src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJ_GeomP1P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_sse4_1_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_sse4_1_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 SSE, 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           tx,ty,tz,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,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  84     real             *ewtab;
  85     __m128           dummy_mask,cutoff_mask;
  86     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  87     __m128           one     = _mm_set1_ps(1.0);
  88     __m128           two     = _mm_set1_ps(2.0);
  89     x                = xx[0];
  90     f                = ff[0];
  91
  92     nri              = nlist->nri;
  93     iinr             = nlist->iinr;
  94     jindex           = nlist->jindex;
  95     jjnr             = nlist->jjnr;
  96     shiftidx         = nlist->shift;
  97     gid              = nlist->gid;
  98     shiftvec         = fr->shift_vec[0];
  99     fshift           = fr->fshift[0];
 100     facel            = _mm_set1_ps(fr->epsfac);
 101     charge           = mdatoms->chargeA;
 102     nvdwtype         = fr->ntype;
 103     vdwparam         = fr->nbfp;
 104     vdwtype          = mdatoms->typeA;
 105
 106     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 107     ewtab            = fr->ic->tabq_coul_FDV0;
 108     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 109     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 110
 111     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = jnrC = jnrD = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115     j_coord_offsetC = 0;
 116     j_coord_offsetD = 0;
 117
 118     outeriter        = 0;
 119     inneriter        = 0;
 120
 121     for(iidx=0;iidx<4*DIM;iidx++)
 122     {
 123         scratch[iidx] = 0.0;
 124     }
 125
 126     /* Start outer loop over neighborlists */
 127     for(iidx=0; iidx<nri; iidx++)
 128     {
 129         /* Load shift vector for this list */
 130         i_shift_offset   = DIM*shiftidx[iidx];
 131
 132         /* Load limits for loop over neighbors */
 133         j_index_start    = jindex[iidx];
 134         j_index_end      = jindex[iidx+1];
 135
 136         /* Get outer coordinate index */
 137         inr              = iinr[iidx];
 138         i_coord_offset   = DIM*inr;
 139
 140         /* Load i particle coords and add shift vector */
 141         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 142
 143         fix0             = _mm_setzero_ps();
 144         fiy0             = _mm_setzero_ps();
 145         fiz0             = _mm_setzero_ps();
 146
 147         /* Load parameters for i particles */
 148         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 149         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 150
 151         /* Reset potential sums */
 152         velecsum         = _mm_setzero_ps();
 153         vvdwsum          = _mm_setzero_ps();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             jnrC             = jjnr[jidx+2];
 163             jnrD             = jjnr[jidx+3];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166             j_coord_offsetC  = DIM*jnrC;
 167             j_coord_offsetD  = DIM*jnrD;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               x+j_coord_offsetC,x+j_coord_offsetD,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_ps(ix0,jx0);
 176             dy00             = _mm_sub_ps(iy0,jy0);
 177             dz00             = _mm_sub_ps(iz0,jz0);
 178
 179             /* Calculate squared distance and things based on it */
 180             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 181
 182             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 183
 184             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 185
 186             /* Load parameters for j particles */
 187             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 188                                                               charge+jnrC+0,charge+jnrD+0);
 189             vdwjidx0A        = 2*vdwtype[jnrA+0];
 190             vdwjidx0B        = 2*vdwtype[jnrB+0];
 191             vdwjidx0C        = 2*vdwtype[jnrC+0];
 192             vdwjidx0D        = 2*vdwtype[jnrD+0];
 193
 194             /**************************
 195              * CALCULATE INTERACTIONS *
 196              **************************/
 197
 198             r00              = _mm_mul_ps(rsq00,rinv00);
 199
 200             /* Compute parameters for interactions between i and j atoms */
 201             qq00             = _mm_mul_ps(iq0,jq0);
 202             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 203                                          vdwparam+vdwioffset0+vdwjidx0B,
 204                                          vdwparam+vdwioffset0+vdwjidx0C,
 205                                          vdwparam+vdwioffset0+vdwjidx0D,
 206                                          &c6_00,&c12_00);
 207
 208             /* EWALD ELECTROSTATICS */
 209
 210             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 211             ewrt             = _mm_mul_ps(r00,ewtabscale);
 212             ewitab           = _mm_cvttps_epi32(ewrt);
 213             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 214             ewitab           = _mm_slli_epi32(ewitab,2);
 215             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 216             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 217             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 218             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 219             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 220             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 221             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 222             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 223             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 224
