src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_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_ElecEw_VdwLJ_GeomP1P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_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     __m128i          ewitab;
  83     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  84     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  85     real             *ewtab;
  86     __m128           dummy_mask,cutoff_mask;
  87     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  88     __m128           one     = _mm_set1_ps(1.0);
  89     __m128           two     = _mm_set1_ps(2.0);
  90     x                = xx[0];
  91     f                = ff[0];
  92
  93     nri              = nlist->nri;
  94     iinr             = nlist->iinr;
  95     jindex           = nlist->jindex;
  96     jjnr             = nlist->jjnr;
  97     shiftidx         = nlist->shift;
  98     gid              = nlist->gid;
  99     shiftvec         = fr->shift_vec[0];
 100     fshift           = fr->fshift[0];
 101     facel            = _mm_set1_ps(fr->epsfac);
 102     charge           = mdatoms->chargeA;
 103     nvdwtype         = fr->ntype;
 104     vdwparam         = fr->nbfp;
 105     vdwtype          = mdatoms->typeA;
 106
 107     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 108     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 109     beta2            = _mm_mul_ps(beta,beta);
 110     beta3            = _mm_mul_ps(beta,beta2);
 111     ewtab            = fr->ic->tabq_coul_FDV0;
 112     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 113     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 114
 115     /* Avoid stupid compiler warnings */
 116     jnrA = jnrB = jnrC = jnrD = 0;
 117     j_coord_offsetA = 0;
 118     j_coord_offsetB = 0;
 119     j_coord_offsetC = 0;
 120     j_coord_offsetD = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     for(iidx=0;iidx<4*DIM;iidx++)
 126     {
 127         scratch[iidx] = 0.0;
 128     }
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _mm_setzero_ps();
 148         fiy0             = _mm_setzero_ps();
 149         fiz0             = _mm_setzero_ps();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 153         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155         /* Reset potential sums */
 156         velecsum         = _mm_setzero_ps();
 157         vvdwsum          = _mm_setzero_ps();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             jnrC             = jjnr[jidx+2];
 167             jnrD             = jjnr[jidx+3];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170             j_coord_offsetC  = DIM*jnrC;
 171             j_coord_offsetD  = DIM*jnrD;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               x+j_coord_offsetC,x+j_coord_offsetD,
 176                                               &jx0,&jy0,&jz0);
 177
 178             /* Calculate displacement vector */
 179             dx00             = _mm_sub_ps(ix0,jx0);
 180             dy00             = _mm_sub_ps(iy0,jy0);
 181             dz00             = _mm_sub_ps(iz0,jz0);
 182
 183             /* Calculate squared distance and things based on it */
 184             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 185
 186             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 187
 188             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 189
 190             /* Load parameters for j particles */
 191             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 192                                                               charge+jnrC+0,charge+jnrD+0);
 193             vdwjidx0A        = 2*vdwtype[jnrA+0];
 194             vdwjidx0B        = 2*vdwtype[jnrB+0];
 195             vdwjidx0C        = 2*vdwtype[jnrC+0];
 196             vdwjidx0D        = 2*vdwtype[jnrD+0];
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             r00              = _mm_mul_ps(rsq00,rinv00);
 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             /* EWALD ELECTROSTATICS */
 213
 214             /* Analytical PME correction */
 215             zeta2            = _mm_mul_ps(beta2,rsq00);
 216             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 217             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 218             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 219             felec            = _mm_mul_ps(qq00,felec);
 220             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 221             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 222             velec            = _mm_mul_ps(qq00,velec);
 223
 224             /* LENNARD-JONES DISPERSION/REPULSION */
 225
 226             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 227             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 228             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 229             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 230             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 231
 232             /* Update potential sum for this i atom from the interaction with this j atom. */
 233             velecsum         = _mm_add_ps(velecsum,velec);
 234             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 235
 236             fscal            = _mm_add_ps(felec,fvdw);
 237
 238              /* Update vectorial force */
 239             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 240             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 241             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 242
 243             fjptrA             = f+j_coord_offsetA;
 244             fjptrB             = f+j_coord_offsetB;
 245             fjptrC             = f+j_coord_offsetC;
 246             fjptrD             = f+j_coord_offsetD;
 247             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 248                                                    _mm_mul_ps(dx00,fscal),
 249                                                    _mm_mul_ps(dy00,fscal),
 250                                                    _mm_mul_ps(dz00,fscal));
 251
 252             /* Inner loop uses 41 flops */
 253         }
 254
 255         if(jidx<j_index_end)
 256         {
 257
 258             /* Get j neighbor index, and coordinate index */
 259             jnrlistA         = jjnr[jidx];
 260             jnrlistB         = jjnr[jidx+1];
 261             jnrlistC         = jjnr[jidx+2];
 262             jnrlistD         = jjnr[jidx+3];
 263             /* Sign of each element will be negative for non-real atoms.
