src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_avx_128_fma_single.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_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 107     real             *vftab;
 108     __m128i          ewitab;
 109     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 111     real             *ewtab;
 112     __m128           dummy_mask,cutoff_mask;
 113     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 114     __m128           one     = _mm_set1_ps(1.0);
 115     __m128           two     = _mm_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm_set1_ps(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     vftab            = kernel_data->table_vdw->data;
 134     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 135
 136     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 137     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 138     beta2            = _mm_mul_ps(beta,beta);
 139     beta3            = _mm_mul_ps(beta,beta2);
 140     ewtab            = fr->ic->tabq_coul_FDV0;
 141     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 142     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 143
 144     /* Setup water-specific parameters */
 145     inr              = nlist->iinr[0];
 146     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 147     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 148     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 149     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = jnrC = jnrD = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155     j_coord_offsetC = 0;
 156     j_coord_offsetD = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     for(iidx=0;iidx<4*DIM;iidx++)
 162     {
 163         scratch[iidx] = 0.0;
 164     }
 165
 166     /* Start outer loop over neighborlists */
 167     for(iidx=0; iidx<nri; iidx++)
 168     {
 169         /* Load shift vector for this list */
 170         i_shift_offset   = DIM*shiftidx[iidx];
 171
 172         /* Load limits for loop over neighbors */
 173         j_index_start    = jindex[iidx];
 174         j_index_end      = jindex[iidx+1];
 175
 176         /* Get outer coordinate index */
 177         inr              = iinr[iidx];
 178         i_coord_offset   = DIM*inr;
 179
 180         /* Load i particle coords and add shift vector */
 181         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 182                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 183
 184         fix0             = _mm_setzero_ps();
 185         fiy0             = _mm_setzero_ps();
 186         fiz0             = _mm_setzero_ps();
 187         fix1             = _mm_setzero_ps();
 188         fiy1             = _mm_setzero_ps();
 189         fiz1             = _mm_setzero_ps();
 190         fix2             = _mm_setzero_ps();
 191         fiy2             = _mm_setzero_ps();
 192         fiz2             = _mm_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm_setzero_ps();
 196         vvdwsum          = _mm_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211
 212             /* load j atom coordinates */
 213             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                               x+j_coord_offsetC,x+j_coord_offsetD,
 215                                               &jx0,&jy0,&jz0);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm_sub_ps(ix0,jx0);
 219             dy00             = _mm_sub_ps(iy0,jy0);
 220             dz00             = _mm_sub_ps(iz0,jz0);
 221             dx10             = _mm_sub_ps(ix1,jx0);
 222             dy10             = _mm_sub_ps(iy1,jy0);
 223             dz10             = _mm_sub_ps(iz1,jz0);
 224             dx20             = _mm_sub_ps(ix2,jx0);
 225             dy20             = _mm_sub_ps(iy2,jy0);
 226             dz20             = _mm_sub_ps(iz2,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 230             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 231             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 232
 233             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 234             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 235             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 236
 237             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 238             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 239             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 240
 241             /* Load parameters for j particles */
 242             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 243                                                               charge+jnrC+0,charge+jnrD+0);
 244             vdwjidx0A        = 2*vdwtype[jnrA+0];
 245             vdwjidx0B        = 2*vdwtype[jnrB+0];
 246             vdwjidx0C        = 2*vdwtype[jnrC+0];
 247             vdwjidx0D        = 2*vdwtype[jnrD+0];
 248
 249             fjx0             = _mm_setzero_ps();
 250             fjy0             = _mm_setzero_ps();
 251             fjz0             = _mm_setzero_ps();
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             r00              = _mm_mul_ps(rsq00,rinv00);
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             qq00             = _mm_mul_ps(iq0,jq0);
 261             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 262                                          vdwparam+vdwioffset0+vdwjidx0B,
 263                                          vdwparam+vdwioffset0+vdwjidx0C,
 264                                          vdwparam+vdwioffset0+vdwjidx0D,
 265                                          &c6_00,&c12_00);
 266
 267             /* Calculate table index by multiplying r with table scale and truncate to integer */
 268             rt               = _mm_mul_ps(r00,vftabscale);
 269             vfitab           = _mm_cvttps_epi32(rt);
 270 #ifdef __XOP__
 271             vfeps            = _mm_frcz_ps(rt);
 272 #else
 273             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 274 #endif
 275             twovfeps         = _mm_add_ps(vfeps,vfeps);
 276             vfitab           = _mm_slli_epi32(vfitab,3);
 277
 278             /* EWALD ELECTROSTATICS */
 279
 280             /* Analytical PME correction */
 281             zeta2            = _mm_mul_ps(beta2,rsq00);
