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) 2012,2013,2014, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128i          ewitab;
 103     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     real             *ewtab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 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     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 134     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 135     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 136     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 163
 164         fix0             = _mm_setzero_pd();
 165         fiy0             = _mm_setzero_pd();
 166         fiz0             = _mm_setzero_pd();
 167         fix1             = _mm_setzero_pd();
 168         fiy1             = _mm_setzero_pd();
 169         fiz1             = _mm_setzero_pd();
 170         fix2             = _mm_setzero_pd();
 171         fiy2             = _mm_setzero_pd();
 172         fiz2             = _mm_setzero_pd();
 173         fix3             = _mm_setzero_pd();
 174         fiy3             = _mm_setzero_pd();
 175         fiz3             = _mm_setzero_pd();
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm_setzero_pd();
 179         vvdwsum          = _mm_setzero_pd();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190
 191             /* load j atom coordinates */
 192             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm_sub_pd(ix0,jx0);
 197             dy00             = _mm_sub_pd(iy0,jy0);
 198             dz00             = _mm_sub_pd(iz0,jz0);
 199             dx10             = _mm_sub_pd(ix1,jx0);
 200             dy10             = _mm_sub_pd(iy1,jy0);
 201             dz10             = _mm_sub_pd(iz1,jz0);
 202             dx20             = _mm_sub_pd(ix2,jx0);
 203             dy20             = _mm_sub_pd(iy2,jy0);
 204             dz20             = _mm_sub_pd(iz2,jz0);
 205             dx30             = _mm_sub_pd(ix3,jx0);
 206             dy30             = _mm_sub_pd(iy3,jy0);
 207             dz30             = _mm_sub_pd(iz3,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 213             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 214
 215             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 216             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 217             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 218
 219             rinvsq00         = gmx_mm_inv_pd(rsq00);
 220             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 221             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 222             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 226             vdwjidx0A        = 2*vdwtype[jnrA+0];
 227             vdwjidx0B        = 2*vdwtype[jnrB+0];
 228
 229             fjx0             = _mm_setzero_pd();
 230             fjy0             = _mm_setzero_pd();
 231             fjz0             = _mm_setzero_pd();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 239                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 240
 241             /* LENNARD-JONES DISPERSION/REPULSION */
 242
 243             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 244             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 245             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 246             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 247             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 248
 249             /* Update potential sum for this i atom from the interaction with this j atom. */
 250             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 251
 252             fscal            = fvdw;
 253
 254             /* Update vectorial force */
 255             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 256             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 257             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 258
 259             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 260             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 261             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             r10              = _mm_mul_pd(rsq10,rinv10);
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq10             = _mm_mul_pd(iq1,jq0);
 271
 272             /* EWALD ELECTROSTATICS */
 273
 274             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 275             ewrt             = _mm_mul_pd(r10,ewtabscale);
 276             ewitab           = _mm_cvttpd_epi32(ewrt);
 277 #ifdef __XOP__
 278             eweps            = _mm_frcz_pd(ewrt);
 279 #else
 280             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 281 #endif
 282             twoeweps         = _mm_add_pd(eweps,eweps);
 283             ewitab           = _mm_slli_epi32(ewitab,2);
 284             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 285             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 286             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 287             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 288             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 289             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 290             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 291             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 292             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 293             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velecsum         = _mm_add_pd(velecsum,velec);
 297
 298             fscal            = felec;
 299
 300             /* Update vectorial force */
 301             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 302             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 303             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 304
 305             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 306             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 307             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 308
 309             /**************************
 310              * CALCULATE INTERACTIONS *
 311              **************************/
 312
 313             r20              = _mm_mul_pd(rsq20,rinv20);
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             qq20             = _mm_mul_pd(iq2,jq0);
 317
