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, 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_ElecRFCut_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_ElecRFCut_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 146     rcutoff_scalar   = fr->rcoulomb;
 147     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 148     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 149
 150     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = jnrC = jnrD = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154     j_coord_offsetC = 0;
 155     j_coord_offsetD = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 182
 183         fix0             = _mm_setzero_ps();
 184         fiy0             = _mm_setzero_ps();
 185         fiz0             = _mm_setzero_ps();
 186         fix1             = _mm_setzero_ps();
 187         fiy1             = _mm_setzero_ps();
 188         fiz1             = _mm_setzero_ps();
 189         fix2             = _mm_setzero_ps();
 190         fiy2             = _mm_setzero_ps();
 191         fiz2             = _mm_setzero_ps();
 192         fix3             = _mm_setzero_ps();
 193         fiy3             = _mm_setzero_ps();
 194         fiz3             = _mm_setzero_ps();
 195
 196         /* Reset potential sums */
 197         velecsum         = _mm_setzero_ps();
 198         vvdwsum          = _mm_setzero_ps();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             jnrC             = jjnr[jidx+2];
 208             jnrD             = jjnr[jidx+3];
 209             j_coord_offsetA  = DIM*jnrA;
 210             j_coord_offsetB  = DIM*jnrB;
 211             j_coord_offsetC  = DIM*jnrC;
 212             j_coord_offsetD  = DIM*jnrD;
 213
 214             /* load j atom coordinates */
 215             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                               x+j_coord_offsetC,x+j_coord_offsetD,
 217                                               &jx0,&jy0,&jz0);
 218
 219             /* Calculate displacement vector */
 220             dx00             = _mm_sub_ps(ix0,jx0);
 221             dy00             = _mm_sub_ps(iy0,jy0);
 222             dz00             = _mm_sub_ps(iz0,jz0);
 223             dx10             = _mm_sub_ps(ix1,jx0);
 224             dy10             = _mm_sub_ps(iy1,jy0);
 225             dz10             = _mm_sub_ps(iz1,jz0);
 226             dx20             = _mm_sub_ps(ix2,jx0);
 227             dy20             = _mm_sub_ps(iy2,jy0);
 228             dz20             = _mm_sub_ps(iz2,jz0);
 229             dx30             = _mm_sub_ps(ix3,jx0);
 230             dy30             = _mm_sub_ps(iy3,jy0);
 231             dz30             = _mm_sub_ps(iz3,jz0);
 232
 233             /* Calculate squared distance and things based on it */
 234             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 235             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 236             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 237             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 238
 239             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 240             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 241             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 242             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 243
 244             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 245             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 246             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 247
 248             /* Load parameters for j particles */
 249             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 250                                                               charge+jnrC+0,charge+jnrD+0);
 251             vdwjidx0A        = 2*vdwtype[jnrA+0];
 252             vdwjidx0B        = 2*vdwtype[jnrB+0];
 253             vdwjidx0C        = 2*vdwtype[jnrC+0];
 254             vdwjidx0D        = 2*vdwtype[jnrD+0];
 255
 256             fjx0             = _mm_setzero_ps();
 257             fjy0             = _mm_setzero_ps();
 258             fjz0             = _mm_setzero_ps();
 259
 260             /**************************
 261              * CALCULATE INTERACTIONS *
 262              **************************/
 263
 264             r00              = _mm_mul_ps(rsq00,rinv00);
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 268                                          vdwparam+vdwioffset0+vdwjidx0B,
 269                                          vdwparam+vdwioffset0+vdwjidx0C,
 270                                          vdwparam+vdwioffset0+vdwjidx0D,
 271                                          &c6_00,&c12_00);
 272
 273             /* Calculate table index by multiplying r with table scale and truncate to integer */
 274             rt               = _mm_mul_ps(r00,vftabscale);
 275             vfitab           = _mm_cvttps_epi32(rt);
 276 #ifdef __XOP__
 277             vfeps            = _mm_frcz_ps(rt);
 278 #else
 279             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 280 #endif
 281             twovfeps         = _mm_add_ps(vfeps,vfeps);
 282             vfitab           = _mm_slli_epi32(vfitab,3);
 283
 284             /* CUBIC SPLINE TABLE DISPERSION */
 285             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 286             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 287             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 288             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 289             _MM_TRANSPOSE4_PS(Y,F,G,H);
 290             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 291             VV               = _mm_macc_ps(vfeps,Fp,Y);
 292             vvdw6            = _mm_mul_ps(c6_00,VV);
 293             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 294             fvdw6            = _mm_mul_ps(c6_00,FF);
 295
