src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_VdwCSTab_GeomW4P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  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           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128i          vfitab;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 108     real             *vftab;
 109     __m128           dummy_mask,cutoff_mask;
 110     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 111     __m128           one     = _mm_set1_ps(1.0);
 112     __m128           two     = _mm_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_ps(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     krf              = _mm_set1_ps(fr->ic->k_rf);
 127     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 128     crf              = _mm_set1_ps(fr->ic->c_rf);
 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     /* Setup water-specific parameters */
 137     inr              = nlist->iinr[0];
 138     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 139     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 140     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 141     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = jnrC = jnrD = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147     j_coord_offsetC = 0;
 148     j_coord_offsetD = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     for(iidx=0;iidx<4*DIM;iidx++)
 154     {
 155         scratch[iidx] = 0.0;
 156     }
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 175
 176         fix0             = _mm_setzero_ps();
 177         fiy0             = _mm_setzero_ps();
 178         fiz0             = _mm_setzero_ps();
 179         fix1             = _mm_setzero_ps();
 180         fiy1             = _mm_setzero_ps();
 181         fiz1             = _mm_setzero_ps();
 182         fix2             = _mm_setzero_ps();
 183         fiy2             = _mm_setzero_ps();
 184         fiz2             = _mm_setzero_ps();
 185         fix3             = _mm_setzero_ps();
 186         fiy3             = _mm_setzero_ps();
 187         fiz3             = _mm_setzero_ps();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm_setzero_ps();
 191         vvdwsum          = _mm_setzero_ps();
 192
 193         /* Start inner kernel loop */
 194         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 195         {
 196
 197             /* Get j neighbor index, and coordinate index */
 198             jnrA             = jjnr[jidx];
 199             jnrB             = jjnr[jidx+1];
 200             jnrC             = jjnr[jidx+2];
 201             jnrD             = jjnr[jidx+3];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204             j_coord_offsetC  = DIM*jnrC;
 205             j_coord_offsetD  = DIM*jnrD;
 206
 207             /* load j atom coordinates */
 208             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 209                                               x+j_coord_offsetC,x+j_coord_offsetD,
 210                                               &jx0,&jy0,&jz0);
 211
 212             /* Calculate displacement vector */
 213             dx00             = _mm_sub_ps(ix0,jx0);
 214             dy00             = _mm_sub_ps(iy0,jy0);
 215             dz00             = _mm_sub_ps(iz0,jz0);
 216             dx10             = _mm_sub_ps(ix1,jx0);
 217             dy10             = _mm_sub_ps(iy1,jy0);
 218             dz10             = _mm_sub_ps(iz1,jz0);
 219             dx20             = _mm_sub_ps(ix2,jx0);
 220             dy20             = _mm_sub_ps(iy2,jy0);
 221             dz20             = _mm_sub_ps(iz2,jz0);
 222             dx30             = _mm_sub_ps(ix3,jx0);
 223             dy30             = _mm_sub_ps(iy3,jy0);
 224             dz30             = _mm_sub_ps(iz3,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 228             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 229             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 230             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 231
 232             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 233             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 234             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 235             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 236
 237             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 238             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 239             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 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             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 261                                          vdwparam+vdwioffset0+vdwjidx0B,
 262                                          vdwparam+vdwioffset0+vdwjidx0C,
 263                                          vdwparam+vdwioffset0+vdwjidx0D,
 264                                          &c6_00,&c12_00);
 265
 266             /* Calculate table index by multiplying r with table scale and truncate to integer */
 267             rt               = _mm_mul_ps(r00,vftabscale);
 268             vfitab           = _mm_cvttps_epi32(rt);
 269 #ifdef __XOP__
 270             vfeps            = _mm_frcz_ps(rt);
 271 #else
 272             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 273 #endif
 274             twovfeps         = _mm_add_ps(vfeps,vfeps);
 275             vfitab           = _mm_slli_epi32(vfitab,3);
 276
 277             /* CUBIC SPLINE TABLE DISPERSION */
 278             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 279             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 280             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 281             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 282             _MM_TRANSPOSE4_PS(Y,F,G,H);
 283             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 284             VV               = _mm_macc_ps(vfeps,Fp,Y);
 285             vvdw6            = _mm_mul_ps(c6_00,VV);
 286             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 287             fvdw6            = _mm_mul_ps(c6_00,FF);
 288
 289             /* CUBIC SPLINE TABLE REPULSION */
 290             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm_mul_ps(c12_00,VV);
 299             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 300             fvdw12           = _mm_mul_ps(c12_00,FF);
 301             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 302             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 306
 307             fscal            = fvdw;
 308
 309              /* Update vectorial force */
 310             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 311             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 312             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 313
 314             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 315             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 316             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq10             = _mm_mul_ps(iq1,jq0);
 324
 325             /* REACTION-FIELD ELECTROSTATICS */
 326             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 327             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velecsum         = _mm_add_ps(velecsum,velec);
 331