 282             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 283             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 284             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 285             felec            = _mm_mul_ps(qq00,felec);
 286             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 287             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 288             velec            = _mm_mul_ps(qq00,velec);
 289
 290             /* CUBIC SPLINE TABLE DISPERSION */
 291             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 292             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 293             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 294             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 295             _MM_TRANSPOSE4_PS(Y,F,G,H);
 296             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 297             VV               = _mm_macc_ps(vfeps,Fp,Y);
 298             vvdw6            = _mm_mul_ps(c6_00,VV);
 299             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 300             fvdw6            = _mm_mul_ps(c6_00,FF);
 301
 302             /* CUBIC SPLINE TABLE REPULSION */
 303             vfitab           = _mm_add_epi32(vfitab,ifour);
 304             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 305             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 306             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 307             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 308             _MM_TRANSPOSE4_PS(Y,F,G,H);
 309             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 310             VV               = _mm_macc_ps(vfeps,Fp,Y);
 311             vvdw12           = _mm_mul_ps(c12_00,VV);
 312             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 313             fvdw12           = _mm_mul_ps(c12_00,FF);
 314             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 315             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velecsum         = _mm_add_ps(velecsum,velec);
 319             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 320
 321             fscal            = _mm_add_ps(felec,fvdw);
 322
 323              /* Update vectorial force */
 324             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 325             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 326             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 327
 328             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 329             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 330             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             r10              = _mm_mul_ps(rsq10,rinv10);
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq10             = _mm_mul_ps(iq1,jq0);
 340
 341             /* EWALD ELECTROSTATICS */
 342
 343             /* Analytical PME correction */
 344             zeta2            = _mm_mul_ps(beta2,rsq10);
 345             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 346             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 347             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 348             felec            = _mm_mul_ps(qq10,felec);
 349             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 350             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 351             velec            = _mm_mul_ps(qq10,velec);
 352
 353             /* Update potential sum for this i atom from the interaction with this j atom. */
 354             velecsum         = _mm_add_ps(velecsum,velec);
 355
 356             fscal            = felec;
 357
 358              /* Update vectorial force */
 359             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 360             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 361             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 362
 363             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 364             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 365             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             r20              = _mm_mul_ps(rsq20,rinv20);
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq20             = _mm_mul_ps(iq2,jq0);
 375
 376             /* EWALD ELECTROSTATICS */
 377
 378             /* Analytical PME correction */
 379             zeta2            = _mm_mul_ps(beta2,rsq20);
 380             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 381             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 382             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 383             felec            = _mm_mul_ps(qq20,felec);
 384             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 385             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 386             velec            = _mm_mul_ps(qq20,velec);
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velecsum         = _mm_add_ps(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393              /* Update vectorial force */
 394             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 395             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 396             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 397
 398             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 399             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 400             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 401
 402             fjptrA             = f+j_coord_offsetA;
 403             fjptrB             = f+j_coord_offsetB;
 404             fjptrC             = f+j_coord_offsetC;
 405             fjptrD             = f+j_coord_offsetD;
 406
 407             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 408
 409             /* Inner loop uses 121 flops */
 410         }
 411
 412         if(jidx<j_index_end)
 413         {
 414
 415             /* Get j neighbor index, and coordinate index */
 416             jnrlistA         = jjnr[jidx];
 417             jnrlistB         = jjnr[jidx+1];
 418             jnrlistC         = jjnr[jidx+2];
 419             jnrlistD         = jjnr[jidx+3];
 420             /* Sign of each element will be negative for non-real atoms.
