src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "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_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_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              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128i          vfitab;
 108     __m128i          ifour       = _mm_set1_epi32(4);
 109     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 110     real             *vftab;
 111     __m128           dummy_mask,cutoff_mask;
 112     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 113     __m128           one     = _mm_set1_ps(1.0);
 114     __m128           two     = _mm_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_ps(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 138     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 139     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 174
 175         fix0             = _mm_setzero_ps();
 176         fiy0             = _mm_setzero_ps();
 177         fiz0             = _mm_setzero_ps();
 178         fix1             = _mm_setzero_ps();
 179         fiy1             = _mm_setzero_ps();
 180         fiz1             = _mm_setzero_ps();
 181         fix2             = _mm_setzero_ps();
 182         fiy2             = _mm_setzero_ps();
 183         fiz2             = _mm_setzero_ps();
 184         fix3             = _mm_setzero_ps();
 185         fiy3             = _mm_setzero_ps();
 186         fiz3             = _mm_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm_setzero_ps();
 190         vvdwsum          = _mm_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               x+j_coord_offsetC,x+j_coord_offsetD,
 209                                               &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm_sub_ps(ix0,jx0);
 213             dy00             = _mm_sub_ps(iy0,jy0);
 214             dz00             = _mm_sub_ps(iz0,jz0);
 215             dx10             = _mm_sub_ps(ix1,jx0);
 216             dy10             = _mm_sub_ps(iy1,jy0);
 217             dz10             = _mm_sub_ps(iz1,jz0);
 218             dx20             = _mm_sub_ps(ix2,jx0);
 219             dy20             = _mm_sub_ps(iy2,jy0);
 220             dz20             = _mm_sub_ps(iz2,jz0);
 221             dx30             = _mm_sub_ps(ix3,jx0);
 222             dy30             = _mm_sub_ps(iy3,jy0);
 223             dz30             = _mm_sub_ps(iz3,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 227             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 228             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 229             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 230
 231             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 232             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 233             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 234             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 235
 236             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 237             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 238             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 239
 240             /* Load parameters for j particles */
 241             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 242                                                               charge+jnrC+0,charge+jnrD+0);
 243             vdwjidx0A        = 2*vdwtype[jnrA+0];
 244             vdwjidx0B        = 2*vdwtype[jnrB+0];
 245             vdwjidx0C        = 2*vdwtype[jnrC+0];
 246             vdwjidx0D        = 2*vdwtype[jnrD+0];
 247
 248             fjx0             = _mm_setzero_ps();
 249             fjy0             = _mm_setzero_ps();
 250             fjz0             = _mm_setzero_ps();
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             r00              = _mm_mul_ps(rsq00,rinv00);
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 260                                          vdwparam+vdwioffset0+vdwjidx0B,
 261                                          vdwparam+vdwioffset0+vdwjidx0C,
 262                                          vdwparam+vdwioffset0+vdwjidx0D,
 263                                          &c6_00,&c12_00);
 264
 265             /* Calculate table index by multiplying r with table scale and truncate to integer */
 266             rt               = _mm_mul_ps(r00,vftabscale);
 267             vfitab           = _mm_cvttps_epi32(rt);
 268 #ifdef __XOP__
 269             vfeps            = _mm_frcz_ps(rt);
 270 #else
 271             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 272 #endif
 273             twovfeps         = _mm_add_ps(vfeps,vfeps);
 274             vfitab           = _mm_slli_epi32(vfitab,3);
 275
 276             /* CUBIC SPLINE TABLE DISPERSION */
 277             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 278             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 279             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 280             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 281             _MM_TRANSPOSE4_PS(Y,F,G,H);
 282             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 283             VV               = _mm_macc_ps(vfeps,Fp,Y);
 284             vvdw6            = _mm_mul_ps(c6_00,VV);
 285             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 286             fvdw6            = _mm_mul_ps(c6_00,FF);
 287
 288             /* CUBIC SPLINE TABLE REPULSION */
 289             vfitab           = _mm_add_epi32(vfitab,ifour);
 290             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 291             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 292             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 293             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 294             _MM_TRANSPOSE4_PS(Y,F,G,H);