 332             fscal            = felec;
 333
 334              /* Update vectorial force */
 335             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 336             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 337             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 338
 339             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 340             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 341             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq20             = _mm_mul_ps(iq2,jq0);
 349
 350             /* REACTION-FIELD ELECTROSTATICS */
 351             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 352             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velecsum         = _mm_add_ps(velecsum,velec);
 356
 357             fscal            = felec;
 358
 359              /* Update vectorial force */
 360             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 361             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 362             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 363
 364             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 365             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 366             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             qq30             = _mm_mul_ps(iq3,jq0);
 374
 375             /* REACTION-FIELD ELECTROSTATICS */
 376             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 377             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velecsum         = _mm_add_ps(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384              /* Update vectorial force */
 385             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 386             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 387             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 388
 389             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 390             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 391             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 392
 393             fjptrA             = f+j_coord_offsetA;
 394             fjptrB             = f+j_coord_offsetB;
 395             fjptrC             = f+j_coord_offsetC;
 396             fjptrD             = f+j_coord_offsetD;
 397
 398             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 399
 400             /* Inner loop uses 164 flops */
 401         }
 402
 403         if(jidx<j_index_end)
 404         {
 405
 406             /* Get j neighbor index, and coordinate index */
 407             jnrlistA         = jjnr[jidx];
 408             jnrlistB         = jjnr[jidx+1];
 409             jnrlistC         = jjnr[jidx+2];
 410             jnrlistD         = jjnr[jidx+3];
 411             /* Sign of each element will be negative for non-real atoms.
 412              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 413              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 414              */
 415             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 416             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 417             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 418             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 419             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 420             j_coord_offsetA  = DIM*jnrA;
 421             j_coord_offsetB  = DIM*jnrB;
 422             j_coord_offsetC  = DIM*jnrC;
 423             j_coord_offsetD  = DIM*jnrD;
 424
 425             /* load j atom coordinates */
 426             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 427                                               x+j_coord_offsetC,x+j_coord_offsetD,
 428                                               &jx0,&jy0,&jz0);
 429
 430             /* Calculate displacement vector */
 431             dx00             = _mm_sub_ps(ix0,jx0);
 432             dy00             = _mm_sub_ps(iy0,jy0);
 433             dz00             = _mm_sub_ps(iz0,jz0);
 434             dx10             = _mm_sub_ps(ix1,jx0);
 435             dy10             = _mm_sub_ps(iy1,jy0);
 436             dz10             = _mm_sub_ps(iz1,jz0);
 437             dx20             = _mm_sub_ps(ix2,jx0);
 438             dy20             = _mm_sub_ps(iy2,jy0);
 439             dz20             = _mm_sub_ps(iz2,jz0);
 440             dx30             = _mm_sub_ps(ix3,jx0);
 441             dy30             = _mm_sub_ps(iy3,jy0);
 442             dz30             = _mm_sub_ps(iz3,jz0);
 443
 444             /* Calculate squared distance and things based on it */
 445             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 446             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 447             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 448             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 449
 450             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 451             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 452             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 453             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 454
 455             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 456             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 457             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 458
 459             /* Load parameters for j particles */
 460             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 461                                                               charge+jnrC+0,charge+jnrD+0);
 462             vdwjidx0A        = 2*vdwtype[jnrA+0];
 463             vdwjidx0B        = 2*vdwtype[jnrB+0];
 464             vdwjidx0C        = 2*vdwtype[jnrC+0];
 465             vdwjidx0D        = 2*vdwtype[jnrD+0];
 466
 467             fjx0             = _mm_setzero_ps();
 468             fjy0             = _mm_setzero_ps();
 469             fjz0             = _mm_setzero_ps();
 470
 471             /**************************
 472              * CALCULATE INTERACTIONS *
 473              **************************/
 474
 475             r00              = _mm_mul_ps(rsq00,rinv00);
 476             r00              = _mm_andnot_ps(dummy_mask,r00);
 477
 478             /* Compute parameters for interactions between i and j atoms */
 479             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 480                                          vdwparam+vdwioffset0+vdwjidx0B,
 481                                          vdwparam+vdwioffset0+vdwjidx0C,
 482                                          vdwparam+vdwioffset0+vdwjidx0D,
 483                                          &c6_00,&c12_00);
 484
 485             /* Calculate table index by multiplying r with table scale and truncate to integer */
 486             rt               = _mm_mul_ps(r00,vftabscale);
 487             vfitab           = _mm_cvttps_epi32(rt);
 488 #ifdef __XOP__
 489             vfeps            = _mm_frcz_ps(rt);
 490 #else
 491             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 492 #endif
 493             twovfeps         = _mm_add_ps(vfeps,vfeps);
 494             vfitab           = _mm_slli_epi32(vfitab,3);
 495
 496             /* CUBIC SPLINE TABLE DISPERSION */