 421              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 422              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 423              */
 424             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 425             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 426             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 427             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 428             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 429             j_coord_offsetA  = DIM*jnrA;
 430             j_coord_offsetB  = DIM*jnrB;
 431             j_coord_offsetC  = DIM*jnrC;
 432             j_coord_offsetD  = DIM*jnrD;
 433
 434             /* load j atom coordinates */
 435             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 436                                               x+j_coord_offsetC,x+j_coord_offsetD,
 437                                               &jx0,&jy0,&jz0);
 438
 439             /* Calculate displacement vector */
 440             dx00             = _mm_sub_ps(ix0,jx0);
 441             dy00             = _mm_sub_ps(iy0,jy0);
 442             dz00             = _mm_sub_ps(iz0,jz0);
 443             dx10             = _mm_sub_ps(ix1,jx0);
 444             dy10             = _mm_sub_ps(iy1,jy0);
 445             dz10             = _mm_sub_ps(iz1,jz0);
 446             dx20             = _mm_sub_ps(ix2,jx0);
 447             dy20             = _mm_sub_ps(iy2,jy0);
 448             dz20             = _mm_sub_ps(iz2,jz0);
 449
 450             /* Calculate squared distance and things based on it */
 451             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 452             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 453             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 454
 455             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 456             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 457             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 458
 459             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 460             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 461             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 462
 463             /* Load parameters for j particles */
 464             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 465                                                               charge+jnrC+0,charge+jnrD+0);
 466             vdwjidx0A        = 2*vdwtype[jnrA+0];
 467             vdwjidx0B        = 2*vdwtype[jnrB+0];
 468             vdwjidx0C        = 2*vdwtype[jnrC+0];
 469             vdwjidx0D        = 2*vdwtype[jnrD+0];
 470
 471             fjx0             = _mm_setzero_ps();
 472             fjy0             = _mm_setzero_ps();
 473             fjz0             = _mm_setzero_ps();
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             r00              = _mm_mul_ps(rsq00,rinv00);
 480             r00              = _mm_andnot_ps(dummy_mask,r00);
 481
 482             /* Compute parameters for interactions between i and j atoms */
 483             qq00             = _mm_mul_ps(iq0,jq0);
 484             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 485                                          vdwparam+vdwioffset0+vdwjidx0B,
 486                                          vdwparam+vdwioffset0+vdwjidx0C,
 487                                          vdwparam+vdwioffset0+vdwjidx0D,
 488                                          &c6_00,&c12_00);
 489
 490             /* Calculate table index by multiplying r with table scale and truncate to integer */
 491             rt               = _mm_mul_ps(r00,vftabscale);
 492             vfitab           = _mm_cvttps_epi32(rt);
 493 #ifdef __XOP__
 494             vfeps            = _mm_frcz_ps(rt);
 495 #else
 496             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 497 #endif
 498             twovfeps         = _mm_add_ps(vfeps,vfeps);
 499             vfitab           = _mm_slli_epi32(vfitab,3);
 500
 501             /* EWALD ELECTROSTATICS */
 502
 503             /* Analytical PME correction */
 504             zeta2            = _mm_mul_ps(beta2,rsq00);
 505             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 506             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 507             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 508             felec            = _mm_mul_ps(qq00,felec);
 509             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 510             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 511             velec            = _mm_mul_ps(qq00,velec);
 512
 513             /* CUBIC SPLINE TABLE DISPERSION */
 514             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 515             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 516             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 517             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 518             _MM_TRANSPOSE4_PS(Y,F,G,H);
 519             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 520             VV               = _mm_macc_ps(vfeps,Fp,Y);
 521             vvdw6            = _mm_mul_ps(c6_00,VV);
 522             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 523             fvdw6            = _mm_mul_ps(c6_00,FF);
 524
 525             /* CUBIC SPLINE TABLE REPULSION */
 526             vfitab           = _mm_add_epi32(vfitab,ifour);
 527             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 528             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 529             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 530             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 531             _MM_TRANSPOSE4_PS(Y,F,G,H);
 532             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 533             VV               = _mm_macc_ps(vfeps,Fp,Y);
 534             vvdw12           = _mm_mul_ps(c12_00,VV);
 535             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 536             fvdw12           = _mm_mul_ps(c12_00,FF);
 537             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 538             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 539
 540             /* Update potential sum for this i atom from the interaction with this j atom. */
 541             velec            = _mm_andnot_ps(dummy_mask,velec);
