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, 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_ElecRF_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_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     krf              = _mm_set1_ps(fr->ic->k_rf);
 129     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 130     crf              = _mm_set1_ps(fr->ic->c_rf);
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134
 135     vftab            = kernel_data->table_vdw->data;
 136     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 137
 138     /* Setup water-specific parameters */
 139     inr              = nlist->iinr[0];
 140     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 141     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 142     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 143     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = jnrC = jnrD = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149     j_coord_offsetC = 0;
 150     j_coord_offsetD = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 177
 178         fix0             = _mm_setzero_ps();
 179         fiy0             = _mm_setzero_ps();
 180         fiz0             = _mm_setzero_ps();
 181         fix1             = _mm_setzero_ps();
 182         fiy1             = _mm_setzero_ps();
 183         fiz1             = _mm_setzero_ps();
 184         fix2             = _mm_setzero_ps();
 185         fiy2             = _mm_setzero_ps();
 186         fiz2             = _mm_setzero_ps();
 187         fix3             = _mm_setzero_ps();
 188         fiy3             = _mm_setzero_ps();
 189         fiz3             = _mm_setzero_ps();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm_setzero_ps();
 193         vvdwsum          = _mm_setzero_ps();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             jnrC             = jjnr[jidx+2];
 203             jnrD             = jjnr[jidx+3];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208
 209             /* load j atom coordinates */
 210             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                               x+j_coord_offsetC,x+j_coord_offsetD,
 212                                               &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx00             = _mm_sub_ps(ix0,jx0);
 216             dy00             = _mm_sub_ps(iy0,jy0);
 217             dz00             = _mm_sub_ps(iz0,jz0);
 218             dx10             = _mm_sub_ps(ix1,jx0);
 219             dy10             = _mm_sub_ps(iy1,jy0);
 220             dz10             = _mm_sub_ps(iz1,jz0);
 221             dx20             = _mm_sub_ps(ix2,jx0);
 222             dy20             = _mm_sub_ps(iy2,jy0);
 223             dz20             = _mm_sub_ps(iz2,jz0);
 224             dx30             = _mm_sub_ps(ix3,jx0);
 225             dy30             = _mm_sub_ps(iy3,jy0);
 226             dz30             = _mm_sub_ps(iz3,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 230             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 231             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 232             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 233
 234             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 235             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 236             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 237             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 238
 239             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 240             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 241             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 242
 243             /* Load parameters for j particles */
 244             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 245                                                               charge+jnrC+0,charge+jnrD+0);
 246             vdwjidx0A        = 2*vdwtype[jnrA+0];
 247             vdwjidx0B        = 2*vdwtype[jnrB+0];
 248             vdwjidx0C        = 2*vdwtype[jnrC+0];
 249             vdwjidx0D        = 2*vdwtype[jnrD+0];
 250
 251             fjx0             = _mm_setzero_ps();
 252             fjy0             = _mm_setzero_ps();
 253             fjz0             = _mm_setzero_ps();
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             r00              = _mm_mul_ps(rsq00,rinv00);
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 263                                          vdwparam+vdwioffset0+vdwjidx0B,
 264                                          vdwparam+vdwioffset0+vdwjidx0C,
 265                                          vdwparam+vdwioffset0+vdwjidx0D,
 266                                          &c6_00,&c12_00);
 267
 268             /* Calculate table index by multiplying r with table scale and truncate to integer */
 269             rt               = _mm_mul_ps(r00,vftabscale);
 270             vfitab           = _mm_cvttps_epi32(rt);
 271 #ifdef __XOP__
 272             vfeps            = _mm_frcz_ps(rt);
 273 #else
 274             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 275 #endif
 276             twovfeps         = _mm_add_ps(vfeps,vfeps);
 277             vfitab           = _mm_slli_epi32(vfitab,3);
 278
 279             /* CUBIC SPLINE TABLE DISPERSION */
 280             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 281             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 282             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 283             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 284             _MM_TRANSPOSE4_PS(Y,F,G,H);
 285             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 286             VV               = _mm_macc_ps(vfeps,Fp,Y);
 287             vvdw6            = _mm_mul_ps(c6_00,VV);
 288             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 289             fvdw6            = _mm_mul_ps(c6_00,FF);
 290
 291             /* CUBIC SPLINE TABLE REPULSION */
 292             vfitab           = _mm_add_epi32(vfitab,ifour);
 293             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 294             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 295             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 296             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 297             _MM_TRANSPOSE4_PS(Y,F,G,H);