 318             /* EWALD ELECTROSTATICS */
 319
 320             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 321             ewrt             = _mm_mul_pd(r20,ewtabscale);
 322             ewitab           = _mm_cvttpd_epi32(ewrt);
 323 #ifdef __XOP__
 324             eweps            = _mm_frcz_pd(ewrt);
 325 #else
 326             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 327 #endif
 328             twoeweps         = _mm_add_pd(eweps,eweps);
 329             ewitab           = _mm_slli_epi32(ewitab,2);
 330             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 331             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 332             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 333             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 334             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 335             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 336             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 337             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 338             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 339             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Update vectorial force */
 347             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 348             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 349             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 350
 351             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 352             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 353             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             r30              = _mm_mul_pd(rsq30,rinv30);
 360
 361             /* Compute parameters for interactions between i and j atoms */
 362             qq30             = _mm_mul_pd(iq3,jq0);
 363
 364             /* EWALD ELECTROSTATICS */
 365
 366             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 367             ewrt             = _mm_mul_pd(r30,ewtabscale);
 368             ewitab           = _mm_cvttpd_epi32(ewrt);
 369 #ifdef __XOP__
 370             eweps            = _mm_frcz_pd(ewrt);
 371 #else
 372             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 373 #endif
 374             twoeweps         = _mm_add_pd(eweps,eweps);
 375             ewitab           = _mm_slli_epi32(ewitab,2);
 376             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 377             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 378             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 379             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 380             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 381             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 382             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 383             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 384             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 385             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velecsum         = _mm_add_pd(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             /* Update vectorial force */
 393             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 394             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 395             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 396
 397             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 398             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 399             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 400
 401             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 402
 403             /* Inner loop uses 170 flops */
 404         }
 405
 406         if(jidx<j_index_end)
 407         {
 408
 409             jnrA             = jjnr[jidx];
 410             j_coord_offsetA  = DIM*jnrA;
 411
 412             /* load j atom coordinates */
 413             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 414                                               &jx0,&jy0,&jz0);
 415
 416             /* Calculate displacement vector */
 417             dx00             = _mm_sub_pd(ix0,jx0);
 418             dy00             = _mm_sub_pd(iy0,jy0);
 419             dz00             = _mm_sub_pd(iz0,jz0);
 420             dx10             = _mm_sub_pd(ix1,jx0);
 421             dy10             = _mm_sub_pd(iy1,jy0);
 422             dz10             = _mm_sub_pd(iz1,jz0);
 423             dx20             = _mm_sub_pd(ix2,jx0);
 424             dy20             = _mm_sub_pd(iy2,jy0);
 425             dz20             = _mm_sub_pd(iz2,jz0);
 426             dx30             = _mm_sub_pd(ix3,jx0);
 427             dy30             = _mm_sub_pd(iy3,jy0);
 428             dz30             = _mm_sub_pd(iz3,jz0);
 429
 430             /* Calculate squared distance and things based on it */
 431             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 432             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 433             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 434             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 435
 436             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 437             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 438             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 439
 440             rinvsq00         = gmx_mm_inv_pd(rsq00);
 441             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 442             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 443             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 444
 445             /* Load parameters for j particles */
 446             jq0              = _mm_load_sd(charge+jnrA+0);
 447             vdwjidx0A        = 2*vdwtype[jnrA+0];
 448
 449             fjx0             = _mm_setzero_pd();
 450             fjy0             = _mm_setzero_pd();
 451             fjz0             = _mm_setzero_pd();
 452
 453             /**************************
 454              * CALCULATE INTERACTIONS *
 455              **************************/
 456
 457             /* Compute parameters for interactions between i and j atoms */
 458             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 459
 460             /* LENNARD-JONES DISPERSION/REPULSION */
 461