 295             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 296             VV               = _mm_macc_ps(vfeps,Fp,Y);
 297             vvdw12           = _mm_mul_ps(c12_00,VV);
 298             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 299             fvdw12           = _mm_mul_ps(c12_00,FF);
 300             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 301             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 305
 306             fscal            = fvdw;
 307
 308              /* Update vectorial force */
 309             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 310             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 311             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 312
 313             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 314             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 315             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq10             = _mm_mul_ps(iq1,jq0);
 323
 324             /* COULOMB ELECTROSTATICS */
 325             velec            = _mm_mul_ps(qq10,rinv10);
 326             felec            = _mm_mul_ps(velec,rinvsq10);
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm_add_ps(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333              /* Update vectorial force */
 334             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 335             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 336             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 337
 338             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 339             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 340             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             /* Compute parameters for interactions between i and j atoms */
 347             qq20             = _mm_mul_ps(iq2,jq0);
 348
 349             /* COULOMB ELECTROSTATICS */
 350             velec            = _mm_mul_ps(qq20,rinv20);
 351             felec            = _mm_mul_ps(velec,rinvsq20);
 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             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 360             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 361             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 362
 363             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 364             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 365             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq30             = _mm_mul_ps(iq3,jq0);
 373
 374             /* COULOMB ELECTROSTATICS */
 375             velec            = _mm_mul_ps(qq30,rinv30);
 376             felec            = _mm_mul_ps(velec,rinvsq30);
 377
 378             /* Update potential sum for this i atom from the interaction with this j atom. */
 379             velecsum         = _mm_add_ps(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383              /* Update vectorial force */
 384             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 385             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 386             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 387
 388             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 389             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 390             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 391
 392             fjptrA             = f+j_coord_offsetA;
 393             fjptrB             = f+j_coord_offsetB;
 394             fjptrC             = f+j_coord_offsetC;
 395             fjptrD             = f+j_coord_offsetD;
 396
 397             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 398
 399             /* Inner loop uses 152 flops */
 400         }
 401
 402         if(jidx<j_index_end)
 403         {
 404
 405             /* Get j neighbor index, and coordinate index */
 406             jnrlistA         = jjnr[jidx];
 407             jnrlistB         = jjnr[jidx+1];
 408             jnrlistC         = jjnr[jidx+2];
 409             jnrlistD         = jjnr[jidx+3];
 410             /* Sign of each element will be negative for non-real atoms.
 411              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 412              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 413              */
 414             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 415             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 416             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 417             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 418             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 419             j_coord_offsetA  = DIM*jnrA;
 420             j_coord_offsetB  = DIM*jnrB;
 421             j_coord_offsetC  = DIM*jnrC;
 422             j_coord_offsetD  = DIM*jnrD;
 423
 424             /* load j atom coordinates */
 425             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 426                                               x+j_coord_offsetC,x+j_coord_offsetD,
 427                                               &jx0,&jy0,&jz0);
 428
 429             /* Calculate displacement vector */
 430             dx00             = _mm_sub_ps(ix0,jx0);
 431             dy00             = _mm_sub_ps(iy0,jy0);
 432             dz00             = _mm_sub_ps(iz0,jz0);
 433             dx10             = _mm_sub_ps(ix1,jx0);
 434             dy10             = _mm_sub_ps(iy1,jy0);
 435             dz10             = _mm_sub_ps(iz1,jz0);