 298             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 299             VV               = _mm_macc_ps(vfeps,Fp,Y);
 300             vvdw12           = _mm_mul_ps(c12_00,VV);
 301             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 302             fvdw12           = _mm_mul_ps(c12_00,FF);
 303             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 304             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 308
 309             fscal            = fvdw;
 310
 311              /* Update vectorial force */
 312             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 313             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 314             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 315
 316             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 317             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 318             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq10             = _mm_mul_ps(iq1,jq0);
 326
 327             /* REACTION-FIELD ELECTROSTATICS */
 328             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 329             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 330
 331             /* Update potential sum for this i atom from the interaction with this j atom. */
 332             velecsum         = _mm_add_ps(velecsum,velec);
 333
 334             fscal            = felec;
 335
 336              /* Update vectorial force */
 337             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 338             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 339             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 340
 341             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 342             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 343             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 344
 345             /**************************
 346              * CALCULATE INTERACTIONS *
 347              **************************/
 348
 349             /* Compute parameters for interactions between i and j atoms */
 350             qq20             = _mm_mul_ps(iq2,jq0);
 351
 352             /* REACTION-FIELD ELECTROSTATICS */
 353             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 354             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velecsum         = _mm_add_ps(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361              /* Update vectorial force */
 362             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 363             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 364             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 365
 366             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 367             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 368             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 369
 370             /**************************
 371              * CALCULATE INTERACTIONS *
 372              **************************/
 373
 374             /* Compute parameters for interactions between i and j atoms */
 375             qq30             = _mm_mul_ps(iq3,jq0);
 376
 377             /* REACTION-FIELD ELECTROSTATICS */
 378             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 379             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 380
 381             /* Update potential sum for this i atom from the interaction with this j atom. */
 382             velecsum         = _mm_add_ps(velecsum,velec);
 383
 384             fscal            = felec;
 385
 386              /* Update vectorial force */
 387             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 388             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 389             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 390
 391             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 392             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 393             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 394
 395             fjptrA             = f+j_coord_offsetA;
 396             fjptrB             = f+j_coord_offsetB;
 397             fjptrC             = f+j_coord_offsetC;
 398             fjptrD             = f+j_coord_offsetD;
 399
 400             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 401
 402             /* Inner loop uses 164 flops */
 403         }
 404
 405         if(jidx<j_index_end)
 406         {
 407
 408             /* Get j neighbor index, and coordinate index */
 409             jnrlistA         = jjnr[jidx];
 410             jnrlistB         = jjnr[jidx+1];
 411             jnrlistC         = jjnr[jidx+2];
 412             jnrlistD         = jjnr[jidx+3];
 413             /* Sign of each element will be negative for non-real atoms.
 414              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 415              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 416              */
 417             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 418             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 419             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 420             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 421             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 422             j_coord_offsetA  = DIM*jnrA;
 423             j_coord_offsetB  = DIM*jnrB;
 424             j_coord_offsetC  = DIM*jnrC;
 425             j_coord_offsetD  = DIM*jnrD;
 426
 427             /* load j atom coordinates */
 428             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 429                                               x+j_coord_offsetC,x+j_coord_offsetD,
 430                                               &jx0,&jy0,&jz0);
 431
 432             /* Calculate displacement vector */
 433             dx00             = _mm_sub_ps(ix0,jx0);
 434             dy00             = _mm_sub_ps(iy0,jy0);
 435             dz00             = _mm_sub_ps(iz0,jz0);
 436             dx10             = _mm_sub_ps(ix1,jx0);
 437             dy10             = _mm_sub_ps(iy1,jy0);
 438             dz10             = _mm_sub_ps(iz1,jz0);
 439             dx20             = _mm_sub_ps(ix2,jx0);
 440             dy20             = _mm_sub_ps(iy2,jy0);