 497             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 498             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 499             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 500             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 501             _MM_TRANSPOSE4_PS(Y,F,G,H);
 502             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 503             VV               = _mm_macc_ps(vfeps,Fp,Y);
 504             vvdw6            = _mm_mul_ps(c6_00,VV);
 505             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 506             fvdw6            = _mm_mul_ps(c6_00,FF);
 507
 508             /* CUBIC SPLINE TABLE REPULSION */
 509             vfitab           = _mm_add_epi32(vfitab,ifour);
 510             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 511             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 512             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 513             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 514             _MM_TRANSPOSE4_PS(Y,F,G,H);
 515             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 516             VV               = _mm_macc_ps(vfeps,Fp,Y);
 517             vvdw12           = _mm_mul_ps(c12_00,VV);
 518             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 519             fvdw12           = _mm_mul_ps(c12_00,FF);
 520             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 521             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 525             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 526
 527             fscal            = fvdw;
 528
 529             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 530
 531              /* Update vectorial force */
 532             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 533             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 534             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 535
 536             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 537             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 538             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             /* Compute parameters for interactions between i and j atoms */
 545             qq10             = _mm_mul_ps(iq1,jq0);
 546
 547             /* REACTION-FIELD ELECTROSTATICS */
 548             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 549             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velec            = _mm_andnot_ps(dummy_mask,velec);
 553             velecsum         = _mm_add_ps(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 558
 559              /* Update vectorial force */
 560             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 561             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 562             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 563
 564             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 565             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 566             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq20             = _mm_mul_ps(iq2,jq0);
 574
 575             /* REACTION-FIELD ELECTROSTATICS */
 576             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 577             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             velec            = _mm_andnot_ps(dummy_mask,velec);
 581             velecsum         = _mm_add_ps(velecsum,velec);
 582
 583             fscal            = felec;
 584
 585             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 586
 587              /* Update vectorial force */
 588             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 589             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 590             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 591
 592             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 593             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 594             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             /* Compute parameters for interactions between i and j atoms */
 601             qq30             = _mm_mul_ps(iq3,jq0);
 602
 603             /* REACTION-FIELD ELECTROSTATICS */
 604             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 605             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             velec            = _mm_andnot_ps(dummy_mask,velec);
 609             velecsum         = _mm_add_ps(velecsum,velec);
 610
 611             fscal            = felec;
 612
 613             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 614
 615              /* Update vectorial force */
 616             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 617             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 618             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 619
 620             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 621             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 622             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 623
 624             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 625             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 626             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 627             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 628
 629             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 630
 631             /* Inner loop uses 165 flops */
 632         }
 633
 634         /* End of innermost loop */
 635
 636         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 637                                               f+i_coord_offset,fshift+i_shift_offset);
 638
 639         ggid                        = gid[iidx];
 640         /* Update potential energies */
 641         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 642         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 643
 644         /* Increment number of inner iterations */
 645         inneriter                  += j_index_end - j_index_start;
 646
 647         /* Outer loop uses 26 flops */
 648     }
 649
 650     /* Increment number of outer iterations */
 651     outeriter        += nri;
 652
 653     /* Update outer/inner flops */
 654
 655     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*165);
 656 }
 657 /*
 658  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 659  * Electrostatics interaction: ReactionField
 660  * VdW interaction:            CubicSplineTable
 661  * Geometry:                   Water4-Particle
 662  * Calculate force/pot:        Force
 663  */
 664 void
 665 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 666                     (t_nblist                    * gmx_restrict       nlist,
 667                      rvec                        * gmx_restrict          xx,
 668                      rvec                        * gmx_restrict          ff,
 669                      t_forcerec                  * gmx_restrict          fr,
 670                      t_mdatoms                   * gmx_restrict     mdatoms,
 671                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 672                      t_nrnb                      * gmx_restrict        nrnb)
 673 {
 674     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 675      * just 0 for non-waters.