 296             /* CUBIC SPLINE TABLE REPULSION */
 297             vfitab           = _mm_add_epi32(vfitab,ifour);
 298             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 299             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 300             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 301             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 302             _MM_TRANSPOSE4_PS(Y,F,G,H);
 303             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 304             VV               = _mm_macc_ps(vfeps,Fp,Y);
 305             vvdw12           = _mm_mul_ps(c12_00,VV);
 306             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 307             fvdw12           = _mm_mul_ps(c12_00,FF);
 308             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 309             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 313
 314             fscal            = fvdw;
 315
 316              /* Update vectorial force */
 317             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 318             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 319             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 320
 321             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 322             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 323             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             if (gmx_mm_any_lt(rsq10,rcutoff2))
 330             {
 331
 332             /* Compute parameters for interactions between i and j atoms */
 333             qq10             = _mm_mul_ps(iq1,jq0);
 334
 335             /* REACTION-FIELD ELECTROSTATICS */
 336             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 337             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 338
 339             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velec            = _mm_and_ps(velec,cutoff_mask);
 343             velecsum         = _mm_add_ps(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             fscal            = _mm_and_ps(fscal,cutoff_mask);
 348
 349              /* Update vectorial force */
 350             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 351             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 352             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 353
 354             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 355             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 356             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 357
 358             }
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             if (gmx_mm_any_lt(rsq20,rcutoff2))
 365             {
 366
 367             /* Compute parameters for interactions between i and j atoms */
 368             qq20             = _mm_mul_ps(iq2,jq0);
 369
 370             /* REACTION-FIELD ELECTROSTATICS */
 371             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 372             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 373
 374             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velec            = _mm_and_ps(velec,cutoff_mask);
 378             velecsum         = _mm_add_ps(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382             fscal            = _mm_and_ps(fscal,cutoff_mask);
 383
 384              /* Update vectorial force */
 385             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 386             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 387             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 388
 389             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 390             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 391             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 392
 393             }
 394
 395             /**************************
 396              * CALCULATE INTERACTIONS *
 397              **************************/
 398
 399             if (gmx_mm_any_lt(rsq30,rcutoff2))
 400             {
 401
 402             /* Compute parameters for interactions between i and j atoms */
 403             qq30             = _mm_mul_ps(iq3,jq0);
 404
 405             /* REACTION-FIELD ELECTROSTATICS */
 406             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 407             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 408
 409             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velec            = _mm_and_ps(velec,cutoff_mask);
 413             velecsum         = _mm_add_ps(velecsum,velec);
 414
 415             fscal            = felec;
 416
 417             fscal            = _mm_and_ps(fscal,cutoff_mask);
 418
 419              /* Update vectorial force */
 420             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 421             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 422             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 423
 424             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 425             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 426             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 427
 428             }
 429
 430             fjptrA             = f+j_coord_offsetA;
 431             fjptrB             = f+j_coord_offsetB;
 432             fjptrC             = f+j_coord_offsetC;
 433             fjptrD             = f+j_coord_offsetD;
 434
 435             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 436
 437             /* Inner loop uses 176 flops */
 438         }
 439
 440         if(jidx<j_index_end)
 441         {
 442
 443             /* Get j neighbor index, and coordinate index */
 444             jnrlistA         = jjnr[jidx];
 445             jnrlistB         = jjnr[jidx+1];
 446             jnrlistC         = jjnr[jidx+2];
 447             jnrlistD         = jjnr[jidx+3];
 448             /* Sign of each element will be negative for non-real atoms.
