_shift_offset   = DIM*shiftidx[iidx];
 763
 764         /* Load limits for loop over neighbors */
 765         j_index_start    = jindex[iidx];
 766         j_index_end      = jindex[iidx+1];
 767
 768         /* Get outer coordinate index */
 769         inr              = iinr[iidx];
 770         i_coord_offset   = DIM*inr;
 771
 772         /* Load i particle coords and add shift vector */
 773         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 774                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 775
 776         fix0             = _mm_setzero_pd();
 777         fiy0             = _mm_setzero_pd();
 778         fiz0             = _mm_setzero_pd();
 779         fix1             = _mm_setzero_pd();
 780         fiy1             = _mm_setzero_pd();
 781         fiz1             = _mm_setzero_pd();
 782         fix2             = _mm_setzero_pd();
 783         fiy2             = _mm_setzero_pd();
 784         fiz2             = _mm_setzero_pd();
 785         fix3             = _mm_setzero_pd();
 786         fiy3             = _mm_setzero_pd();
 787         fiz3             = _mm_setzero_pd();
 788
 789         /* Start inner kernel loop */
 790         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 791         {
 792
 793             /* Get j neighbor index, and coordinate index */
 794             jnrA             = jjnr[jidx];
 795             jnrB             = jjnr[jidx+1];
 796             j_coord_offsetA  = DIM*jnrA;
 797             j_coord_offsetB  = DIM*jnrB;
 798
 799             /* load j atom coordinates */
 800             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 801                                               &jx0,&jy0,&jz0);
 802
 803             /* Calculate displacement vector */
 804             dx00             = _mm_sub_pd(ix0,jx0);
 805             dy00             = _mm_sub_pd(iy0,jy0);
 806             dz00             = _mm_sub_pd(iz0,jz0);
 807             dx10             = _mm_sub_pd(ix1,jx0);
 808             dy10             = _mm_sub_pd(iy1,jy0);
 809             dz10             = _mm_sub_pd(iz1,jz0);
 810             dx20             = _mm_sub_pd(ix2,jx0);
 811             dy20             = _mm_sub_pd(iy2,jy0);
 812             dz20             = _mm_sub_pd(iz2,jz0);
 813             dx30             = _mm_sub_pd(ix3,jx0);
 814             dy30             = _mm_sub_pd(iy3,jy0);
 815             dz30             = _mm_sub_pd(iz3,jz0);
 816
 817             /* Calculate squared distance and things based on it */
 818             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 819             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 820             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 821             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 822
 823             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 824             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 825             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 826
 827             rinvsq00         = gmx_mm_inv_pd(rsq00);
 828             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 829             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 830             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 831
 832             /* Load parameters for j particles */
 833             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 834             vdwjidx0A        = 2*vdwtype[jnrA+0];
 835             vdwjidx0B        = 2*vdwtype[jnrB+0];
 836
 837             fjx0             = _mm_setzero_pd();
 838             fjy0             = _mm_setzero_pd();
 839             fjz0             = _mm_setzero_pd();
 840
 841             /**************************
 842              * CALCULATE INTERACTIONS *
 843              **************************/
 844
 845             /* Compute parameters for interactions between i and j atoms */
 846             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 847                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 848
 849             /* LENNARD-JONES DISPERSION/REPULSION */
 850
 851             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 852             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 853
 854             fscal            = fvdw;
 855
 856             /* Update vectorial force */
 857             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 858             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 859             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 860
 861             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 862             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 863             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 864
 865             /**************************
 866              * CALCULATE INTERACTIONS *
 867              **************************/
 868
 869             r10              = _mm_mul_pd(rsq10,rinv10);
 870
 871             /* Compute parameters for interactions between i and j atoms */
 872             qq10             = _mm_mul_pd(iq1,jq0);
 873
 874             /* EWALD ELECTROSTATICS */
 875
 876             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 877             ewrt             = _mm_mul_pd(r10,ewtabscale);
 878             ewitab           = _mm_cvttpd_epi32(ewrt);
 879 #ifdef __XOP__
 880             eweps            = _mm_frcz_pd(ewrt);
 881 #else
 882             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 883 #endif
 884             twoeweps         = _mm_add_pd(eweps,eweps);
 885             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 886                                          &ewtabF,&ewtabFn);
 887             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 888             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 889
 890             fscal            = felec;
 891
 892             /* Update vectorial force */
 893             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 894             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 895             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 896