 225             /* LENNARD-JONES DISPERSION/REPULSION */
 226
 227             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 228             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 229             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 230             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 231             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 232
 233             /* Update potential sum for this i atom from the interaction with this j atom. */
 234             velecsum         = _mm_add_ps(velecsum,velec);
 235             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 236
 237             fscal            = _mm_add_ps(felec,fvdw);
 238
 239             /* Calculate temporary vectorial force */
 240             tx               = _mm_mul_ps(fscal,dx00);
 241             ty               = _mm_mul_ps(fscal,dy00);
 242             tz               = _mm_mul_ps(fscal,dz00);
 243
 244             /* Update vectorial force */
 245             fix0             = _mm_add_ps(fix0,tx);
 246             fiy0             = _mm_add_ps(fiy0,ty);
 247             fiz0             = _mm_add_ps(fiz0,tz);
 248
 249             fjptrA             = f+j_coord_offsetA;
 250             fjptrB             = f+j_coord_offsetB;
 251             fjptrC             = f+j_coord_offsetC;
 252             fjptrD             = f+j_coord_offsetD;
 253             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 254
 255             /* Inner loop uses 53 flops */
 256         }
 257
 258         if(jidx<j_index_end)
 259         {
 260
 261             /* Get j neighbor index, and coordinate index */
 262             jnrlistA         = jjnr[jidx];
 263             jnrlistB         = jjnr[jidx+1];
 264             jnrlistC         = jjnr[jidx+2];
 265             jnrlistD         = jjnr[jidx+3];
 266             /* Sign of each element will be negative for non-real atoms.
 267              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 268              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 269              */
 270             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 271             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 272             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 273             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 274             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 275             j_coord_offsetA  = DIM*jnrA;
 276             j_coord_offsetB  = DIM*jnrB;
 277             j_coord_offsetC  = DIM*jnrC;
 278             j_coord_offsetD  = DIM*jnrD;
 279
 280             /* load j atom coordinates */
 281             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 282                                               x+j_coord_offsetC,x+j_coord_offsetD,
 283                                               &jx0,&jy0,&jz0);
 284
 285             /* Calculate displacement vector */
 286             dx00             = _mm_sub_ps(ix0,jx0);
 287             dy00             = _mm_sub_ps(iy0,jy0);
 288             dz00             = _mm_sub_ps(iz0,jz0);
 289
 290             /* Calculate squared distance and things based on it */
 291             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 292
 293             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 294
 295             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 296
 297             /* Load parameters for j particles */
 298             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 299                                                               charge+jnrC+0,charge+jnrD+0);
 300             vdwjidx0A        = 2*vdwtype[jnrA+0];
 301             vdwjidx0B        = 2*vdwtype[jnrB+0];
 302             vdwjidx0C        = 2*vdwtype[jnrC+0];
 303             vdwjidx0D        = 2*vdwtype[jnrD+0];
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             r00              = _mm_mul_ps(rsq00,rinv00);
 310             r00              = _mm_andnot_ps(dummy_mask,r00);
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq00             = _mm_mul_ps(iq0,jq0);
 314             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 315                                          vdwparam+vdwioffset0+vdwjidx0B,
 316                                          vdwparam+vdwioffset0+vdwjidx0C,
 317                                          vdwparam+vdwioffset0+vdwjidx0D,
 318                                          &c6_00,&c12_00);
 319
 320             /* EWALD ELECTROSTATICS */
 321
 322             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 323             ewrt             = _mm_mul_ps(r00,ewtabscale);
 324             ewitab           = _mm_cvttps_epi32(ewrt);
 325             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 326             ewitab           = _mm_slli_epi32(ewitab,2);
 327             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 328             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 329             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 330             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 331             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 332             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 333             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 334             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 335             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 336
 337             /* LENNARD-JONES DISPERSION/REPULSION */
 338