 436             dx20             = _mm_sub_ps(ix2,jx0);
 437             dy20             = _mm_sub_ps(iy2,jy0);
 438             dz20             = _mm_sub_ps(iz2,jz0);
 439             dx30             = _mm_sub_ps(ix3,jx0);
 440             dy30             = _mm_sub_ps(iy3,jy0);
 441             dz30             = _mm_sub_ps(iz3,jz0);
 442
 443             /* Calculate squared distance and things based on it */
 444             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 445             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 446             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 447             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 448
 449             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 450             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 451             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 452             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 453
 454             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 455             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 456             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 457
 458             /* Load parameters for j particles */
 459             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 460                                                               charge+jnrC+0,charge+jnrD+0);
 461             vdwjidx0A        = 2*vdwtype[jnrA+0];
 462             vdwjidx0B        = 2*vdwtype[jnrB+0];
 463             vdwjidx0C        = 2*vdwtype[jnrC+0];
 464             vdwjidx0D        = 2*vdwtype[jnrD+0];
 465
 466             fjx0             = _mm_setzero_ps();
 467             fjy0             = _mm_setzero_ps();
 468             fjz0             = _mm_setzero_ps();
 469
 470             /**************************
 471              * CALCULATE INTERACTIONS *
 472              **************************/
 473
 474             r00              = _mm_mul_ps(rsq00,rinv00);
 475             r00              = _mm_andnot_ps(dummy_mask,r00);
 476
 477             /* Compute parameters for interactions between i and j atoms */
 478             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 479                                          vdwparam+vdwioffset0+vdwjidx0B,
 480                                          vdwparam+vdwioffset0+vdwjidx0C,
 481                                          vdwparam+vdwioffset0+vdwjidx0D,
 482                                          &c6_00,&c12_00);
 483
 484             /* Calculate table index by multiplying r with table scale and truncate to integer */
 485             rt               = _mm_mul_ps(r00,vftabscale);
 486             vfitab           = _mm_cvttps_epi32(rt);
 487 #ifdef __XOP__
 488             vfeps            = _mm_frcz_ps(rt);
 489 #else
 490             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 491 #endif
 492             twovfeps         = _mm_add_ps(vfeps,vfeps);
 493             vfitab           = _mm_slli_epi32(vfitab,3);
 494
 495             /* CUBIC SPLINE TABLE DISPERSION */
 496             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 497             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 498             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 499             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 500             _MM_TRANSPOSE4_PS(Y,F,G,H);
 501             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 502             VV               = _mm_macc_ps(vfeps,Fp,Y);
 503             vvdw6            = _mm_mul_ps(c6_00,VV);
 504             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 505             fvdw6            = _mm_mul_ps(c6_00,FF);
 506
 507             /* CUBIC SPLINE TABLE REPULSION */
 508             vfitab           = _mm_add_epi32(vfitab,ifour);
 509             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 510             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 511             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 512             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 513             _MM_TRANSPOSE4_PS(Y,F,G,H);
 514             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 515             VV               = _mm_macc_ps(vfeps,Fp,Y);
 516             vvdw12           = _mm_mul_ps(c12_00,VV);
 517             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 518             fvdw12           = _mm_mul_ps(c12_00,FF);
 519             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 520             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 524             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 525
 526             fscal            = fvdw;
 527
 528             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 529
 530              /* Update vectorial force */
 531             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 532             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 533             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 534
 535             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 536             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 537             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 538
 539             /**************************
 540              * CALCULATE INTERACTIONS *
 541              **************************/
 542
 543             /* Compute parameters for interactions between i and j atoms */
 544             qq10             = _mm_mul_ps(iq1,jq0);
 545
 546             /* COULOMB ELECTROSTATICS */
 547             velec            = _mm_mul_ps(qq10,rinv10);
 548             felec            = _mm_mul_ps(velec,rinvsq10);
