src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_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_ElecCoul_VdwLJ_GeomP1P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_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     nvdwtype         = fr->ntype;
 100     vdwparam         = fr->nbfp;
 101     vdwtype          = mdatoms->typeA;
 102
 103     /* Avoid stupid compiler warnings */
 104     jnrA = jnrB = jnrC = jnrD = 0;
 105     j_coord_offsetA = 0;
 106     j_coord_offsetB = 0;
 107     j_coord_offsetC = 0;
 108     j_coord_offsetD = 0;
 109
 110     outeriter        = 0;
 111     inneriter        = 0;
 112
 113     for(iidx=0;iidx<4*DIM;iidx++)
 114     {
 115         scratch[iidx] = 0.0;
 116     }
 117
 118     /* Start outer loop over neighborlists */
 119     for(iidx=0; iidx<nri; iidx++)
 120     {
 121         /* Load shift vector for this list */
 122         i_shift_offset   = DIM*shiftidx[iidx];
 123
 124         /* Load limits for loop over neighbors */
 125         j_index_start    = jindex[iidx];
 126         j_index_end      = jindex[iidx+1];
 127
 128         /* Get outer coordinate index */
 129         inr              = iinr[iidx];
 130         i_coord_offset   = DIM*inr;
 131
 132         /* Load i particle coords and add shift vector */
 133         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 134
 135         fix0             = _mm_setzero_ps();
 136         fiy0             = _mm_setzero_ps();
 137         fiz0             = _mm_setzero_ps();
 138
 139         /* Load parameters for i particles */
 140         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 141         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 142
 143         /* Reset potential sums */
 144         velecsum         = _mm_setzero_ps();
 145         vvdwsum          = _mm_setzero_ps();
 146
 147         /* Start inner kernel loop */
 148         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 149         {
 150
 151             /* Get j neighbor index, and coordinate index */
 152             jnrA             = jjnr[jidx];
 153             jnrB             = jjnr[jidx+1];
 154             jnrC             = jjnr[jidx+2];
 155             jnrD             = jjnr[jidx+3];
 156             j_coord_offsetA  = DIM*jnrA;
 157             j_coord_offsetB  = DIM*jnrB;
 158             j_coord_offsetC  = DIM*jnrC;
 159             j_coord_offsetD  = DIM*jnrD;
 160
 161             /* load j atom coordinates */
 162             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 163                                               x+j_coord_offsetC,x+j_coord_offsetD,
 164                                               &jx0,&jy0,&jz0);
 165
 166             /* Calculate displacement vector */
 167             dx00             = _mm_sub_ps(ix0,jx0);
 168             dy00             = _mm_sub_ps(iy0,jy0);
 169             dz00             = _mm_sub_ps(iz0,jz0);
 170
 171             /* Calculate squared distance and things based on it */
 172             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 173
 174             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 175
 176             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 177
 178             /* Load parameters for j particles */
 179             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 180                                                               charge+jnrC+0,charge+jnrD+0);
 181             vdwjidx0A        = 2*vdwtype[jnrA+0];
 182             vdwjidx0B        = 2*vdwtype[jnrB+0];
 183             vdwjidx0C        = 2*vdwtype[jnrC+0];
 184             vdwjidx0D        = 2*vdwtype[jnrD+0];
 185
 186             /**************************
 187              * CALCULATE INTERACTIONS *
 188              **************************/
 189
 190             /* Compute parameters for interactions between i and j atoms */
 191             qq00             = _mm_mul_ps(iq0,jq0);
 192             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 193                                          vdwparam+vdwioffset0+vdwjidx0B,
 194                                          vdwparam+vdwioffset0+vdwjidx0C,
 195                                          vdwparam+vdwioffset0+vdwjidx0D,
 196                                          &c6_00,&c12_00);
 197
 198             /* COULOMB ELECTROSTATICS */
 199             velec            = _mm_mul_ps(qq00,rinv00);
 200             felec            = _mm_mul_ps(velec,rinvsq00);
 201
 202             /* LENNARD-JONES DISPERSION/REPULSION */
 203
 204             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 205             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 206             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 207             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 208             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 209
 210             /* Update potential sum for this i atom from the interaction with this j atom. */
 211             velecsum         = _mm_add_ps(velecsum,velec);
 212             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 213
 214             fscal            = _mm_add_ps(felec,fvdw);
 215
 216              /* Update vectorial force */
 217             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 218             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 219             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 220
 221             fjptrA             = f+j_coord_offsetA;
 222             fjptrB             = f+j_coord_offsetB;
 223             fjptrC             = f+j_coord_offsetC;
 224             fjptrD             = f+j_coord_offsetD;
 225             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 226                                                    _mm_mul_ps(dx00,fscal),
 227                                                    _mm_mul_ps(dy00,fscal),
 228                                                    _mm_mul_ps(dz00,fscal));
 229
 230             /* Inner loop uses 43 flops */
 231         }
 232
 233         if(jidx<j_index_end)
 234         {
 235
 236             /* Get j neighbor index, and coordinate index */
 237             jnrlistA         = jjnr[jidx];
 238             jnrlistB         = jjnr[jidx+1];
 239             jnrlistC         = jjnr[jidx+2];
 240             jnrlistD         = jjnr[jidx+3];
 241             /* Sign of each element will be negative for non-real atoms.