 339             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 340             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 341             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 342             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 343             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velec            = _mm_andnot_ps(dummy_mask,velec);
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 349             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 350
 351             fscal            = _mm_add_ps(felec,fvdw);
 352
 353             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 354
 355             /* Calculate temporary vectorial force */
 356             tx               = _mm_mul_ps(fscal,dx00);
 357             ty               = _mm_mul_ps(fscal,dy00);
 358             tz               = _mm_mul_ps(fscal,dz00);
 359
 360             /* Update vectorial force */
 361             fix0             = _mm_add_ps(fix0,tx);
 362             fiy0             = _mm_add_ps(fiy0,ty);
 363             fiz0             = _mm_add_ps(fiz0,tz);
 364
 365             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 366             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 367             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 368             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 369             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 370
 371             /* Inner loop uses 54 flops */
 372         }
 373
 374         /* End of innermost loop */
 375
 376         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 377                                               f+i_coord_offset,fshift+i_shift_offset);
 378
 379         ggid                        = gid[iidx];
 380         /* Update potential energies */
 381         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 382         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 383
 384         /* Increment number of inner iterations */
 385         inneriter                  += j_index_end - j_index_start;
 386
 387         /* Outer loop uses 9 flops */
 388     }
 389
 390     /* Increment number of outer iterations */
 391     outeriter        += nri;
 392
 393     /* Update outer/inner flops */
 394
 395     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
 396 }
 397 /*
 398  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse4_1_single
 399  * Electrostatics interaction: Ewald
 400  * VdW interaction:            LennardJones
 401  * Geometry:                   Particle-Particle
 402  * Calculate force/pot:        Force
 403  */
 404 void
 405 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse4_1_single
 406                     (t_nblist * gmx_restrict                nlist,
 407                      rvec * gmx_restrict                    xx,
 408                      rvec * gmx_restrict                    ff,
 409                      t_forcerec * gmx_restrict              fr,
 410                      t_mdatoms * gmx_restrict               mdatoms,
 411                      nb_kernel_data_t * gmx_restrict        kernel_data,
 412                      t_nrnb * gmx_restrict                  nrnb)
 413 {
 414     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 415      * just 0 for non-waters.
 416      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 417      * jnr indices corresponding to data put in the four positions in the SIMD register.
 418      */
 419     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 420     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 421     int              jnrA,jnrB,jnrC,jnrD;
 422     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 423     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 424     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 425     real             rcutoff_scalar;
 426     real             *shiftvec,*fshift,*x,*f;
 427     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 428     real             scratch[4*DIM];
 429     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 430     int              vdwioffset0;
 431     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 432     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 433     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 434     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 435     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 436     real             *charge;
 437     int              nvdwtype;
 438     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 439     int              *vdwtype;
 440     real             *vdwparam;
 441     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 442     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 443     __m128i          ewitab;
 444     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 445     real             *ewtab;
 446     __m128           dummy_mask,cutoff_mask;
 447     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 448     __m128           one     = _mm_set1_ps(1.0);
 449     __m128           two     = _mm_set1_ps(2.0);
 450     x                = xx[0];
 451     f                = ff[0];
 452
 453     nri              = nlist->nri;
 454     iinr             = nlist->iinr;
 455     jindex           = nlist->jindex;
 456     jjnr             = nlist->jjnr;
 457     shiftidx         = nlist->shift;
 458     gid              = nlist->gid;
 459     shiftvec         = fr->shift_vec[0];
 460     fshift           = fr->fshift[0];
 461     facel            = _mm_set1_ps(fr->epsfac);
 462     charge           = mdatoms->chargeA;
 463     nvdwtype         = fr->ntype;
 464     vdwparam         = fr->nbfp;
 465     vdwtype          = mdatoms->typeA;
 466
 467     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 468     ewtab            = fr->ic->tabq_coul_F;
 469     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 470     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 471