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             velec            = _mm_andnot_ps(dummy_mask,velec);
 552             velecsum         = _mm_add_ps(velecsum,velec);
 553
 554             fscal            = felec;
 555
 556             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 557
 558              /* Update vectorial force */
 559             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 560             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 561             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 562
 563             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 564             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 565             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 566
 567             /**************************
 568              * CALCULATE INTERACTIONS *
 569              **************************/
 570
 571             /* Compute parameters for interactions between i and j atoms */
 572             qq20             = _mm_mul_ps(iq2,jq0);
 573
 574             /* COULOMB ELECTROSTATICS */
 575             velec            = _mm_mul_ps(qq20,rinv20);
 576             felec            = _mm_mul_ps(velec,rinvsq20);
 577
 578             /* Update potential sum for this i atom from the interaction with this j atom. */
 579             velec            = _mm_andnot_ps(dummy_mask,velec);
 580             velecsum         = _mm_add_ps(velecsum,velec);
 581
 582             fscal            = felec;
 583
 584             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 585
 586              /* Update vectorial force */
 587             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 588             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 589             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 590
 591             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 592             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 593             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 594
 595             /**************************
 596              * CALCULATE INTERACTIONS *
 597              **************************/
 598
 599             /* Compute parameters for interactions between i and j atoms */
 600             qq30             = _mm_mul_ps(iq3,jq0);
 601
 602             /* COULOMB ELECTROSTATICS */
 603             velec            = _mm_mul_ps(qq30,rinv30);
 604             felec            = _mm_mul_ps(velec,rinvsq30);
 605
 606             /* Update potential sum for this i atom from the interaction with this j atom. */
 607             velec            = _mm_andnot_ps(dummy_mask,velec);
 608             velecsum         = _mm_add_ps(velecsum,velec);
 609
 610             fscal            = felec;
 611
 612             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 613
 614              /* Update vectorial force */
 615             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 616             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 617             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 618
 619             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 620             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 621             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 622
 623             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 624             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 625             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 626             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 627
 628             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 629
 630             /* Inner loop uses 153 flops */
 631         }
 632
 633         /* End of innermost loop */
 634
 635         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 636                                               f+i_coord_offset,fshift+i_shift_offset);
 637
 638         ggid                        = gid[iidx];
 639         /* Update potential energies */
 640         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 641         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 642
 643         /* Increment number of inner iterations */
 644         inneriter                  += j_index_end - j_index_start;
 645
 646         /* Outer loop uses 26 flops */
 647     }
 648
 649     /* Increment number of outer iterations */
 650     outeriter        += nri;
 651
 652     /* Update outer/inner flops */
 653
 654     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*153);
 655 }
 656 /*
 657  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 658  * Electrostatics interaction: Coulomb
 659  * VdW interaction:            CubicSplineTable
 660  * Geometry:                   Water4-Particle
 661  * Calculate force/pot:        Force
 662  */
 663 void
 664 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 665                     (t_nblist                    * gmx_restrict       nlist,
 666                      rvec                        * gmx_restrict          xx,
 667                      rvec                        * gmx_restrict          ff,
 668                      t_forcerec                  * gmx_restrict          fr,
 669                      t_mdatoms                   * gmx_restrict     mdatoms,
 670                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 671                      t_nrnb                      * gmx_restrict        nrnb)
 672 {
 673     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 674      * just 0 for non-waters.
 675      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 676      * jnr indices corresponding to data put in the four positions in the SIMD register.
