src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_VdwLJ_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_ElecRF_VdwLJ_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_ElecRF_VdwLJ_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     /* Avoid stupid compiler warnings */
 107     jnrA = jnrB = jnrC = jnrD = 0;
 108     j_coord_offsetA = 0;
 109     j_coord_offsetB = 0;
 110     j_coord_offsetC = 0;
 111     j_coord_offsetD = 0;
 112
 113     outeriter        = 0;
 114     inneriter        = 0;
 115
 116     for(iidx=0;iidx<4*DIM;iidx++)
 117     {
 118         scratch[iidx] = 0.0;
 119     }
 120
 121     /* Start outer loop over neighborlists */
 122     for(iidx=0; iidx<nri; iidx++)
 123     {
 124         /* Load shift vector for this list */
 125         i_shift_offset   = DIM*shiftidx[iidx];
 126
 127         /* Load limits for loop over neighbors */
 128         j_index_start    = jindex[iidx];
 129         j_index_end      = jindex[iidx+1];
 130
 131         /* Get outer coordinate index */
 132         inr              = iinr[iidx];
 133         i_coord_offset   = DIM*inr;
 134
 135         /* Load i particle coords and add shift vector */
 136         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 137
 138         fix0             = _mm_setzero_ps();
 139         fiy0             = _mm_setzero_ps();
 140         fiz0             = _mm_setzero_ps();
 141
 142         /* Load parameters for i particles */
 143         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 144         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 145
 146         /* Reset potential sums */
 147         velecsum         = _mm_setzero_ps();
 148         vvdwsum          = _mm_setzero_ps();
 149
 150         /* Start inner kernel loop */
 151         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 152         {
 153
 154             /* Get j neighbor index, and coordinate index */
 155             jnrA             = jjnr[jidx];
 156             jnrB             = jjnr[jidx+1];
 157             jnrC             = jjnr[jidx+2];
 158             jnrD             = jjnr[jidx+3];
 159             j_coord_offsetA  = DIM*jnrA;
 160             j_coord_offsetB  = DIM*jnrB;
 161             j_coord_offsetC  = DIM*jnrC;
 162             j_coord_offsetD  = DIM*jnrD;
 163
 164             /* load j atom coordinates */
 165             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 166                                               x+j_coord_offsetC,x+j_coord_offsetD,
 167                                               &jx0,&jy0,&jz0);
 168
 169             /* Calculate displacement vector */
 170             dx00             = _mm_sub_ps(ix0,jx0);
 171             dy00             = _mm_sub_ps(iy0,jy0);
 172             dz00             = _mm_sub_ps(iz0,jz0);
 173
 174             /* Calculate squared distance and things based on it */
 175             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 176
 177             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 178
 179             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 180
 181             /* Load parameters for j particles */
 182             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 183                                                               charge+jnrC+0,charge+jnrD+0);
 184             vdwjidx0A        = 2*vdwtype[jnrA+0];
 185             vdwjidx0B        = 2*vdwtype[jnrB+0];
 186             vdwjidx0C        = 2*vdwtype[jnrC+0];
 187             vdwjidx0D        = 2*vdwtype[jnrD+0];
 188
 189             /**************************
 190              * CALCULATE INTERACTIONS *
 191              **************************/
 192
 193             /* Compute parameters for interactions between i and j atoms */
 194             qq00             = _mm_mul_ps(iq0,jq0);
 195             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 196                                          vdwparam+vdwioffset0+vdwjidx0B,
 197                                          vdwparam+vdwioffset0+vdwjidx0C,
 198                                          vdwparam+vdwioffset0+vdwjidx0D,
 199                                          &c6_00,&c12_00);
 200
 201             /* REACTION-FIELD ELECTROSTATICS */
 202             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 203             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 204
 205             /* LENNARD-JONES DISPERSION/REPULSION */
 206
 207             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 208             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 209             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 210             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 211             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 212
 213             /* Update potential sum for this i atom from the interaction with this j atom. */
 214             velecsum         = _mm_add_ps(velecsum,velec);
 215             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 216
 217             fscal            = _mm_add_ps(felec,fvdw);
 218
 219              /* Update vectorial force */
 220             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 221             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 222             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 223
 224             fjptrA             = f+j_coord_offsetA;
 225             fjptrB             = f+j_coord_offsetB;
 226             fjptrC             = f+j_coord_offsetC;
 227             fjptrD             = f+j_coord_offsetD;
 228             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 229                                                    _mm_mul_ps(dx00,fscal),
 230                                                    _mm_mul_ps(dy00,fscal),
 231                                                    _mm_mul_ps(dz00,fscal));
 232
 233             /* Inner loop uses 47 flops */
 234         }
 235
 236         if(jidx<j_index_end)
 237         {
 238
 239             /* Get j neighbor index, and coordinate index */
 240             jnrlistA         = jjnr[jidx];
 241             jnrlistB         = jjnr[jidx+1];
 242             jnrlistC         = jjnr[jidx+2];
 243             jnrlistD         = jjnr[jidx+3];
 244             /* Sign of each element will be negative for non-real atoms.
