src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJEw_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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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_VdwLJEw_GeomW4P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_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     __m128d           c6grid_00;
 103     __m128d           c6grid_10;
 104     __m128d           c6grid_20;
 105     __m128d           c6grid_30;
 106     real             *vdwgridparam;
 107     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 108     __m128d           one_half  = _mm_set1_pd(0.5);
 109     __m128d           minus_one = _mm_set1_pd(-1.0);
 110     __m128i          ewitab;
 111     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 112     real             *ewtab;
 113     __m128d          dummy_mask,cutoff_mask;
 114     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 115     __m128d          one     = _mm_set1_pd(1.0);
 116     __m128d          two     = _mm_set1_pd(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm_set1_pd(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133     vdwgridparam     = fr->ljpme_c6grid;
 134     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 135     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 136     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 137
 138     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 146     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 147     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 148     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 149
 150     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 175
 176         fix0             = _mm_setzero_pd();
 177         fiy0             = _mm_setzero_pd();
 178         fiz0             = _mm_setzero_pd();
 179         fix1             = _mm_setzero_pd();
 180         fiy1             = _mm_setzero_pd();
 181         fiz1             = _mm_setzero_pd();
 182         fix2             = _mm_setzero_pd();
 183         fiy2             = _mm_setzero_pd();
 184         fiz2             = _mm_setzero_pd();
 185         fix3             = _mm_setzero_pd();
 186         fiy3             = _mm_setzero_pd();
 187         fiz3             = _mm_setzero_pd();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm_setzero_pd();
 191         vvdwsum          = _mm_setzero_pd();
 192
 193         /* Start inner kernel loop */
 194         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 195         {
 196
 197             /* Get j neighbor index, and coordinate index */
 198             jnrA             = jjnr[jidx];
 199             jnrB             = jjnr[jidx+1];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_pd(ix0,jx0);
 209             dy00             = _mm_sub_pd(iy0,jy0);
 210             dz00             = _mm_sub_pd(iz0,jz0);
 211             dx10             = _mm_sub_pd(ix1,jx0);
 212             dy10             = _mm_sub_pd(iy1,jy0);
 213             dz10             = _mm_sub_pd(iz1,jz0);
 214             dx20             = _mm_sub_pd(ix2,jx0);
 215             dy20             = _mm_sub_pd(iy2,jy0);
 216             dz20             = _mm_sub_pd(iz2,jz0);
 217             dx30             = _mm_sub_pd(ix3,jx0);
 218             dy30             = _mm_sub_pd(iy3,jy0);
 219             dz30             = _mm_sub_pd(iz3,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 223             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 224             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 225             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 226
 227             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 228             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 229             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 230             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 231
 232             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 233             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 234             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 235             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 239             vdwjidx0A        = 2*vdwtype[jnrA+0];
 240             vdwjidx0B        = 2*vdwtype[jnrB+0];
 241
 242             fjx0             = _mm_setzero_pd();
 243             fjy0             = _mm_setzero_pd();
 244             fjz0             = _mm_setzero_pd();
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             r00              = _mm_mul_pd(rsq00,rinv00);
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 254                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 255             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 256                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 257
 258             /* Analytical LJ-PME */
 259             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 260             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 261             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 262             exponent         = gmx_simd_exp_d(ewcljrsq);
 263             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 264             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 265             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 266             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 267             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 268             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 269             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 270             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 274
 275             fscal            = fvdw;
 276
 277             /* Update vectorial force */
 278             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 279             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 280             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 281
 282             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 283             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 284             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 285
 286             /**************************
 287              * CALCULATE INTERACTIONS *
 288              **************************/
 289
 290             r10              = _mm_mul_pd(rsq10,rinv10);
 291
 292             /* Compute parameters for interactions between i and j atoms */
 293             qq10             = _mm_mul_pd(iq1,jq0);
 294
 295             /* EWALD ELECTROSTATICS */
 296
 297             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 298             ewrt             = _mm_mul_pd(r10,ewtabscale);
 299             ewitab           = _mm_cvttpd_epi32(ewrt);
 300 #ifdef __XOP__
 301             eweps            = _mm_frcz_pd(ewrt);
 302 #else
 303             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 304 #endif
 305             twoeweps         = _mm_add_pd(eweps,eweps);
 306             ewitab           = _mm_slli_epi32(ewitab,2);
