src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_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_ElecRFCut_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_ElecRFCut_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 144     rcutoff_scalar   = fr->rcoulomb;
 145     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 146     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = jnrC = jnrD = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152     j_coord_offsetC = 0;
 153     j_coord_offsetD = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 180
 181         fix0             = _mm_setzero_ps();
 182         fiy0             = _mm_setzero_ps();
 183         fiz0             = _mm_setzero_ps();
 184         fix1             = _mm_setzero_ps();
 185         fiy1             = _mm_setzero_ps();
 186         fiz1             = _mm_setzero_ps();
 187         fix2             = _mm_setzero_ps();
 188         fiy2             = _mm_setzero_ps();
 189         fiz2             = _mm_setzero_ps();
 190         fix3             = _mm_setzero_ps();
 191         fiy3             = _mm_setzero_ps();
 192         fiz3             = _mm_setzero_ps();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm_setzero_ps();
 196         vvdwsum          = _mm_setzero_ps();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211
 212             /* load j atom coordinates */
 213             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                               x+j_coord_offsetC,x+j_coord_offsetD,
 215                                               &jx0,&jy0,&jz0);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm_sub_ps(ix0,jx0);
 219             dy00             = _mm_sub_ps(iy0,jy0);
 220             dz00             = _mm_sub_ps(iz0,jz0);
 221             dx10             = _mm_sub_ps(ix1,jx0);
 222             dy10             = _mm_sub_ps(iy1,jy0);
 223             dz10             = _mm_sub_ps(iz1,jz0);
 224             dx20             = _mm_sub_ps(ix2,jx0);
 225             dy20             = _mm_sub_ps(iy2,jy0);
 226             dz20             = _mm_sub_ps(iz2,jz0);
 227             dx30             = _mm_sub_ps(ix3,jx0);
 228             dy30             = _mm_sub_ps(iy3,jy0);
 229             dz30             = _mm_sub_ps(iz3,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 233             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 234             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 235             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 236
 237             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 238             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 239             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 240             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 241
 242             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 243             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 244             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 245
 246             /* Load parameters for j particles */
 247             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 248                                                               charge+jnrC+0,charge+jnrD+0);
 249             vdwjidx0A        = 2*vdwtype[jnrA+0];
 250             vdwjidx0B        = 2*vdwtype[jnrB+0];
 251             vdwjidx0C        = 2*vdwtype[jnrC+0];
 252             vdwjidx0D        = 2*vdwtype[jnrD+0];
 253
 254             fjx0             = _mm_setzero_ps();
 255             fjy0             = _mm_setzero_ps();
 256             fjz0             = _mm_setzero_ps();
 257
 258             /**************************
 259              * CALCULATE INTERACTIONS *
 260              **************************/
 261
 262             r00              = _mm_mul_ps(rsq00,rinv00);
 263
 264             /* Compute parameters for interactions between i and j atoms */
 265             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 266                                          vdwparam+vdwioffset0+vdwjidx0B,
 267                                          vdwparam+vdwioffset0+vdwjidx0C,
 268                                          vdwparam+vdwioffset0+vdwjidx0D,
 269                                          &c6_00,&c12_00);
 270
 271             /* Calculate table index by multiplying r with table scale and truncate to integer */
 272             rt               = _mm_mul_ps(r00,vftabscale);
 273             vfitab           = _mm_cvttps_epi32(rt);
 274 #ifdef __XOP__
 275             vfeps            = _mm_frcz_ps(rt);
 276 #else
 277             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 278 #endif
 279             twovfeps         = _mm_add_ps(vfeps,vfeps);
 280             vfitab           = _mm_slli_epi32(vfitab,3);
 281
 282             /* CUBIC SPLINE TABLE DISPERSION */
 283             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 284             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 285             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 286             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 287             _MM_TRANSPOSE4_PS(Y,F,G,H);
 288             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 289             VV               = _mm_macc_ps(vfeps,Fp,Y);
 290             vvdw6            = _mm_mul_ps(c6_00,VV);
 291             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 292             fvdw6            = _mm_mul_ps(c6_00,FF);
 293
 294             /* CUBIC SPLINE TABLE REPULSION */
 295             vfitab           = _mm_add_epi32(vfitab,ifour);
 296             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 297             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 298             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 299             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 300             _MM_TRANSPOSE4_PS(Y,F,G,H);