 449              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 450              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 451              */
 452             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 453             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 454             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 455             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 456             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 457             j_coord_offsetA  = DIM*jnrA;
 458             j_coord_offsetB  = DIM*jnrB;
 459             j_coord_offsetC  = DIM*jnrC;
 460             j_coord_offsetD  = DIM*jnrD;
 461
 462             /* load j atom coordinates */
 463             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 464                                               x+j_coord_offsetC,x+j_coord_offsetD,
 465                                               &jx0,&jy0,&jz0);
 466
 467             /* Calculate displacement vector */
 468             dx00             = _mm_sub_ps(ix0,jx0);
 469             dy00             = _mm_sub_ps(iy0,jy0);
 470             dz00             = _mm_sub_ps(iz0,jz0);
 471             dx10             = _mm_sub_ps(ix1,jx0);
 472             dy10             = _mm_sub_ps(iy1,jy0);
 473             dz10             = _mm_sub_ps(iz1,jz0);
 474             dx20             = _mm_sub_ps(ix2,jx0);
 475             dy20             = _mm_sub_ps(iy2,jy0);
 476             dz20             = _mm_sub_ps(iz2,jz0);
 477             dx30             = _mm_sub_ps(ix3,jx0);
 478             dy30             = _mm_sub_ps(iy3,jy0);
 479             dz30             = _mm_sub_ps(iz3,jz0);
 480
 481             /* Calculate squared distance and things based on it */
 482             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 483             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 484             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 485             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 486
 487             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 488             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 489             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 490             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 491
 492             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 493             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 494             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 495
 496             /* Load parameters for j particles */
 497             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 498                                                               charge+jnrC+0,charge+jnrD+0);
 499             vdwjidx0A        = 2*vdwtype[jnrA+0];
 500             vdwjidx0B        = 2*vdwtype[jnrB+0];
 501             vdwjidx0C        = 2*vdwtype[jnrC+0];
 502             vdwjidx0D        = 2*vdwtype[jnrD+0];
 503
 504             fjx0             = _mm_setzero_ps();
 505             fjy0             = _mm_setzero_ps();
 506             fjz0             = _mm_setzero_ps();
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             r00              = _mm_mul_ps(rsq00,rinv00);
 513             r00              = _mm_andnot_ps(dummy_mask,r00);
 514
 515             /* Compute parameters for interactions between i and j atoms */
 516             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 517                                          vdwparam+vdwioffset0+vdwjidx0B,
 518                                          vdwparam+vdwioffset0+vdwjidx0C,
 519                                          vdwparam+vdwioffset0+vdwjidx0D,
 520                                          &c6_00,&c12_00);
 521
 522             /* Calculate table index by multiplying r with table scale and truncate to integer */
 523             rt               = _mm_mul_ps(r00,vftabscale);
 524             vfitab           = _mm_cvttps_epi32(rt);
 525 #ifdef __XOP__
 526             vfeps            = _mm_frcz_ps(rt);
 527 #else
 528             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 529 #endif
 530             twovfeps         = _mm_add_ps(vfeps,vfeps);
 531             vfitab           = _mm_slli_epi32(vfitab,3);
 532
 533             /* CUBIC SPLINE TABLE DISPERSION */
 534             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 535             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 536             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 537             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 538             _MM_TRANSPOSE4_PS(Y,F,G,H);
 539             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 540             VV               = _mm_macc_ps(vfeps,Fp,Y);
 541             vvdw6            = _mm_mul_ps(c6_00,VV);
 542             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 543             fvdw6            = _mm_mul_ps(c6_00,FF);
 544
 545             /* CUBIC SPLINE TABLE REPULSION */
 546             vfitab           = _mm_add_epi32(vfitab,ifour);
 547             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 548             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 549             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 550             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 551             _MM_TRANSPOSE4_PS(Y,F,G,H);
 552             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 553             VV               = _mm_macc_ps(vfeps,Fp,Y);
 554             vvdw12           = _mm_mul_ps(c12_00,VV);
 555             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 556             fvdw12           = _mm_mul_ps(c12_00,FF);
 557             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 558             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 562             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 563
 564             fscal            = fvdw;
 565
 566             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 567
 568              /* Update vectorial force */
 569             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 570             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 571             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 572
 573             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 574             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 575             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             if (gmx_mm_any_lt(rsq10,rcutoff2))
 582             {
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq10             = _mm_mul_ps(iq1,jq0);