 462             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 463             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 464             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 465             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 466             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 467
 468             /* Update potential sum for this i atom from the interaction with this j atom. */
 469             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 470             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 471
 472             fscal            = fvdw;
 473
 474             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 475
 476             /* Update vectorial force */
 477             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 478             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 479             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 480
 481             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 482             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 483             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 484
 485             /**************************
 486              * CALCULATE INTERACTIONS *
 487              **************************/
 488
 489             r10              = _mm_mul_pd(rsq10,rinv10);
 490
 491             /* Compute parameters for interactions between i and j atoms */
 492             qq10             = _mm_mul_pd(iq1,jq0);
 493
 494             /* EWALD ELECTROSTATICS */
 495
 496             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 497             ewrt             = _mm_mul_pd(r10,ewtabscale);
 498             ewitab           = _mm_cvttpd_epi32(ewrt);
 499 #ifdef __XOP__
 500             eweps            = _mm_frcz_pd(ewrt);
 501 #else
 502             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 503 #endif
 504             twoeweps         = _mm_add_pd(eweps,eweps);
 505             ewitab           = _mm_slli_epi32(ewitab,2);
 506             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 507             ewtabD           = _mm_setzero_pd();
 508             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 509             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 510             ewtabFn          = _mm_setzero_pd();
 511             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 512             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 513             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 514             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 515             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 516
 517             /* Update potential sum for this i atom from the interaction with this j atom. */
 518             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 519             velecsum         = _mm_add_pd(velecsum,velec);
 520
 521             fscal            = felec;
 522
 523             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 524
 525             /* Update vectorial force */
 526             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 527             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 528             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 529
 530             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 531             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 532             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             r20              = _mm_mul_pd(rsq20,rinv20);
 539
 540             /* Compute parameters for interactions between i and j atoms */
 541             qq20             = _mm_mul_pd(iq2,jq0);
 542
 543             /* EWALD ELECTROSTATICS */
 544
 545             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 546             ewrt             = _mm_mul_pd(r20,ewtabscale);
 547             ewitab           = _mm_cvttpd_epi32(ewrt);
 548 #ifdef __XOP__
 549             eweps            = _mm_frcz_pd(ewrt);
 550 #else
 551             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 552 #endif
 553             twoeweps         = _mm_add_pd(eweps,eweps);
 554             ewitab           = _mm_slli_epi32(ewitab,2);
 555             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 556             ewtabD           = _mm_setzero_pd();
 557             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 558             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 559             ewtabFn          = _mm_setzero_pd();
 560             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 561             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 562             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 563             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 564             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 565
 566             /* Update potential sum for this i atom from the interaction with this j atom. */
 567             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 568             velecsum         = _mm_add_pd(velecsum,velec);
 569
 570             fscal            = felec;
 571
 572             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 573
 574             /* Update vectorial force */
 575             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 576             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 577             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 578
 579             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 580             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 581             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 582
 583             /**************************
 584              * CALCULATE INTERACTIONS *
 585              **************************/
 586
 587             r30              = _mm_mul_pd(rsq30,rinv30);
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq30             = _mm_mul_pd(iq3,jq0);
 591
 592             /* EWALD ELECTROSTATICS */
 593
 594             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 595             ewrt             = _mm_mul_pd(r30,ewtabscale);
 596             ewitab           = _mm_cvttpd_epi32(ewrt);
 597 #ifdef __XOP__
 598             eweps            = _mm_frcz_pd(ewrt);
 599 #else
 600             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 601 #endif
 602             twoeweps         = _mm_add_pd(eweps,eweps);