 301             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 302             VV               = _mm_macc_ps(vfeps,Fp,Y);
 303             vvdw12           = _mm_mul_ps(c12_00,VV);
 304             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 305             fvdw12           = _mm_mul_ps(c12_00,FF);
 306             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 307             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 311
 312             fscal            = fvdw;
 313
 314              /* Update vectorial force */
 315             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 316             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 317             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 318
 319             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 320             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 321             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 322
 323             /**************************
 324              * CALCULATE INTERACTIONS *
 325              **************************/
 326
 327             if (gmx_mm_any_lt(rsq10,rcutoff2))
 328             {
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             qq10             = _mm_mul_ps(iq1,jq0);
 332
 333             /* REACTION-FIELD ELECTROSTATICS */
 334             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 335             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 336
 337             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             velec            = _mm_and_ps(velec,cutoff_mask);
 341             velecsum         = _mm_add_ps(velecsum,velec);
 342
 343             fscal            = felec;
 344
 345             fscal            = _mm_and_ps(fscal,cutoff_mask);
 346
 347              /* Update vectorial force */
 348             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 349             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 350             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 351
 352             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 353             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 354             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 355
 356             }
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             if (gmx_mm_any_lt(rsq20,rcutoff2))
 363             {
 364
 365             /* Compute parameters for interactions between i and j atoms */
 366             qq20             = _mm_mul_ps(iq2,jq0);
 367
 368             /* REACTION-FIELD ELECTROSTATICS */
 369             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 370             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 371
 372             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velec            = _mm_and_ps(velec,cutoff_mask);
 376             velecsum         = _mm_add_ps(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             fscal            = _mm_and_ps(fscal,cutoff_mask);
 381
 382              /* Update vectorial force */
 383             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 384             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 385             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 386
 387             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 388             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 389             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 390
 391             }
 392
 393             /**************************
 394              * CALCULATE INTERACTIONS *
 395              **************************/
 396
 397             if (gmx_mm_any_lt(rsq30,rcutoff2))
 398             {
 399
 400             /* Compute parameters for interactions between i and j atoms */
 401             qq30             = _mm_mul_ps(iq3,jq0);
 402
 403             /* REACTION-FIELD ELECTROSTATICS */
 404             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 405             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 406
 407             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velec            = _mm_and_ps(velec,cutoff_mask);
 411             velecsum         = _mm_add_ps(velecsum,velec);
 412
 413             fscal            = felec;
 414
 415             fscal            = _mm_and_ps(fscal,cutoff_mask);
 416
 417              /* Update vectorial force */
 418             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 419             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 420             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 421
 422             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 423             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 424             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 425
 426             }
 427
 428             fjptrA             = f+j_coord_offsetA;
 429             fjptrB             = f+j_coord_offsetB;
 430             fjptrC             = f+j_coord_offsetC;
 431             fjptrD             = f+j_coord_offsetD;
 432
 433             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 434
 435             /* Inner loop uses 176 flops */
 436         }
 437
 438         if(jidx<j_index_end)
 439         {
 440
 441             /* Get j neighbor index, and coordinate index */
 442             jnrlistA         = jjnr[jidx];
 443             jnrlistB         = jjnr[jidx+1];
 444             jnrlistC         = jjnr[jidx+2];
 445             jnrlistD         = jjnr[jidx+3];
 446             /* Sign of each element will be negative for non-real atoms.