 441             dz20             = _mm_sub_ps(iz2,jz0);
 442             dx30             = _mm_sub_ps(ix3,jx0);
 443             dy30             = _mm_sub_ps(iy3,jy0);
 444             dz30             = _mm_sub_ps(iz3,jz0);
 445
 446             /* Calculate squared distance and things based on it */
 447             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 448             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 449             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 450             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 451
 452             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 453             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 454             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 455             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 456
 457             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 458             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 459             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 460
 461             /* Load parameters for j particles */
 462             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 463                                                               charge+jnrC+0,charge+jnrD+0);
 464             vdwjidx0A        = 2*vdwtype[jnrA+0];
 465             vdwjidx0B        = 2*vdwtype[jnrB+0];
 466             vdwjidx0C        = 2*vdwtype[jnrC+0];
 467             vdwjidx0D        = 2*vdwtype[jnrD+0];
 468
 469             fjx0             = _mm_setzero_ps();
 470             fjy0             = _mm_setzero_ps();
 471             fjz0             = _mm_setzero_ps();
 472
 473             /**************************
 474              * CALCULATE INTERACTIONS *
 475              **************************/
 476
 477             r00              = _mm_mul_ps(rsq00,rinv00);
 478             r00              = _mm_andnot_ps(dummy_mask,r00);
 479
 480             /* Compute parameters for interactions between i and j atoms */
 481             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 482                                          vdwparam+vdwioffset0+vdwjidx0B,
 483                                          vdwparam+vdwioffset0+vdwjidx0C,
 484                                          vdwparam+vdwioffset0+vdwjidx0D,
 485                                          &c6_00,&c12_00);
 486
 487             /* Calculate table index by multiplying r with table scale and truncate to integer */
 488             rt               = _mm_mul_ps(r00,vftabscale);
 489             vfitab           = _mm_cvttps_epi32(rt);
 490 #ifdef __XOP__
 491             vfeps            = _mm_frcz_ps(rt);
 492 #else
 493             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 494 #endif
 495             twovfeps         = _mm_add_ps(vfeps,vfeps);
 496             vfitab           = _mm_slli_epi32(vfitab,3);
 497
 498             /* CUBIC SPLINE TABLE DISPERSION */
 499             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 500             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 501             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 502             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 503             _MM_TRANSPOSE4_PS(Y,F,G,H);
 504             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 505             VV               = _mm_macc_ps(vfeps,Fp,Y);
 506             vvdw6            = _mm_mul_ps(c6_00,VV);
 507             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 508             fvdw6            = _mm_mul_ps(c6_00,FF);
 509
 510             /* CUBIC SPLINE TABLE REPULSION */
 511             vfitab           = _mm_add_epi32(vfitab,ifour);
 512             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 513             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 514             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 515             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 516             _MM_TRANSPOSE4_PS(Y,F,G,H);
 517             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 518             VV               = _mm_macc_ps(vfeps,Fp,Y);
 519             vvdw12           = _mm_mul_ps(c12_00,VV);
 520             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 521             fvdw12           = _mm_mul_ps(c12_00,FF);
 522             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 523             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 527             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 528
 529             fscal            = fvdw;
 530
 531             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 532
 533              /* Update vectorial force */
 534             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 535             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 536             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 537
 538             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 539             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 540             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 541
 542             /**************************
 543              * CALCULATE INTERACTIONS *
 544              **************************/
 545
 546             /* Compute parameters for interactions between i and j atoms */
 547             qq10             = _mm_mul_ps(iq1,jq0);
 548
 549             /* REACTION-FIELD ELECTROSTATICS */
 550             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 551             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 552
 553             /* Update potential sum for this i atom from the interaction with this j atom. */
 554             velec            = _mm_andnot_ps(dummy_mask,velec);
 555             velecsum         = _mm_add_ps(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 560
 561              /* Update vectorial force */
 562             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 563             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 564             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 565
 566             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 567             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 568             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 569