 603             ewitab           = _mm_slli_epi32(ewitab,2);
 604             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 605             ewtabD           = _mm_setzero_pd();
 606             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 607             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 608             ewtabFn          = _mm_setzero_pd();
 609             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 610             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 611             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 612             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 613             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 614
 615             /* Update potential sum for this i atom from the interaction with this j atom. */
 616             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 617             velecsum         = _mm_add_pd(velecsum,velec);
 618
 619             fscal            = felec;
 620
 621             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 622
 623             /* Update vectorial force */
 624             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 625             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 626             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 627
 628             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 629             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 630             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 631
 632             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 633
 634             /* Inner loop uses 170 flops */
 635         }
 636
 637         /* End of innermost loop */
 638
 639         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 640                                               f+i_coord_offset,fshift+i_shift_offset);
 641
 642         ggid                        = gid[iidx];
 643         /* Update potential energies */
 644         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 645         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 646
 647         /* Increment number of inner iterations */
 648         inneriter                  += j_index_end - j_index_start;
 649
 650         /* Outer loop uses 26 flops */
 651     }
 652
 653     /* Increment number of outer iterations */
 654     outeriter        += nri;
 655
 656     /* Update outer/inner flops */
 657
 658     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*170);
 659 }
 660 /*
 661  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_128_fma_double
 662  * Electrostatics interaction: Ewald
 663  * VdW interaction:            LennardJones
 664  * Geometry:                   Water4-Particle
 665  * Calculate force/pot:        Force
 666  */
 667 void
 668 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_128_fma_double
 669                     (t_nblist                    * gmx_restrict       nlist,
 670                      rvec                        * gmx_restrict          xx,
 671                      rvec                        * gmx_restrict          ff,
 672                      t_forcerec                  * gmx_restrict          fr,
 673                      t_mdatoms                   * gmx_restrict     mdatoms,
 674                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 675                      t_nrnb                      * gmx_restrict        nrnb)
 676 {
 677     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 678      * just 0 for non-waters.
 679      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 680      * jnr indices corresponding to data put in the four positions in the SIMD register.
 681      */
 682     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 683     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 684     int              jnrA,jnrB;
 685     int              j_coord_offsetA,j_coord_offsetB;
 686     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 687     real             rcutoff_scalar;
 688     real             *shiftvec,*fshift,*x,*f;
 689     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 690     int              vdwioffset0;
 691     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 692     int              vdwioffset1;
 693     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 694     int              vdwioffset2;
 695     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 696     int              vdwioffset3;
 697     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 698     int              vdwjidx0A,vdwjidx0B;
 699     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 700     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 701     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 702     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 703     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 704     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 705     real             *charge;
 706     int              nvdwtype;
 707     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 708     int              *vdwtype;
 709     real             *vdwparam;
 710     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 711     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 712     __m128i          ewitab;
 713     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 714     real             *ewtab;
 715     __m128d          dummy_mask,cutoff_mask;
 716     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 717     __m128d          one     = _mm_set1_pd(1.0);
 718     __m128d          two     = _mm_set1_pd(2.0);
 719     x                = xx[0];
 720     f                = ff[0];
 721
 722     nri              = nlist->nri;
 723     iinr             = nlist->iinr;
 724     jindex           = nlist->jindex;
 725     jjnr             = nlist->jjnr;
 726     shiftidx         = nlist->shift;
 727     gid              = nlist->gid;
 728     shiftvec         = fr->shift_vec[0];
 729     fshift           = fr->fshift[0];
 730     facel            = _mm_set1_pd(fr->epsfac);
 731     charge           = mdatoms->chargeA;
 732     nvdwtype         = fr->ntype;
 733     vdwparam         = fr->nbfp;
 734     vdwtype          = mdatoms->typeA;
 735
 736     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 737     ewtab            = fr->ic->tabq_coul_F;
 738     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 739     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 740