 264              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 265              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 266              */
 267             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 268             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 269             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 270             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 271             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 272             j_coord_offsetA  = DIM*jnrA;
 273             j_coord_offsetB  = DIM*jnrB;
 274             j_coord_offsetC  = DIM*jnrC;
 275             j_coord_offsetD  = DIM*jnrD;
 276
 277             /* load j atom coordinates */
 278             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 279                                               x+j_coord_offsetC,x+j_coord_offsetD,
 280                                               &jx0,&jy0,&jz0);
 281
 282             /* Calculate displacement vector */
 283             dx00             = _mm_sub_ps(ix0,jx0);
 284             dy00             = _mm_sub_ps(iy0,jy0);
 285             dz00             = _mm_sub_ps(iz0,jz0);
 286
 287             /* Calculate squared distance and things based on it */
 288             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 289
 290             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 291
 292             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 293
 294             /* Load parameters for j particles */
 295             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 296                                                               charge+jnrC+0,charge+jnrD+0);
 297             vdwjidx0A        = 2*vdwtype[jnrA+0];
 298             vdwjidx0B        = 2*vdwtype[jnrB+0];
 299             vdwjidx0C        = 2*vdwtype[jnrC+0];
 300             vdwjidx0D        = 2*vdwtype[jnrD+0];
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             r00              = _mm_mul_ps(rsq00,rinv00);
 307             r00              = _mm_andnot_ps(dummy_mask,r00);
 308
 309             /* Compute parameters for interactions between i and j atoms */
 310             qq00             = _mm_mul_ps(iq0,jq0);
 311             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 312                                          vdwparam+vdwioffset0+vdwjidx0B,
 313                                          vdwparam+vdwioffset0+vdwjidx0C,
 314                                          vdwparam+vdwioffset0+vdwjidx0D,
 315                                          &c6_00,&c12_00);
 316
 317             /* EWALD ELECTROSTATICS */
 318
 319             /* Analytical PME correction */
 320             zeta2            = _mm_mul_ps(beta2,rsq00);
 321             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 322             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 323             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 324             felec            = _mm_mul_ps(qq00,felec);
 325             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 326             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 327             velec            = _mm_mul_ps(qq00,velec);
 328
 329             /* LENNARD-JONES DISPERSION/REPULSION */
 330
 331             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 332             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 333             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 334             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 335             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velec            = _mm_andnot_ps(dummy_mask,velec);
 339             velecsum         = _mm_add_ps(velecsum,velec);
 340             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 341             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 342
 343             fscal            = _mm_add_ps(felec,fvdw);
 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             /* Inner loop uses 42 flops */
 362         }
 363
 364         /* End of innermost loop */
 365
 366         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 367                                               f+i_coord_offset,fshift+i_shift_offset);
 368
 369         ggid                        = gid[iidx];
 370         /* Update potential energies */
 371         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 372         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 373
 374         /* Increment number of inner iterations */
 375         inneriter                  += j_index_end - j_index_start;
 376
 377         /* Outer loop uses 9 flops */
 378     }
 379
 380     /* Increment number of outer iterations */
 381     outeriter        += nri;
 382
 383     /* Update outer/inner flops */
 384
 385     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*42);
 386 }
 387 /*
 388  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_128_fma_single
 389  * Electrostatics interaction: Ewald
 390  * VdW interaction:            LennardJones
 391  * Geometry:                   Particle-Particle
 392  * Calculate force/pot:        Force
 393  */
 394 void
 395 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_128_fma_single
 396                     (t_nblist * gmx_restrict                nlist,
 397                      rvec * gmx_restrict                    xx,
 398                      rvec * gmx_restrict                    ff,
 399                      t_forcerec * gmx_restrict              fr,
 400                      t_mdatoms * gmx_restrict               mdatoms,
 401                      nb_kernel_data_t * gmx_restrict        kernel_data,
 402                      t_nrnb * gmx_restrict                  nrnb)
 403 {
 404     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 405      * just 0 for non-waters.
 406      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 407      * jnr indices corresponding to data put in the four positions in the SIMD register.