 245              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 246              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 247              */
 248             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 249             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 250             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 251             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 252             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 253             j_coord_offsetA  = DIM*jnrA;
 254             j_coord_offsetB  = DIM*jnrB;
 255             j_coord_offsetC  = DIM*jnrC;
 256             j_coord_offsetD  = DIM*jnrD;
 257
 258             /* load j atom coordinates */
 259             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 260                                               x+j_coord_offsetC,x+j_coord_offsetD,
 261                                               &jx0,&jy0,&jz0);
 262
 263             /* Calculate displacement vector */
 264             dx00             = _mm_sub_ps(ix0,jx0);
 265             dy00             = _mm_sub_ps(iy0,jy0);
 266             dz00             = _mm_sub_ps(iz0,jz0);
 267
 268             /* Calculate squared distance and things based on it */
 269             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 270
 271             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 272
 273             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 274
 275             /* Load parameters for j particles */
 276             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 277                                                               charge+jnrC+0,charge+jnrD+0);
 278             vdwjidx0A        = 2*vdwtype[jnrA+0];
 279             vdwjidx0B        = 2*vdwtype[jnrB+0];
 280             vdwjidx0C        = 2*vdwtype[jnrC+0];
 281             vdwjidx0D        = 2*vdwtype[jnrD+0];
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq00             = _mm_mul_ps(iq0,jq0);
 289             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 290                                          vdwparam+vdwioffset0+vdwjidx0B,
 291                                          vdwparam+vdwioffset0+vdwjidx0C,
 292                                          vdwparam+vdwioffset0+vdwjidx0D,
 293                                          &c6_00,&c12_00);
 294
 295             /* REACTION-FIELD ELECTROSTATICS */
 296             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 297             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 298
 299             /* LENNARD-JONES DISPERSION/REPULSION */
 300
 301             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 302             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 303             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 304             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 305             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velec            = _mm_andnot_ps(dummy_mask,velec);
 309             velecsum         = _mm_add_ps(velecsum,velec);
 310             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 311             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 312
 313             fscal            = _mm_add_ps(felec,fvdw);
 314
 315             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 316
 317              /* Update vectorial force */
 318             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 319             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 320             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 321
 322             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 323             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 324             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 325             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 326             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 327                                                    _mm_mul_ps(dx00,fscal),
 328                                                    _mm_mul_ps(dy00,fscal),
 329                                                    _mm_mul_ps(dz00,fscal));
 330
 331             /* Inner loop uses 47 flops */
 332         }
 333
 334         /* End of innermost loop */
 335
 336         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 337                                               f+i_coord_offset,fshift+i_shift_offset);
 338
 339         ggid                        = gid[iidx];
 340         /* Update potential energies */
 341         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 342         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 343
 344         /* Increment number of inner iterations */
 345         inneriter                  += j_index_end - j_index_start;
 346
 347         /* Outer loop uses 9 flops */
 348     }
 349
 350     /* Increment number of outer iterations */
 351     outeriter        += nri;
 352
 353     /* Update outer/inner flops */
 354
 355     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*47);
 356 }
 357 /*
 358  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_single
 359  * Electrostatics interaction: ReactionField
 360  * VdW interaction:            LennardJones
 361  * Geometry:                   Particle-Particle
 362  * Calculate force/pot:        Force
 363  */
 364 void
 365 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_avx_128_fma_single
 366                     (t_nblist * gmx_restrict                nlist,
 367                      rvec * gmx_restrict                    xx,
 368                      rvec * gmx_restrict                    ff,
 369                      t_forcerec * gmx_restrict              fr,
 370                      t_mdatoms * gmx_restrict               mdatoms,
 371                      nb_kernel_data_t * gmx_restrict        kernel_data,
 372                      t_nrnb * gmx_restrict                  nrnb)
 373 {
 374     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 375      * just 0 for non-waters.
 376      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 377      * jnr indices corresponding to data put in the four positions in the SIMD register.