 570             /**************************
 571              * CALCULATE INTERACTIONS *
 572              **************************/
 573
 574             /* Compute parameters for interactions between i and j atoms */
 575             qq20             = _mm_mul_ps(iq2,jq0);
 576
 577             /* REACTION-FIELD ELECTROSTATICS */
 578             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 579             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm_andnot_ps(dummy_mask,velec);
 583             velecsum         = _mm_add_ps(velecsum,velec);
 584
 585             fscal            = felec;
 586
 587             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 588
 589              /* Update vectorial force */
 590             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 591             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 592             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 593
 594             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 595             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 596             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             /* Compute parameters for interactions between i and j atoms */
 603             qq30             = _mm_mul_ps(iq3,jq0);
 604
 605             /* REACTION-FIELD ELECTROSTATICS */
 606             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 607             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 608
 609             /* Update potential sum for this i atom from the interaction with this j atom. */
 610             velec            = _mm_andnot_ps(dummy_mask,velec);
 611             velecsum         = _mm_add_ps(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 616
 617              /* Update vectorial force */
 618             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 619             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 620             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 621
 622             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 623             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 624             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 625
 626             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 627             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 628             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 629             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 630
 631             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 632
 633             /* Inner loop uses 165 flops */
 634         }
 635
 636         /* End of innermost loop */
 637
 638         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 639                                               f+i_coord_offset,fshift+i_shift_offset);
 640
 641         ggid                        = gid[iidx];
 642         /* Update potential energies */
 643         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 644         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 645
 646         /* Increment number of inner iterations */
 647         inneriter                  += j_index_end - j_index_start;
 648
 649         /* Outer loop uses 26 flops */
 650     }
 651
 652     /* Increment number of outer iterations */
 653     outeriter        += nri;
 654
 655     /* Update outer/inner flops */
 656
 657     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*165);
 658 }
 659 /*
 660  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 661  * Electrostatics interaction: ReactionField
 662  * VdW interaction:            CubicSplineTable
 663  * Geometry:                   Water4-Particle
 664  * Calculate force/pot:        Force
 665  */
 666 void
 667 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 668                     (t_nblist                    * gmx_restrict       nlist,
 669                      rvec                        * gmx_restrict          xx,
 670                      rvec                        * gmx_restrict          ff,
 671                      t_forcerec                  * gmx_restrict          fr,
 672                      t_mdatoms                   * gmx_restrict     mdatoms,
 673                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 674                      t_nrnb                      * gmx_restrict        nrnb)
 675 {
 676     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 677      * just 0 for non-waters.
 678      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 679      * jnr indices corresponding to data put in the four positions in the SIMD register.
 680      */
 681     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 682     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 683     int              jnrA,jnrB,jnrC,jnrD;
 684     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 685     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 686     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 687     real             rcutoff_scalar;
 688     real             *shiftvec,*fshift,*x,*f;
 689     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 690     real             scratch[4*DIM];
 691     __m128           fscal,rcutoff,rcutoff2,jidxall;
 692     int              vdwioffset0;
 693     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 694     int              vdwioffset1;
 695     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 696     int              vdwioffset2;
 697     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 698     int              vdwioffset3;
 699     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 700     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 701     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 702     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 703     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 704     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 705     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 706     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 707     real             *charge;
 708     int              nvdwtype;
 709     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 710     int              *vdwtype;
 711     real             *vdwparam;