 447              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 448              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 449              */
 450             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 451             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 452             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 453             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 454             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 455             j_coord_offsetA  = DIM*jnrA;
 456             j_coord_offsetB  = DIM*jnrB;
 457             j_coord_offsetC  = DIM*jnrC;
 458             j_coord_offsetD  = DIM*jnrD;
 459
 460             /* load j atom coordinates */
 461             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 462                                               x+j_coord_offsetC,x+j_coord_offsetD,
 463                                               &jx0,&jy0,&jz0);
 464
 465             /* Calculate displacement vector */
 466             dx00             = _mm_sub_ps(ix0,jx0);
 467             dy00             = _mm_sub_ps(iy0,jy0);
 468             dz00             = _mm_sub_ps(iz0,jz0);
 469             dx10             = _mm_sub_ps(ix1,jx0);
 470             dy10             = _mm_sub_ps(iy1,jy0);
 471             dz10             = _mm_sub_ps(iz1,jz0);
 472             dx20             = _mm_sub_ps(ix2,jx0);
 473             dy20             = _mm_sub_ps(iy2,jy0);
 474             dz20             = _mm_sub_ps(iz2,jz0);
 475             dx30             = _mm_sub_ps(ix3,jx0);
 476             dy30             = _mm_sub_ps(iy3,jy0);
 477             dz30             = _mm_sub_ps(iz3,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 481             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 482             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 483             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 484
 485             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 486             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 487             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 488             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 489
 490             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 491             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 492             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 493
 494             /* Load parameters for j particles */
 495             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 496                                                               charge+jnrC+0,charge+jnrD+0);
 497             vdwjidx0A        = 2*vdwtype[jnrA+0];
 498             vdwjidx0B        = 2*vdwtype[jnrB+0];
 499             vdwjidx0C        = 2*vdwtype[jnrC+0];
 500             vdwjidx0D        = 2*vdwtype[jnrD+0];
 501
 502             fjx0             = _mm_setzero_ps();
 503             fjy0             = _mm_setzero_ps();
 504             fjz0             = _mm_setzero_ps();
 505
 506             /**************************
 507              * CALCULATE INTERACTIONS *
 508              **************************/
 509
 510             r00              = _mm_mul_ps(rsq00,rinv00);
 511             r00              = _mm_andnot_ps(dummy_mask,r00);
 512
 513             /* Compute parameters for interactions between i and j atoms */
 514             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 515                                          vdwparam+vdwioffset0+vdwjidx0B,
 516                                          vdwparam+vdwioffset0+vdwjidx0C,
 517                                          vdwparam+vdwioffset0+vdwjidx0D,
 518                                          &c6_00,&c12_00);
 519
 520             /* Calculate table index by multiplying r with table scale and truncate to integer */
 521             rt               = _mm_mul_ps(r00,vftabscale);
 522             vfitab           = _mm_cvttps_epi32(rt);
 523 #ifdef __XOP__
 524             vfeps            = _mm_frcz_ps(rt);
 525 #else
 526             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 527 #endif
 528             twovfeps         = _mm_add_ps(vfeps,vfeps);
 529             vfitab           = _mm_slli_epi32(vfitab,3);
 530
 531             /* CUBIC SPLINE TABLE DISPERSION */
 532             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 533             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 534             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 535             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 536             _MM_TRANSPOSE4_PS(Y,F,G,H);
 537             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 538             VV               = _mm_macc_ps(vfeps,Fp,Y);
 539             vvdw6            = _mm_mul_ps(c6_00,VV);
 540             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 541             fvdw6            = _mm_mul_ps(c6_00,FF);
 542
 543             /* CUBIC SPLINE TABLE REPULSION */
 544             vfitab           = _mm_add_epi32(vfitab,ifour);
 545             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 546             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 547             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 548             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 549             _MM_TRANSPOSE4_PS(Y,F,G,H);
 550             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 551             VV               = _mm_macc_ps(vfeps,Fp,Y);
 552             vvdw12           = _mm_mul_ps(c12_00,VV);
 553             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 554             fvdw12           = _mm_mul_ps(c12_00,FF);
 555             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 556             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 560             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 561
 562             fscal            = fvdw;
 563
 564             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 565
 566              /* Update vectorial force */
 567             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 568             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 569             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 570
 571             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 572             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 573             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             if (gmx_mm_any_lt(rsq10,rcutoff2))
 580             {
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq10             = _mm_mul_ps(iq1,jq0);
