src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_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_VdwLJSh_GeomP1P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  97     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  98     __m128           dummy_mask,cutoff_mask;
  99     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 100     __m128           one     = _mm_set1_ps(1.0);
 101     __m128           two     = _mm_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_ps(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = _mm_set1_ps(fr->ic->k_rf);
 116     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 117     crf              = _mm_set1_ps(fr->ic->c_rf);
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->rcoulomb;
 124     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 125     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 126
 127     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 128     rvdw             = _mm_set1_ps(fr->rvdw);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     for(iidx=0;iidx<4*DIM;iidx++)
 141     {
 142         scratch[iidx] = 0.0;
 143     }
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 161
 162         fix0             = _mm_setzero_ps();
 163         fiy0             = _mm_setzero_ps();
 164         fiz0             = _mm_setzero_ps();
 165
 166         /* Load parameters for i particles */
 167         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 168         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_ps();
 172         vvdwsum          = _mm_setzero_ps();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             jnrC             = jjnr[jidx+2];
 182             jnrD             = jjnr[jidx+3];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185             j_coord_offsetC  = DIM*jnrC;
 186             j_coord_offsetD  = DIM*jnrD;
 187
 188             /* load j atom coordinates */
 189             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 190                                               x+j_coord_offsetC,x+j_coord_offsetD,
 191                                               &jx0,&jy0,&jz0);
 192
 193             /* Calculate displacement vector */
 194             dx00             = _mm_sub_ps(ix0,jx0);
 195             dy00             = _mm_sub_ps(iy0,jy0);
 196             dz00             = _mm_sub_ps(iz0,jz0);
 197
 198             /* Calculate squared distance and things based on it */
 199             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 200
 201             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 202
 203             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 204
 205             /* Load parameters for j particles */
 206             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 207                                                               charge+jnrC+0,charge+jnrD+0);
 208             vdwjidx0A        = 2*vdwtype[jnrA+0];
 209             vdwjidx0B        = 2*vdwtype[jnrB+0];
 210             vdwjidx0C        = 2*vdwtype[jnrC+0];
 211             vdwjidx0D        = 2*vdwtype[jnrD+0];
 212
 213             /**************************
 214              * CALCULATE INTERACTIONS *
 215              **************************/
 216
 217             if (gmx_mm_any_lt(rsq00,rcutoff2))
 218             {
 219
 220             /* Compute parameters for interactions between i and j atoms */
 221             qq00             = _mm_mul_ps(iq0,jq0);
 222             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 223                                          vdwparam+vdwioffset0+vdwjidx0B,
 224                                          vdwparam+vdwioffset0+vdwjidx0C,
 225                                          vdwparam+vdwioffset0+vdwjidx0D,
 226                                          &c6_00,&c12_00);
 227
 228             /* REACTION-FIELD ELECTROSTATICS */
 229             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 230             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 231
 232             /* LENNARD-JONES DISPERSION/REPULSION */
 233
 234             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 235             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 236             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 237             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 238                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 239             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 240
 241             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             velec            = _mm_and_ps(velec,cutoff_mask);
 245             velecsum         = _mm_add_ps(velecsum,velec);
 246             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 247             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 248
 249             fscal            = _mm_add_ps(felec,fvdw);
 250
 251             fscal            = _mm_and_ps(fscal,cutoff_mask);
 252
 253              /* Update vectorial force */
 254             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 255             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 256             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 257
 258             fjptrA             = f+j_coord_offsetA;
 259             fjptrB             = f+j_coord_offsetB;
 260             fjptrC             = f+j_coord_offsetC;
 261             fjptrD             = f+j_coord_offsetD;
 262             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 263                                                    _mm_mul_ps(dx00,fscal),
 264                                                    _mm_mul_ps(dy00,fscal),
 265                                                    _mm_mul_ps(dz00,fscal));
 266
 267             }
 268
 269             /* Inner loop uses 57 flops */
 270         }
 271
 272         if(jidx<j_index_end)
 273         {
 274
 275             /* Get j neighbor index, and coordinate index */
 276             jnrlistA         = jjnr[jidx];
 277             jnrlistB         = jjnr[jidx+1];
 278             jnrlistC         = jjnr[jidx+2];
 279             jnrlistD         = jjnr[jidx+3];
 280             /* Sign of each element will be negative for non-real atoms.