 676      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 677      * jnr indices corresponding to data put in the four positions in the SIMD register.
 678      */
 679     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 680     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 681     int              jnrA,jnrB,jnrC,jnrD;
 682     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 683     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 684     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 685     real             rcutoff_scalar;
 686     real             *shiftvec,*fshift,*x,*f;
 687     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 688     real             scratch[4*DIM];
 689     __m128           fscal,rcutoff,rcutoff2,jidxall;
 690     int              vdwioffset0;
 691     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 692     int              vdwioffset1;
 693     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 694     int              vdwioffset2;
 695     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 696     int              vdwioffset3;
 697     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 698     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 699     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 700     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 701     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 702     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 703     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 704     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 705     real             *charge;
 706     int              nvdwtype;
 707     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 708     int              *vdwtype;
 709     real             *vdwparam;
 710     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 711     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 712     __m128i          vfitab;
 713     __m128i          ifour       = _mm_set1_epi32(4);
 714     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 715     real             *vftab;
 716     __m128           dummy_mask,cutoff_mask;
 717     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 718     __m128           one     = _mm_set1_ps(1.0);
 719     __m128           two     = _mm_set1_ps(2.0);
 720     x                = xx[0];
 721     f                = ff[0];
 722
 723     nri              = nlist->nri;
 724     iinr             = nlist->iinr;
 725     jindex           = nlist->jindex;
 726     jjnr             = nlist->jjnr;
 727     shiftidx         = nlist->shift;
 728     gid              = nlist->gid;
 729     shiftvec         = fr->shift_vec[0];
 730     fshift           = fr->fshift[0];
 731     facel            = _mm_set1_ps(fr->epsfac);
 732     charge           = mdatoms->chargeA;
 733     krf              = _mm_set1_ps(fr->ic->k_rf);
 734     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 735     crf              = _mm_set1_ps(fr->ic->c_rf);
 736     nvdwtype         = fr->ntype;
 737     vdwparam         = fr->nbfp;
 738     vdwtype          = mdatoms->typeA;
 739
 740     vftab            = kernel_data->table_vdw->data;
 741     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 742
 743     /* Setup water-specific parameters */
 744     inr              = nlist->iinr[0];
 745     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 746     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 747     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 748     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 749
 750     /* Avoid stupid compiler warnings */
 751     jnrA = jnrB = jnrC = jnrD = 0;
 752     j_coord_offsetA = 0;
 753     j_coord_offsetB = 0;
 754     j_coord_offsetC = 0;
 755     j_coord_offsetD = 0;
 756
 757     outeriter        = 0;
 758     inneriter        = 0;
 759
 760     for(iidx=0;iidx<4*DIM;iidx++)
 761     {
 762         scratch[iidx] = 0.0;
 763     }
 764
 765     /* Start outer loop over neighborlists */
 766     for(iidx=0; iidx<nri; iidx++)
 767     {
 768         /* Load shift vector for this list */
 769         i_shift_offset   = DIM*shiftidx[iidx];
 770
 771         /* Load limits for loop over neighbors */
 772         j_index_start    = jindex[iidx];
 773         j_index_end      = jindex[iidx+1];
 774
 775         /* Get outer coordinate index */
 776         inr              = iinr[iidx];
 777         i_coord_offset   = DIM*inr;
 778
 779         /* Load i particle coords and add shift vector */
 780         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 781                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 782
 783         fix0             = _mm_setzero_ps();
 784         fiy0             = _mm_setzero_ps();
 785         fiz0             = _mm_setzero_ps();
 786         fix1             = _mm_setzero_ps();
 787         fiy1             = _mm_setzero_ps();
 788         fiz1             = _mm_setzero_ps();
 789         fix2             = _mm_setzero_ps();
 790         fiy2             = _mm_setzero_ps();
 791         fiz2             = _mm_setzero_ps();
 792         fix3             = _mm_setzero_ps();
 793         fiy3             = _mm_setzero_ps();
 794         fiz3             = _mm_setzero_ps();
 795
 796         /* Start inner kernel loop */
 797         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 798         {
 799
 800             /* Get j neighbor index, and coordinate index */
 801             jnrA             = jjnr[jidx];
 802             jnrB             = jjnr[jidx+1];
 803             jnrC             = jjnr[jidx+2];
 804             jnrD             = jjnr[jidx+3];
 805             j_coord_offsetA  = DIM*jnrA;
 806             j_coord_offsetB  = DIM*jnrB;
 807             j_coord_offsetC  = DIM*jnrC;
 808             j_coord_offsetD  = DIM*jnrD;
 809
 810             /* load j atom coordinates */