 586
 587             /* REACTION-FIELD ELECTROSTATICS */
 588             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 589             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 590
 591             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velec            = _mm_and_ps(velec,cutoff_mask);
 595             velec            = _mm_andnot_ps(dummy_mask,velec);
 596             velecsum         = _mm_add_ps(velecsum,velec);
 597
 598             fscal            = felec;
 599
 600             fscal            = _mm_and_ps(fscal,cutoff_mask);
 601
 602             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 603
 604              /* Update vectorial force */
 605             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 606             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 607             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 608
 609             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 610             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 611             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 612
 613             }
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             if (gmx_mm_any_lt(rsq20,rcutoff2))
 620             {
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq20             = _mm_mul_ps(iq2,jq0);
 624
 625             /* REACTION-FIELD ELECTROSTATICS */
 626             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 627             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 628
 629             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 630
 631             /* Update potential sum for this i atom from the interaction with this j atom. */
 632             velec            = _mm_and_ps(velec,cutoff_mask);
 633             velec            = _mm_andnot_ps(dummy_mask,velec);
 634             velecsum         = _mm_add_ps(velecsum,velec);
 635
 636             fscal            = felec;
 637
 638             fscal            = _mm_and_ps(fscal,cutoff_mask);
 639
 640             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 641
 642              /* Update vectorial force */
 643             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 644             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 645             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 646
 647             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 648             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 649             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 650
 651             }
 652
 653             /**************************
 654              * CALCULATE INTERACTIONS *
 655              **************************/
 656
 657             if (gmx_mm_any_lt(rsq30,rcutoff2))
 658             {
 659
 660             /* Compute parameters for interactions between i and j atoms */
 661             qq30             = _mm_mul_ps(iq3,jq0);
 662
 663             /* REACTION-FIELD ELECTROSTATICS */
 664             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 665             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 666
 667             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 668
 669             /* Update potential sum for this i atom from the interaction with this j atom. */
 670             velec            = _mm_and_ps(velec,cutoff_mask);
 671             velec            = _mm_andnot_ps(dummy_mask,velec);
 672             velecsum         = _mm_add_ps(velecsum,velec);
 673
 674             fscal            = felec;
 675
 676             fscal            = _mm_and_ps(fscal,cutoff_mask);
 677
 678             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 679
 680              /* Update vectorial force */
 681             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 682             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 683             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 684
 685             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 686             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 687             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 688
 689             }
 690
 691             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 692             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 693             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 694             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 695
 696             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 697
 698             /* Inner loop uses 177 flops */
 699         }
 700
 701         /* End of innermost loop */
 702
 703         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 704                                               f+i_coord_offset,fshift+i_shift_offset);
 705
 706         ggid                        = gid[iidx];
 707         /* Update potential energies */
 708         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 709         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 710
 711         /* Increment number of inner iterations */
 712         inneriter                  += j_index_end - j_index_start;
 713
 714         /* Outer loop uses 26 flops */
 715     }
 716
 717     /* Increment number of outer iterations */
 718     outeriter        += nri;
 719
 720     /* Update outer/inner flops */
 721
 722     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*177);
 723 }
 724 /*
 725  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 726  * Electrostatics interaction: ReactionField
 727  * VdW interaction:            CubicSplineTable
 728  * Geometry:                   Water4-Particle
 729  * Calculate force/pot:        Force
 730  */
 731 void
 732 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 733                     (t_nblist                    * gmx_restrict       nlist,
 734                      rvec                        * gmx_restrict          xx,
 735                      rvec                        * gmx_restrict          ff,
 736                      t_forcerec                  * gmx_restrict          fr,
 737                      t_mdatoms                   * gmx_restrict     mdatoms,
 738                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 739                      t_nrnb                      * gmx_restrict        nrnb)
 740 {
 741     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 742      * just 0 for non-waters.
 743      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 744      * jnr indices corresponding to data put in the four positions in the SIMD register.