 408      */
 409     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 410     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 411     int              jnrA,jnrB,jnrC,jnrD;
 412     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 413     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 414     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 415     real             rcutoff_scalar;
 416     real             *shiftvec,*fshift,*x,*f;
 417     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 418     real             scratch[4*DIM];
 419     __m128           fscal,rcutoff,rcutoff2,jidxall;
 420     int              vdwioffset0;
 421     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 422     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 423     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 424     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 425     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 426     real             *charge;
 427     int              nvdwtype;
 428     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 429     int              *vdwtype;
 430     real             *vdwparam;
 431     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 432     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 433     __m128i          ewitab;
 434     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 435     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 436     real             *ewtab;
 437     __m128           dummy_mask,cutoff_mask;
 438     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 439     __m128           one     = _mm_set1_ps(1.0);
 440     __m128           two     = _mm_set1_ps(2.0);
 441     x                = xx[0];
 442     f                = ff[0];
 443
 444     nri              = nlist->nri;
 445     iinr             = nlist->iinr;
 446     jindex           = nlist->jindex;
 447     jjnr             = nlist->jjnr;
 448     shiftidx         = nlist->shift;
 449     gid              = nlist->gid;
 450     shiftvec         = fr->shift_vec[0];
 451     fshift           = fr->fshift[0];
 452     facel            = _mm_set1_ps(fr->epsfac);
 453     charge           = mdatoms->chargeA;
 454     nvdwtype         = fr->ntype;
 455     vdwparam         = fr->nbfp;
 456     vdwtype          = mdatoms->typeA;
 457
 458     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 459     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 460     beta2            = _mm_mul_ps(beta,beta);
 461     beta3            = _mm_mul_ps(beta,beta2);
 462     ewtab            = fr->ic->tabq_coul_F;
 463     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 464     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 465
 466     /* Avoid stupid compiler warnings */
 467     jnrA = jnrB = jnrC = jnrD = 0;
 468     j_coord_offsetA = 0;
 469     j_coord_offsetB = 0;
 470     j_coord_offsetC = 0;
 471     j_coord_offsetD = 0;
 472
 473     outeriter        = 0;
 474     inneriter        = 0;
 475
 476     for(iidx=0;iidx<4*DIM;iidx++)
 477     {
 478         scratch[iidx] = 0.0;
 479     }
 480
 481     /* Start outer loop over neighborlists */
 482     for(iidx=0; iidx<nri; iidx++)
 483     {
 484         /* Load shift vector for this list */
 485         i_shift_offset   = DIM*shiftidx[iidx];
 486
 487         /* Load limits for loop over neighbors */
 488         j_index_start    = jindex[iidx];
 489         j_index_end      = jindex[iidx+1];
 490
 491         /* Get outer coordinate index */
 492         inr              = iinr[iidx];
 493         i_coord_offset   = DIM*inr;
 494
 495         /* Load i particle coords and add shift vector */
 496         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 497
 498         fix0             = _mm_setzero_ps();
 499         fiy0             = _mm_setzero_ps();
 500         fiz0             = _mm_setzero_ps();
 501
 502         /* Load parameters for i particles */
 503         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 504         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 505
 506         /* Start inner kernel loop */
 507         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 508         {
 509
 510             /* Get j neighbor index, and coordinate index */
 511             jnrA             = jjnr[jidx];
 512             jnrB             = jjnr[jidx+1];
 513             jnrC             = jjnr[jidx+2];
 514             jnrD             = jjnr[jidx+3];
 515             j_coord_offsetA  = DIM*jnrA;
 516             j_coord_offsetB  = DIM*jnrB;
 517             j_coord_offsetC  = DIM*jnrC;
 518             j_coord_offsetD  = DIM*jnrD;
 519
 520             /* load j atom coordinates */
 521             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 522                                               x+j_coord_offsetC,x+j_coord_offsetD,
 523                                               &jx0,&jy0,&jz0);
 524
 525             /* Calculate displacement vector */
 526             dx00             = _mm_sub_ps(ix0,jx0);
 527             dy00             = _mm_sub_ps(iy0,jy0);
 528             dz00             = _mm_sub_ps(iz0,jz0);
 529
 530             /* Calculate squared distance and things based on it */
 531             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 532
 533             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 534
 535             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 536
 537             /* Load parameters for j particles */
 538             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 539                                                               charge+jnrC+0,charge+jnrD+0);
 540             vdwjidx0A        = 2*vdwtype[jnrA+0];
 541             vdwjidx0B        = 2*vdwtype[jnrB+0];
 542             vdwjidx0C        = 2*vdwtype[jnrC+0];
 543             vdwjidx0D        = 2*vdwtype[jnrD+0];