 584
 585             /* REACTION-FIELD ELECTROSTATICS */
 586             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 587             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 588
 589             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velec            = _mm_and_ps(velec,cutoff_mask);
 593             velec            = _mm_andnot_ps(dummy_mask,velec);
 594             velecsum         = _mm_add_ps(velecsum,velec);
 595
 596             fscal            = felec;
 597
 598             fscal            = _mm_and_ps(fscal,cutoff_mask);
 599
 600             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 601
 602              /* Update vectorial force */
 603             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 604             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 605             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 606
 607             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 608             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 609             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 610
 611             }
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             if (gmx_mm_any_lt(rsq20,rcutoff2))
 618             {
 619
 620             /* Compute parameters for interactions between i and j atoms */
 621             qq20             = _mm_mul_ps(iq2,jq0);
 622
 623             /* REACTION-FIELD ELECTROSTATICS */
 624             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 625             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 626
 627             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 628
 629             /* Update potential sum for this i atom from the interaction with this j atom. */
 630             velec            = _mm_and_ps(velec,cutoff_mask);
 631             velec            = _mm_andnot_ps(dummy_mask,velec);
 632             velecsum         = _mm_add_ps(velecsum,velec);
 633
 634             fscal            = felec;
 635
 636             fscal            = _mm_and_ps(fscal,cutoff_mask);
 637
 638             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 639
 640              /* Update vectorial force */
 641             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 642             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 643             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 644
 645             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 646             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 647             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 648
 649             }
 650
 651             /**************************
 652              * CALCULATE INTERACTIONS *
 653              **************************/
 654
 655             if (gmx_mm_any_lt(rsq30,rcutoff2))
 656             {
 657
 658             /* Compute parameters for interactions between i and j atoms */
 659             qq30             = _mm_mul_ps(iq3,jq0);
 660
 661             /* REACTION-FIELD ELECTROSTATICS */
 662             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 663             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 664
 665             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 666
 667             /* Update potential sum for this i atom from the interaction with this j atom. */
 668             velec            = _mm_and_ps(velec,cutoff_mask);
 669             velec            = _mm_andnot_ps(dummy_mask,velec);
 670             velecsum         = _mm_add_ps(velecsum,velec);
 671
 672             fscal            = felec;
 673
 674             fscal            = _mm_and_ps(fscal,cutoff_mask);
 675
 676             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 677
 678              /* Update vectorial force */
 679             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 680             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 681             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 682
 683             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 684             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 685             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 686
 687             }
 688
 689             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 690             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 691             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 692             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 693
 694             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 695
 696             /* Inner loop uses 177 flops */
 697         }
 698
 699         /* End of innermost loop */
 700
 701         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 702                                               f+i_coord_offset,fshift+i_shift_offset);
 703
 704         ggid                        = gid[iidx];
 705         /* Update potential energies */
 706         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 707         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 708
 709         /* Increment number of inner iterations */
 710         inneriter                  += j_index_end - j_index_start;
 711
 712         /* Outer loop uses 26 flops */
 713     }
 714
 715     /* Increment number of outer iterations */
 716     outeriter        += nri;
 717
 718     /* Update outer/inner flops */
 719
 720     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*177);
 721 }
 722 /*
 723  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 724  * Electrostatics interaction: ReactionField
 725  * VdW interaction:            CubicSplineTable
 726  * Geometry:                   Water4-Particle
 727  * Calculate force/pot:        Force
 728  */
 729 void
 730 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 731                     (t_nblist                    * gmx_restrict       nlist,
 732                      rvec                        * gmx_restrict          xx,
 733                      rvec                        * gmx_restrict          ff,
 734                      t_forcerec                  * gmx_restrict          fr,
 735                      t_mdatoms                   * gmx_restrict     mdatoms,
 736                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 737                      t_nrnb                      * gmx_restrict        nrnb)
 738 {
 739     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 740      * just 0 for non-waters.
 741      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 742      * jnr indices corresponding to data put in the four positions in the SIMD register.