 745      */
 746     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 747     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 748     int              jnrA,jnrB,jnrC,jnrD;
 749     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 750     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 751     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 752     real             rcutoff_scalar;
 753     real             *shiftvec,*fshift,*x,*f;
 754     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 755     real             scratch[4*DIM];
 756     __m128           fscal,rcutoff,rcutoff2,jidxall;
 757     int              vdwioffset0;
 758     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 759     int              vdwioffset1;
 760     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 761     int              vdwioffset2;
 762     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 763     int              vdwioffset3;
 764     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 765     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 766     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 767     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 768     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 769     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 770     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 771     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 772     real             *charge;
 773     int              nvdwtype;
 774     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 775     int              *vdwtype;
 776     real             *vdwparam;
 777     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 778     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 779     __m128i          vfitab;
 780     __m128i          ifour       = _mm_set1_epi32(4);
 781     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 782     real             *vftab;
 783     __m128           dummy_mask,cutoff_mask;
 784     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 785     __m128           one     = _mm_set1_ps(1.0);
 786     __m128           two     = _mm_set1_ps(2.0);
 787     x                = xx[0];
 788     f                = ff[0];
 789
 790     nri              = nlist->nri;
 791     iinr             = nlist->iinr;
 792     jindex           = nlist->jindex;
 793     jjnr             = nlist->jjnr;
 794     shiftidx         = nlist->shift;
 795     gid              = nlist->gid;
 796     shiftvec         = fr->shift_vec[0];
 797     fshift           = fr->fshift[0];
 798     facel            = _mm_set1_ps(fr->epsfac);
 799     charge           = mdatoms->chargeA;
 800     krf              = _mm_set1_ps(fr->ic->k_rf);
 801     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 802     crf              = _mm_set1_ps(fr->ic->c_rf);
 803     nvdwtype         = fr->ntype;
 804     vdwparam         = fr->nbfp;
 805     vdwtype          = mdatoms->typeA;
 806
 807     vftab            = kernel_data->table_vdw->data;
 808     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 809
 810     /* Setup water-specific parameters */
 811     inr              = nlist->iinr[0];
 812     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 813     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 814     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 815     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 816
 817     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 818     rcutoff_scalar   = fr->rcoulomb;
 819     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 820     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 821
 822     /* Avoid stupid compiler warnings */
 823     jnrA = jnrB = jnrC = jnrD = 0;
 824     j_coord_offsetA = 0;
 825     j_coord_offsetB = 0;
 826     j_coord_offsetC = 0;
 827     j_coord_offsetD = 0;
 828
 829     outeriter        = 0;
 830     inneriter        = 0;
 831
 832     for(iidx=0;iidx<4*DIM;iidx++)
 833     {
 834         scratch[iidx] = 0.0;
 835     }
 836
 837     /* Start outer loop over neighborlists */
 838     for(iidx=0; iidx<nri; iidx++)
 839     {
 840         /* Load shift vector for this list */
 841         i_shift_offset   = DIM*shiftidx[iidx];
 842
 843         /* Load limits for loop over neighbors */
 844         j_index_start    = jindex[iidx];
 845         j_index_end      = jindex[iidx+1];
 846
 847         /* Get outer coordinate index */
 848         inr              = iinr[iidx];
 849         i_coord_offset   = DIM*inr;
 850
 851         /* Load i particle coords and add shift vector */
 852         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 853                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 854
 855         fix0             = _mm_setzero_ps();
 856         fiy0             = _mm_setzero_ps();
 857         fiz0             = _mm_setzero_ps();
 858         fix1             = _mm_setzero_ps();
 859         fiy1             = _mm_setzero_ps();
 860         fiz1             = _mm_setzero_ps();
 861         fix2             = _mm_setzero_ps();
 862         fiy2             = _mm_setzero_ps();
 863         fiz2             = _mm_setzero_ps();
 864         fix3             = _mm_setzero_ps();
 865         fiy3             = _mm_setzero_ps();
 866         fiz3             = _mm_setzero_ps();
 867
 868         /* Start inner kernel loop */
 869         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 870         {
 871
 872             /* Get j neighbor index, and coordinate index */
 873             jnrA             = jjnr[jidx];
 874             jnrB             = jjnr[jidx+1];
 875             jnrC             = jjnr[jidx+2];
 876             jnrD             = jjnr[jidx+3];
 877             j_coord_offsetA  = DIM*jnrA;
 878             j_coord_offsetB  = DIM*jnrB;
 879             j_coord_offsetC  = DIM*jnrC;
 880             j_coord_offsetD  = DIM*jnrD;
 881
 882             /* load j atom coordinates */
 883             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 884                                               x+j_coord_offsetC,x+j_coord_offsetD,
 885                                               &jx0,&jy0,&jz0);
 886
 887             /* Calculate displacement vector */
 888             dx00             = _mm_sub_ps(ix0,jx0);
 889             dy00             = _mm_sub_ps(iy0,jy0);
 890             dz00             = _mm_sub_ps(iz0,jz0);
 891             dx10             = _mm_sub_ps(ix1,jx0);
 892             dy10             = _mm_sub_ps(iy1,jy0);
 893             dz10             = _mm_sub_ps(iz1,jz0);