 677      */
 678     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 679     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 680     int              jnrA,jnrB,jnrC,jnrD;
 681     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 682     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 683     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 684     real             rcutoff_scalar;
 685     real             *shiftvec,*fshift,*x,*f;
 686     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 687     real             scratch[4*DIM];
 688     __m128           fscal,rcutoff,rcutoff2,jidxall;
 689     int              vdwioffset0;
 690     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 691     int              vdwioffset1;
 692     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 693     int              vdwioffset2;
 694     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 695     int              vdwioffset3;
 696     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 697     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 698     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 699     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 700     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 701     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 702     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 703     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 704     real             *charge;
 705     int              nvdwtype;
 706     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 707     int              *vdwtype;
 708     real             *vdwparam;
 709     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 710     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 711     __m128i          vfitab;
 712     __m128i          ifour       = _mm_set1_epi32(4);
 713     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 714     real             *vftab;
 715     __m128           dummy_mask,cutoff_mask;
 716     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 717     __m128           one     = _mm_set1_ps(1.0);
 718     __m128           two     = _mm_set1_ps(2.0);
 719     x                = xx[0];
 720     f                = ff[0];
 721
 722     nri              = nlist->nri;
 723     iinr             = nlist->iinr;
 724     jindex           = nlist->jindex;
 725     jjnr             = nlist->jjnr;
 726     shiftidx         = nlist->shift;
 727     gid              = nlist->gid;
 728     shiftvec         = fr->shift_vec[0];
 729     fshift           = fr->fshift[0];
 730     facel            = _mm_set1_ps(fr->epsfac);
 731     charge           = mdatoms->chargeA;
 732     nvdwtype         = fr->ntype;
 733     vdwparam         = fr->nbfp;
 734     vdwtype          = mdatoms->typeA;
 735
 736     vftab            = kernel_data->table_vdw->data;
 737     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 738
 739     /* Setup water-specific parameters */
 740     inr              = nlist->iinr[0];
 741     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 742     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 743     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 744     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 745
 746     /* Avoid stupid compiler warnings */
 747     jnrA = jnrB = jnrC = jnrD = 0;
 748     j_coord_offsetA = 0;
 749     j_coord_offsetB = 0;
 750     j_coord_offsetC = 0;
 751     j_coord_offsetD = 0;
 752
 753     outeriter        = 0;
 754     inneriter        = 0;
 755
 756     for(iidx=0;iidx<4*DIM;iidx++)
 757     {
 758         scratch[iidx] = 0.0;
 759     }
 760
 761     /* Start outer loop over neighborlists */
 762     for(iidx=0; iidx<nri; iidx++)
 763     {
 764         /* Load shift vector for this list */
 765         i_shift_offset   = DIM*shiftidx[iidx];
 766
 767         /* Load limits for loop over neighbors */
 768         j_index_start    = jindex[iidx];
 769         j_index_end      = jindex[iidx+1];
 770
 771         /* Get outer coordinate index */
 772         inr              = iinr[iidx];
 773         i_coord_offset   = DIM*inr;
 774
 775         /* Load i particle coords and add shift vector */
 776         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 777                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 778
 779         fix0             = _mm_setzero_ps();
 780         fiy0             = _mm_setzero_ps();
 781         fiz0             = _mm_setzero_ps();
 782         fix1             = _mm_setzero_ps();
 783         fiy1             = _mm_setzero_ps();
 784         fiz1             = _mm_setzero_ps();
 785         fix2             = _mm_setzero_ps();
 786         fiy2             = _mm_setzero_ps();
 787         fiz2             = _mm_setzero_ps();
 788         fix3             = _mm_setzero_ps();
 789         fiy3             = _mm_setzero_ps();
 790         fiz3             = _mm_setzero_ps();
 791
 792         /* Start inner kernel loop */
 793         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 794         {
 795
 796             /* Get j neighbor index, and coordinate index */
 797             jnrA             = jjnr[jidx];
 798             jnrB             = jjnr[jidx+1];
 799             jnrC             = jjnr[jidx+2];
 800             jnrD             = jjnr[jidx+3];
 801             j_coord_offsetA  = DIM*jnrA;
 802             j_coord_offsetB  = DIM*jnrB;
 803             j_coord_offsetC  = DIM*jnrC;
 804             j_coord_offsetD  = DIM*jnrD;
 805
 806             /* load j atom coordinates */
 807             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 808                                               x+j_coord_offsetC,x+j_coord_offsetD,
 809                                               &jx0,&jy0,&jz0);
 810
 811             /* Calculate displacement vector */
 812             dx00             = _mm_sub_ps(ix0,jx0);
 813             dy00             = _mm_sub_ps(iy0,jy0);
 814             dz00             = _mm_sub_ps(iz0,jz0);
 815             dx10             = _mm_sub_ps(ix1,jx0);
 816             dy10             = _mm_sub_ps(iy1,jy0);
 817             dz10             = _mm_sub_ps(iz1,jz0);
 818             dx20             = _mm_sub_ps(ix2,jx0);
 819             dy20             = _mm_sub_ps(iy2,jy0);