 743      */
 744     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 745     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 746     int              jnrA,jnrB,jnrC,jnrD;
 747     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 748     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 749     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 750     real             rcutoff_scalar;
 751     real             *shiftvec,*fshift,*x,*f;
 752     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 753     real             scratch[4*DIM];
 754     __m128           fscal,rcutoff,rcutoff2,jidxall;
 755     int              vdwioffset0;
 756     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 757     int              vdwioffset1;
 758     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 759     int              vdwioffset2;
 760     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 761     int              vdwioffset3;
 762     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 763     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 764     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 765     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 766     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 767     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 768     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 769     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 770     real             *charge;
 771     int              nvdwtype;
 772     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 773     int              *vdwtype;
 774     real             *vdwparam;
 775     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 776     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 777     __m128i          vfitab;
 778     __m128i          ifour       = _mm_set1_epi32(4);
 779     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 780     real             *vftab;
 781     __m128           dummy_mask,cutoff_mask;
 782     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 783     __m128           one     = _mm_set1_ps(1.0);
 784     __m128           two     = _mm_set1_ps(2.0);
 785     x                = xx[0];
 786     f                = ff[0];
 787
 788     nri              = nlist->nri;
 789     iinr             = nlist->iinr;
 790     jindex           = nlist->jindex;
 791     jjnr             = nlist->jjnr;
 792     shiftidx         = nlist->shift;
 793     gid              = nlist->gid;
 794     shiftvec         = fr->shift_vec[0];
 795     fshift           = fr->fshift[0];
 796     facel            = _mm_set1_ps(fr->epsfac);
 797     charge           = mdatoms->chargeA;
 798     krf              = _mm_set1_ps(fr->ic->k_rf);
 799     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 800     crf              = _mm_set1_ps(fr->ic->c_rf);
 801     nvdwtype         = fr->ntype;
 802     vdwparam         = fr->nbfp;
 803     vdwtype          = mdatoms->typeA;
 804
 805     vftab            = kernel_data->table_vdw->data;
 806     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 807
 808     /* Setup water-specific parameters */
 809     inr              = nlist->iinr[0];
 810     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 811     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 812     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 813     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 814
 815     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 816     rcutoff_scalar   = fr->rcoulomb;
 817     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 818     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 819
 820     /* Avoid stupid compiler warnings */
 821     jnrA = jnrB = jnrC = jnrD = 0;
 822     j_coord_offsetA = 0;
 823     j_coord_offsetB = 0;
 824     j_coord_offsetC = 0;
 825     j_coord_offsetD = 0;
 826
 827     outeriter        = 0;
 828     inneriter        = 0;
 829
 830     for(iidx=0;iidx<4*DIM;iidx++)
 831     {
 832         scratch[iidx] = 0.0;
 833     }
 834
 835     /* Start outer loop over neighborlists */
 836     for(iidx=0; iidx<nri; iidx++)
 837     {
 838         /* Load shift vector for this list */
 839         i_shift_offset   = DIM*shiftidx[iidx];
 840
 841         /* Load limits for loop over neighbors */
 842         j_index_start    = jindex[iidx];
 843         j_index_end      = jindex[iidx+1];
 844
 845         /* Get outer coordinate index */
 846         inr              = iinr[iidx];
 847         i_coord_offset   = DIM*inr;
 848
 849         /* Load i particle coords and add shift vector */
 850         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 851                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 852
 853         fix0             = _mm_setzero_ps();
 854         fiy0             = _mm_setzero_ps();
 855         fiz0             = _mm_setzero_ps();
 856         fix1             = _mm_setzero_ps();
 857         fiy1             = _mm_setzero_ps();
 858         fiz1             = _mm_setzero_ps();
 859         fix2             = _mm_setzero_ps();
 860         fiy2             = _mm_setzero_ps();
 861         fiz2             = _mm_setzero_ps();
 862         fix3             = _mm_setzero_ps();
 863         fiy3             = _mm_setzero_ps();
 864         fiz3             = _mm_setzero_ps();
 865
 866         /* Start inner kernel loop */
 867         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 868         {
 869
 870             /* Get j neighbor index, and coordinate index */
 871             jnrA             = jjnr[jidx];
 872             jnrB             = jjnr[jidx+1];
 873             jnrC             = jjnr[jidx+2];
 874             jnrD             = jjnr[jidx+3];
 875             j_coord_offsetA  = DIM*jnrA;
 876             j_coord_offsetB  = DIM*jnrB;
 877             j_coord_offsetC  = DIM*jnrC;
 878             j_coord_offsetD  = DIM*jnrD;
 879
 880             /* load j atom coordinates */
 881             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 882                                               x+j_coord_offsetC,x+j_coord_offsetD,
 883                                               &jx0,&jy0,&jz0);
 884
 885             /* Calculate displacement vector */
 886             dx00             = _mm_sub_ps(ix0,jx0);
 887             dy00             = _mm_sub_ps(iy0,jy0);
 888             dz00             = _mm_sub_ps(iz0,jz0);
 889             dx10             = _mm_sub_ps(ix1,jx0);
 890             dy10             = _mm_sub_ps(iy1,jy0);
 891             dz10             = _mm_sub_ps(iz1,jz0);