 542             velecsum         = _mm_add_ps(velecsum,velec);
 543             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 544             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 545
 546             fscal            = _mm_add_ps(felec,fvdw);
 547
 548             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 549
 550              /* Update vectorial force */
 551             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 552             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 553             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 554
 555             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 556             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 557             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             r10              = _mm_mul_ps(rsq10,rinv10);
 564             r10              = _mm_andnot_ps(dummy_mask,r10);
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq10             = _mm_mul_ps(iq1,jq0);
 568
 569             /* EWALD ELECTROSTATICS */
 570
 571             /* Analytical PME correction */
 572             zeta2            = _mm_mul_ps(beta2,rsq10);
 573             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 574             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 575             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 576             felec            = _mm_mul_ps(qq10,felec);
 577             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 578             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 579             velec            = _mm_mul_ps(qq10,velec);
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm_andnot_ps(dummy_mask,velec);
 583             velecsum         = _mm_add_ps(velecsum,velec);
 584
 585             fscal            = felec;
 586
 587             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 588
 589              /* Update vectorial force */
 590             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 591             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 592             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 593
 594             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 595             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 596             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             r20              = _mm_mul_ps(rsq20,rinv20);
 603             r20              = _mm_andnot_ps(dummy_mask,r20);
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq20             = _mm_mul_ps(iq2,jq0);
 607
 608             /* EWALD ELECTROSTATICS */
 609
 610             /* Analytical PME correction */
 611             zeta2            = _mm_mul_ps(beta2,rsq20);
 612             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 613             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 614             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 615             felec            = _mm_mul_ps(qq20,felec);
 616             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 617             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 618             velec            = _mm_mul_ps(qq20,velec);
 619
 620             /* Update potential sum for this i atom from the interaction with this j atom. */
 621             velec            = _mm_andnot_ps(dummy_mask,velec);
 622             velecsum         = _mm_add_ps(velecsum,velec);
 623
 624             fscal            = felec;
 625
 626             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 627
 628              /* Update vectorial force */
 629             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 630             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 631             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 632
 633             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 634             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 635             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 636
 637             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 638             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 639             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 640             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 641
 642             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 643
 644             /* Inner loop uses 124 flops */
 645         }
 646
 647         /* End of innermost loop */
 648
 649         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 650                                               f+i_coord_offset,fshift+i_shift_offset);
 651
 652         ggid                        = gid[iidx];
 653         /* Update potential energies */
 654         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 655         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 656
 657         /* Increment number of inner iterations */
 658         inneriter                  += j_index_end - j_index_start;
 659
 660         /* Outer loop uses 20 flops */
 661     }
 662
 663     /* Increment number of outer iterations */
 664     outeriter        += nri;
 665
 666     /* Update outer/inner flops */
 667
 668     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*124);
 669 }
 670 /*
 671  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 672  * Electrostatics interaction: Ewald
 673  * VdW interaction:            CubicSplineTable
 674  * Geometry:                   Water3-Particle
 675  * Calculate force/pot:        Force
 676  */
 677 void
 678 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 679                     (t_nblist                    * gmx_restrict       nlist,
 680                      rvec                        * gmx_restrict          xx,
 681                      rvec                        * gmx_restrict          ff,
 682                      t_forcerec                  * gmx_restrict          fr,
 683                      t_mdatoms                   * gmx_restrict     mdatoms,
 684                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 685                      t_nrnb                      * gmx_restrict        nrnb)
 686 {
 687     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 688      * just 0 for non-waters.
 689      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 690      * jnr indices corresponding to data put in the four positions in the SIMD register.