 712     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 713     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 714     __m128i          vfitab;
 715     __m128i          ifour       = _mm_set1_epi32(4);
 716     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 717     real             *vftab;
 718     __m128           dummy_mask,cutoff_mask;
 719     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 720     __m128           one     = _mm_set1_ps(1.0);
 721     __m128           two     = _mm_set1_ps(2.0);
 722     x                = xx[0];
 723     f                = ff[0];
 724
 725     nri              = nlist->nri;
 726     iinr             = nlist->iinr;
 727     jindex           = nlist->jindex;
 728     jjnr             = nlist->jjnr;
 729     shiftidx         = nlist->shift;
 730     gid              = nlist->gid;
 731     shiftvec         = fr->shift_vec[0];
 732     fshift           = fr->fshift[0];
 733     facel            = _mm_set1_ps(fr->epsfac);
 734     charge           = mdatoms->chargeA;
 735     krf              = _mm_set1_ps(fr->ic->k_rf);
 736     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 737     crf              = _mm_set1_ps(fr->ic->c_rf);
 738     nvdwtype         = fr->ntype;
 739     vdwparam         = fr->nbfp;
 740     vdwtype          = mdatoms->typeA;
 741
 742     vftab            = kernel_data->table_vdw->data;
 743     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 744
 745     /* Setup water-specific parameters */
 746     inr              = nlist->iinr[0];
 747     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 748     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 749     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 750     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 751
 752     /* Avoid stupid compiler warnings */
 753     jnrA = jnrB = jnrC = jnrD = 0;
 754     j_coord_offsetA = 0;
 755     j_coord_offsetB = 0;
 756     j_coord_offsetC = 0;
 757     j_coord_offsetD = 0;
 758
 759     outeriter        = 0;
 760     inneriter        = 0;
 761
 762     for(iidx=0;iidx<4*DIM;iidx++)
 763     {
 764         scratch[iidx] = 0.0;
 765     }
 766
 767     /* Start outer loop over neighborlists */
 768     for(iidx=0; iidx<nri; iidx++)
 769     {
 770         /* Load shift vector for this list */
 771         i_shift_offset   = DIM*shiftidx[iidx];
 772
 773         /* Load limits for loop over neighbors */
 774         j_index_start    = jindex[iidx];
 775         j_index_end      = jindex[iidx+1];
 776
 777         /* Get outer coordinate index */
 778         inr              = iinr[iidx];
 779         i_coord_offset   = DIM*inr;
 780
 781         /* Load i particle coords and add shift vector */
 782         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 783                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 784
 785         fix0             = _mm_setzero_ps();
 786         fiy0             = _mm_setzero_ps();
 787         fiz0             = _mm_setzero_ps();
 788         fix1             = _mm_setzero_ps();
 789         fiy1             = _mm_setzero_ps();
 790         fiz1             = _mm_setzero_ps();
 791         fix2             = _mm_setzero_ps();
 792         fiy2             = _mm_setzero_ps();
 793         fiz2             = _mm_setzero_ps();
 794         fix3             = _mm_setzero_ps();
 795         fiy3             = _mm_setzero_ps();
 796         fiz3             = _mm_setzero_ps();
 797
 798         /* Start inner kernel loop */
 799         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 800         {
 801
 802             /* Get j neighbor index, and coordinate index */
 803             jnrA             = jjnr[jidx];
 804             jnrB             = jjnr[jidx+1];
 805             jnrC             = jjnr[jidx+2];
 806             jnrD             = jjnr[jidx+3];
 807             j_coord_offsetA  = DIM*jnrA;
 808             j_coord_offsetB  = DIM*jnrB;
 809             j_coord_offsetC  = DIM*jnrC;
 810             j_coord_offsetD  = DIM*jnrD;
 811
 812             /* load j atom coordinates */
 813             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 814                                               x+j_coord_offsetC,x+j_coord_offsetD,
 815                                               &jx0,&jy0,&jz0);
 816
 817             /* Calculate displacement vector */
 818             dx00             = _mm_sub_ps(ix0,jx0);
 819             dy00             = _mm_sub_ps(iy0,jy0);
 820             dz00             = _mm_sub_ps(iz0,jz0);
 821             dx10             = _mm_sub_ps(ix1,jx0);
 822             dy10             = _mm_sub_ps(iy1,jy0);
 823             dz10             = _mm_sub_ps(iz1,jz0);
 824             dx20             = _mm_sub_ps(ix2,jx0);
 825             dy20             = _mm_sub_ps(iy2,jy0);
 826             dz20             = _mm_sub_ps(iz2,jz0);
 827             dx30             = _mm_sub_ps(ix3,jx0);
 828             dy30             = _mm_sub_ps(iy3,jy0);
 829             dz30             = _mm_sub_ps(iz3,jz0);
 830
 831             /* Calculate squared distance and things based on it */
 832             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 833             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 834             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 835             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 836
 837             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 838             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 839             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 840             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 841
 842             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 843             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 844             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 845
 846             /* Load parameters for j particles */
 847             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 848                                                               charge+jnrC+0,charge+jnrD+0);
 849             vdwjidx0A        = 2*vdwtype[jnrA+0];
 850             vdwjidx0B        = 2*vdwtype[jnrB+0];
 851             vdwjidx0C        = 2*vdwtype[jnrC+0];
 852             vdwjidx0D        = 2*vdwtype[jnrD+0];
 853