 281              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 282              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 283              */
 284             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 285             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 286             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 287             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 288             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 289             j_coord_offsetA  = DIM*jnrA;
 290             j_coord_offsetB  = DIM*jnrB;
 291             j_coord_offsetC  = DIM*jnrC;
 292             j_coord_offsetD  = DIM*jnrD;
 293
 294             /* load j atom coordinates */
 295             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 296                                               x+j_coord_offsetC,x+j_coord_offsetD,
 297                                               &jx0,&jy0,&jz0);
 298
 299             /* Calculate displacement vector */
 300             dx00             = _mm_sub_ps(ix0,jx0);
 301             dy00             = _mm_sub_ps(iy0,jy0);
 302             dz00             = _mm_sub_ps(iz0,jz0);
 303
 304             /* Calculate squared distance and things based on it */
 305             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 306
 307             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 308
 309             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 310
 311             /* Load parameters for j particles */
 312             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 313                                                               charge+jnrC+0,charge+jnrD+0);
 314             vdwjidx0A        = 2*vdwtype[jnrA+0];
 315             vdwjidx0B        = 2*vdwtype[jnrB+0];
 316             vdwjidx0C        = 2*vdwtype[jnrC+0];
 317             vdwjidx0D        = 2*vdwtype[jnrD+0];
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             if (gmx_mm_any_lt(rsq00,rcutoff2))
 324             {
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq00             = _mm_mul_ps(iq0,jq0);
 328             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 329                                          vdwparam+vdwioffset0+vdwjidx0B,
 330                                          vdwparam+vdwioffset0+vdwjidx0C,
 331                                          vdwparam+vdwioffset0+vdwjidx0D,
 332                                          &c6_00,&c12_00);
 333
 334             /* REACTION-FIELD ELECTROSTATICS */
 335             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 336             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 337
 338             /* LENNARD-JONES DISPERSION/REPULSION */
 339
 340             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 341             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 342             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 343             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 344                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 345             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 346
 347             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velec            = _mm_and_ps(velec,cutoff_mask);
 351             velec            = _mm_andnot_ps(dummy_mask,velec);
 352             velecsum         = _mm_add_ps(velecsum,velec);
 353             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 354             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 355             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 356
 357             fscal            = _mm_add_ps(felec,fvdw);
 358
 359             fscal            = _mm_and_ps(fscal,cutoff_mask);
 360
 361             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 362
 363              /* Update vectorial force */
 364             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 365             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 366             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 367
 368             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 369             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 370             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 371             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 372             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 373                                                    _mm_mul_ps(dx00,fscal),
 374                                                    _mm_mul_ps(dy00,fscal),
 375                                                    _mm_mul_ps(dz00,fscal));
 376
 377             }
 378
 379             /* Inner loop uses 57 flops */
 380         }
 381
 382         /* End of innermost loop */
 383
 384         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 385                                               f+i_coord_offset,fshift+i_shift_offset);
 386
 387         ggid                        = gid[iidx];
 388         /* Update potential energies */
 389         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 390         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 391
 392         /* Increment number of inner iterations */
 393         inneriter                  += j_index_end - j_index_start;
 394
 395         /* Outer loop uses 9 flops */
 396     }
 397
 398     /* Increment number of outer iterations */
 399     outeriter        += nri;
 400
 401     /* Update outer/inner flops */
 402
 403     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*57);
 404 }
 405 /*
 406  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_128_fma_single
 407  * Electrostatics interaction: ReactionField
 408  * VdW interaction:            LennardJones
 409  * Geometry:                   Particle-Particle
 410  * Calculate force/pot:        Force
 411  */
 412 void
 413 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_128_fma_single
 414                     (t_nblist                    * gmx_restrict       nlist,
 415                      rvec                        * gmx_restrict          xx,
 416                      rvec                        * gmx_restrict          ff,
 417                      t_forcerec                  * gmx_restrict          fr,
 418                      t_mdatoms                   * gmx_restrict     mdatoms,
 419                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 420                      t_nrnb                      * gmx_restrict        nrnb)
 421 {
 422     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 423      * just 0 for non-waters.
 424      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 425      * jnr indices corresponding to data put in the four positions in the SIMD register.