 691      */
 692     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 693     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 694     int              jnrA,jnrB,jnrC,jnrD;
 695     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 696     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 697     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 698     real             rcutoff_scalar;
 699     real             *shiftvec,*fshift,*x,*f;
 700     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 701     real             scratch[4*DIM];
 702     __m128           fscal,rcutoff,rcutoff2,jidxall;
 703     int              vdwioffset0;
 704     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 705     int              vdwioffset1;
 706     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 707     int              vdwioffset2;
 708     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 709     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 710     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 711     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 712     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 713     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 714     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 715     real             *charge;
 716     int              nvdwtype;
 717     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 718     int              *vdwtype;
 719     real             *vdwparam;
 720     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 721     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 722     __m128i          vfitab;
 723     __m128i          ifour       = _mm_set1_epi32(4);
 724     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 725     real             *vftab;
 726     __m128i          ewitab;
 727     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 728     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 729     real             *ewtab;
 730     __m128           dummy_mask,cutoff_mask;
 731     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 732     __m128           one     = _mm_set1_ps(1.0);
 733     __m128           two     = _mm_set1_ps(2.0);
 734     x                = xx[0];
 735     f                = ff[0];
 736
 737     nri              = nlist->nri;
 738     iinr             = nlist->iinr;
 739     jindex           = nlist->jindex;
 740     jjnr             = nlist->jjnr;
 741     shiftidx         = nlist->shift;
 742     gid              = nlist->gid;
 743     shiftvec         = fr->shift_vec[0];
 744     fshift           = fr->fshift[0];
 745     facel            = _mm_set1_ps(fr->epsfac);
 746     charge           = mdatoms->chargeA;
 747     nvdwtype         = fr->ntype;
 748     vdwparam         = fr->nbfp;
 749     vdwtype          = mdatoms->typeA;
 750
 751     vftab            = kernel_data->table_vdw->data;
 752     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 753
 754     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 755     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 756     beta2            = _mm_mul_ps(beta,beta);
 757     beta3            = _mm_mul_ps(beta,beta2);
 758     ewtab            = fr->ic->tabq_coul_F;
 759     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 760     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 761
 762     /* Setup water-specific parameters */
 763     inr              = nlist->iinr[0];
 764     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 765     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 766     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 767     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 768
 769     /* Avoid stupid compiler warnings */
 770     jnrA = jnrB = jnrC = jnrD = 0;
 771     j_coord_offsetA = 0;
 772     j_coord_offsetB = 0;
 773     j_coord_offsetC = 0;
 774     j_coord_offsetD = 0;
 775
 776     outeriter        = 0;
 777     inneriter        = 0;
 778
 779     for(iidx=0;iidx<4*DIM;iidx++)
 780     {
 781         scratch[iidx] = 0.0;
 782     }
 783
 784     /* Start outer loop over neighborlists */
 785     for(iidx=0; iidx<nri; iidx++)
 786     {
 787         /* Load shift vector for this list */
 788         i_shift_offset   = DIM*shiftidx[iidx];
 789
 790         /* Load limits for loop over neighbors */
 791         j_index_start    = jindex[iidx];
 792         j_index_end      = jindex[iidx+1];
 793
 794         /* Get outer coordinate index */
 795         inr              = iinr[iidx];
 796         i_coord_offset   = DIM*inr;
 797
 798         /* Load i particle coords and add shift vector */
 799         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 800                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 801
 802         fix0             = _mm_setzero_ps();
 803         fiy0             = _mm_setzero_ps();
 804         fiz0             = _mm_setzero_ps();
 805         fix1             = _mm_setzero_ps();
 806         fiy1             = _mm_setzero_ps();
 807         fiz1             = _mm_setzero_ps();
 808         fix2             = _mm_setzero_ps();
 809         fiy2             = _mm_setzero_ps();
 810         fiz2             = _mm_setzero_ps();
 811
 812         /* Start inner kernel loop */
 813         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 814         {
 815
 816             /* Get j neighbor index, and coordinate index */
 817             jnrA             = jjnr[jidx];
 818             jnrB             = jjnr[jidx+1];
 819             jnrC             = jjnr[jidx+2];
 820             jnrD             = jjnr[jidx+3];
 821             j_coord_offsetA  = DIM*jnrA;
 822             j_coord_offsetB  = DIM*jnrB;
 823             j_coord_offsetC  = DIM*jnrC;
 824             j_coord_offsetD  = DIM*jnrD;
 825
 826             /* load j atom coordinates */
 827             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 828                                               x+j_coord_offsetC,x+j_coord_offsetD,
 829                                               &jx0,&jy0,&jz0);
 830
 831             /* Calculate displacement vector */
 832             dx00             = _mm_sub_ps(ix0,jx0);
 833             dy00             = _mm_sub_ps(iy0,jy0);