 741     /* Setup water-specific parameters */
 742     inr              = nlist->iinr[0];
 743     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 744     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 745     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 746     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 747
 748     /* Avoid stupid compiler warnings */
 749     jnrA = jnrB = 0;
 750     j_coord_offsetA = 0;
 751     j_coord_offsetB = 0;
 752
 753     outeriter        = 0;
 754     inneriter        = 0;
 755
 756     /* Start outer loop over neighborlists */
 757     for(iidx=0; iidx<nri; iidx++)
 758     {
 759         /* Load shift vector for this list */
 760         i_shift_offset   = DIM*shiftidx[iidx];
 761
 762         /* Load limits for loop over neighbors */
 763         j_index_start    = jindex[iidx];
 764         j_index_end      = jindex[iidx+1];
 765
 766         /* Get outer coordinate index */
 767         inr              = iinr[iidx];
 768         i_coord_offset   = DIM*inr;
 769
 770         /* Load i particle coords and add shift vector */
 771         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 772                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 773
 774         fix0             = _mm_setzero_pd();
 775         fiy0             = _mm_setzero_pd();
 776         fiz0             = _mm_setzero_pd();
 777         fix1             = _mm_setzero_pd();
 778         fiy1             = _mm_setzero_pd();
 779         fiz1             = _mm_setzero_pd();
 780         fix2             = _mm_setzero_pd();
 781         fiy2             = _mm_setzero_pd();
 782         fiz2             = _mm_setzero_pd();
 783         fix3             = _mm_setzero_pd();
 784         fiy3             = _mm_setzero_pd();
 785         fiz3             = _mm_setzero_pd();
 786
 787         /* Start inner kernel loop */
 788         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 789         {
 790
 791             /* Get j neighbor index, and coordinate index */
 792             jnrA             = jjnr[jidx];
 793             jnrB             = jjnr[jidx+1];
 794             j_coord_offsetA  = DIM*jnrA;
 795             j_coord_offsetB  = DIM*jnrB;
 796
 797             /* load j atom coordinates */
 798             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 799                                               &jx0,&jy0,&jz0);
 800
 801             /* Calculate displacement vector */
 802             dx00             = _mm_sub_pd(ix0,jx0);
 803             dy00             = _mm_sub_pd(iy0,jy0);
 804             dz00             = _mm_sub_pd(iz0,jz0);
 805             dx10             = _mm_sub_pd(ix1,jx0);
 806             dy10             = _mm_sub_pd(iy1,jy0);
 807             dz10             = _mm_sub_pd(iz1,jz0);
 808             dx20             = _mm_sub_pd(ix2,jx0);
 809             dy20             = _mm_sub_pd(iy2,jy0);
 810             dz20             = _mm_sub_pd(iz2,jz0);
 811             dx30             = _mm_sub_pd(ix3,jx0);
 812             dy30             = _mm_sub_pd(iy3,jy0);
 813             dz30             = _mm_sub_pd(iz3,jz0);
 814
 815             /* Calculate squared distance and things based on it */
 816             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 817             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 818             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 819             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 820
 821             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 822             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 823             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 824
 825             rinvsq00         = gmx_mm_inv_pd(rsq00);
 826             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 827             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 828             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 829
 830             /* Load parameters for j particles */
 831             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 832             vdwjidx0A        = 2*vdwtype[jnrA+0];
 833             vdwjidx0B        = 2*vdwtype[jnrB+0];
 834
 835             fjx0             = _mm_setzero_pd();
 836             fjy0             = _mm_setzero_pd();
 837             fjz0             = _mm_setzero_pd();
 838
 839             /**************************
 840              * CALCULATE INTERACTIONS *
 841              **************************/
 842
 843             /* Compute parameters for interactions between i and j atoms */
 844             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 845                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 846
 847             /* LENNARD-JONES DISPERSION/REPULSION */
 848
 849             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 850             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 851
 852             fscal            = fvdw;
 853
 854             /* Update vectorial force */
 855             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 856             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 857             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 858
 859             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 860             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 861             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 862
 863             /**************************
 864              * CALCULATE INTERACTIONS *
 865              **************************/
 866
 867             r10              = _mm_mul_pd(rsq10,rinv10);
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq10             = _mm_mul_pd(iq1,jq0);
 871
 872             /* EWALD ELECTROSTATICS */
 873
 874             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 875             ewrt             = _mm_mul_pd(r10,ewtabscale);
 876             ewitab           = _mm_cvttpd_epi32(ewrt);
 877 #ifdef __XOP__
 878             eweps            = _mm_frcz_pd(ewrt);
 879 #else
 880             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 881 #endif
 882             twoeweps         = _mm_add_pd(eweps,eweps);
 883             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 884                                          &ewtabF,&ewtabFn);
