src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  72     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  73     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  74     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  75     real             *charge;
  76     int              nvdwtype;
  77     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  78     int              *vdwtype;
  79     real             *vdwparam;
  80     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  81     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  82     __m128           dummy_mask,cutoff_mask;
  83     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  84     __m128           one     = _mm_set1_ps(1.0);
  85     __m128           two     = _mm_set1_ps(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     facel            = _mm_set1_ps(fr->epsfac);
  98     charge           = mdatoms->chargeA;
  99     krf              = _mm_set1_ps(fr->ic->k_rf);
 100     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 101     crf              = _mm_set1_ps(fr->ic->c_rf);
 102     nvdwtype         = fr->ntype;
 103     vdwparam         = fr->nbfp;
 104     vdwtype          = mdatoms->typeA;
 105
 106     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 107     rcutoff_scalar   = fr->rcoulomb;
 108     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 109     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 110
 111     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 112     rvdw             = _mm_set1_ps(fr->rvdw);
 113
 114     /* Avoid stupid compiler warnings */
 115     jnrA = jnrB = jnrC = jnrD = 0;
 116     j_coord_offsetA = 0;
 117     j_coord_offsetB = 0;
 118     j_coord_offsetC = 0;
 119     j_coord_offsetD = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     for(iidx=0;iidx<4*DIM;iidx++)
 125     {
 126         scratch[iidx] = 0.0;
 127     }
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 145
 146         fix0             = _mm_setzero_ps();
 147         fiy0             = _mm_setzero_ps();
 148         fiz0             = _mm_setzero_ps();
 149
 150         /* Load parameters for i particles */
 151         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 152         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm_setzero_ps();
 156         vvdwsum          = _mm_setzero_ps();
 157
 158         /* Start inner kernel loop */
 159         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 160         {
 161
 162             /* Get j neighbor index, and coordinate index */
 163             jnrA             = jjnr[jidx];
 164             jnrB             = jjnr[jidx+1];
 165             jnrC             = jjnr[jidx+2];
 166             jnrD             = jjnr[jidx+3];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169             j_coord_offsetC  = DIM*jnrC;
 170             j_coord_offsetD  = DIM*jnrD;
 171
 172             /* load j atom coordinates */
 173             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 174                                               x+j_coord_offsetC,x+j_coord_offsetD,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx00             = _mm_sub_ps(ix0,jx0);
 179             dy00             = _mm_sub_ps(iy0,jy0);
 180             dz00             = _mm_sub_ps(iz0,jz0);
 181
 182             /* Calculate squared distance and things based on it */
 183             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 184
 185             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 186
 187             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 188
 189             /* Load parameters for j particles */
 190             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 191                                                               charge+jnrC+0,charge+jnrD+0);
 192             vdwjidx0A        = 2*vdwtype[jnrA+0];
 193             vdwjidx0B        = 2*vdwtype[jnrB+0];
 194             vdwjidx0C        = 2*vdwtype[jnrC+0];
 195             vdwjidx0D        = 2*vdwtype[jnrD+0];
 196
 197             /**************************
 198              * CALCULATE INTERACTIONS *
 199              **************************/
 200
 201             if (gmx_mm_any_lt(rsq00,rcutoff2))
 202             {
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             qq00             = _mm_mul_ps(iq0,jq0);
 206             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 207                                          vdwparam+vdwioffset0+vdwjidx0B,
 208                                          vdwparam+vdwioffset0+vdwjidx0C,
 209                                          vdwparam+vdwioffset0+vdwjidx0D,
 210                                          &c6_00,&c12_00);
 211
 212             /* REACTION-FIELD ELECTROSTATICS */
 213             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 214             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 215
 216             /* LENNARD-JONES DISPERSION/REPULSION */
 217
 218             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 219             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 220             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 221             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 222                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 223             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 224
 225             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             velec            = _mm_and_ps(velec,cutoff_mask);
 229             velecsum         = _mm_add_ps(velecsum,velec);
 230             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 231             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 232
 233             fscal            = _mm_add_ps(felec,fvdw);
 234
 235             fscal            = _mm_and_ps(fscal,cutoff_mask);
 236
 237              /* Update vectorial force */
 238             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 239             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 240             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 241
 242             fjptrA             = f+j_coord_offsetA;
 243             fjptrB             = f+j_coord_offsetB;
 244             fjptrC             = f+j_coord_offsetC;
 245             fjptrD             = f+j_coord_offsetD;
 246             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 247                                                    _mm_mul_ps(dx00,fscal),
 248                                                    _mm_mul_ps(dy00,fscal),
 249                                                    _mm_mul_ps(dz00,fscal));
 250
 251             }
 252
 253             /* Inner loop uses 57 flops */
 254         }
 255
 256         if(jidx<j_index_end)
 257         {
 258
 259             /* Get j neighbor index, and coordinate index */
 260             jnrlistA         = jjnr[jidx];
 261             jnrlistB         = jjnr[jidx+1];
 262             jnrlistC         = jjnr[jidx+2];
 263             jnrlistD         = jjnr[jidx+3];
 264             /* Sign of each element will be negative for non-real atoms.