 472     /* Avoid stupid compiler warnings */
 473     jnrA = jnrB = jnrC = jnrD = 0;
 474     j_coord_offsetA = 0;
 475     j_coord_offsetB = 0;
 476     j_coord_offsetC = 0;
 477     j_coord_offsetD = 0;
 478
 479     outeriter        = 0;
 480     inneriter        = 0;
 481
 482     for(iidx=0;iidx<4*DIM;iidx++)
 483     {
 484         scratch[iidx] = 0.0;
 485     }
 486
 487     /* Start outer loop over neighborlists */
 488     for(iidx=0; iidx<nri; iidx++)
 489     {
 490         /* Load shift vector for this list */
 491         i_shift_offset   = DIM*shiftidx[iidx];
 492
 493         /* Load limits for loop over neighbors */
 494         j_index_start    = jindex[iidx];
 495         j_index_end      = jindex[iidx+1];
 496
 497         /* Get outer coordinate index */
 498         inr              = iinr[iidx];
 499         i_coord_offset   = DIM*inr;
 500
 501         /* Load i particle coords and add shift vector */
 502         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 503
 504         fix0             = _mm_setzero_ps();
 505         fiy0             = _mm_setzero_ps();
 506         fiz0             = _mm_setzero_ps();
 507
 508         /* Load parameters for i particles */
 509         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 510         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 511
 512         /* Start inner kernel loop */
 513         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 514         {
 515
 516             /* Get j neighbor index, and coordinate index */
 517             jnrA             = jjnr[jidx];
 518             jnrB             = jjnr[jidx+1];
 519             jnrC             = jjnr[jidx+2];
 520             jnrD             = jjnr[jidx+3];
 521             j_coord_offsetA  = DIM*jnrA;
 522             j_coord_offsetB  = DIM*jnrB;
 523             j_coord_offsetC  = DIM*jnrC;
 524             j_coord_offsetD  = DIM*jnrD;
 525
 526             /* load j atom coordinates */
 527             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 528                                               x+j_coord_offsetC,x+j_coord_offsetD,
 529                                               &jx0,&jy0,&jz0);
 530
 531             /* Calculate displacement vector */
 532             dx00             = _mm_sub_ps(ix0,jx0);
 533             dy00             = _mm_sub_ps(iy0,jy0);
 534             dz00             = _mm_sub_ps(iz0,jz0);
 535
 536             /* Calculate squared distance and things based on it */
 537             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 538
 539             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 540
 541             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 542
 543             /* Load parameters for j particles */
 544             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 545                                                               charge+jnrC+0,charge+jnrD+0);
 546             vdwjidx0A        = 2*vdwtype[jnrA+0];
 547             vdwjidx0B        = 2*vdwtype[jnrB+0];
 548             vdwjidx0C        = 2*vdwtype[jnrC+0];
 549             vdwjidx0D        = 2*vdwtype[jnrD+0];
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             r00              = _mm_mul_ps(rsq00,rinv00);
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             qq00             = _mm_mul_ps(iq0,jq0);
 559             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 560                                          vdwparam+vdwioffset0+vdwjidx0B,
 561                                          vdwparam+vdwioffset0+vdwjidx0C,
 562                                          vdwparam+vdwioffset0+vdwjidx0D,
 563                                          &c6_00,&c12_00);
 564
 565             /* EWALD ELECTROSTATICS */
 566
 567             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 568             ewrt             = _mm_mul_ps(r00,ewtabscale);
 569             ewitab           = _mm_cvttps_epi32(ewrt);
 570             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 571             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 572                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 573                                          &ewtabF,&ewtabFn);
 574             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 575             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 576
 577             /* LENNARD-JONES DISPERSION/REPULSION */
 578
 579             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 580             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 581
 582             fscal            = _mm_add_ps(felec,fvdw);
 583
 584             /* Calculate temporary vectorial force */
 585             tx               = _mm_mul_ps(fscal,dx00);
 586             ty               = _mm_mul_ps(fscal,dy00);
 587             tz               = _mm_mul_ps(fscal,dz00);
 588
 589             /* Update vectorial force */
 590             fix0             = _mm_add_ps(fix0,tx);
 591             fiy0             = _mm_add_ps(fiy0,ty);
 592             fiz0             = _mm_add_ps(fiz0,tz);
 593
 594             fjptrA             = f+j_coord_offsetA;
 595             fjptrB             = f+j_coord_offsetB;
 596             fjptrC             = f+j_coord_offsetC;
 597             fjptrD             = f+j_coord_offsetD;
 598             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 599
 600             /* Inner loop uses 43 flops */
 601         }
 602
 603         if(jidx<j_index_end)
 604         {
 605
 606             /* Get j neighbor index, and coordinate index */
 607             jnrlistA         = jjnr[jidx];
 608             jnrlistB         = jjnr[jidx+1];
 609             jnrlistC         = jjnr[jidx+2];
 610             jnrlistD         = jjnr[jidx+3];
 611             /* Sign of each element will be negative for non-real atoms.