 820             dz20             = _mm_sub_ps(iz2,jz0);
 821             dx30             = _mm_sub_ps(ix3,jx0);
 822             dy30             = _mm_sub_ps(iy3,jy0);
 823             dz30             = _mm_sub_ps(iz3,jz0);
 824
 825             /* Calculate squared distance and things based on it */
 826             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 827             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 828             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 829             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 830
 831             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 832             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 833             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 834             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 835
 836             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 837             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 838             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 839
 840             /* Load parameters for j particles */
 841             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 842                                                               charge+jnrC+0,charge+jnrD+0);
 843             vdwjidx0A        = 2*vdwtype[jnrA+0];
 844             vdwjidx0B        = 2*vdwtype[jnrB+0];
 845             vdwjidx0C        = 2*vdwtype[jnrC+0];
 846             vdwjidx0D        = 2*vdwtype[jnrD+0];
 847
 848             fjx0             = _mm_setzero_ps();
 849             fjy0             = _mm_setzero_ps();
 850             fjz0             = _mm_setzero_ps();
 851
 852             /**************************
 853              * CALCULATE INTERACTIONS *
 854              **************************/
 855
 856             r00              = _mm_mul_ps(rsq00,rinv00);
 857
 858             /* Compute parameters for interactions between i and j atoms */
 859             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 860                                          vdwparam+vdwioffset0+vdwjidx0B,
 861                                          vdwparam+vdwioffset0+vdwjidx0C,
 862                                          vdwparam+vdwioffset0+vdwjidx0D,
 863                                          &c6_00,&c12_00);
 864
 865             /* Calculate table index by multiplying r with table scale and truncate to integer */
 866             rt               = _mm_mul_ps(r00,vftabscale);
 867             vfitab           = _mm_cvttps_epi32(rt);
 868 #ifdef __XOP__
 869             vfeps            = _mm_frcz_ps(rt);
 870 #else
 871             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 872 #endif
 873             twovfeps         = _mm_add_ps(vfeps,vfeps);
 874             vfitab           = _mm_slli_epi32(vfitab,3);
 875
 876             /* CUBIC SPLINE TABLE DISPERSION */
 877             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 878             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 879             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 880             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 881             _MM_TRANSPOSE4_PS(Y,F,G,H);
 882             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 883             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 884             fvdw6            = _mm_mul_ps(c6_00,FF);
 885
 886             /* CUBIC SPLINE TABLE REPULSION */
 887             vfitab           = _mm_add_epi32(vfitab,ifour);
 888             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 889             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 890             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 891             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 892             _MM_TRANSPOSE4_PS(Y,F,G,H);
 893             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 894             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 895             fvdw12           = _mm_mul_ps(c12_00,FF);
 896             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 897
 898             fscal            = fvdw;
 899
 900              /* Update vectorial force */
 901             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 902             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 903             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 904
 905             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 906             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 907             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 908
 909             /**************************
 910              * CALCULATE INTERACTIONS *
 911              **************************/
 912
 913             /* Compute parameters for interactions between i and j atoms */
 914             qq10             = _mm_mul_ps(iq1,jq0);
 915
 916             /* COULOMB ELECTROSTATICS */
 917             velec            = _mm_mul_ps(qq10,rinv10);
 918             felec            = _mm_mul_ps(velec,rinvsq10);
 919
 920             fscal            = felec;
 921
 922              /* Update vectorial force */
 923             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 924             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 925             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 926
 927             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 928             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 929             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 930
 931             /**************************
 932              * CALCULATE INTERACTIONS *
 933              **************************/
 934
 935             /* Compute parameters for interactions between i and j atoms */
 936             qq20             = _mm_mul_ps(iq2,jq0);
 937
 938             /* COULOMB ELECTROSTATICS */
 939             velec            = _mm_mul_ps(qq20,rinv20);
 940             felec            = _mm_mul_ps(velec,rinvsq20);
 941
 942             fscal            = felec;
 943
 944              /* Update vectorial force */