 897             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 898             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 899             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             r20              = _mm_mul_pd(rsq20,rinv20);
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             qq20             = _mm_mul_pd(iq2,jq0);
 909
 910             /* EWALD ELECTROSTATICS */
 911
 912             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 913             ewrt             = _mm_mul_pd(r20,ewtabscale);
 914             ewitab           = _mm_cvttpd_epi32(ewrt);
 915 #ifdef __XOP__
 916             eweps            = _mm_frcz_pd(ewrt);
 917 #else
 918             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 919 #endif
 920             twoeweps         = _mm_add_pd(eweps,eweps);
 921             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 922                                          &ewtabF,&ewtabFn);
 923             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 924             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 925
 926             fscal            = felec;
 927
 928             /* Update vectorial force */
 929             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 930             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 931             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 932
 933             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 934             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 935             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 936
 937             /**************************
 938              * CALCULATE INTERACTIONS *
 939              **************************/
 940
 941             r30              = _mm_mul_pd(rsq30,rinv30);
 942
 943             /* Compute parameters for interactions between i and j atoms */
 944             qq30             = _mm_mul_pd(iq3,jq0);
 945
 946             /* EWALD ELECTROSTATICS */
 947
 948             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 949             ewrt             = _mm_mul_pd(r30,ewtabscale);
 950             ewitab           = _mm_cvttpd_epi32(ewrt);
 951 #ifdef __XOP__
 952             eweps            = _mm_frcz_pd(ewrt);
 953 #else
 954             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 955 #endif
 956             twoeweps         = _mm_add_pd(eweps,eweps);
 957             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 958                                          &ewtabF,&ewtabFn);
 959             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 960             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 961
 962             fscal            = felec;
 963
 964             /* Update vectorial force */
 965             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 966             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 967             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 968
 969             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 970             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 971             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 972
 973             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 974
 975             /* Inner loop uses 150 flops */
 976         }
 977
 978         if(jidx<j_index_end)
 979         {
 980
 981             jnrA             = jjnr[jidx];
 982             j_coord_offsetA  = DIM*jnrA;
 983
 984             /* load j atom coordinates */
 985             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 986                                               &jx0,&jy0,&jz0);
 987
 988             /* Calculate displacement vector */
 989             dx00             = _mm_sub_pd(ix0,jx0);
 990             dy00             = _mm_sub_pd(iy0,jy0);
 991             dz00             = _mm_sub_pd(iz0,jz0);
 992             dx10             = _mm_sub_pd(ix1,jx0);
 993             dy10             = _mm_sub_pd(iy1,jy0);
 994             dz10             = _mm_sub_pd(iz1,jz0);
 995             dx20             = _mm_sub_pd(ix2,jx0);
 996             dy20             = _mm_sub_pd(iy2,jy0);
 997             dz20             = _mm_sub_pd(iz2,jz0);
 998             dx30             = _mm_sub_pd(ix3,jx0);
 999             dy30             = _mm_sub_pd(iy3,jy0);
1000             dz30             = _mm_sub_pd(iz3,jz0);
1001
1002             /* Calculate squared distance and things based on it */
1003             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1004             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1005             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1006             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1007
1008             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1009             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1010             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1011
1012             rinvsq00         = gmx_mm_inv_pd(rsq00);
1013             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1014             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1015             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1016
1017             /* Load parameters for j particles */
1018             jq0              = _mm_load_sd(charge+jnrA+0);
1019             vdwjidx0A        = 2*vdwtype[jnrA+0];
1020
1021             fjx0             = _mm_setzero_pd();
1022             fjy0             = _mm_setzero_pd();
1023             fjz0             = _mm_setzero_pd();
1024
1025             /**************************
1026              * CALCULATE INTERACTIONS *
1027              **************************/
1028
1029             /* Compute parameters for interactions between i and j atoms */
1030             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1031
1032             /* LENNARD-JONES DISPERSION/REPULSION */
1033
1034             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1035             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1036
1037             fscal            = fvdw;