 426      */
 427     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 428     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 429     int              jnrA,jnrB,jnrC,jnrD;
 430     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 431     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 432     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 433     real             rcutoff_scalar;
 434     real             *shiftvec,*fshift,*x,*f;
 435     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 436     real             scratch[4*DIM];
 437     __m128           fscal,rcutoff,rcutoff2,jidxall;
 438     int              vdwioffset0;
 439     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 440     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 441     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 442     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 443     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 444     real             *charge;
 445     int              nvdwtype;
 446     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 447     int              *vdwtype;
 448     real             *vdwparam;
 449     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 450     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 451     __m128           dummy_mask,cutoff_mask;
 452     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 453     __m128           one     = _mm_set1_ps(1.0);
 454     __m128           two     = _mm_set1_ps(2.0);
 455     x                = xx[0];
 456     f                = ff[0];
 457
 458     nri              = nlist->nri;
 459     iinr             = nlist->iinr;
 460     jindex           = nlist->jindex;
 461     jjnr             = nlist->jjnr;
 462     shiftidx         = nlist->shift;
 463     gid              = nlist->gid;
 464     shiftvec         = fr->shift_vec[0];
 465     fshift           = fr->fshift[0];
 466     facel            = _mm_set1_ps(fr->epsfac);
 467     charge           = mdatoms->chargeA;
 468     krf              = _mm_set1_ps(fr->ic->k_rf);
 469     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 470     crf              = _mm_set1_ps(fr->ic->c_rf);
 471     nvdwtype         = fr->ntype;
 472     vdwparam         = fr->nbfp;
 473     vdwtype          = mdatoms->typeA;
 474
 475     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 476     rcutoff_scalar   = fr->rcoulomb;
 477     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 478     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 479
 480     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 481     rvdw             = _mm_set1_ps(fr->rvdw);
 482
 483     /* Avoid stupid compiler warnings */
 484     jnrA = jnrB = jnrC = jnrD = 0;
 485     j_coord_offsetA = 0;
 486     j_coord_offsetB = 0;
 487     j_coord_offsetC = 0;
 488     j_coord_offsetD = 0;
 489
 490     outeriter        = 0;
 491     inneriter        = 0;
 492
 493     for(iidx=0;iidx<4*DIM;iidx++)
 494     {
 495         scratch[iidx] = 0.0;
 496     }
 497
 498     /* Start outer loop over neighborlists */
 499     for(iidx=0; iidx<nri; iidx++)
 500     {
 501         /* Load shift vector for this list */
 502         i_shift_offset   = DIM*shiftidx[iidx];
 503
 504         /* Load limits for loop over neighbors */
 505         j_index_start    = jindex[iidx];
 506         j_index_end      = jindex[iidx+1];
 507
 508         /* Get outer coordinate index */
 509         inr              = iinr[iidx];
 510         i_coord_offset   = DIM*inr;
 511
 512         /* Load i particle coords and add shift vector */
 513         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 514
 515         fix0             = _mm_setzero_ps();
 516         fiy0             = _mm_setzero_ps();
 517         fiz0             = _mm_setzero_ps();
 518
 519         /* Load parameters for i particles */
 520         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 521         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 522
 523         /* Start inner kernel loop */
 524         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 525         {
 526
 527             /* Get j neighbor index, and coordinate index */
 528             jnrA             = jjnr[jidx];
 529             jnrB             = jjnr[jidx+1];
 530             jnrC             = jjnr[jidx+2];
 531             jnrD             = jjnr[jidx+3];
 532             j_coord_offsetA  = DIM*jnrA;
 533             j_coord_offsetB  = DIM*jnrB;
 534             j_coord_offsetC  = DIM*jnrC;
 535             j_coord_offsetD  = DIM*jnrD;
 536
 537             /* load j atom coordinates */
 538             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 539                                               x+j_coord_offsetC,x+j_coord_offsetD,
 540                                               &jx0,&jy0,&jz0);
 541
 542             /* Calculate displacement vector */
 543             dx00             = _mm_sub_ps(ix0,jx0);
 544             dy00             = _mm_sub_ps(iy0,jy0);
 545             dz00             = _mm_sub_ps(iz0,jz0);
 546
 547             /* Calculate squared distance and things based on it */
 548             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 549
 550             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 551
 552             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 553
 554             /* Load parameters for j particles */
 555             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 556                                                               charge+jnrC+0,charge+jnrD+0);
 557             vdwjidx0A        = 2*vdwtype[jnrA+0];
 558             vdwjidx0B        = 2*vdwtype[jnrB+0];
 559             vdwjidx0C        = 2*vdwtype[jnrC+0];
 560             vdwjidx0D        = 2*vdwtype[jnrD+0];
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             if (gmx_mm_any_lt(rsq00,rcutoff2))
 567             {
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq00             = _mm_mul_ps(iq0,jq0);
 571             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 572                                          vdwparam+vdwioffset0+vdwjidx0B,
 573                                          vdwparam+vdwioffset0+vdwjidx0C,
 574                                          vdwparam+vdwioffset0+vdwjidx0D,
 575                                          &c6_00,&c12_00);
 576
 577             /* REACTION-FIELD ELECTROSTATICS */