 854             fjx0             = _mm_setzero_ps();
 855             fjy0             = _mm_setzero_ps();
 856             fjz0             = _mm_setzero_ps();
 857
 858             /**************************
 859              * CALCULATE INTERACTIONS *
 860              **************************/
 861
 862             r00              = _mm_mul_ps(rsq00,rinv00);
 863
 864             /* Compute parameters for interactions between i and j atoms */
 865             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 866                                          vdwparam+vdwioffset0+vdwjidx0B,
 867                                          vdwparam+vdwioffset0+vdwjidx0C,
 868                                          vdwparam+vdwioffset0+vdwjidx0D,
 869                                          &c6_00,&c12_00);
 870
 871             /* Calculate table index by multiplying r with table scale and truncate to integer */
 872             rt               = _mm_mul_ps(r00,vftabscale);
 873             vfitab           = _mm_cvttps_epi32(rt);
 874 #ifdef __XOP__
 875             vfeps            = _mm_frcz_ps(rt);
 876 #else
 877             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 878 #endif
 879             twovfeps         = _mm_add_ps(vfeps,vfeps);
 880             vfitab           = _mm_slli_epi32(vfitab,3);
 881
 882             /* CUBIC SPLINE TABLE DISPERSION */
 883             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 884             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 885             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 886             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 887             _MM_TRANSPOSE4_PS(Y,F,G,H);
 888             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 889             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 890             fvdw6            = _mm_mul_ps(c6_00,FF);
 891
 892             /* CUBIC SPLINE TABLE REPULSION */
 893             vfitab           = _mm_add_epi32(vfitab,ifour);
 894             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 895             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 896             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 897             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 898             _MM_TRANSPOSE4_PS(Y,F,G,H);
 899             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 900             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 901             fvdw12           = _mm_mul_ps(c12_00,FF);
 902             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 903
 904             fscal            = fvdw;
 905
 906              /* Update vectorial force */
 907             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 908             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 909             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 910
 911             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 912             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 913             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 914
 915             /**************************
 916              * CALCULATE INTERACTIONS *
 917              **************************/
 918
 919             /* Compute parameters for interactions between i and j atoms */
 920             qq10             = _mm_mul_ps(iq1,jq0);
 921
 922             /* REACTION-FIELD ELECTROSTATICS */
 923             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 924
 925             fscal            = felec;
 926
 927              /* Update vectorial force */
 928             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 929             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 930             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 931
 932             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 933             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 934             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             /* Compute parameters for interactions between i and j atoms */
 941             qq20             = _mm_mul_ps(iq2,jq0);
 942
 943             /* REACTION-FIELD ELECTROSTATICS */
 944             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 945
 946             fscal            = felec;
 947
 948              /* Update vectorial force */
 949             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 950             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 951             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 952
 953             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 954             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 955             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             /* Compute parameters for interactions between i and j atoms */
 962             qq30             = _mm_mul_ps(iq3,jq0);
 963
 964             /* REACTION-FIELD ELECTROSTATICS */
 965             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 966
 967             fscal            = felec;
 968
 969              /* Update vectorial force */
 970             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 971             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 972             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 973
 974             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 975             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 976             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 977
 978             fjptrA             = f+j_coord_offsetA;
 979             fjptrB             = f+j_coord_offsetB;
 980             fjptrC             = f+j_coord_offsetC;
 981             fjptrD             = f+j_coord_offsetD;
 982
 983             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 984
 985             /* Inner loop uses 141 flops */
 986         }
 987
 988         if(jidx<j_index_end)
 989         {
 990
 991             /* Get j neighbor index, and coordinate index */
 992             jnrlistA         = jjnr[jidx];
 993             jnrlistB         = jjnr[jidx+1];
 994             jnrlistC         = jjnr[jidx+2];
 995             jnrlistD         = jjnr[jidx+3];
 996             /* Sign of each element will be negative for non-real atoms.