 378      */
 379     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 380     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 381     int              jnrA,jnrB,jnrC,jnrD;
 382     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 383     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 384     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 385     real             rcutoff_scalar;
 386     real             *shiftvec,*fshift,*x,*f;
 387     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 388     real             scratch[4*DIM];
 389     __m128           fscal,rcutoff,rcutoff2,jidxall;
 390     int              vdwioffset0;
 391     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 392     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 393     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 394     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 395     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 396     real             *charge;
 397     int              nvdwtype;
 398     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 399     int              *vdwtype;
 400     real             *vdwparam;
 401     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 402     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 403     __m128           dummy_mask,cutoff_mask;
 404     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 405     __m128           one     = _mm_set1_ps(1.0);
 406     __m128           two     = _mm_set1_ps(2.0);
 407     x                = xx[0];
 408     f                = ff[0];
 409
 410     nri              = nlist->nri;
 411     iinr             = nlist->iinr;
 412     jindex           = nlist->jindex;
 413     jjnr             = nlist->jjnr;
 414     shiftidx         = nlist->shift;
 415     gid              = nlist->gid;
 416     shiftvec         = fr->shift_vec[0];
 417     fshift           = fr->fshift[0];
 418     facel            = _mm_set1_ps(fr->epsfac);
 419     charge           = mdatoms->chargeA;
 420     krf              = _mm_set1_ps(fr->ic->k_rf);
 421     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 422     crf              = _mm_set1_ps(fr->ic->c_rf);
 423     nvdwtype         = fr->ntype;
 424     vdwparam         = fr->nbfp;
 425     vdwtype          = mdatoms->typeA;
 426
 427     /* Avoid stupid compiler warnings */
 428     jnrA = jnrB = jnrC = jnrD = 0;
 429     j_coord_offsetA = 0;
 430     j_coord_offsetB = 0;
 431     j_coord_offsetC = 0;
 432     j_coord_offsetD = 0;
 433
 434     outeriter        = 0;
 435     inneriter        = 0;
 436
 437     for(iidx=0;iidx<4*DIM;iidx++)
 438     {
 439         scratch[iidx] = 0.0;
 440     }
 441
 442     /* Start outer loop over neighborlists */
 443     for(iidx=0; iidx<nri; iidx++)
 444     {
 445         /* Load shift vector for this list */
 446         i_shift_offset   = DIM*shiftidx[iidx];
 447
 448         /* Load limits for loop over neighbors */
 449         j_index_start    = jindex[iidx];
 450         j_index_end      = jindex[iidx+1];
 451
 452         /* Get outer coordinate index */
 453         inr              = iinr[iidx];
 454         i_coord_offset   = DIM*inr;
 455
 456         /* Load i particle coords and add shift vector */
 457         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 458
 459         fix0             = _mm_setzero_ps();
 460         fiy0             = _mm_setzero_ps();
 461         fiz0             = _mm_setzero_ps();
 462
 463         /* Load parameters for i particles */
 464         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 465         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 466
 467         /* Start inner kernel loop */
 468         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 469         {
 470
 471             /* Get j neighbor index, and coordinate index */
 472             jnrA             = jjnr[jidx];
 473             jnrB             = jjnr[jidx+1];
 474             jnrC             = jjnr[jidx+2];
 475             jnrD             = jjnr[jidx+3];
 476             j_coord_offsetA  = DIM*jnrA;
 477             j_coord_offsetB  = DIM*jnrB;
 478             j_coord_offsetC  = DIM*jnrC;
 479             j_coord_offsetD  = DIM*jnrD;
 480
 481             /* load j atom coordinates */
 482             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 483                                               x+j_coord_offsetC,x+j_coord_offsetD,
 484                                               &jx0,&jy0,&jz0);
 485
 486             /* Calculate displacement vector */
 487             dx00             = _mm_sub_ps(ix0,jx0);
 488             dy00             = _mm_sub_ps(iy0,jy0);
 489             dz00             = _mm_sub_ps(iz0,jz0);
 490
 491             /* Calculate squared distance and things based on it */
 492             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 493
 494             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 495
 496             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 497
 498             /* Load parameters for j particles */
 499             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 500                                                               charge+jnrC+0,charge+jnrD+0);
 501             vdwjidx0A        = 2*vdwtype[jnrA+0];
 502             vdwjidx0B        = 2*vdwtype[jnrB+0];
 503             vdwjidx0C        = 2*vdwtype[jnrC+0];
 504             vdwjidx0D        = 2*vdwtype[jnrD+0];
 505
 506             /**************************
 507              * CALCULATE INTERACTIONS *
 508              **************************/
 509
 510             /* Compute parameters for interactions between i and j atoms */
 511             qq00             = _mm_mul_ps(iq0,jq0);
 512             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 513                                          vdwparam+vdwioffset0+vdwjidx0B,
 514                                          vdwparam+vdwioffset0+vdwjidx0C,
 515                                          vdwparam+vdwioffset0+vdwjidx0D,
 516                                          &c6_00,&c12_00);
 517
 518             /* REACTION-FIELD ELECTROSTATICS */