 307             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 308             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 309             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 310             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 311             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 312             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 313             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 314             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 315             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 316             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velecsum         = _mm_add_pd(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             /* Update vectorial force */
 324             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 325             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 326             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 327
 328             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 329             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 330             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             r20              = _mm_mul_pd(rsq20,rinv20);
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq20             = _mm_mul_pd(iq2,jq0);
 340
 341             /* EWALD ELECTROSTATICS */
 342
 343             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 344             ewrt             = _mm_mul_pd(r20,ewtabscale);
 345             ewitab           = _mm_cvttpd_epi32(ewrt);
 346 #ifdef __XOP__
 347             eweps            = _mm_frcz_pd(ewrt);
 348 #else
 349             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 350 #endif
 351             twoeweps         = _mm_add_pd(eweps,eweps);
 352             ewitab           = _mm_slli_epi32(ewitab,2);
 353             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 354             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 355             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 356             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 357             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 358             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 359             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 360             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 361             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 362             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velecsum         = _mm_add_pd(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369             /* Update vectorial force */
 370             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 371             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 372             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 373
 374             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 375             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 376             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             r30              = _mm_mul_pd(rsq30,rinv30);
 383
 384             /* Compute parameters for interactions between i and j atoms */
 385             qq30             = _mm_mul_pd(iq3,jq0);
 386
 387             /* EWALD ELECTROSTATICS */
 388
 389             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 390             ewrt             = _mm_mul_pd(r30,ewtabscale);
 391             ewitab           = _mm_cvttpd_epi32(ewrt);
 392 #ifdef __XOP__
 393             eweps            = _mm_frcz_pd(ewrt);
 394 #else
 395             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 396 #endif
 397             twoeweps         = _mm_add_pd(eweps,eweps);
 398             ewitab           = _mm_slli_epi32(ewitab,2);
 399             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 400             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 401             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 402             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 403             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 404             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 405             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 406             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 407             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 408             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 409
 410             /* Update potential sum for this i atom from the interaction with this j atom. */
 411             velecsum         = _mm_add_pd(velecsum,velec);
 412
 413             fscal            = felec;
 414
 415             /* Update vectorial force */
 416             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 417             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 418             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 419
 420             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 421             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 422             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 423
 424             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 425
 426             /* Inner loop uses 185 flops */
 427         }
 428
 429         if(jidx<j_index_end)
 430         {
 431
 432             jnrA             = jjnr[jidx];
 433             j_coord_offsetA  = DIM*jnrA;
 434
 435             /* load j atom coordinates */
 436             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 437                                               &jx0,&jy0,&jz0);
 438
 439             /* Calculate displacement vector */
 440             dx00             = _mm_sub_pd(ix0,jx0);
 441             dy00             = _mm_sub_pd(iy0,jy0);
 442             dz00             = _mm_sub_pd(iz0,jz0);
 443             dx10             = _mm_sub_pd(ix1,jx0);
 444             dy10             = _mm_sub_pd(iy1,jy0);
 445             dz10             = _mm_sub_pd(iz1,jz0);
 446             dx20             = _mm_sub_pd(ix2,jx0);
 447             dy20             = _mm_sub_pd(iy2,jy0);
 448             dz20             = _mm_sub_pd(iz2,jz0);
 449             dx30             = _mm_sub_pd(ix3,jx0);
 450             dy30             = _mm_sub_pd(iy3,jy0);
 451             dz30             = _mm_sub_pd(iz3,jz0);
 452
 453             /* Calculate squared distance and things based on it */
 454             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 455             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 456             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 457             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 458
 459             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 460             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 461             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 462             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 463