 578             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 579
 580             /* LENNARD-JONES DISPERSION/REPULSION */
 581
 582             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 583             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 584
 585             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 586
 587             fscal            = _mm_add_ps(felec,fvdw);
 588
 589             fscal            = _mm_and_ps(fscal,cutoff_mask);
 590
 591              /* Update vectorial force */
 592             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 593             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 594             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 595
 596             fjptrA             = f+j_coord_offsetA;
 597             fjptrB             = f+j_coord_offsetB;
 598             fjptrC             = f+j_coord_offsetC;
 599             fjptrD             = f+j_coord_offsetD;
 600             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 601                                                    _mm_mul_ps(dx00,fscal),
 602                                                    _mm_mul_ps(dy00,fscal),
 603                                                    _mm_mul_ps(dz00,fscal));
 604
 605             }
 606
 607             /* Inner loop uses 40 flops */
 608         }
 609
 610         if(jidx<j_index_end)
 611         {
 612
 613             /* Get j neighbor index, and coordinate index */
 614             jnrlistA         = jjnr[jidx];
 615             jnrlistB         = jjnr[jidx+1];
 616             jnrlistC         = jjnr[jidx+2];
 617             jnrlistD         = jjnr[jidx+3];
 618             /* Sign of each element will be negative for non-real atoms.
 619              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 620              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 621              */
 622             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 623             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 624             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 625             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 626             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 627             j_coord_offsetA  = DIM*jnrA;
 628             j_coord_offsetB  = DIM*jnrB;
 629             j_coord_offsetC  = DIM*jnrC;
 630             j_coord_offsetD  = DIM*jnrD;
 631
 632             /* load j atom coordinates */
 633             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 634                                               x+j_coord_offsetC,x+j_coord_offsetD,
 635                                               &jx0,&jy0,&jz0);
 636
 637             /* Calculate displacement vector */
 638             dx00             = _mm_sub_ps(ix0,jx0);
 639             dy00             = _mm_sub_ps(iy0,jy0);
 640             dz00             = _mm_sub_ps(iz0,jz0);
 641
 642             /* Calculate squared distance and things based on it */
 643             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 644
 645             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 646
 647             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 648
 649             /* Load parameters for j particles */
 650             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 651                                                               charge+jnrC+0,charge+jnrD+0);
 652             vdwjidx0A        = 2*vdwtype[jnrA+0];
 653             vdwjidx0B        = 2*vdwtype[jnrB+0];
 654             vdwjidx0C        = 2*vdwtype[jnrC+0];
 655             vdwjidx0D        = 2*vdwtype[jnrD+0];
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             if (gmx_mm_any_lt(rsq00,rcutoff2))
 662             {
 663
 664             /* Compute parameters for interactions between i and j atoms */
 665             qq00             = _mm_mul_ps(iq0,jq0);
 666             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 667                                          vdwparam+vdwioffset0+vdwjidx0B,
 668                                          vdwparam+vdwioffset0+vdwjidx0C,
 669                                          vdwparam+vdwioffset0+vdwjidx0D,
 670                                          &c6_00,&c12_00);
 671
 672             /* REACTION-FIELD ELECTROSTATICS */
 673             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 674
 675             /* LENNARD-JONES DISPERSION/REPULSION */
 676
 677             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 678             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 679
 680             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 681
 682             fscal            = _mm_add_ps(felec,fvdw);
 683
 684             fscal            = _mm_and_ps(fscal,cutoff_mask);
 685
 686             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 687
 688              /* Update vectorial force */
 689             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 690             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 691             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 692
 693             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 694             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 695             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 696             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 697             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 698                                                    _mm_mul_ps(dx00,fscal),
 699                                                    _mm_mul_ps(dy00,fscal),
 700                                                    _mm_mul_ps(dz00,fscal));
 701
 702             }
 703
 704             /* Inner loop uses 40 flops */
 705         }
 706
 707         /* End of innermost loop */
 708
 709         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 710                                               f+i_coord_offset,fshift+i_shift_offset);
 711
 712         /* Increment number of inner iterations */
 713         inneriter                  += j_index_end - j_index_start;
 714
 715         /* Outer loop uses 7 flops */
 716     }
 717
 718     /* Increment number of outer iterations */
 719     outeriter        += nri;
 720
 721     /* Update outer/inner flops */
 722
 723     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*40);
 724 }