 834             dz00             = _mm_sub_ps(iz0,jz0);
 835             dx10             = _mm_sub_ps(ix1,jx0);
 836             dy10             = _mm_sub_ps(iy1,jy0);
 837             dz10             = _mm_sub_ps(iz1,jz0);
 838             dx20             = _mm_sub_ps(ix2,jx0);
 839             dy20             = _mm_sub_ps(iy2,jy0);
 840             dz20             = _mm_sub_ps(iz2,jz0);
 841
 842             /* Calculate squared distance and things based on it */
 843             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 844             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 845             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 846
 847             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 848             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 849             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 850
 851             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 852             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 853             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 854
 855             /* Load parameters for j particles */
 856             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 857                                                               charge+jnrC+0,charge+jnrD+0);
 858             vdwjidx0A        = 2*vdwtype[jnrA+0];
 859             vdwjidx0B        = 2*vdwtype[jnrB+0];
 860             vdwjidx0C        = 2*vdwtype[jnrC+0];
 861             vdwjidx0D        = 2*vdwtype[jnrD+0];
 862
 863             fjx0             = _mm_setzero_ps();
 864             fjy0             = _mm_setzero_ps();
 865             fjz0             = _mm_setzero_ps();
 866
 867             /**************************
 868              * CALCULATE INTERACTIONS *
 869              **************************/
 870
 871             r00              = _mm_mul_ps(rsq00,rinv00);
 872
 873             /* Compute parameters for interactions between i and j atoms */
 874             qq00             = _mm_mul_ps(iq0,jq0);
 875             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 876                                          vdwparam+vdwioffset0+vdwjidx0B,
 877                                          vdwparam+vdwioffset0+vdwjidx0C,
 878                                          vdwparam+vdwioffset0+vdwjidx0D,
 879                                          &c6_00,&c12_00);
 880
 881             /* Calculate table index by multiplying r with table scale and truncate to integer */
 882             rt               = _mm_mul_ps(r00,vftabscale);
 883             vfitab           = _mm_cvttps_epi32(rt);
 884 #ifdef __XOP__
 885             vfeps            = _mm_frcz_ps(rt);
 886 #else
 887             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 888 #endif
 889             twovfeps         = _mm_add_ps(vfeps,vfeps);
 890             vfitab           = _mm_slli_epi32(vfitab,3);
 891
 892             /* EWALD ELECTROSTATICS */
 893
 894             /* Analytical PME correction */
 895             zeta2            = _mm_mul_ps(beta2,rsq00);
 896             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 897             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 898             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 899             felec            = _mm_mul_ps(qq00,felec);
 900
 901             /* CUBIC SPLINE TABLE DISPERSION */
 902             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 903             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 904             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 905             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 906             _MM_TRANSPOSE4_PS(Y,F,G,H);
 907             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 908             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 909             fvdw6            = _mm_mul_ps(c6_00,FF);
 910
 911             /* CUBIC SPLINE TABLE REPULSION */
 912             vfitab           = _mm_add_epi32(vfitab,ifour);
 913             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 914             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 915             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 916             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 917             _MM_TRANSPOSE4_PS(Y,F,G,H);
 918             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 919             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 920             fvdw12           = _mm_mul_ps(c12_00,FF);
 921             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 922
 923             fscal            = _mm_add_ps(felec,fvdw);
 924
 925              /* Update vectorial force */
 926             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 927             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 928             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 929
 930             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 931             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 932             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 933
 934             /**************************
 935              * CALCULATE INTERACTIONS *
 936              **************************/
 937
 938             r10              = _mm_mul_ps(rsq10,rinv10);
 939
 940             /* Compute parameters for interactions between i and j atoms */
 941             qq10             = _mm_mul_ps(iq1,jq0);
 942
 943             /* EWALD ELECTROSTATICS */
 944
 945             /* Analytical PME correction */
 946             zeta2            = _mm_mul_ps(beta2,rsq10);
 947             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 948             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 949             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 950             felec            = _mm_mul_ps(qq10,felec);
 951
 952             fscal            = felec;
 953
 954              /* Update vectorial force */
 955             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 956             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 957             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 958