 464             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 465             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 466             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 467             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 468
 469             /* Load parameters for j particles */
 470             jq0              = _mm_load_sd(charge+jnrA+0);
 471             vdwjidx0A        = 2*vdwtype[jnrA+0];
 472
 473             fjx0             = _mm_setzero_pd();
 474             fjy0             = _mm_setzero_pd();
 475             fjz0             = _mm_setzero_pd();
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             r00              = _mm_mul_pd(rsq00,rinv00);
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 485             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 486
 487             /* Analytical LJ-PME */
 488             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 489             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 490             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 491             exponent         = gmx_simd_exp_d(ewcljrsq);
 492             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 493             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 494             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 495             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 496             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 497             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 498             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 499             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 500
 501             /* Update potential sum for this i atom from the interaction with this j atom. */
 502             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 503             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 504
 505             fscal            = fvdw;
 506
 507             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 508
 509             /* Update vectorial force */
 510             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 511             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 512             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 513
 514             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 515             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 516             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             r10              = _mm_mul_pd(rsq10,rinv10);
 523
 524             /* Compute parameters for interactions between i and j atoms */
 525             qq10             = _mm_mul_pd(iq1,jq0);
 526
 527             /* EWALD ELECTROSTATICS */
 528
 529             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 530             ewrt             = _mm_mul_pd(r10,ewtabscale);
 531             ewitab           = _mm_cvttpd_epi32(ewrt);
 532 #ifdef __XOP__
 533             eweps            = _mm_frcz_pd(ewrt);
 534 #else
 535             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 536 #endif
 537             twoeweps         = _mm_add_pd(eweps,eweps);
 538             ewitab           = _mm_slli_epi32(ewitab,2);
 539             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 540             ewtabD           = _mm_setzero_pd();
 541             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 542             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 543             ewtabFn          = _mm_setzero_pd();
 544             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 545             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 546             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 547             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 548             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 552             velecsum         = _mm_add_pd(velecsum,velec);
 553
 554             fscal            = felec;
 555
 556             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 557
 558             /* Update vectorial force */
 559             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 560             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 561             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 562
 563             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 564             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 565             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 566
 567             /**************************
 568              * CALCULATE INTERACTIONS *
 569              **************************/
 570
 571             r20              = _mm_mul_pd(rsq20,rinv20);
 572
 573             /* Compute parameters for interactions between i and j atoms */
 574             qq20             = _mm_mul_pd(iq2,jq0);
 575
 576             /* EWALD ELECTROSTATICS */
 577
 578             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 579             ewrt             = _mm_mul_pd(r20,ewtabscale);
 580             ewitab           = _mm_cvttpd_epi32(ewrt);
 581 #ifdef __XOP__
 582             eweps            = _mm_frcz_pd(ewrt);
 583 #else
 584             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 585 #endif
 586             twoeweps         = _mm_add_pd(eweps,eweps);
 587             ewitab           = _mm_slli_epi32(ewitab,2);
 588             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 589             ewtabD           = _mm_setzero_pd();
 590             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 591             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 592             ewtabFn          = _mm_setzero_pd();
 593             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 594             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 595             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 596             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 597             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 598
 599             /* Update potential sum for this i atom from the interaction with this j atom. */
 600             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 601             velecsum         = _mm_add_pd(velecsum,velec);
 602
 603             fscal            = felec;
 604
 605             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 606
 607             /* Update vectorial force */
 608             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 609             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 610             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 611
 612             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 613             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 614             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             r30              = _mm_mul_pd(rsq30,rinv30);