 997              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 998              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 999              */
1000             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1001             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1002             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1003             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1004             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1005             j_coord_offsetA  = DIM*jnrA;
1006             j_coord_offsetB  = DIM*jnrB;
1007             j_coord_offsetC  = DIM*jnrC;
1008             j_coord_offsetD  = DIM*jnrD;
1009
1010             /* load j atom coordinates */
1011             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1012                                               x+j_coord_offsetC,x+j_coord_offsetD,
1013                                               &jx0,&jy0,&jz0);
1014
1015             /* Calculate displacement vector */
1016             dx00             = _mm_sub_ps(ix0,jx0);
1017             dy00             = _mm_sub_ps(iy0,jy0);
1018             dz00             = _mm_sub_ps(iz0,jz0);
1019             dx10             = _mm_sub_ps(ix1,jx0);
1020             dy10             = _mm_sub_ps(iy1,jy0);
1021             dz10             = _mm_sub_ps(iz1,jz0);
1022             dx20             = _mm_sub_ps(ix2,jx0);
1023             dy20             = _mm_sub_ps(iy2,jy0);
1024             dz20             = _mm_sub_ps(iz2,jz0);
1025             dx30             = _mm_sub_ps(ix3,jx0);
1026             dy30             = _mm_sub_ps(iy3,jy0);
1027             dz30             = _mm_sub_ps(iz3,jz0);
1028
1029             /* Calculate squared distance and things based on it */
1030             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1031             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1032             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1033             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1034
1035             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1036             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1037             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1038             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1039
1040             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1041             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1042             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1043
1044             /* Load parameters for j particles */
1045             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1046                                                               charge+jnrC+0,charge+jnrD+0);
1047             vdwjidx0A        = 2*vdwtype[jnrA+0];
1048             vdwjidx0B        = 2*vdwtype[jnrB+0];
1049             vdwjidx0C        = 2*vdwtype[jnrC+0];
1050             vdwjidx0D        = 2*vdwtype[jnrD+0];
1051
1052             fjx0             = _mm_setzero_ps();
1053             fjy0             = _mm_setzero_ps();
1054             fjz0             = _mm_setzero_ps();
1055
1056             /**************************
1057              * CALCULATE INTERACTIONS *
1058              **************************/
1059
1060             r00              = _mm_mul_ps(rsq00,rinv00);
1061             r00              = _mm_andnot_ps(dummy_mask,r00);
1062
1063             /* Compute parameters for interactions between i and j atoms */
1064             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1065                                          vdwparam+vdwioffset0+vdwjidx0B,
1066                                          vdwparam+vdwioffset0+vdwjidx0C,
1067                                          vdwparam+vdwioffset0+vdwjidx0D,
1068                                          &c6_00,&c12_00);
1069
1070             /* Calculate table index by multiplying r with table scale and truncate to integer */
1071             rt               = _mm_mul_ps(r00,vftabscale);
1072             vfitab           = _mm_cvttps_epi32(rt);
1073 #ifdef __XOP__
1074             vfeps            = _mm_frcz_ps(rt);
1075 #else
1076             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1077 #endif
1078             twovfeps         = _mm_add_ps(vfeps,vfeps);
1079             vfitab           = _mm_slli_epi32(vfitab,3);
1080
1081             /* CUBIC SPLINE TABLE DISPERSION */
1082             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1083             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1084             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1085             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1086             _MM_TRANSPOSE4_PS(Y,F,G,H);
1087             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1088             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1089             fvdw6            = _mm_mul_ps(c6_00,FF);
1090
1091             /* CUBIC SPLINE TABLE REPULSION */
1092             vfitab           = _mm_add_epi32(vfitab,ifour);
1093             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1094             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1095             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1096             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1097             _MM_TRANSPOSE4_PS(Y,F,G,H);
1098             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1099             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1100             fvdw12           = _mm_mul_ps(c12_00,FF);
1101             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1102
1103             fscal            = fvdw;
1104
1105             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1106
1107              /* Update vectorial force */
1108             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1109             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1110             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1111
1112             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1113             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1114             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1115
1116             /**************************
1117              * CALCULATE INTERACTIONS *
1118              **************************/
1119
1120             /* Compute parameters for interactions between i and j atoms */
1121             qq10             = _mm_mul_ps(iq1,jq0);
1122
1123             /* REACTION-FIELD ELECTROSTATICS */
1124             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1125
1126             fscal            = felec;
1127
1128             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1129
1130              /* Update vectorial force */
1131             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1132             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1133             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1134
1135             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1136             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1137             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1138
1139             /**************************
1140              * CALCULATE INTERACTIONS *
1141              **************************/
1142
1143             /* Compute parameters for interactions between i and j atoms */
1144             qq20             = _mm_mul_ps(iq2,jq0);
1145
1146             /* REACTION-FIELD ELECTROSTATICS */
1147             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1148
1149             fscal            = felec;
1150
1151             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1152
1153              /* Update vectorial force */
1154             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1155             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1156             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1157
1158             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1159             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1160             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1161
1162             /**************************
1163              * CALCULATE INTERACTIONS *
1164              **************************/
1165
1166             /* Compute parameters for interactions between i and j atoms */
1167             qq30             = _mm_mul_ps(iq3,jq0);
1168
1169             /* REACTION-FIELD ELECTROSTATICS */
1170             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
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 }