 811             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 812                                               x+j_coord_offsetC,x+j_coord_offsetD,
 813                                               &jx0,&jy0,&jz0);
 814
 815             /* Calculate displacement vector */
 816             dx00             = _mm_sub_ps(ix0,jx0);
 817             dy00             = _mm_sub_ps(iy0,jy0);
 818             dz00             = _mm_sub_ps(iz0,jz0);
 819             dx10             = _mm_sub_ps(ix1,jx0);
 820             dy10             = _mm_sub_ps(iy1,jy0);
 821             dz10             = _mm_sub_ps(iz1,jz0);
 822             dx20             = _mm_sub_ps(ix2,jx0);
 823             dy20             = _mm_sub_ps(iy2,jy0);
 824             dz20             = _mm_sub_ps(iz2,jz0);
 825             dx30             = _mm_sub_ps(ix3,jx0);
 826             dy30             = _mm_sub_ps(iy3,jy0);
 827             dz30             = _mm_sub_ps(iz3,jz0);
 828
 829             /* Calculate squared distance and things based on it */
 830             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 831             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 832             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 833             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 834
 835             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 836             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 837             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 838             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 839
 840             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 841             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 842             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 843
 844             /* Load parameters for j particles */
 845             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 846                                                               charge+jnrC+0,charge+jnrD+0);
 847             vdwjidx0A        = 2*vdwtype[jnrA+0];
 848             vdwjidx0B        = 2*vdwtype[jnrB+0];
 849             vdwjidx0C        = 2*vdwtype[jnrC+0];
 850             vdwjidx0D        = 2*vdwtype[jnrD+0];
 851
 852             fjx0             = _mm_setzero_ps();
 853             fjy0             = _mm_setzero_ps();
 854             fjz0             = _mm_setzero_ps();
 855
 856             /**************************
 857              * CALCULATE INTERACTIONS *
 858              **************************/
 859
 860             r00              = _mm_mul_ps(rsq00,rinv00);
 861
 862             /* Compute parameters for interactions between i and j atoms */
 863             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 864                                          vdwparam+vdwioffset0+vdwjidx0B,
 865                                          vdwparam+vdwioffset0+vdwjidx0C,
 866                                          vdwparam+vdwioffset0+vdwjidx0D,
 867                                          &c6_00,&c12_00);
 868
 869             /* Calculate table index by multiplying r with table scale and truncate to integer */
 870             rt               = _mm_mul_ps(r00,vftabscale);
 871             vfitab           = _mm_cvttps_epi32(rt);
 872 #ifdef __XOP__
 873             vfeps            = _mm_frcz_ps(rt);
 874 #else
 875             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 876 #endif
 877             twovfeps         = _mm_add_ps(vfeps,vfeps);
 878             vfitab           = _mm_slli_epi32(vfitab,3);
 879
 880             /* CUBIC SPLINE TABLE DISPERSION */
 881             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 882             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 883             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 884             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 885             _MM_TRANSPOSE4_PS(Y,F,G,H);
 886             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 887             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 888             fvdw6            = _mm_mul_ps(c6_00,FF);
 889
 890             /* CUBIC SPLINE TABLE REPULSION */
 891             vfitab           = _mm_add_epi32(vfitab,ifour);
 892             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 893             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 894             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 895             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 896             _MM_TRANSPOSE4_PS(Y,F,G,H);
 897             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 898             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 899             fvdw12           = _mm_mul_ps(c12_00,FF);
 900             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 901
 902             fscal            = fvdw;
 903
 904              /* Update vectorial force */
 905             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 906             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 907             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 908
 909             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 910             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 911             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             /* Compute parameters for interactions between i and j atoms */
 918             qq10             = _mm_mul_ps(iq1,jq0);
 919
 920             /* REACTION-FIELD ELECTROSTATICS */
 921             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 922
 923             fscal            = felec;
 924
 925              /* Update vectorial force */
 926             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 927             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 928             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 929