 894             dx20             = _mm_sub_ps(ix2,jx0);
 895             dy20             = _mm_sub_ps(iy2,jy0);
 896             dz20             = _mm_sub_ps(iz2,jz0);
 897             dx30             = _mm_sub_ps(ix3,jx0);
 898             dy30             = _mm_sub_ps(iy3,jy0);
 899             dz30             = _mm_sub_ps(iz3,jz0);
 900
 901             /* Calculate squared distance and things based on it */
 902             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 903             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 904             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 905             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 906
 907             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 908             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 909             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 910             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 911
 912             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 913             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 914             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 915
 916             /* Load parameters for j particles */
 917             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 918                                                               charge+jnrC+0,charge+jnrD+0);
 919             vdwjidx0A        = 2*vdwtype[jnrA+0];
 920             vdwjidx0B        = 2*vdwtype[jnrB+0];
 921             vdwjidx0C        = 2*vdwtype[jnrC+0];
 922             vdwjidx0D        = 2*vdwtype[jnrD+0];
 923
 924             fjx0             = _mm_setzero_ps();
 925             fjy0             = _mm_setzero_ps();
 926             fjz0             = _mm_setzero_ps();
 927
 928             /**************************
 929              * CALCULATE INTERACTIONS *
 930              **************************/
 931
 932             r00              = _mm_mul_ps(rsq00,rinv00);
 933
 934             /* Compute parameters for interactions between i and j atoms */
 935             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 936                                          vdwparam+vdwioffset0+vdwjidx0B,
 937                                          vdwparam+vdwioffset0+vdwjidx0C,
 938                                          vdwparam+vdwioffset0+vdwjidx0D,
 939                                          &c6_00,&c12_00);
 940
 941             /* Calculate table index by multiplying r with table scale and truncate to integer */
 942             rt               = _mm_mul_ps(r00,vftabscale);
 943             vfitab           = _mm_cvttps_epi32(rt);
 944 #ifdef __XOP__
 945             vfeps            = _mm_frcz_ps(rt);
 946 #else
 947             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 948 #endif
 949             twovfeps         = _mm_add_ps(vfeps,vfeps);
 950             vfitab           = _mm_slli_epi32(vfitab,3);
 951
 952             /* CUBIC SPLINE TABLE DISPERSION */
 953             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 954             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 955             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 956             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 957             _MM_TRANSPOSE4_PS(Y,F,G,H);
 958             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 959             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 960             fvdw6            = _mm_mul_ps(c6_00,FF);
 961
 962             /* CUBIC SPLINE TABLE REPULSION */
 963             vfitab           = _mm_add_epi32(vfitab,ifour);
 964             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 965             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 966             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 967             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 968             _MM_TRANSPOSE4_PS(Y,F,G,H);
 969             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 970             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 971             fvdw12           = _mm_mul_ps(c12_00,FF);
 972             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 973
 974             fscal            = fvdw;
 975
 976              /* Update vectorial force */
 977             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 978             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 979             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 980
 981             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 982             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 983             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             if (gmx_mm_any_lt(rsq10,rcutoff2))
 990             {
 991
 992             /* Compute parameters for interactions between i and j atoms */
 993             qq10             = _mm_mul_ps(iq1,jq0);
 994
 995             /* REACTION-FIELD ELECTROSTATICS */
 996             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 997
 998             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 999
1000             fscal            = felec;
1001
1002             fscal            = _mm_and_ps(fscal,cutoff_mask);
1003
1004              /* Update vectorial force */
1005             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1006             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1007             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1008
1009             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1010             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1011             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1012
1013             }
1014
1015             /**************************
1016              * CALCULATE INTERACTIONS *
1017              **************************/
1018
1019             if (gmx_mm_any_lt(rsq20,rcutoff2))
1020             {
1021
1022             /* Compute parameters for interactions between i and j atoms */
1023             qq20             = _mm_mul_ps(iq2,jq0);
1024
1025             /* REACTION-FIELD ELECTROSTATICS */
1026             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1027
1028             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1029
1030             fscal            = felec;
1031
1032             fscal            = _mm_and_ps(fscal,cutoff_mask);
1033
1034              /* Update vectorial force */