 242              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 243              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 244              */
 245             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 246             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 247             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 248             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 249             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 250             j_coord_offsetA  = DIM*jnrA;
 251             j_coord_offsetB  = DIM*jnrB;
 252             j_coord_offsetC  = DIM*jnrC;
 253             j_coord_offsetD  = DIM*jnrD;
 254
 255             /* load j atom coordinates */
 256             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 257                                               x+j_coord_offsetC,x+j_coord_offsetD,
 258                                               &jx0,&jy0,&jz0);
 259
 260             /* Calculate displacement vector */
 261             dx00             = _mm_sub_ps(ix0,jx0);
 262             dy00             = _mm_sub_ps(iy0,jy0);
 263             dz00             = _mm_sub_ps(iz0,jz0);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 267
 268             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 269
 270             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 271
 272             /* Load parameters for j particles */
 273             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 274                                                               charge+jnrC+0,charge+jnrD+0);
 275             vdwjidx0A        = 2*vdwtype[jnrA+0];
 276             vdwjidx0B        = 2*vdwtype[jnrB+0];
 277             vdwjidx0C        = 2*vdwtype[jnrC+0];
 278             vdwjidx0D        = 2*vdwtype[jnrD+0];
 279
 280             /**************************
 281              * CALCULATE INTERACTIONS *
 282              **************************/
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             qq00             = _mm_mul_ps(iq0,jq0);
 286             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 287                                          vdwparam+vdwioffset0+vdwjidx0B,
 288                                          vdwparam+vdwioffset0+vdwjidx0C,
 289                                          vdwparam+vdwioffset0+vdwjidx0D,
 290                                          &c6_00,&c12_00);
 291
 292             /* COULOMB ELECTROSTATICS */
 293             velec            = _mm_mul_ps(qq00,rinv00);
 294             felec            = _mm_mul_ps(velec,rinvsq00);
 295
 296             /* LENNARD-JONES DISPERSION/REPULSION */
 297
 298             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 299             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 300             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 301             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 302             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velec            = _mm_andnot_ps(dummy_mask,velec);
 306             velecsum         = _mm_add_ps(velecsum,velec);
 307             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 308             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 309
 310             fscal            = _mm_add_ps(felec,fvdw);
 311
 312             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 313
 314              /* Update vectorial force */
 315             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 316             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 317             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 318
 319             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 320             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 321             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 322             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 323             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 324                                                    _mm_mul_ps(dx00,fscal),
 325                                                    _mm_mul_ps(dy00,fscal),
 326                                                    _mm_mul_ps(dz00,fscal));
 327
 328             /* Inner loop uses 43 flops */
 329         }
 330
 331         /* End of innermost loop */
 332
 333         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 334                                               f+i_coord_offset,fshift+i_shift_offset);
 335
 336         ggid                        = gid[iidx];
 337         /* Update potential energies */
 338         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 339         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 340
 341         /* Increment number of inner iterations */
 342         inneriter                  += j_index_end - j_index_start;
 343
 344         /* Outer loop uses 9 flops */
 345     }
 346
 347     /* Increment number of outer iterations */
 348     outeriter        += nri;
 349
 350     /* Update outer/inner flops */
 351
 352     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*43);
 353 }
 354 /*
 355  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_128_fma_single
 356  * Electrostatics interaction: Coulomb
 357  * VdW interaction:            LennardJones
 358  * Geometry:                   Particle-Particle
 359  * Calculate force/pot:        Force
 360  */
 361 void
 362 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_avx_128_fma_single
 363                     (t_nblist * gmx_restrict                nlist,
 364                      rvec * gmx_restrict                    xx,
 365                      rvec * gmx_restrict                    ff,
 366                      t_forcerec * gmx_restrict              fr,
 367                      t_mdatoms * gmx_restrict               mdatoms,
 368                      nb_kernel_data_t * gmx_restrict        kernel_data,
 369                      t_nrnb * gmx_restrict                  nrnb)
 370 {
 371     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 372      * just 0 for non-waters.