 612              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 613              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 614              */
 615             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 616             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 617             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 618             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 619             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 620             j_coord_offsetA  = DIM*jnrA;
 621             j_coord_offsetB  = DIM*jnrB;
 622             j_coord_offsetC  = DIM*jnrC;
 623             j_coord_offsetD  = DIM*jnrD;
 624
 625             /* load j atom coordinates */
 626             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 627                                               x+j_coord_offsetC,x+j_coord_offsetD,
 628                                               &jx0,&jy0,&jz0);
 629
 630             /* Calculate displacement vector */
 631             dx00             = _mm_sub_ps(ix0,jx0);
 632             dy00             = _mm_sub_ps(iy0,jy0);
 633             dz00             = _mm_sub_ps(iz0,jz0);
 634
 635             /* Calculate squared distance and things based on it */
 636             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 637
 638             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 639
 640             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 641
 642             /* Load parameters for j particles */
 643             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 644                                                               charge+jnrC+0,charge+jnrD+0);
 645             vdwjidx0A        = 2*vdwtype[jnrA+0];
 646             vdwjidx0B        = 2*vdwtype[jnrB+0];
 647             vdwjidx0C        = 2*vdwtype[jnrC+0];
 648             vdwjidx0D        = 2*vdwtype[jnrD+0];
 649
 650             /**************************
 651              * CALCULATE INTERACTIONS *
 652              **************************/
 653
 654             r00              = _mm_mul_ps(rsq00,rinv00);
 655             r00              = _mm_andnot_ps(dummy_mask,r00);
 656
 657             /* Compute parameters for interactions between i and j atoms */
 658             qq00             = _mm_mul_ps(iq0,jq0);
 659             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 660                                          vdwparam+vdwioffset0+vdwjidx0B,
 661                                          vdwparam+vdwioffset0+vdwjidx0C,
 662                                          vdwparam+vdwioffset0+vdwjidx0D,
 663                                          &c6_00,&c12_00);
 664
 665             /* EWALD ELECTROSTATICS */
 666
 667             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 668             ewrt             = _mm_mul_ps(r00,ewtabscale);
 669             ewitab           = _mm_cvttps_epi32(ewrt);
 670             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 671             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 672                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 673                                          &ewtabF,&ewtabFn);
 674             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 675             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 676
 677             /* LENNARD-JONES DISPERSION/REPULSION */
 678
 679             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 680             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 681
 682             fscal            = _mm_add_ps(felec,fvdw);
 683
 684             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 685
 686             /* Calculate temporary vectorial force */
 687             tx               = _mm_mul_ps(fscal,dx00);
 688             ty               = _mm_mul_ps(fscal,dy00);
 689             tz               = _mm_mul_ps(fscal,dz00);
 690
 691             /* Update vectorial force */
 692             fix0             = _mm_add_ps(fix0,tx);
 693             fiy0             = _mm_add_ps(fiy0,ty);
 694             fiz0             = _mm_add_ps(fiz0,tz);
 695
 696             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 697             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 698             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 699             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 700             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 701
 702             /* Inner loop uses 44 flops */
 703         }
 704
 705         /* End of innermost loop */
 706
 707         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 708                                               f+i_coord_offset,fshift+i_shift_offset);
 709
 710         /* Increment number of inner iterations */
 711         inneriter                  += j_index_end - j_index_start;
 712
 713         /* Outer loop uses 7 flops */
 714     }
 715
 716     /* Increment number of outer iterations */
 717     outeriter        += nri;
 718
 719     /* Update outer/inner flops */
 720
 721     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*44);
 722 }