1035             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1036             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1037             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1038
1039             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1040             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1041             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1042
1043             }
1044
1045             /**************************
1046              * CALCULATE INTERACTIONS *
1047              **************************/
1048
1049             if (gmx_mm_any_lt(rsq30,rcutoff2))
1050             {
1051
1052             /* Compute parameters for interactions between i and j atoms */
1053             qq30             = _mm_mul_ps(iq3,jq0);
1054
1055             /* REACTION-FIELD ELECTROSTATICS */
1056             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1057
1058             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1059
1060             fscal            = felec;
1061
1062             fscal            = _mm_and_ps(fscal,cutoff_mask);
1063
1064              /* Update vectorial force */
1065             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1066             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1067             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1068
1069             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1070             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1071             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1072
1073             }
1074
1075             fjptrA             = f+j_coord_offsetA;
1076             fjptrB             = f+j_coord_offsetB;
1077             fjptrC             = f+j_coord_offsetC;
1078             fjptrD             = f+j_coord_offsetD;
1079
1080             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1081
1082             /* Inner loop uses 150 flops */
1083         }
1084
1085         if(jidx<j_index_end)
1086         {
1087
1088             /* Get j neighbor index, and coordinate index */
1089             jnrlistA         = jjnr[jidx];
1090             jnrlistB         = jjnr[jidx+1];
1091             jnrlistC         = jjnr[jidx+2];
1092             jnrlistD         = jjnr[jidx+3];
1093             /* Sign of each element will be negative for non-real atoms.
1094              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1095              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1096              */
1097             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1098             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1099             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1100             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1101             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1102             j_coord_offsetA  = DIM*jnrA;
1103             j_coord_offsetB  = DIM*jnrB;
1104             j_coord_offsetC  = DIM*jnrC;
1105             j_coord_offsetD  = DIM*jnrD;
1106
1107             /* load j atom coordinates */
1108             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1109                                               x+j_coord_offsetC,x+j_coord_offsetD,
1110                                               &jx0,&jy0,&jz0);
1111
1112             /* Calculate displacement vector */
1113             dx00             = _mm_sub_ps(ix0,jx0);
1114             dy00             = _mm_sub_ps(iy0,jy0);
1115             dz00             = _mm_sub_ps(iz0,jz0);
1116             dx10             = _mm_sub_ps(ix1,jx0);
1117             dy10             = _mm_sub_ps(iy1,jy0);
1118             dz10             = _mm_sub_ps(iz1,jz0);
1119             dx20             = _mm_sub_ps(ix2,jx0);
1120             dy20             = _mm_sub_ps(iy2,jy0);
1121             dz20             = _mm_sub_ps(iz2,jz0);
1122             dx30             = _mm_sub_ps(ix3,jx0);
1123             dy30             = _mm_sub_ps(iy3,jy0);
1124             dz30             = _mm_sub_ps(iz3,jz0);
1125
1126             /* Calculate squared distance and things based on it */
1127             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1128             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1129             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1130             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1131
1132             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1133             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1134             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1135             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1136
1137             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1138             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1139             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1140
1141             /* Load parameters for j particles */
1142             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1143                                                               charge+jnrC+0,charge+jnrD+0);
1144             vdwjidx0A        = 2*vdwtype[jnrA+0];
1145             vdwjidx0B        = 2*vdwtype[jnrB+0];
1146             vdwjidx0C        = 2*vdwtype[jnrC+0];
1147             vdwjidx0D        = 2*vdwtype[jnrD+0];
1148
1149             fjx0             = _mm_setzero_ps();
1150             fjy0             = _mm_setzero_ps();
1151             fjz0             = _mm_setzero_ps();
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             r00              = _mm_mul_ps(rsq00,rinv00);
1158             r00              = _mm_andnot_ps(dummy_mask,r00);
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1162                                          vdwparam+vdwioffset0+vdwjidx0B,
1163                                          vdwparam+vdwioffset0+vdwjidx0C,
1164                                          vdwparam+vdwioffset0+vdwjidx0D,
1165                                          &c6_00,&c12_00);
1166
1167             /* Calculate table index by multiplying r with table scale and truncate to integer */
1168             rt               = _mm_mul_ps(r00,vftabscale);
1169             vfitab           = _mm_cvttps_epi32(rt);
1170 #ifdef __XOP__
1171             vfeps            = _mm_frcz_ps(rt);
1172 #else
1173             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1174 #endif
1175             twovfeps         = _mm_add_ps(vfeps,vfeps);
1176             vfitab           = _mm_slli_epi32(vfitab,3);
1177
1178             /* CUBIC SPLINE TABLE DISPERSION */