 945             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 946             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 947             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 948
 949             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 950             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 951             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 952
 953             /**************************
 954              * CALCULATE INTERACTIONS *
 955              **************************/
 956
 957             /* Compute parameters for interactions between i and j atoms */
 958             qq30             = _mm_mul_ps(iq3,jq0);
 959
 960             /* COULOMB ELECTROSTATICS */
 961             velec            = _mm_mul_ps(qq30,rinv30);
 962             felec            = _mm_mul_ps(velec,rinvsq30);
 963
 964             fscal            = felec;
 965
 966              /* Update vectorial force */
 967             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 968             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 969             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 970
 971             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 972             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 973             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 974
 975             fjptrA             = f+j_coord_offsetA;
 976             fjptrB             = f+j_coord_offsetB;
 977             fjptrC             = f+j_coord_offsetC;
 978             fjptrD             = f+j_coord_offsetD;
 979
 980             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 981
 982             /* Inner loop uses 141 flops */
 983         }
 984
 985         if(jidx<j_index_end)
 986         {
 987
 988             /* Get j neighbor index, and coordinate index */
 989             jnrlistA         = jjnr[jidx];
 990             jnrlistB         = jjnr[jidx+1];
 991             jnrlistC         = jjnr[jidx+2];
 992             jnrlistD         = jjnr[jidx+3];
 993             /* Sign of each element will be negative for non-real atoms.
 994              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 995              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 996              */
 997             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 998             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 999             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1000             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1001             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1002             j_coord_offsetA  = DIM*jnrA;
1003             j_coord_offsetB  = DIM*jnrB;
1004             j_coord_offsetC  = DIM*jnrC;
1005             j_coord_offsetD  = DIM*jnrD;
1006
1007             /* load j atom coordinates */
1008             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1009                                               x+j_coord_offsetC,x+j_coord_offsetD,
1010                                               &jx0,&jy0,&jz0);
1011
1012             /* Calculate displacement vector */
1013             dx00             = _mm_sub_ps(ix0,jx0);
1014             dy00             = _mm_sub_ps(iy0,jy0);
1015             dz00             = _mm_sub_ps(iz0,jz0);
1016             dx10             = _mm_sub_ps(ix1,jx0);
1017             dy10             = _mm_sub_ps(iy1,jy0);
1018             dz10             = _mm_sub_ps(iz1,jz0);
1019             dx20             = _mm_sub_ps(ix2,jx0);
1020             dy20             = _mm_sub_ps(iy2,jy0);
1021             dz20             = _mm_sub_ps(iz2,jz0);
1022             dx30             = _mm_sub_ps(ix3,jx0);
1023             dy30             = _mm_sub_ps(iy3,jy0);
1024             dz30             = _mm_sub_ps(iz3,jz0);
1025
1026             /* Calculate squared distance and things based on it */
1027             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1028             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1029             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1030             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1031
1032             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1033             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1034             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1035             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1036
1037             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1038             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1039             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1040
1041             /* Load parameters for j particles */
1042             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1043                                                               charge+jnrC+0,charge+jnrD+0);
1044             vdwjidx0A        = 2*vdwtype[jnrA+0];
1045             vdwjidx0B        = 2*vdwtype[jnrB+0];
1046             vdwjidx0C        = 2*vdwtype[jnrC+0];
1047             vdwjidx0D        = 2*vdwtype[jnrD+0];
1048
1049             fjx0             = _mm_setzero_ps();
1050             fjy0             = _mm_setzero_ps();
1051             fjz0             = _mm_setzero_ps();
1052
1053             /**************************
1054              * CALCULATE INTERACTIONS *
1055              **************************/
1056
1057             r00              = _mm_mul_ps(rsq00,rinv00);
1058             r00              = _mm_andnot_ps(dummy_mask,r00);
1059
1060             /* Compute parameters for interactions between i and j atoms */
1061             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1062                                          vdwparam+vdwioffset0+vdwjidx0B,
1063                                          vdwparam+vdwioffset0+vdwjidx0C,
1064                                          vdwparam+vdwioffset0+vdwjidx0D,
1065                                          &c6_00,&c12_00);
1066
1067             /* Calculate table index by multiplying r with table scale and truncate to integer */
1068             rt               = _mm_mul_ps(r00,vftabscale);
1069             vfitab           = _mm_cvttps_epi32(rt);
1070 #ifdef __XOP__
1071             vfeps            = _mm_frcz_ps(rt);