 265              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 266              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 267              */
 268             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 269             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 270             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 271             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 272             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 273             j_coord_offsetA  = DIM*jnrA;
 274             j_coord_offsetB  = DIM*jnrB;
 275             j_coord_offsetC  = DIM*jnrC;
 276             j_coord_offsetD  = DIM*jnrD;
 277
 278             /* load j atom coordinates */
 279             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 280                                               x+j_coord_offsetC,x+j_coord_offsetD,
 281                                               &jx0,&jy0,&jz0);
 282
 283             /* Calculate displacement vector */
 284             dx00             = _mm_sub_ps(ix0,jx0);
 285             dy00             = _mm_sub_ps(iy0,jy0);
 286             dz00             = _mm_sub_ps(iz0,jz0);
 287
 288             /* Calculate squared distance and things based on it */
 289             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 290
 291             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 292
 293             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 294
 295             /* Load parameters for j particles */
 296             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 297                                                               charge+jnrC+0,charge+jnrD+0);
 298             vdwjidx0A        = 2*vdwtype[jnrA+0];
 299             vdwjidx0B        = 2*vdwtype[jnrB+0];
 300             vdwjidx0C        = 2*vdwtype[jnrC+0];
 301             vdwjidx0D        = 2*vdwtype[jnrD+0];
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             if (gmx_mm_any_lt(rsq00,rcutoff2))
 308             {
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq00             = _mm_mul_ps(iq0,jq0);
 312             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 313                                          vdwparam+vdwioffset0+vdwjidx0B,
 314                                          vdwparam+vdwioffset0+vdwjidx0C,
 315                                          vdwparam+vdwioffset0+vdwjidx0D,
 316                                          &c6_00,&c12_00);
 317
 318             /* REACTION-FIELD ELECTROSTATICS */
 319             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 320             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 321
 322             /* LENNARD-JONES DISPERSION/REPULSION */
 323
 324             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 325             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 326             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 327             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 328                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 329             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 330
 331             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velec            = _mm_and_ps(velec,cutoff_mask);
 335             velec            = _mm_andnot_ps(dummy_mask,velec);
 336             velecsum         = _mm_add_ps(velecsum,velec);
 337             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 338             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 339             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 340
 341             fscal            = _mm_add_ps(felec,fvdw);
 342
 343             fscal            = _mm_and_ps(fscal,cutoff_mask);
 344
 345             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 346
 347              /* Update vectorial force */
 348             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 349             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 350             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 351
 352             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 353             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 354             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 355             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 356             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 357                                                    _mm_mul_ps(dx00,fscal),
 358                                                    _mm_mul_ps(dy00,fscal),
 359                                                    _mm_mul_ps(dz00,fscal));
 360
 361             }
 362
 363             /* Inner loop uses 57 flops */
 364         }
 365
 366         /* End of innermost loop */
 367
 368         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 369                                               f+i_coord_offset,fshift+i_shift_offset);
 370
 371         ggid                        = gid[iidx];
 372         /* Update potential energies */
 373         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 374         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 375
 376         /* Increment number of inner iterations */
 377         inneriter                  += j_index_end - j_index_start;
 378
 379         /* Outer loop uses 9 flops */
 380     }
 381
 382     /* Increment number of outer iterations */
 383     outeriter        += nri;
 384
 385     /* Update outer/inner flops */
 386
 387     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*57);
 388 }
 389 /*
 390  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_128_fma_single
 391  * Electrostatics interaction: ReactionField
 392  * VdW interaction:            LennardJones
 393  * Geometry:                   Particle-Particle
 394  * Calculate force/pot:        Force
 395  */
 396 void
 397 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_128_fma_single
 398                     (t_nblist * gmx_restrict                nlist,
 399                      rvec * gmx_restrict                    xx,
 400                      rvec * gmx_restrict                    ff,
 401                      t_forcerec * gmx_restrict              fr,
 402                      t_mdatoms * gmx_restrict               mdatoms,
 403                      nb_kernel_data_t * gmx_restrict        kernel_data,
 404                      t_nrnb * gmx_restrict                  nrnb)
 405 {
 406     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 407      * just 0 for non-waters.
 408      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 409      * jnr indices corresponding to data put in the four positions in the SIMD register.