 930             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 931             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 932             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 933
 934             /**************************
 935              * CALCULATE INTERACTIONS *
 936              **************************/
 937
 938             /* Compute parameters for interactions between i and j atoms */
 939             qq20             = _mm_mul_ps(iq2,jq0);
 940
 941             /* REACTION-FIELD ELECTROSTATICS */
 942             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 943
 944             fscal            = felec;
 945
 946              /* Update vectorial force */
 947             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 948             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 949             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 950
 951             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 952             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 953             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             /* Compute parameters for interactions between i and j atoms */
 960             qq30             = _mm_mul_ps(iq3,jq0);
 961
 962             /* REACTION-FIELD ELECTROSTATICS */
 963             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 964
 965             fscal            = felec;
 966
 967              /* Update vectorial force */
 968             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 969             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 970             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 971
 972             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 973             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 974             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 975
 976             fjptrA             = f+j_coord_offsetA;
 977             fjptrB             = f+j_coord_offsetB;
 978             fjptrC             = f+j_coord_offsetC;
 979             fjptrD             = f+j_coord_offsetD;
 980
 981             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 982
 983             /* Inner loop uses 141 flops */
 984         }
 985
 986         if(jidx<j_index_end)
 987         {
 988
 989             /* Get j neighbor index, and coordinate index */
 990             jnrlistA         = jjnr[jidx];
 991             jnrlistB         = jjnr[jidx+1];
 992             jnrlistC         = jjnr[jidx+2];
 993             jnrlistD         = jjnr[jidx+3];
 994             /* Sign of each element will be negative for non-real atoms.
 995              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 996              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 997              */
 998             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 999             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1000             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1001             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1002             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1003             j_coord_offsetA  = DIM*jnrA;
1004             j_coord_offsetB  = DIM*jnrB;
1005             j_coord_offsetC  = DIM*jnrC;
1006             j_coord_offsetD  = DIM*jnrD;
1007
1008             /* load j atom coordinates */
1009             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1010                                               x+j_coord_offsetC,x+j_coord_offsetD,
1011                                               &jx0,&jy0,&jz0);
1012
1013             /* Calculate displacement vector */
1014             dx00             = _mm_sub_ps(ix0,jx0);
1015             dy00             = _mm_sub_ps(iy0,jy0);
1016             dz00             = _mm_sub_ps(iz0,jz0);
1017             dx10             = _mm_sub_ps(ix1,jx0);
1018             dy10             = _mm_sub_ps(iy1,jy0);
1019             dz10             = _mm_sub_ps(iz1,jz0);
1020             dx20             = _mm_sub_ps(ix2,jx0);
1021             dy20             = _mm_sub_ps(iy2,jy0);
1022             dz20             = _mm_sub_ps(iz2,jz0);
1023             dx30             = _mm_sub_ps(ix3,jx0);
1024             dy30             = _mm_sub_ps(iy3,jy0);
1025             dz30             = _mm_sub_ps(iz3,jz0);
1026
1027             /* Calculate squared distance and things based on it */
1028             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1029             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1030             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1031             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1032
1033             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1034             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1035             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1036             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1037
1038             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1039             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1040             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1041
1042             /* Load parameters for j particles */
1043             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1044                                                               charge+jnrC+0,charge+jnrD+0);
1045             vdwjidx0A        = 2*vdwtype[jnrA+0];
1046             vdwjidx0B        = 2*vdwtype[jnrB+0];
1047             vdwjidx0C        = 2*vdwtype[jnrC+0];
1048             vdwjidx0D        = 2*vdwtype[jnrD+0];
1049
1050             fjx0             = _mm_setzero_ps();
1051             fjy0             = _mm_setzero_ps();
1052             fjz0             = _mm_setzero_ps();
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             r00              = _mm_mul_ps(rsq00,rinv00);