 885             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 886             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 887
 888             fscal            = felec;
 889
 890             /* Update vectorial force */
 891             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 892             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 893             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 894
 895             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 896             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 897             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 898
 899             /**************************
 900              * CALCULATE INTERACTIONS *
 901              **************************/
 902
 903             r20              = _mm_mul_pd(rsq20,rinv20);
 904
 905             /* Compute parameters for interactions between i and j atoms */
 906             qq20             = _mm_mul_pd(iq2,jq0);
 907
 908             /* EWALD ELECTROSTATICS */
 909
 910             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 911             ewrt             = _mm_mul_pd(r20,ewtabscale);
 912             ewitab           = _mm_cvttpd_epi32(ewrt);
 913 #ifdef __XOP__
 914             eweps            = _mm_frcz_pd(ewrt);
 915 #else
 916             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 917 #endif
 918             twoeweps         = _mm_add_pd(eweps,eweps);
 919             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 920                                          &ewtabF,&ewtabFn);
 921             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 922             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 923
 924             fscal            = felec;
 925
 926             /* Update vectorial force */
 927             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 928             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 929             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 930
 931             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 932             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 933             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 934
 935             /**************************
 936              * CALCULATE INTERACTIONS *
 937              **************************/
 938
 939             r30              = _mm_mul_pd(rsq30,rinv30);
 940
 941             /* Compute parameters for interactions between i and j atoms */
 942             qq30             = _mm_mul_pd(iq3,jq0);
 943
 944             /* EWALD ELECTROSTATICS */
 945
 946             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 947             ewrt             = _mm_mul_pd(r30,ewtabscale);
 948             ewitab           = _mm_cvttpd_epi32(ewrt);
 949 #ifdef __XOP__
 950             eweps            = _mm_frcz_pd(ewrt);
 951 #else
 952             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 953 #endif
 954             twoeweps         = _mm_add_pd(eweps,eweps);
 955             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 956                                          &ewtabF,&ewtabFn);
 957             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 958             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 959
 960             fscal            = felec;
 961
 962             /* Update vectorial force */
 963             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 964             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 965             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 966
 967             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 968             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 969             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 970
 971             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 972
 973             /* Inner loop uses 150 flops */
 974         }
 975
 976         if(jidx<j_index_end)
 977         {
 978
 979             jnrA             = jjnr[jidx];
 980             j_coord_offsetA  = DIM*jnrA;
 981
 982             /* load j atom coordinates */
 983             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 984                                               &jx0,&jy0,&jz0);
 985
 986             /* Calculate displacement vector */
 987             dx00             = _mm_sub_pd(ix0,jx0);
 988             dy00             = _mm_sub_pd(iy0,jy0);
 989             dz00             = _mm_sub_pd(iz0,jz0);
 990             dx10             = _mm_sub_pd(ix1,jx0);
 991             dy10             = _mm_sub_pd(iy1,jy0);
 992             dz10             = _mm_sub_pd(iz1,jz0);
 993             dx20             = _mm_sub_pd(ix2,jx0);
 994             dy20             = _mm_sub_pd(iy2,jy0);
 995             dz20             = _mm_sub_pd(iz2,jz0);
 996             dx30             = _mm_sub_pd(ix3,jx0);
 997             dy30             = _mm_sub_pd(iy3,jy0);
 998             dz30             = _mm_sub_pd(iz3,jz0);
 999
1000             /* Calculate squared distance and things based on it */
1001             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1002             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1003             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1004             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1005
1006             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1007             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1008             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1009
1010             rinvsq00         = gmx_mm_inv_pd(rsq00);
1011             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1012             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1013             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1014
1015             /* Load parameters for j particles */
1016             jq0              = _mm_load_sd(charge+jnrA+0);
1017             vdwjidx0A        = 2*vdwtype[jnrA+0];
1018
1019             fjx0             = _mm_setzero_pd();
1020             fjy0             = _mm_setzero_pd();
1021             fjz0             = _mm_setzero_pd();
1022
1023             /**************************
1024              * CALCULATE INTERACTIONS *
1025              **************************/
1026
1027             /* Compute parameters for interactions between i and j atoms */