 519             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 520
 521             /* LENNARD-JONES DISPERSION/REPULSION */
 522
 523             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 524             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 525
 526             fscal            = _mm_add_ps(felec,fvdw);
 527
 528              /* Update vectorial force */
 529             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 530             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 531             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 532
 533             fjptrA             = f+j_coord_offsetA;
 534             fjptrB             = f+j_coord_offsetB;
 535             fjptrC             = f+j_coord_offsetC;
 536             fjptrD             = f+j_coord_offsetD;
 537             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 538                                                    _mm_mul_ps(dx00,fscal),
 539                                                    _mm_mul_ps(dy00,fscal),
 540                                                    _mm_mul_ps(dz00,fscal));
 541
 542             /* Inner loop uses 37 flops */
 543         }
 544
 545         if(jidx<j_index_end)
 546         {
 547
 548             /* Get j neighbor index, and coordinate index */
 549             jnrlistA         = jjnr[jidx];
 550             jnrlistB         = jjnr[jidx+1];
 551             jnrlistC         = jjnr[jidx+2];
 552             jnrlistD         = jjnr[jidx+3];
 553             /* Sign of each element will be negative for non-real atoms.
 554              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 555              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 556              */
 557             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 558             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 559             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 560             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 561             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 562             j_coord_offsetA  = DIM*jnrA;
 563             j_coord_offsetB  = DIM*jnrB;
 564             j_coord_offsetC  = DIM*jnrC;
 565             j_coord_offsetD  = DIM*jnrD;
 566
 567             /* load j atom coordinates */
 568             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 569                                               x+j_coord_offsetC,x+j_coord_offsetD,
 570                                               &jx0,&jy0,&jz0);
 571
 572             /* Calculate displacement vector */
 573             dx00             = _mm_sub_ps(ix0,jx0);
 574             dy00             = _mm_sub_ps(iy0,jy0);
 575             dz00             = _mm_sub_ps(iz0,jz0);
 576
 577             /* Calculate squared distance and things based on it */
 578             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 579
 580             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 581
 582             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 583
 584             /* Load parameters for j particles */
 585             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 586                                                               charge+jnrC+0,charge+jnrD+0);
 587             vdwjidx0A        = 2*vdwtype[jnrA+0];
 588             vdwjidx0B        = 2*vdwtype[jnrB+0];
 589             vdwjidx0C        = 2*vdwtype[jnrC+0];
 590             vdwjidx0D        = 2*vdwtype[jnrD+0];
 591
 592             /**************************
 593              * CALCULATE INTERACTIONS *
 594              **************************/
 595
 596             /* Compute parameters for interactions between i and j atoms */
 597             qq00             = _mm_mul_ps(iq0,jq0);
 598             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 599                                          vdwparam+vdwioffset0+vdwjidx0B,
 600                                          vdwparam+vdwioffset0+vdwjidx0C,
 601                                          vdwparam+vdwioffset0+vdwjidx0D,
 602                                          &c6_00,&c12_00);
 603
 604             /* REACTION-FIELD ELECTROSTATICS */
 605             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 606
 607             /* LENNARD-JONES DISPERSION/REPULSION */
 608
 609             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 610             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 611
 612             fscal            = _mm_add_ps(felec,fvdw);
 613
 614             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 615
 616              /* Update vectorial force */
 617             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 618             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 619             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 620
 621             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 622             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 623             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 624             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 625             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 626                                                    _mm_mul_ps(dx00,fscal),
 627                                                    _mm_mul_ps(dy00,fscal),
 628                                                    _mm_mul_ps(dz00,fscal));
 629
 630             /* Inner loop uses 37 flops */
 631         }
 632
 633         /* End of innermost loop */
 634
 635         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 636                                               f+i_coord_offset,fshift+i_shift_offset);
 637
 638         /* Increment number of inner iterations */
 639         inneriter                  += j_index_end - j_index_start;
 640
 641         /* Outer loop uses 7 flops */
 642     }
 643
 644     /* Increment number of outer iterations */
 645     outeriter        += nri;
 646
 647     /* Update outer/inner flops */
 648
 649     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 650 }