 410      */
 411     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 412     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 413     int              jnrA,jnrB,jnrC,jnrD;
 414     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 415     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 416     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 417     real             rcutoff_scalar;
 418     real             *shiftvec,*fshift,*x,*f;
 419     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 420     real             scratch[4*DIM];
 421     __m128           fscal,rcutoff,rcutoff2,jidxall;
 422     int              vdwioffset0;
 423     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 424     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 425     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 426     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 427     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 428     real             *charge;
 429     int              nvdwtype;
 430     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 431     int              *vdwtype;
 432     real             *vdwparam;
 433     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 434     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 435     __m128           dummy_mask,cutoff_mask;
 436     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 437     __m128           one     = _mm_set1_ps(1.0);
 438     __m128           two     = _mm_set1_ps(2.0);
 439     x                = xx[0];
 440     f                = ff[0];
 441
 442     nri              = nlist->nri;
 443     iinr             = nlist->iinr;
 444     jindex           = nlist->jindex;
 445     jjnr             = nlist->jjnr;
 446     shiftidx         = nlist->shift;
 447     gid              = nlist->gid;
 448     shiftvec         = fr->shift_vec[0];
 449     fshift           = fr->fshift[0];
 450     facel            = _mm_set1_ps(fr->epsfac);
 451     charge           = mdatoms->chargeA;
 452     krf              = _mm_set1_ps(fr->ic->k_rf);
 453     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 454     crf              = _mm_set1_ps(fr->ic->c_rf);
 455     nvdwtype         = fr->ntype;
 456     vdwparam         = fr->nbfp;
 457     vdwtype          = mdatoms->typeA;
 458
 459     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 460     rcutoff_scalar   = fr->rcoulomb;
 461     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 462     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 463
 464     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 465     rvdw             = _mm_set1_ps(fr->rvdw);
 466
 467     /* Avoid stupid compiler warnings */
 468     jnrA = jnrB = jnrC = jnrD = 0;
 469     j_coord_offsetA = 0;
 470     j_coord_offsetB = 0;
 471     j_coord_offsetC = 0;
 472     j_coord_offsetD = 0;
 473
 474     outeriter        = 0;
 475     inneriter        = 0;
 476
 477     for(iidx=0;iidx<4*DIM;iidx++)
 478     {
 479         scratch[iidx] = 0.0;
 480     }
 481
 482     /* Start outer loop over neighborlists */
 483     for(iidx=0; iidx<nri; iidx++)
 484     {
 485         /* Load shift vector for this list */
 486         i_shift_offset   = DIM*shiftidx[iidx];
 487
 488         /* Load limits for loop over neighbors */
 489         j_index_start    = jindex[iidx];
 490         j_index_end      = jindex[iidx+1];
 491
 492         /* Get outer coordinate index */
 493         inr              = iinr[iidx];
 494         i_coord_offset   = DIM*inr;
 495
 496         /* Load i particle coords and add shift vector */
 497         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 498
 499         fix0             = _mm_setzero_ps();
 500         fiy0             = _mm_setzero_ps();
 501         fiz0             = _mm_setzero_ps();
 502
 503         /* Load parameters for i particles */
 504         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 505         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 506
 507         /* Start inner kernel loop */
 508         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 509         {
 510
 511             /* Get j neighbor index, and coordinate index */
 512             jnrA             = jjnr[jidx];
 513             jnrB             = jjnr[jidx+1];
 514             jnrC             = jjnr[jidx+2];
 515             jnrD             = jjnr[jidx+3];
 516             j_coord_offsetA  = DIM*jnrA;
 517             j_coord_offsetB  = DIM*jnrB;
 518             j_coord_offsetC  = DIM*jnrC;
 519             j_coord_offsetD  = DIM*jnrD;
 520
 521             /* load j atom coordinates */
 522             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 523                                               x+j_coord_offsetC,x+j_coord_offsetD,
 524                                               &jx0,&jy0,&jz0);
 525
 526             /* Calculate displacement vector */
 527             dx00             = _mm_sub_ps(ix0,jx0);
 528             dy00             = _mm_sub_ps(iy0,jy0);
 529             dz00             = _mm_sub_ps(iz0,jz0);
 530
 531             /* Calculate squared distance and things based on it */
 532             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 533
 534             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 535
 536             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 537
 538             /* Load parameters for j particles */
 539             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 540                                                               charge+jnrC+0,charge+jnrD+0);
 541             vdwjidx0A        = 2*vdwtype[jnrA+0];
 542             vdwjidx0B        = 2*vdwtype[jnrB+0];
 543             vdwjidx0C        = 2*vdwtype[jnrC+0];
 544             vdwjidx0D        = 2*vdwtype[jnrD+0];
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             if (gmx_mm_any_lt(rsq00,rcutoff2))
 551             {
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq00             = _mm_mul_ps(iq0,jq0);
 555             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 556                                          vdwparam+vdwioffset0+vdwjidx0B,
 557                                          vdwparam+vdwioffset0+vdwjidx0C,
 558                                          vdwparam+vdwioffset0+vdwjidx0D,
 559                                          &c6_00,&c12_00);
 560
 561             /* REACTION-FIELD ELECTROSTATICS */
 562             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 563