1179             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1180             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1181             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1182             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1183             _MM_TRANSPOSE4_PS(Y,F,G,H);
1184             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1185             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1186             fvdw6            = _mm_mul_ps(c6_00,FF);
1187
1188             /* CUBIC SPLINE TABLE REPULSION */
1189             vfitab           = _mm_add_epi32(vfitab,ifour);
1190             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1191             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1192             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1193             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1194             _MM_TRANSPOSE4_PS(Y,F,G,H);
1195             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1196             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1197             fvdw12           = _mm_mul_ps(c12_00,FF);
1198             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1199
1200             fscal            = fvdw;
1201
1202             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1203
1204              /* Update vectorial force */
1205             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1206             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1207             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1208
1209             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1210             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1211             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1212
1213             /**************************
1214              * CALCULATE INTERACTIONS *
1215              **************************/
1216
1217             if (gmx_mm_any_lt(rsq10,rcutoff2))
1218             {
1219
1220             /* Compute parameters for interactions between i and j atoms */
1221             qq10             = _mm_mul_ps(iq1,jq0);
1222
1223             /* REACTION-FIELD ELECTROSTATICS */
1224             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1225
1226             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1227
1228             fscal            = felec;
1229
1230             fscal            = _mm_and_ps(fscal,cutoff_mask);
1231
1232             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1233
1234              /* Update vectorial force */
1235             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1236             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1237             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1238
1239             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1240             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1241             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1242
1243             }
1244
1245             /**************************
1246              * CALCULATE INTERACTIONS *
1247              **************************/
1248
1249             if (gmx_mm_any_lt(rsq20,rcutoff2))
1250             {
1251
1252             /* Compute parameters for interactions between i and j atoms */
1253             qq20             = _mm_mul_ps(iq2,jq0);
1254
1255             /* REACTION-FIELD ELECTROSTATICS */
1256             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1257
1258             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1259
1260             fscal            = felec;
1261
1262             fscal            = _mm_and_ps(fscal,cutoff_mask);
1263
1264             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1265
1266              /* Update vectorial force */
1267             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1268             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1269             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1270
1271             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1272             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1273             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1274
1275             }
1276
1277             /**************************
1278              * CALCULATE INTERACTIONS *
1279              **************************/
1280
1281             if (gmx_mm_any_lt(rsq30,rcutoff2))
1282             {
1283
1284             /* Compute parameters for interactions between i and j atoms */
1285             qq30             = _mm_mul_ps(iq3,jq0);
1286
1287             /* REACTION-FIELD ELECTROSTATICS */
1288             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1289
1290             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1291
1292             fscal            = felec;
1293
1294             fscal            = _mm_and_ps(fscal,cutoff_mask);
1295
1296             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1297
1298              /* Update vectorial force */
1299             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1300             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1301             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1302
1303             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1304             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1305             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1306
1307             }
1308
1309             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1310             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1311             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1312             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1313
1314             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1315
1316             /* Inner loop uses 151 flops */
1317         }
1318
1319         /* End of innermost loop */
1320
1321         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1322                                               f+i_coord_offset,fshift+i_shift_offset);
1323
1324         /* Increment number of inner iterations */
1325         inneriter                  += j_index_end - j_index_start;
1326
1327         /* Outer loop uses 24 flops */
1328     }
1329
1330     /* Increment number of outer iterations */
1331     outeriter        += nri;
1332
1333     /* Update outer/inner flops */
1334
1335     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*151);
1336 }