1072 #else
1073             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1074 #endif
1075             twovfeps         = _mm_add_ps(vfeps,vfeps);
1076             vfitab           = _mm_slli_epi32(vfitab,3);
1077
1078             /* CUBIC SPLINE TABLE DISPERSION */
1079             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1080             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1081             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1082             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1083             _MM_TRANSPOSE4_PS(Y,F,G,H);
1084             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1085             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1086             fvdw6            = _mm_mul_ps(c6_00,FF);
1087
1088             /* CUBIC SPLINE TABLE REPULSION */
1089             vfitab           = _mm_add_epi32(vfitab,ifour);
1090             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1091             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1092             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1093             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1094             _MM_TRANSPOSE4_PS(Y,F,G,H);
1095             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1096             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1097             fvdw12           = _mm_mul_ps(c12_00,FF);
1098             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1099
1100             fscal            = fvdw;
1101
1102             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1103
1104              /* Update vectorial force */
1105             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1106             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1107             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1108
1109             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1110             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1111             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1112
1113             /**************************
1114              * CALCULATE INTERACTIONS *
1115              **************************/
1116
1117             /* Compute parameters for interactions between i and j atoms */
1118             qq10             = _mm_mul_ps(iq1,jq0);
1119
1120             /* COULOMB ELECTROSTATICS */
1121             velec            = _mm_mul_ps(qq10,rinv10);
1122             felec            = _mm_mul_ps(velec,rinvsq10);
1123
1124             fscal            = felec;
1125
1126             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1127
1128              /* Update vectorial force */
1129             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1130             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1131             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1132
1133             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1134             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1135             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1136
1137             /**************************
1138              * CALCULATE INTERACTIONS *
1139              **************************/
1140
1141             /* Compute parameters for interactions between i and j atoms */
1142             qq20             = _mm_mul_ps(iq2,jq0);
1143
1144             /* COULOMB ELECTROSTATICS */
1145             velec            = _mm_mul_ps(qq20,rinv20);
1146             felec            = _mm_mul_ps(velec,rinvsq20);
1147
1148             fscal            = felec;
1149
1150             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1151
1152              /* Update vectorial force */
1153             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1154             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1155             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1156
1157             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1158             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1159             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1160
1161             /**************************
1162              * CALCULATE INTERACTIONS *
1163              **************************/
1164
1165             /* Compute parameters for interactions between i and j atoms */
1166             qq30             = _mm_mul_ps(iq3,jq0);
1167
1168             /* COULOMB ELECTROSTATICS */
1169             velec            = _mm_mul_ps(qq30,rinv30);
1170             felec            = _mm_mul_ps(velec,rinvsq30);
1171
1172             fscal            = felec;
1173
1174             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1175
1176              /* Update vectorial force */
1177             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1178             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1179             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1180
1181             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1182             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1183             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1184
1185             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1186             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1187             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1188             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1189
1190             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1191
1192             /* Inner loop uses 142 flops */
1193         }
1194
1195         /* End of innermost loop */
1196
1197         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1198                                               f+i_coord_offset,fshift+i_shift_offset);
1199
1200         /* Increment number of inner iterations */
1201         inneriter                  += j_index_end - j_index_start;
1202
1203         /* Outer loop uses 24 flops */
1204     }
1205
1206     /* Increment number of outer iterations */
1207     outeriter        += nri;
1208
1209     /* Update outer/inner flops */
1210
1211     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*142);
1212 }