 373      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 374      * jnr indices corresponding to data put in the four positions in the SIMD register.
 375      */
 376     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 377     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 378     int              jnrA,jnrB,jnrC,jnrD;
 379     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 380     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 381     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 382     real             rcutoff_scalar;
 383     real             *shiftvec,*fshift,*x,*f;
 384     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 385     real             scratch[4*DIM];
 386     __m128           fscal,rcutoff,rcutoff2,jidxall;
 387     int              vdwioffset0;
 388     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 389     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 390     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 391     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 392     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 393     real             *charge;
 394     int              nvdwtype;
 395     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 396     int              *vdwtype;
 397     real             *vdwparam;
 398     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 399     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 400     __m128           dummy_mask,cutoff_mask;
 401     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 402     __m128           one     = _mm_set1_ps(1.0);
 403     __m128           two     = _mm_set1_ps(2.0);
 404     x                = xx[0];
 405     f                = ff[0];
 406
 407     nri              = nlist->nri;
 408     iinr             = nlist->iinr;
 409     jindex           = nlist->jindex;
 410     jjnr             = nlist->jjnr;
 411     shiftidx         = nlist->shift;
 412     gid              = nlist->gid;
 413     shiftvec         = fr->shift_vec[0];
 414     fshift           = fr->fshift[0];
 415     facel            = _mm_set1_ps(fr->epsfac);
 416     charge           = mdatoms->chargeA;
 417     nvdwtype         = fr->ntype;
 418     vdwparam         = fr->nbfp;
 419     vdwtype          = mdatoms->typeA;
 420
 421     /* Avoid stupid compiler warnings */
 422     jnrA = jnrB = jnrC = jnrD = 0;
 423     j_coord_offsetA = 0;
 424     j_coord_offsetB = 0;
 425     j_coord_offsetC = 0;
 426     j_coord_offsetD = 0;
 427
 428     outeriter        = 0;
 429     inneriter        = 0;
 430
 431     for(iidx=0;iidx<4*DIM;iidx++)
 432     {
 433         scratch[iidx] = 0.0;
 434     }
 435
 436     /* Start outer loop over neighborlists */
 437     for(iidx=0; iidx<nri; iidx++)
 438     {
 439         /* Load shift vector for this list */
 440         i_shift_offset   = DIM*shiftidx[iidx];
 441
 442         /* Load limits for loop over neighbors */
 443         j_index_start    = jindex[iidx];
 444         j_index_end      = jindex[iidx+1];
 445
 446         /* Get outer coordinate index */
 447         inr              = iinr[iidx];
 448         i_coord_offset   = DIM*inr;
 449
 450         /* Load i particle coords and add shift vector */
 451         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 452
 453         fix0             = _mm_setzero_ps();
 454         fiy0             = _mm_setzero_ps();
 455         fiz0             = _mm_setzero_ps();
 456
 457         /* Load parameters for i particles */
 458         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 459         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 460
 461         /* Start inner kernel loop */
 462         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 463         {
 464
 465             /* Get j neighbor index, and coordinate index */
 466             jnrA             = jjnr[jidx];
 467             jnrB             = jjnr[jidx+1];
 468             jnrC             = jjnr[jidx+2];
 469             jnrD             = jjnr[jidx+3];
 470             j_coord_offsetA  = DIM*jnrA;
 471             j_coord_offsetB  = DIM*jnrB;
 472             j_coord_offsetC  = DIM*jnrC;
 473             j_coord_offsetD  = DIM*jnrD;
 474
 475             /* load j atom coordinates */
 476             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 477                                               x+j_coord_offsetC,x+j_coord_offsetD,
 478                                               &jx0,&jy0,&jz0);
 479
 480             /* Calculate displacement vector */
 481             dx00             = _mm_sub_ps(ix0,jx0);
 482             dy00             = _mm_sub_ps(iy0,jy0);
 483             dz00             = _mm_sub_ps(iz0,jz0);
 484
 485             /* Calculate squared distance and things based on it */
 486             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 487
 488             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 489
 490             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 491
 492             /* Load parameters for j particles */
 493             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 494                                                               charge+jnrC+0,charge+jnrD+0);
 495             vdwjidx0A        = 2*vdwtype[jnrA+0];
 496             vdwjidx0B        = 2*vdwtype[jnrB+0];
 497             vdwjidx0C        = 2*vdwtype[jnrC+0];
 498             vdwjidx0D        = 2*vdwtype[jnrD+0];
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             qq00             = _mm_mul_ps(iq0,jq0);
 506             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 507                                          vdwparam+vdwioffset0+vdwjidx0B,
 508                                          vdwparam+vdwioffset0+vdwjidx0C,
 509                                          vdwparam+vdwioffset0+vdwjidx0D,
 510                                          &c6_00,&c12_00);
 511
 512             /* COULOMB ELECTROSTATICS */
 513             velec            = _mm_mul_ps(qq00,rinv00);