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             r00              = _mm_mul_ps(rsq00,rinv00);
 550
 551             /* Compute parameters for interactions between i and j atoms */
 552             qq00             = _mm_mul_ps(iq0,jq0);
 553             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 554                                          vdwparam+vdwioffset0+vdwjidx0B,
 555                                          vdwparam+vdwioffset0+vdwjidx0C,
 556                                          vdwparam+vdwioffset0+vdwjidx0D,
 557                                          &c6_00,&c12_00);
 558
 559             /* EWALD ELECTROSTATICS */
 560
 561             /* Analytical PME correction */
 562             zeta2            = _mm_mul_ps(beta2,rsq00);
 563             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 564             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 565             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 566             felec            = _mm_mul_ps(qq00,felec);
 567
 568             /* LENNARD-JONES DISPERSION/REPULSION */
 569
 570             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 571             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 572
 573             fscal            = _mm_add_ps(felec,fvdw);
 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             /* Inner loop uses 35 flops */
 590         }
 591
 592         if(jidx<j_index_end)
 593         {
 594
 595             /* Get j neighbor index, and coordinate index */
 596             jnrlistA         = jjnr[jidx];
 597             jnrlistB         = jjnr[jidx+1];
 598             jnrlistC         = jjnr[jidx+2];
 599             jnrlistD         = jjnr[jidx+3];
 600             /* Sign of each element will be negative for non-real atoms.
 601              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 602              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 603              */
 604             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 605             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 606             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 607             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 608             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 609             j_coord_offsetA  = DIM*jnrA;
 610             j_coord_offsetB  = DIM*jnrB;
 611             j_coord_offsetC  = DIM*jnrC;
 612             j_coord_offsetD  = DIM*jnrD;
 613
 614             /* load j atom coordinates */
 615             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 616                                               x+j_coord_offsetC,x+j_coord_offsetD,
 617                                               &jx0,&jy0,&jz0);
 618
 619             /* Calculate displacement vector */
 620             dx00             = _mm_sub_ps(ix0,jx0);
 621             dy00             = _mm_sub_ps(iy0,jy0);
 622             dz00             = _mm_sub_ps(iz0,jz0);
 623
 624             /* Calculate squared distance and things based on it */
 625             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 626
 627             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 628
 629             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 630
 631             /* Load parameters for j particles */
 632             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 633                                                               charge+jnrC+0,charge+jnrD+0);
 634             vdwjidx0A        = 2*vdwtype[jnrA+0];
 635             vdwjidx0B        = 2*vdwtype[jnrB+0];
 636             vdwjidx0C        = 2*vdwtype[jnrC+0];
 637             vdwjidx0D        = 2*vdwtype[jnrD+0];
 638
 639             /**************************
 640              * CALCULATE INTERACTIONS *
 641              **************************/
 642
 643             r00              = _mm_mul_ps(rsq00,rinv00);
 644             r00              = _mm_andnot_ps(dummy_mask,r00);
 645
 646             /* Compute parameters for interactions between i and j atoms */
 647             qq00             = _mm_mul_ps(iq0,jq0);
 648             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 649                                          vdwparam+vdwioffset0+vdwjidx0B,
 650                                          vdwparam+vdwioffset0+vdwjidx0C,
 651                                          vdwparam+vdwioffset0+vdwjidx0D,
 652                                          &c6_00,&c12_00);
 653
 654             /* EWALD ELECTROSTATICS */
 655
 656             /* Analytical PME correction */
 657             zeta2            = _mm_mul_ps(beta2,rsq00);
 658             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 659             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 660             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 661             felec            = _mm_mul_ps(qq00,felec);
 662
 663             /* LENNARD-JONES DISPERSION/REPULSION */
 664
 665             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 666             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 667
 668             fscal            = _mm_add_ps(felec,fvdw);
 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             /* Inner loop uses 36 flops */
 687         }
 688
 689         /* End of innermost loop */
 690
 691         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 692                                               f+i_coord_offset,fshift+i_shift_offset);
 693
 694         /* Increment number of inner iterations */
 695         inneriter                  += j_index_end - j_index_start;
 696
 697         /* Outer loop uses 7 flops */
 698     }
 699
 700     /* Increment number of outer iterations */
 701     outeriter        += nri;
 702
 703     /* Update outer/inner flops */
 704
 705     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*36);
 706 }