 892             dx20             = _mm_sub_ps(ix2,jx0);
 893             dy20             = _mm_sub_ps(iy2,jy0);
 894             dz20             = _mm_sub_ps(iz2,jz0);
 895             dx30             = _mm_sub_ps(ix3,jx0);
 896             dy30             = _mm_sub_ps(iy3,jy0);
 897             dz30             = _mm_sub_ps(iz3,jz0);
 898
 899             /* Calculate squared distance and things based on it */
 900             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 901             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 902             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 903             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 904
 905             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 906             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 907             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 908             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 909
 910             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 911             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 912             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 913
 914             /* Load parameters for j particles */
 915             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 916                                                               charge+jnrC+0,charge+jnrD+0);
 917             vdwjidx0A        = 2*vdwtype[jnrA+0];
 918             vdwjidx0B        = 2*vdwtype[jnrB+0];
 919             vdwjidx0C        = 2*vdwtype[jnrC+0];
 920             vdwjidx0D        = 2*vdwtype[jnrD+0];
 921
 922             fjx0             = _mm_setzero_ps();
 923             fjy0             = _mm_setzero_ps();
 924             fjz0             = _mm_setzero_ps();
 925
 926             /**************************
 927              * CALCULATE INTERACTIONS *
 928              **************************/
 929
 930             r00              = _mm_mul_ps(rsq00,rinv00);
 931
 932             /* Compute parameters for interactions between i and j atoms */
 933             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 934                                          vdwparam+vdwioffset0+vdwjidx0B,
 935                                          vdwparam+vdwioffset0+vdwjidx0C,
 936                                          vdwparam+vdwioffset0+vdwjidx0D,
 937                                          &c6_00,&c12_00);
 938
 939             /* Calculate table index by multiplying r with table scale and truncate to integer */
 940             rt               = _mm_mul_ps(r00,vftabscale);
 941             vfitab           = _mm_cvttps_epi32(rt);
 942 #ifdef __XOP__
 943             vfeps            = _mm_frcz_ps(rt);
 944 #else
 945             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 946 #endif
 947             twovfeps         = _mm_add_ps(vfeps,vfeps);
 948             vfitab           = _mm_slli_epi32(vfitab,3);
 949
 950             /* CUBIC SPLINE TABLE DISPERSION */
 951             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 952             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 953             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 954             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 955             _MM_TRANSPOSE4_PS(Y,F,G,H);
 956             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 957             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 958             fvdw6            = _mm_mul_ps(c6_00,FF);
 959
 960             /* CUBIC SPLINE TABLE REPULSION */
 961             vfitab           = _mm_add_epi32(vfitab,ifour);
 962             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 963             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 964             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 965             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 966             _MM_TRANSPOSE4_PS(Y,F,G,H);
 967             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 968             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 969             fvdw12           = _mm_mul_ps(c12_00,FF);
 970             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 971
 972             fscal            = fvdw;
 973
 974              /* Update vectorial force */
 975             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 976             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 977             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 978
 979             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 980             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 981             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             if (gmx_mm_any_lt(rsq10,rcutoff2))
 988             {
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq10             = _mm_mul_ps(iq1,jq0);
 992
 993             /* REACTION-FIELD ELECTROSTATICS */
 994             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 995
 996             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 997
 998             fscal            = felec;
 999
1000             fscal            = _mm_and_ps(fscal,cutoff_mask);
1001
1002              /* Update vectorial force */
1003             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1004             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1005             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1006
1007             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1008             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1009             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1010
1011             }
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             if (gmx_mm_any_lt(rsq20,rcutoff2))
1018             {
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             qq20             = _mm_mul_ps(iq2,jq0);
1022
1023             /* REACTION-FIELD ELECTROSTATICS */
1024             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1025
1026             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1027
1028             fscal            = felec;
1029
1030             fscal            = _mm_and_ps(fscal,cutoff_mask);
1031
1032              /* Update vectorial force */
1033             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1034             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1035             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1036