1038
1039             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1040
1041             /* Update vectorial force */
1042             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1043             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1044             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1045
1046             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1047             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1048             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1049
1050             /**************************
1051              * CALCULATE INTERACTIONS *
1052              **************************/
1053
1054             r10              = _mm_mul_pd(rsq10,rinv10);
1055
1056             /* Compute parameters for interactions between i and j atoms */
1057             qq10             = _mm_mul_pd(iq1,jq0);
1058
1059             /* EWALD ELECTROSTATICS */
1060
1061             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1062             ewrt             = _mm_mul_pd(r10,ewtabscale);
1063             ewitab           = _mm_cvttpd_epi32(ewrt);
1064 #ifdef __XOP__
1065             eweps            = _mm_frcz_pd(ewrt);
1066 #else
1067             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1068 #endif
1069             twoeweps         = _mm_add_pd(eweps,eweps);
1070             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1071             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1072             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1073
1074             fscal            = felec;
1075
1076             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1077
1078             /* Update vectorial force */
1079             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1080             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1081             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1082
1083             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1084             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1085             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1086
1087             /**************************
1088              * CALCULATE INTERACTIONS *
1089              **************************/
1090
1091             r20              = _mm_mul_pd(rsq20,rinv20);
1092
1093             /* Compute parameters for interactions between i and j atoms */
1094             qq20             = _mm_mul_pd(iq2,jq0);
1095
1096             /* EWALD ELECTROSTATICS */
1097
1098             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1099             ewrt             = _mm_mul_pd(r20,ewtabscale);
1100             ewitab           = _mm_cvttpd_epi32(ewrt);
1101 #ifdef __XOP__
1102             eweps            = _mm_frcz_pd(ewrt);
1103 #else
1104             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1105 #endif
1106             twoeweps         = _mm_add_pd(eweps,eweps);
1107             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1108             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1109             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1110
1111             fscal            = felec;
1112
1113             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1114
1115             /* Update vectorial force */
1116             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1117             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1118             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1119
1120             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1121             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1122             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1123
1124             /**************************
1125              * CALCULATE INTERACTIONS *
1126              **************************/
1127
1128             r30              = _mm_mul_pd(rsq30,rinv30);
1129
1130             /* Compute parameters for interactions between i and j atoms */
1131             qq30             = _mm_mul_pd(iq3,jq0);
1132
1133             /* EWALD ELECTROSTATICS */
1134
1135             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1136             ewrt             = _mm_mul_pd(r30,ewtabscale);
1137             ewitab           = _mm_cvttpd_epi32(ewrt);
1138 #ifdef __XOP__
1139             eweps            = _mm_frcz_pd(ewrt);
1140 #else
1141             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1142 #endif
1143             twoeweps         = _mm_add_pd(eweps,eweps);
1144             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1145             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1146             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1147
1148             fscal            = felec;
1149
1150             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1151
1152             /* Update vectorial force */
1153             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1154             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1155             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1156
1157             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1158             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1159             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1160
1161             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1162
1163             /* Inner loop uses 150 flops */
1164         }
1165
1166         /* End of innermost loop */
1167
1168         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1169                                               f+i_coord_offset,fshift+i_shift_offset);
1170
1171         /* Increment number of inner iterations */
1172         inneriter                  += j_index_end - j_index_start;
1173
1174         /* Outer loop uses 24 flops */
1175     }
1176
1177     /* Increment number of outer iterations */
1178     outeriter        += nri;
1179
1180     /* Update outer/inner flops */
1181
1182     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1183 }