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq30             = _mm_mul_pd(iq3,jq0);
 624
 625             /* EWALD ELECTROSTATICS */
 626
 627             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 628             ewrt             = _mm_mul_pd(r30,ewtabscale);
 629             ewitab           = _mm_cvttpd_epi32(ewrt);
 630 #ifdef __XOP__
 631             eweps            = _mm_frcz_pd(ewrt);
 632 #else
 633             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 634 #endif
 635             twoeweps         = _mm_add_pd(eweps,eweps);
 636             ewitab           = _mm_slli_epi32(ewitab,2);
 637             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 638             ewtabD           = _mm_setzero_pd();
 639             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 640             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 641             ewtabFn          = _mm_setzero_pd();
 642             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 643             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 644             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 645             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 646             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 647
 648             /* Update potential sum for this i atom from the interaction with this j atom. */
 649             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 650             velecsum         = _mm_add_pd(velecsum,velec);
 651
 652             fscal            = felec;
 653
 654             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 655
 656             /* Update vectorial force */
 657             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 658             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 659             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 660
 661             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 662             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 663             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 664
 665             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 666
 667             /* Inner loop uses 185 flops */
 668         }
 669
 670         /* End of innermost loop */
 671
 672         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 673                                               f+i_coord_offset,fshift+i_shift_offset);
 674
 675         ggid                        = gid[iidx];
 676         /* Update potential energies */
 677         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 678         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 679
 680         /* Increment number of inner iterations */
 681         inneriter                  += j_index_end - j_index_start;
 682
 683         /* Outer loop uses 26 flops */
 684     }
 685
 686     /* Increment number of outer iterations */
 687     outeriter        += nri;
 688
 689     /* Update outer/inner flops */
 690
 691     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*185);
 692 }
 693 /*
 694  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_double
 695  * Electrostatics interaction: Ewald
 696  * VdW interaction:            LJEwald
 697  * Geometry:                   Water4-Particle
 698  * Calculate force/pot:        Force
 699  */
 700 void
 701 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_double
 702                     (t_nblist                    * gmx_restrict       nlist,
 703                      rvec                        * gmx_restrict          xx,
 704                      rvec                        * gmx_restrict          ff,
 705                      t_forcerec                  * gmx_restrict          fr,
 706                      t_mdatoms                   * gmx_restrict     mdatoms,
 707                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 708                      t_nrnb                      * gmx_restrict        nrnb)
 709 {
 710     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 711      * just 0 for non-waters.
 712      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 713      * jnr indices corresponding to data put in the four positions in the SIMD register.
 714      */
 715     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 716     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 717     int              jnrA,jnrB;
 718     int              j_coord_offsetA,j_coord_offsetB;
 719     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 720     real             rcutoff_scalar;
 721     real             *shiftvec,*fshift,*x,*f;
 722     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 723     int              vdwioffset0;
 724     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 725     int              vdwioffset1;
 726     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 727     int              vdwioffset2;
 728     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 729     int              vdwioffset3;
 730     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 731     int              vdwjidx0A,vdwjidx0B;
 732     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 733     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 734     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 735     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 736     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 737     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 738     real             *charge;
 739     int              nvdwtype;
 740     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 741     int              *vdwtype;
 742     real             *vdwparam;
 743     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 744     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 745     __m128d           c6grid_00;
 746     __m128d           c6grid_10;
 747     __m128d           c6grid_20;
 748     __m128d           c6grid_30;
 749     real             *vdwgridparam;
 750     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 751     __m128d           one_half  = _mm_set1_pd(0.5);
 752     __m128d           minus_one = _mm_set1_pd(-1.0);
 753     __m128i          ewitab;
 754     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 755     real             *ewtab;
 756     __m128d          dummy_mask,cutoff_mask;
 757     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 758     __m128d          one     = _mm_set1_pd(1.0);
 759     __m128d          two     = _mm_set1_pd(2.0);
 760     x                = xx[0];
 761     f                = ff[0];
 762
 763     nri              = nlist->nri;
 764     iinr             = nlist->iinr;
 765     jindex           = nlist->jindex;
 766     jjnr             = nlist->jjnr;
 767     shiftidx         = nlist->shift;
 768     gid              = nlist->gid;
 769     shiftvec         = fr->shift_vec[0];
 770     fshift           = fr->fshift[0];
 771     facel            = _mm_set1_pd(fr->epsfac);
 772     charge           = mdatoms->chargeA;
 773     nvdwtype         = fr->ntype;
 774     vdwparam         = fr->nbfp;