 959             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 960             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 961             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             r20              = _mm_mul_ps(rsq20,rinv20);
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq20             = _mm_mul_ps(iq2,jq0);
 971
 972             /* EWALD ELECTROSTATICS */
 973
 974             /* Analytical PME correction */
 975             zeta2            = _mm_mul_ps(beta2,rsq20);
 976             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 977             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 978             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 979             felec            = _mm_mul_ps(qq20,felec);
 980
 981             fscal            = felec;
 982
 983              /* Update vectorial force */
 984             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 985             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 986             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 987
 988             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 989             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 990             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 991
 992             fjptrA             = f+j_coord_offsetA;
 993             fjptrB             = f+j_coord_offsetB;
 994             fjptrC             = f+j_coord_offsetC;
 995             fjptrD             = f+j_coord_offsetD;
 996
 997             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 998
 999             /* Inner loop uses 110 flops */
1000         }
1001
1002         if(jidx<j_index_end)
1003         {
1004
1005             /* Get j neighbor index, and coordinate index */
1006             jnrlistA         = jjnr[jidx];
1007             jnrlistB         = jjnr[jidx+1];
1008             jnrlistC         = jjnr[jidx+2];
1009             jnrlistD         = jjnr[jidx+3];
1010             /* Sign of each element will be negative for non-real atoms.
1011              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1012              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1013              */
1014             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1015             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1016             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1017             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1018             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1019             j_coord_offsetA  = DIM*jnrA;
1020             j_coord_offsetB  = DIM*jnrB;
1021             j_coord_offsetC  = DIM*jnrC;
1022             j_coord_offsetD  = DIM*jnrD;
1023
1024             /* load j atom coordinates */
1025             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1026                                               x+j_coord_offsetC,x+j_coord_offsetD,
1027                                               &jx0,&jy0,&jz0);
1028
1029             /* Calculate displacement vector */
1030             dx00             = _mm_sub_ps(ix0,jx0);
1031             dy00             = _mm_sub_ps(iy0,jy0);
1032             dz00             = _mm_sub_ps(iz0,jz0);
1033             dx10             = _mm_sub_ps(ix1,jx0);
1034             dy10             = _mm_sub_ps(iy1,jy0);
1035             dz10             = _mm_sub_ps(iz1,jz0);
1036             dx20             = _mm_sub_ps(ix2,jx0);
1037             dy20             = _mm_sub_ps(iy2,jy0);
1038             dz20             = _mm_sub_ps(iz2,jz0);
1039
1040             /* Calculate squared distance and things based on it */
1041             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1042             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1043             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1044
1045             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1046             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1047             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1048
1049             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1050             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1051             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1052
1053             /* Load parameters for j particles */
1054             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1055                                                               charge+jnrC+0,charge+jnrD+0);
1056             vdwjidx0A        = 2*vdwtype[jnrA+0];
1057             vdwjidx0B        = 2*vdwtype[jnrB+0];
1058             vdwjidx0C        = 2*vdwtype[jnrC+0];
1059             vdwjidx0D        = 2*vdwtype[jnrD+0];
1060
1061             fjx0             = _mm_setzero_ps();
1062             fjy0             = _mm_setzero_ps();
1063             fjz0             = _mm_setzero_ps();
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             r00              = _mm_mul_ps(rsq00,rinv00);
1070             r00              = _mm_andnot_ps(dummy_mask,r00);
1071
1072             /* Compute parameters for interactions between i and j atoms */
1073             qq00             = _mm_mul_ps(iq0,jq0);
1074             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1075                                          vdwparam+vdwioffset0+vdwjidx0B,
1076                                          vdwparam+vdwioffset0+vdwjidx0C,
1077                                          vdwparam+vdwioffset0+vdwjidx0D,
1078                                          &c6_00,&c12_00);
1079
1080             /* Calculate table index by multiplying r with table scale and truncate to integer */
1081             rt               = _mm_mul_ps(r00,vftabscale);
1082             vfitab           = _mm_cvttps_epi32(rt);
1083 #ifdef __XOP__
1084             vfeps            = _mm_frcz_ps(rt);
1085 #else
1086             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1087 #endif
1088             twovfeps         = _mm_add_ps(vfeps,vfeps);
1089             vfitab           = _mm_slli_epi32(vfitab,3);
1090