1037             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1038             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1039             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1040
1041             }
1042
1043             /**************************
1044              * CALCULATE INTERACTIONS *
1045              **************************/
1046
1047             if (gmx_mm_any_lt(rsq30,rcutoff2))
1048             {
1049
1050             /* Compute parameters for interactions between i and j atoms */
1051             qq30             = _mm_mul_ps(iq3,jq0);
1052
1053             /* REACTION-FIELD ELECTROSTATICS */
1054             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1055
1056             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1057
1058             fscal            = felec;
1059
1060             fscal            = _mm_and_ps(fscal,cutoff_mask);
1061
1062              /* Update vectorial force */
1063             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1064             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1065             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1066
1067             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1068             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1069             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1070
1071             }
1072
1073             fjptrA             = f+j_coord_offsetA;
1074             fjptrB             = f+j_coord_offsetB;
1075             fjptrC             = f+j_coord_offsetC;
1076             fjptrD             = f+j_coord_offsetD;
1077
1078             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1079
1080             /* Inner loop uses 150 flops */
1081         }
1082
1083         if(jidx<j_index_end)
1084         {
1085
1086             /* Get j neighbor index, and coordinate index */
1087             jnrlistA         = jjnr[jidx];
1088             jnrlistB         = jjnr[jidx+1];
1089             jnrlistC         = jjnr[jidx+2];
1090             jnrlistD         = jjnr[jidx+3];
1091             /* Sign of each element will be negative for non-real atoms.
1092              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1093              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1094              */
1095             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1096             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1097             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1098             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1099             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1100             j_coord_offsetA  = DIM*jnrA;
1101             j_coord_offsetB  = DIM*jnrB;
1102             j_coord_offsetC  = DIM*jnrC;
1103             j_coord_offsetD  = DIM*jnrD;
1104
1105             /* load j atom coordinates */
1106             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1107                                               x+j_coord_offsetC,x+j_coord_offsetD,
1108                                               &jx0,&jy0,&jz0);
1109
1110             /* Calculate displacement vector */
1111             dx00             = _mm_sub_ps(ix0,jx0);
1112             dy00             = _mm_sub_ps(iy0,jy0);
1113             dz00             = _mm_sub_ps(iz0,jz0);
1114             dx10             = _mm_sub_ps(ix1,jx0);
1115             dy10             = _mm_sub_ps(iy1,jy0);
1116             dz10             = _mm_sub_ps(iz1,jz0);
1117             dx20             = _mm_sub_ps(ix2,jx0);
1118             dy20             = _mm_sub_ps(iy2,jy0);
1119             dz20             = _mm_sub_ps(iz2,jz0);
1120             dx30             = _mm_sub_ps(ix3,jx0);
1121             dy30             = _mm_sub_ps(iy3,jy0);
1122             dz30             = _mm_sub_ps(iz3,jz0);
1123
1124             /* Calculate squared distance and things based on it */
1125             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1126             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1127             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1128             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1129
1130             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1131             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1132             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1133             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1134
1135             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1136             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1137             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1138
1139             /* Load parameters for j particles */
1140             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1141                                                               charge+jnrC+0,charge+jnrD+0);
1142             vdwjidx0A        = 2*vdwtype[jnrA+0];
1143             vdwjidx0B        = 2*vdwtype[jnrB+0];
1144             vdwjidx0C        = 2*vdwtype[jnrC+0];
1145             vdwjidx0D        = 2*vdwtype[jnrD+0];
1146
1147             fjx0             = _mm_setzero_ps();
1148             fjy0             = _mm_setzero_ps();
1149             fjz0             = _mm_setzero_ps();
1150
1151             /**************************
1152              * CALCULATE INTERACTIONS *
1153              **************************/
1154
1155             r00              = _mm_mul_ps(rsq00,rinv00);
1156             r00              = _mm_andnot_ps(dummy_mask,r00);
1157
1158             /* Compute parameters for interactions between i and j atoms */
1159             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1160                                          vdwparam+vdwioffset0+vdwjidx0B,
1161                                          vdwparam+vdwioffset0+vdwjidx0C,
1162                                          vdwparam+vdwioffset0+vdwjidx0D,
1163                                          &c6_00,&c12_00);
1164
1165             /* Calculate table index by multiplying r with table scale and truncate to integer */
1166             rt               = _mm_mul_ps(r00,vftabscale);
1167             vfitab           = _mm_cvttps_epi32(rt);
1168 #ifdef __XOP__
1169             vfeps            = _mm_frcz_ps(rt);
1170 #else
1171             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1172 #endif
1173             twovfeps         = _mm_add_ps(vfeps,vfeps);
1174             vfitab           = _mm_slli_epi32(vfitab,3);
1175
1176             /* CUBIC SPLINE TABLE DISPERSION */