 775     vdwtype          = mdatoms->typeA;
 776     vdwgridparam     = fr->ljpme_c6grid;
 777     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 778     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 779     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 780
 781     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 782     ewtab            = fr->ic->tabq_coul_F;
 783     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 784     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 785
 786     /* Setup water-specific parameters */
 787     inr              = nlist->iinr[0];
 788     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 789     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 790     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 791     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 792
 793     /* Avoid stupid compiler warnings */
 794     jnrA = jnrB = 0;
 795     j_coord_offsetA = 0;
 796     j_coord_offsetB = 0;
 797
 798     outeriter        = 0;
 799     inneriter        = 0;
 800
 801     /* Start outer loop over neighborlists */
 802     for(iidx=0; iidx<nri; iidx++)
 803     {
 804         /* Load shift vector for this list */
 805         i_shift_offset   = DIM*shiftidx[iidx];
 806
 807         /* Load limits for loop over neighbors */
 808         j_index_start    = jindex[iidx];
 809         j_index_end      = jindex[iidx+1];
 810
 811         /* Get outer coordinate index */
 812         inr              = iinr[iidx];
 813         i_coord_offset   = DIM*inr;
 814
 815         /* Load i particle coords and add shift vector */
 816         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 817                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 818
 819         fix0             = _mm_setzero_pd();
 820         fiy0             = _mm_setzero_pd();
 821         fiz0             = _mm_setzero_pd();
 822         fix1             = _mm_setzero_pd();
 823         fiy1             = _mm_setzero_pd();
 824         fiz1             = _mm_setzero_pd();
 825         fix2             = _mm_setzero_pd();
 826         fiy2             = _mm_setzero_pd();
 827         fiz2             = _mm_setzero_pd();
 828         fix3             = _mm_setzero_pd();
 829         fiy3             = _mm_setzero_pd();
 830         fiz3             = _mm_setzero_pd();
 831
 832         /* Start inner kernel loop */
 833         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 834         {
 835
 836             /* Get j neighbor index, and coordinate index */
 837             jnrA             = jjnr[jidx];
 838             jnrB             = jjnr[jidx+1];
 839             j_coord_offsetA  = DIM*jnrA;
 840             j_coord_offsetB  = DIM*jnrB;
 841
 842             /* load j atom coordinates */
 843             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 844                                               &jx0,&jy0,&jz0);
 845
 846             /* Calculate displacement vector */
 847             dx00             = _mm_sub_pd(ix0,jx0);
 848             dy00             = _mm_sub_pd(iy0,jy0);
 849             dz00             = _mm_sub_pd(iz0,jz0);
 850             dx10             = _mm_sub_pd(ix1,jx0);
 851             dy10             = _mm_sub_pd(iy1,jy0);
 852             dz10             = _mm_sub_pd(iz1,jz0);
 853             dx20             = _mm_sub_pd(ix2,jx0);
 854             dy20             = _mm_sub_pd(iy2,jy0);
 855             dz20             = _mm_sub_pd(iz2,jz0);
 856             dx30             = _mm_sub_pd(ix3,jx0);
 857             dy30             = _mm_sub_pd(iy3,jy0);
 858             dz30             = _mm_sub_pd(iz3,jz0);
 859
 860             /* Calculate squared distance and things based on it */
 861             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 862             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 863             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 864             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 865
 866             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 867             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 868             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 869             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 870
 871             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 872             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 873             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 874             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 875
 876             /* Load parameters for j particles */
 877             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 878             vdwjidx0A        = 2*vdwtype[jnrA+0];
 879             vdwjidx0B        = 2*vdwtype[jnrB+0];
 880
 881             fjx0             = _mm_setzero_pd();
 882             fjy0             = _mm_setzero_pd();
 883             fjz0             = _mm_setzero_pd();
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             r00              = _mm_mul_pd(rsq00,rinv00);
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 893                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 894             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 895                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 896
 897             /* Analytical LJ-PME */
 898             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 899             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 900             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 901             exponent         = gmx_simd_exp_d(ewcljrsq);
 902             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 903             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 904             /* f6A = 6 * C6grid * (1 - poly) */
 905             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 906             /* f6B = C6grid * exponent * beta^6 */
 907             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 908             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 909             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 910
 911             fscal            = fvdw;
 912
 913             /* Update vectorial force */
 914             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 915             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 916             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 917
 918             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 919             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 920             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 921
 922             /**************************
 923              * CALCULATE INTERACTIONS *
 924              **************************/
 925
 926             r10              = _mm_mul_pd(rsq10,rinv10);
 927
 928             /* Compute parameters for interactions between i and j atoms */
 929             qq10             = _mm_mul_pd(iq1,jq0);
 930
 931             /* EWALD ELECTROSTATICS */