1059             r00              = _mm_andnot_ps(dummy_mask,r00);
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1063                                          vdwparam+vdwioffset0+vdwjidx0B,
1064                                          vdwparam+vdwioffset0+vdwjidx0C,
1065                                          vdwparam+vdwioffset0+vdwjidx0D,
1066                                          &c6_00,&c12_00);
1067
1068             /* Calculate table index by multiplying r with table scale and truncate to integer */
1069             rt               = _mm_mul_ps(r00,vftabscale);
1070             vfitab           = _mm_cvttps_epi32(rt);
1071 #ifdef __XOP__
1072             vfeps            = _mm_frcz_ps(rt);
1073 #else
1074             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1075 #endif
1076             twovfeps         = _mm_add_ps(vfeps,vfeps);
1077             vfitab           = _mm_slli_epi32(vfitab,3);
1078
1079             /* CUBIC SPLINE TABLE DISPERSION */
1080             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1081             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1082             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1083             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1084             _MM_TRANSPOSE4_PS(Y,F,G,H);
1085             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1086             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1087             fvdw6            = _mm_mul_ps(c6_00,FF);
1088
1089             /* CUBIC SPLINE TABLE REPULSION */
1090             vfitab           = _mm_add_epi32(vfitab,ifour);
1091             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1092             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1093             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1094             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1095             _MM_TRANSPOSE4_PS(Y,F,G,H);
1096             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1097             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1098             fvdw12           = _mm_mul_ps(c12_00,FF);
1099             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1100
1101             fscal            = fvdw;
1102
1103             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1104
1105              /* Update vectorial force */
1106             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1107             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1108             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1109
1110             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1111             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1112             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1113
1114             /**************************
1115              * CALCULATE INTERACTIONS *
1116              **************************/
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq10             = _mm_mul_ps(iq1,jq0);
1120
1121             /* REACTION-FIELD ELECTROSTATICS */
1122             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
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             /* REACTION-FIELD ELECTROSTATICS */
1145             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1146
1147             fscal            = felec;
1148
1149             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1150
1151              /* Update vectorial force */
1152             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1153             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1154             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1155
1156             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1157             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1158             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1159
1160             /**************************
1161              * CALCULATE INTERACTIONS *
1162              **************************/
1163
1164             /* Compute parameters for interactions between i and j atoms */
1165             qq30             = _mm_mul_ps(iq3,jq0);
1166
1167             /* REACTION-FIELD ELECTROSTATICS */
1168             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1169
1170             fscal            = felec;
1171
1172             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1173
1174              /* Update vectorial force */
1175             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1176             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1177             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1178
1179             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1180             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1181             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1182
1183             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1184             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1185             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1186             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1187
1188             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1189
1190             /* Inner loop uses 142 flops */
1191         }
1192
1193         /* End of innermost loop */
1194
1195         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1196                                               f+i_coord_offset,fshift+i_shift_offset);
1197
1198         /* Increment number of inner iterations */
1199         inneriter                  += j_index_end - j_index_start;
1200
1201         /* Outer loop uses 24 flops */
1202     }
1203
1204     /* Increment number of outer iterations */
1205     outeriter        += nri;
1206
1207     /* Update outer/inner flops */
1208
1209     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*142);
1210 }