1028             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1029
1030             /* LENNARD-JONES DISPERSION/REPULSION */
1031
1032             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1033             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1034
1035             fscal            = fvdw;
1036
1037             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1038
1039             /* Update vectorial force */
1040             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1041             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1042             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1043
1044             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1045             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1046             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1047
1048             /**************************
1049              * CALCULATE INTERACTIONS *
1050              **************************/
1051
1052             r10              = _mm_mul_pd(rsq10,rinv10);
1053
1054             /* Compute parameters for interactions between i and j atoms */
1055             qq10             = _mm_mul_pd(iq1,jq0);
1056
1057             /* EWALD ELECTROSTATICS */
1058
1059             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1060             ewrt             = _mm_mul_pd(r10,ewtabscale);
1061             ewitab           = _mm_cvttpd_epi32(ewrt);
1062 #ifdef __XOP__
1063             eweps            = _mm_frcz_pd(ewrt);
1064 #else
1065             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1066 #endif
1067             twoeweps         = _mm_add_pd(eweps,eweps);
1068             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1069             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1070             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1071
1072             fscal            = felec;
1073
1074             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1075
1076             /* Update vectorial force */
1077             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1078             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1079             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1080
1081             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1082             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1083             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1084
1085             /**************************
1086              * CALCULATE INTERACTIONS *
1087              **************************/
1088
1089             r20              = _mm_mul_pd(rsq20,rinv20);
1090
1091             /* Compute parameters for interactions between i and j atoms */
1092             qq20             = _mm_mul_pd(iq2,jq0);
1093
1094             /* EWALD ELECTROSTATICS */
1095
1096             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1097             ewrt             = _mm_mul_pd(r20,ewtabscale);
1098             ewitab           = _mm_cvttpd_epi32(ewrt);
1099 #ifdef __XOP__
1100             eweps            = _mm_frcz_pd(ewrt);
1101 #else
1102             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1103 #endif
1104             twoeweps         = _mm_add_pd(eweps,eweps);
1105             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1106             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1107             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1108
1109             fscal            = felec;
1110
1111             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1112
1113             /* Update vectorial force */
1114             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1115             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1116             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1117
1118             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1119             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1120             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1121
1122             /**************************
1123              * CALCULATE INTERACTIONS *
1124              **************************/
1125
1126             r30              = _mm_mul_pd(rsq30,rinv30);
1127
1128             /* Compute parameters for interactions between i and j atoms */
1129             qq30             = _mm_mul_pd(iq3,jq0);
1130
1131             /* EWALD ELECTROSTATICS */
1132
1133             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1134             ewrt             = _mm_mul_pd(r30,ewtabscale);
1135             ewitab           = _mm_cvttpd_epi32(ewrt);
1136 #ifdef __XOP__
1137             eweps            = _mm_frcz_pd(ewrt);
1138 #else
1139             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1140 #endif
1141             twoeweps         = _mm_add_pd(eweps,eweps);
1142             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1143             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1144             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1145
1146             fscal            = felec;
1147
1148             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1149
1150             /* Update vectorial force */
1151             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1152             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1153             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1154
1155             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1156             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1157             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1158
1159             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1160
1161             /* Inner loop uses 150 flops */
1162         }
1163
1164         /* End of innermost loop */
1165
1166         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1167                                               f+i_coord_offset,fshift+i_shift_offset);
1168
1169         /* Increment number of inner iterations */
1170         inneriter                  += j_index_end - j_index_start;
1171
1172         /* Outer loop uses 24 flops */
1173     }
1174
1175     /* Increment number of outer iterations */
1176     outeriter        += nri;
1177
1178     /* Update outer/inner flops */
1179
1180     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1181 }