 514             felec            = _mm_mul_ps(velec,rinvsq00);
 515
 516             /* LENNARD-JONES DISPERSION/REPULSION */
 517
 518             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 519             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 520
 521             fscal            = _mm_add_ps(felec,fvdw);
 522
 523              /* Update vectorial force */
 524             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 525             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 526             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 527
 528             fjptrA             = f+j_coord_offsetA;
 529             fjptrB             = f+j_coord_offsetB;
 530             fjptrC             = f+j_coord_offsetC;
 531             fjptrD             = f+j_coord_offsetD;
 532             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 533                                                    _mm_mul_ps(dx00,fscal),
 534                                                    _mm_mul_ps(dy00,fscal),
 535                                                    _mm_mul_ps(dz00,fscal));
 536
 537             /* Inner loop uses 37 flops */
 538         }
 539
 540         if(jidx<j_index_end)
 541         {
 542
 543             /* Get j neighbor index, and coordinate index */
 544             jnrlistA         = jjnr[jidx];
 545             jnrlistB         = jjnr[jidx+1];
 546             jnrlistC         = jjnr[jidx+2];
 547             jnrlistD         = jjnr[jidx+3];
 548             /* Sign of each element will be negative for non-real atoms.
 549              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 550              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 551              */
 552             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 553             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 554             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 555             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 556             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 557             j_coord_offsetA  = DIM*jnrA;
 558             j_coord_offsetB  = DIM*jnrB;
 559             j_coord_offsetC  = DIM*jnrC;
 560             j_coord_offsetD  = DIM*jnrD;
 561
 562             /* load j atom coordinates */
 563             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 564                                               x+j_coord_offsetC,x+j_coord_offsetD,
 565                                               &jx0,&jy0,&jz0);
 566
 567             /* Calculate displacement vector */
 568             dx00             = _mm_sub_ps(ix0,jx0);
 569             dy00             = _mm_sub_ps(iy0,jy0);
 570             dz00             = _mm_sub_ps(iz0,jz0);
 571
 572             /* Calculate squared distance and things based on it */
 573             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 574
 575             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 576
 577             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 578
 579             /* Load parameters for j particles */
 580             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 581                                                               charge+jnrC+0,charge+jnrD+0);
 582             vdwjidx0A        = 2*vdwtype[jnrA+0];
 583             vdwjidx0B        = 2*vdwtype[jnrB+0];
 584             vdwjidx0C        = 2*vdwtype[jnrC+0];
 585             vdwjidx0D        = 2*vdwtype[jnrD+0];
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq00             = _mm_mul_ps(iq0,jq0);
 593             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 594                                          vdwparam+vdwioffset0+vdwjidx0B,
 595                                          vdwparam+vdwioffset0+vdwjidx0C,
 596                                          vdwparam+vdwioffset0+vdwjidx0D,
 597                                          &c6_00,&c12_00);
 598
 599             /* COULOMB ELECTROSTATICS */
 600             velec            = _mm_mul_ps(qq00,rinv00);
 601             felec            = _mm_mul_ps(velec,rinvsq00);
 602
 603             /* LENNARD-JONES DISPERSION/REPULSION */
 604
 605             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 606             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 607
 608             fscal            = _mm_add_ps(felec,fvdw);
 609
 610             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 611
 612              /* Update vectorial force */
 613             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 614             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 615             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 616
 617             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 618             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 619             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 620             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 621             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 622                                                    _mm_mul_ps(dx00,fscal),
 623                                                    _mm_mul_ps(dy00,fscal),
 624                                                    _mm_mul_ps(dz00,fscal));
 625
 626             /* Inner loop uses 37 flops */
 627         }
 628
 629         /* End of innermost loop */
 630
 631         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 632                                               f+i_coord_offset,fshift+i_shift_offset);
 633
 634         /* Increment number of inner iterations */
 635         inneriter                  += j_index_end - j_index_start;
 636
 637         /* Outer loop uses 7 flops */
 638     }
 639
 640     /* Increment number of outer iterations */
 641     outeriter        += nri;
 642
 643     /* Update outer/inner flops */
 644
 645     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 646 }