 564             /* LENNARD-JONES DISPERSION/REPULSION */
 565
 566             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 567             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 568
 569             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 570
 571             fscal            = _mm_add_ps(felec,fvdw);
 572
 573             fscal            = _mm_and_ps(fscal,cutoff_mask);
 574
 575              /* Update vectorial force */
 576             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 577             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 578             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 579
 580             fjptrA             = f+j_coord_offsetA;
 581             fjptrB             = f+j_coord_offsetB;
 582             fjptrC             = f+j_coord_offsetC;
 583             fjptrD             = f+j_coord_offsetD;
 584             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 585                                                    _mm_mul_ps(dx00,fscal),
 586                                                    _mm_mul_ps(dy00,fscal),
 587                                                    _mm_mul_ps(dz00,fscal));
 588
 589             }
 590
 591             /* Inner loop uses 40 flops */
 592         }
 593
 594         if(jidx<j_index_end)
 595         {
 596
 597             /* Get j neighbor index, and coordinate index */
 598             jnrlistA         = jjnr[jidx];
 599             jnrlistB         = jjnr[jidx+1];
 600             jnrlistC         = jjnr[jidx+2];
 601             jnrlistD         = jjnr[jidx+3];
 602             /* Sign of each element will be negative for non-real atoms.
 603              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 604              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 605              */
 606             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 607             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 608             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 609             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 610             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 611             j_coord_offsetA  = DIM*jnrA;
 612             j_coord_offsetB  = DIM*jnrB;
 613             j_coord_offsetC  = DIM*jnrC;
 614             j_coord_offsetD  = DIM*jnrD;
 615
 616             /* load j atom coordinates */
 617             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 618                                               x+j_coord_offsetC,x+j_coord_offsetD,
 619                                               &jx0,&jy0,&jz0);
 620
 621             /* Calculate displacement vector */
 622             dx00             = _mm_sub_ps(ix0,jx0);
 623             dy00             = _mm_sub_ps(iy0,jy0);
 624             dz00             = _mm_sub_ps(iz0,jz0);
 625
 626             /* Calculate squared distance and things based on it */
 627             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 628
 629             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 630
 631             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 632
 633             /* Load parameters for j particles */
 634             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 635                                                               charge+jnrC+0,charge+jnrD+0);
 636             vdwjidx0A        = 2*vdwtype[jnrA+0];
 637             vdwjidx0B        = 2*vdwtype[jnrB+0];
 638             vdwjidx0C        = 2*vdwtype[jnrC+0];
 639             vdwjidx0D        = 2*vdwtype[jnrD+0];
 640
 641             /**************************
 642              * CALCULATE INTERACTIONS *
 643              **************************/
 644
 645             if (gmx_mm_any_lt(rsq00,rcutoff2))
 646             {
 647
 648             /* Compute parameters for interactions between i and j atoms */
 649             qq00             = _mm_mul_ps(iq0,jq0);
 650             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 651                                          vdwparam+vdwioffset0+vdwjidx0B,
 652                                          vdwparam+vdwioffset0+vdwjidx0C,
 653                                          vdwparam+vdwioffset0+vdwjidx0D,
 654                                          &c6_00,&c12_00);
 655
 656             /* REACTION-FIELD ELECTROSTATICS */
 657             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 658
 659             /* LENNARD-JONES DISPERSION/REPULSION */
 660
 661             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 662             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 663
 664             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 665
 666             fscal            = _mm_add_ps(felec,fvdw);
 667
 668             fscal            = _mm_and_ps(fscal,cutoff_mask);
 669
 670             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 671
 672              /* Update vectorial force */
 673             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 674             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 675             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 676
 677             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 678             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 679             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 680             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 681             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 682                                                    _mm_mul_ps(dx00,fscal),
 683                                                    _mm_mul_ps(dy00,fscal),
 684                                                    _mm_mul_ps(dz00,fscal));
 685
 686             }
 687
 688             /* Inner loop uses 40 flops */
 689         }
 690
 691         /* End of innermost loop */
 692
 693         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 694                                               f+i_coord_offset,fshift+i_shift_offset);
 695
 696         /* Increment number of inner iterations */
 697         inneriter                  += j_index_end - j_index_start;
 698
 699         /* Outer loop uses 7 flops */
 700     }
 701
 702     /* Increment number of outer iterations */
 703     outeriter        += nri;
 704
 705     /* Update outer/inner flops */
 706
 707     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*40);
 708 }