 932
 933             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 934             ewrt             = _mm_mul_pd(r10,ewtabscale);
 935             ewitab           = _mm_cvttpd_epi32(ewrt);
 936 #ifdef __XOP__
 937             eweps            = _mm_frcz_pd(ewrt);
 938 #else
 939             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 940 #endif
 941             twoeweps         = _mm_add_pd(eweps,eweps);
 942             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 943                                          &ewtabF,&ewtabFn);
 944             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 945             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 946
 947             fscal            = felec;
 948
 949             /* Update vectorial force */
 950             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 951             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 952             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 953
 954             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 955             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 956             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 957
 958             /**************************
 959              * CALCULATE INTERACTIONS *
 960              **************************/
 961
 962             r20              = _mm_mul_pd(rsq20,rinv20);
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             qq20             = _mm_mul_pd(iq2,jq0);
 966
 967             /* EWALD ELECTROSTATICS */
 968
 969             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 970             ewrt             = _mm_mul_pd(r20,ewtabscale);
 971             ewitab           = _mm_cvttpd_epi32(ewrt);
 972 #ifdef __XOP__
 973             eweps            = _mm_frcz_pd(ewrt);
 974 #else
 975             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 976 #endif
 977             twoeweps         = _mm_add_pd(eweps,eweps);
 978             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 979                                          &ewtabF,&ewtabFn);
 980             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 981             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 982
 983             fscal            = felec;
 984
 985             /* Update vectorial force */
 986             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 987             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 988             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 989
 990             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 991             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 992             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             r30              = _mm_mul_pd(rsq30,rinv30);
 999
1000             /* Compute parameters for interactions between i and j atoms */
1001             qq30             = _mm_mul_pd(iq3,jq0);
1002
1003             /* EWALD ELECTROSTATICS */
1004
1005             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1006             ewrt             = _mm_mul_pd(r30,ewtabscale);
1007             ewitab           = _mm_cvttpd_epi32(ewrt);
1008 #ifdef __XOP__
1009             eweps            = _mm_frcz_pd(ewrt);
1010 #else
1011             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1012 #endif
1013             twoeweps         = _mm_add_pd(eweps,eweps);
1014             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1015                                          &ewtabF,&ewtabFn);
1016             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1017             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1018
1019             fscal            = felec;
1020
1021             /* Update vectorial force */
1022             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1023             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1024             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1025
1026             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1027             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1028             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1029
1030             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1031
1032             /* Inner loop uses 167 flops */
1033         }
1034
1035         if(jidx<j_index_end)
1036         {
1037
1038             jnrA             = jjnr[jidx];
1039             j_coord_offsetA  = DIM*jnrA;
1040
1041             /* load j atom coordinates */
1042             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1043                                               &jx0,&jy0,&jz0);
1044
1045             /* Calculate displacement vector */
1046             dx00             = _mm_sub_pd(ix0,jx0);
1047             dy00             = _mm_sub_pd(iy0,jy0);
1048             dz00             = _mm_sub_pd(iz0,jz0);
1049             dx10             = _mm_sub_pd(ix1,jx0);
1050             dy10             = _mm_sub_pd(iy1,jy0);
1051             dz10             = _mm_sub_pd(iz1,jz0);
1052             dx20             = _mm_sub_pd(ix2,jx0);
1053             dy20             = _mm_sub_pd(iy2,jy0);
1054             dz20             = _mm_sub_pd(iz2,jz0);
1055             dx30             = _mm_sub_pd(ix3,jx0);
1056             dy30             = _mm_sub_pd(iy3,jy0);
1057             dz30             = _mm_sub_pd(iz3,jz0);
1058
1059             /* Calculate squared distance and things based on it */
1060             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1061             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1062             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1063             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1064
1065             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1066             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1067             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1068             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1069
1070             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1071             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1072             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1073             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1074
1075             /* Load parameters for j particles */
1076             jq0              = _mm_load_sd(charge+jnrA+0);
1077             vdwjidx0A        = 2*vdwtype[jnrA+0];
1078
1079             fjx0             = _mm_setzero_pd();
1080             fjy0             = _mm_setzero_pd();
1081             fjz0             = _mm_setzero_pd();
1082
1083             /**************************
1084              * CALCULATE INTERACTIONS *
1085              **************************/
1086
1087             r00              = _mm_mul_pd(rsq00,rinv00);
1088
1089             /* Compute parameters for interactions between i and j atoms */
1090             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1091             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1092