1177             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1178             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1179             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1180             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1181             _MM_TRANSPOSE4_PS(Y,F,G,H);
1182             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1183             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1184             fvdw6            = _mm_mul_ps(c6_00,FF);
1185
1186             /* CUBIC SPLINE TABLE REPULSION */
1187             vfitab           = _mm_add_epi32(vfitab,ifour);
1188             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1189             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1190             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1191             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1192             _MM_TRANSPOSE4_PS(Y,F,G,H);
1193             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1194             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1195             fvdw12           = _mm_mul_ps(c12_00,FF);
1196             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1197
1198             fscal            = fvdw;
1199
1200             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1201
1202              /* Update vectorial force */
1203             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1204             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1205             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1206
1207             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1208             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1209             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1210
1211             /**************************
1212              * CALCULATE INTERACTIONS *
1213              **************************/
1214
1215             if (gmx_mm_any_lt(rsq10,rcutoff2))
1216             {
1217
1218             /* Compute parameters for interactions between i and j atoms */
1219             qq10             = _mm_mul_ps(iq1,jq0);
1220
1221             /* REACTION-FIELD ELECTROSTATICS */
1222             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1223
1224             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1225
1226             fscal            = felec;
1227
1228             fscal            = _mm_and_ps(fscal,cutoff_mask);
1229
1230             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1231
1232              /* Update vectorial force */
1233             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1234             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1235             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1236
1237             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1238             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1239             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1240
1241             }
1242
1243             /**************************
1244              * CALCULATE INTERACTIONS *
1245              **************************/
1246
1247             if (gmx_mm_any_lt(rsq20,rcutoff2))
1248             {
1249
1250             /* Compute parameters for interactions between i and j atoms */
1251             qq20             = _mm_mul_ps(iq2,jq0);
1252
1253             /* REACTION-FIELD ELECTROSTATICS */
1254             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1255
1256             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1257
1258             fscal            = felec;
1259
1260             fscal            = _mm_and_ps(fscal,cutoff_mask);
1261
1262             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1263
1264              /* Update vectorial force */
1265             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1266             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1267             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1268
1269             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1270             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1271             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1272
1273             }
1274
1275             /**************************
1276              * CALCULATE INTERACTIONS *
1277              **************************/
1278
1279             if (gmx_mm_any_lt(rsq30,rcutoff2))
1280             {
1281
1282             /* Compute parameters for interactions between i and j atoms */
1283             qq30             = _mm_mul_ps(iq3,jq0);
1284
1285             /* REACTION-FIELD ELECTROSTATICS */
1286             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1287
1288             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1289
1290             fscal            = felec;
1291
1292             fscal            = _mm_and_ps(fscal,cutoff_mask);
1293
1294             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1295
1296              /* Update vectorial force */
1297             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1298             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1299             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1300
1301             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1302             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1303             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1304
1305             }
1306
1307             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1308             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1309             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1310             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1311
1312             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1313
1314             /* Inner loop uses 151 flops */
1315         }
1316
1317         /* End of innermost loop */
1318
1319         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1320                                               f+i_coord_offset,fshift+i_shift_offset);
1321
1322         /* Increment number of inner iterations */
1323         inneriter                  += j_index_end - j_index_start;
1324
1325         /* Outer loop uses 24 flops */
1326     }
1327
1328     /* Increment number of outer iterations */
1329     outeriter        += nri;
1330
1331     /* Update outer/inner flops */
1332
1333     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*151);
1334 }