1091             /* EWALD ELECTROSTATICS */
1092
1093             /* Analytical PME correction */
1094             zeta2            = _mm_mul_ps(beta2,rsq00);
1095             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1096             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1097             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1098             felec            = _mm_mul_ps(qq00,felec);
1099
1100             /* CUBIC SPLINE TABLE DISPERSION */
1101             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1102             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1103             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1104             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1105             _MM_TRANSPOSE4_PS(Y,F,G,H);
1106             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1107             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1108             fvdw6            = _mm_mul_ps(c6_00,FF);
1109
1110             /* CUBIC SPLINE TABLE REPULSION */
1111             vfitab           = _mm_add_epi32(vfitab,ifour);
1112             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1113             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1114             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1115             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1116             _MM_TRANSPOSE4_PS(Y,F,G,H);
1117             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1118             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1119             fvdw12           = _mm_mul_ps(c12_00,FF);
1120             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1121
1122             fscal            = _mm_add_ps(felec,fvdw);
1123
1124             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1125
1126              /* Update vectorial force */
1127             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1128             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1129             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1130
1131             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1132             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1133             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1134
1135             /**************************
1136              * CALCULATE INTERACTIONS *
1137              **************************/
1138
1139             r10              = _mm_mul_ps(rsq10,rinv10);
1140             r10              = _mm_andnot_ps(dummy_mask,r10);
1141
1142             /* Compute parameters for interactions between i and j atoms */
1143             qq10             = _mm_mul_ps(iq1,jq0);
1144
1145             /* EWALD ELECTROSTATICS */
1146
1147             /* Analytical PME correction */
1148             zeta2            = _mm_mul_ps(beta2,rsq10);
1149             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1150             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1151             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1152             felec            = _mm_mul_ps(qq10,felec);
1153
1154             fscal            = felec;
1155
1156             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1157
1158              /* Update vectorial force */
1159             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1160             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1161             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1162
1163             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1164             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1165             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1166
1167             /**************************
1168              * CALCULATE INTERACTIONS *
1169              **************************/
1170
1171             r20              = _mm_mul_ps(rsq20,rinv20);
1172             r20              = _mm_andnot_ps(dummy_mask,r20);
1173
1174             /* Compute parameters for interactions between i and j atoms */
1175             qq20             = _mm_mul_ps(iq2,jq0);
1176
1177             /* EWALD ELECTROSTATICS */
1178
1179             /* Analytical PME correction */
1180             zeta2            = _mm_mul_ps(beta2,rsq20);
1181             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1182             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1183             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1184             felec            = _mm_mul_ps(qq20,felec);
1185
1186             fscal            = felec;
1187
1188             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1189
1190              /* Update vectorial force */
1191             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1192             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1193             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1194
1195             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1196             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1197             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1198
1199             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1200             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1201             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1202             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1203
1204             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1205
1206             /* Inner loop uses 113 flops */
1207         }
1208
1209         /* End of innermost loop */
1210
1211         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1212                                               f+i_coord_offset,fshift+i_shift_offset);
1213
1214         /* Increment number of inner iterations */
1215         inneriter                  += j_index_end - j_index_start;
1216
1217         /* Outer loop uses 18 flops */
1218     }
1219
1220     /* Increment number of outer iterations */
1221     outeriter        += nri;
1222
1223     /* Update outer/inner flops */
1224
1225     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*113);
1226 }