1093             /* Analytical LJ-PME */
1094             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1095             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1096             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1097             exponent         = gmx_simd_exp_d(ewcljrsq);
1098             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1099             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1100             /* f6A = 6 * C6grid * (1 - poly) */
1101             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1102             /* f6B = C6grid * exponent * beta^6 */
1103             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1104             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1105             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1106
1107             fscal            = fvdw;
1108
1109             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1110
1111             /* Update vectorial force */
1112             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1113             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1114             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1115
1116             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1117             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1118             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1119
1120             /**************************
1121              * CALCULATE INTERACTIONS *
1122              **************************/
1123
1124             r10              = _mm_mul_pd(rsq10,rinv10);
1125
1126             /* Compute parameters for interactions between i and j atoms */
1127             qq10             = _mm_mul_pd(iq1,jq0);
1128
1129             /* EWALD ELECTROSTATICS */
1130
1131             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1132             ewrt             = _mm_mul_pd(r10,ewtabscale);
1133             ewitab           = _mm_cvttpd_epi32(ewrt);
1134 #ifdef __XOP__
1135             eweps            = _mm_frcz_pd(ewrt);
1136 #else
1137             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1138 #endif
1139             twoeweps         = _mm_add_pd(eweps,eweps);
1140             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1141             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1142             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1143
1144             fscal            = felec;
1145
1146             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1147
1148             /* Update vectorial force */
1149             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1150             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1151             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1152
1153             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1154             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1155             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1156
1157             /**************************
1158              * CALCULATE INTERACTIONS *
1159              **************************/
1160
1161             r20              = _mm_mul_pd(rsq20,rinv20);
1162
1163             /* Compute parameters for interactions between i and j atoms */
1164             qq20             = _mm_mul_pd(iq2,jq0);
1165
1166             /* EWALD ELECTROSTATICS */
1167
1168             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1169             ewrt             = _mm_mul_pd(r20,ewtabscale);
1170             ewitab           = _mm_cvttpd_epi32(ewrt);
1171 #ifdef __XOP__
1172             eweps            = _mm_frcz_pd(ewrt);
1173 #else
1174             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1175 #endif
1176             twoeweps         = _mm_add_pd(eweps,eweps);
1177             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1178             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1179             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1180
1181             fscal            = felec;
1182
1183             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1184
1185             /* Update vectorial force */
1186             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1187             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1188             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1189
1190             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1191             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1192             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1193
1194             /**************************
1195              * CALCULATE INTERACTIONS *
1196              **************************/
1197
1198             r30              = _mm_mul_pd(rsq30,rinv30);
1199
1200             /* Compute parameters for interactions between i and j atoms */
1201             qq30             = _mm_mul_pd(iq3,jq0);
1202
1203             /* EWALD ELECTROSTATICS */
1204
1205             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1206             ewrt             = _mm_mul_pd(r30,ewtabscale);
1207             ewitab           = _mm_cvttpd_epi32(ewrt);
1208 #ifdef __XOP__
1209             eweps            = _mm_frcz_pd(ewrt);
1210 #else
1211             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1212 #endif
1213             twoeweps         = _mm_add_pd(eweps,eweps);
1214             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1215             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1216             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1217
1218             fscal            = felec;
1219
1220             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1221
1222             /* Update vectorial force */
1223             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1224             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1225             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1226
1227             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1228             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1229             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1230
1231             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1232
1233             /* Inner loop uses 167 flops */
1234         }
1235
1236         /* End of innermost loop */
1237
1238         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1239                                               f+i_coord_offset,fshift+i_shift_offset);
1240
1241         /* Increment number of inner iterations */
1242         inneriter                  += j_index_end - j_index_start;
1243
1244         /* Outer loop uses 24 flops */
1245     }
1246
1247     /* Increment number of outer iterations */
1248     outeriter        += nri;
1249
1250     /* Update outer/inner flops */
1251
1252     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*167);
1253 }