src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwLJ_GeomP1P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_sse2_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_sse2_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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  69     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  78     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  79     __m128i          ewitab;
  80     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  81     real             *ewtab;
  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     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 104     ewtab            = fr->ic->tabq_coul_FDV0;
 105     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 106     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 107
 108     /* Avoid stupid compiler warnings */
 109     jnrA = jnrB = jnrC = jnrD = 0;
 110     j_coord_offsetA = 0;
 111     j_coord_offsetB = 0;
 112     j_coord_offsetC = 0;
 113     j_coord_offsetD = 0;
 114
 115     outeriter        = 0;
 116     inneriter        = 0;
 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         shX              = shiftvec[i_shift_offset+XX];
 124         shY              = shiftvec[i_shift_offset+YY];
 125         shZ              = shiftvec[i_shift_offset+ZZ];
 126
 127         /* Load limits for loop over neighbors */
 128         j_index_start    = jindex[iidx];
 129         j_index_end      = jindex[iidx+1];
 130
 131         /* Get outer coordinate index */
 132         inr              = iinr[iidx];
 133         i_coord_offset   = DIM*inr;
 134
 135         /* Load i particle coords and add shift vector */
 136         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 137         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 138         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 139
 140         fix0             = _mm_setzero_ps();
 141         fiy0             = _mm_setzero_ps();
 142         fiz0             = _mm_setzero_ps();
 143
 144         /* Load parameters for i particles */
 145         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 146         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148         /* Reset potential sums */
 149         velecsum         = _mm_setzero_ps();
 150         vvdwsum          = _mm_setzero_ps();
 151
 152         /* Start inner kernel loop */
 153         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 154         {
 155
 156             /* Get j neighbor index, and coordinate index */
 157             jnrA             = jjnr[jidx];
 158             jnrB             = jjnr[jidx+1];
 159             jnrC             = jjnr[jidx+2];
 160             jnrD             = jjnr[jidx+3];
 161
 162             j_coord_offsetA  = DIM*jnrA;
 163             j_coord_offsetB  = DIM*jnrB;
 164             j_coord_offsetC  = DIM*jnrC;
 165             j_coord_offsetD  = DIM*jnrD;
 166
 167             /* load j atom coordinates */
 168             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               x+j_coord_offsetC,x+j_coord_offsetD,
 170                                               &jx0,&jy0,&jz0);
 171
 172             /* Calculate displacement vector */
 173             dx00             = _mm_sub_ps(ix0,jx0);
 174             dy00             = _mm_sub_ps(iy0,jy0);
 175             dz00             = _mm_sub_ps(iz0,jz0);
 176
 177             /* Calculate squared distance and things based on it */
 178             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 179
 180             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 181
 182             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 183
 184             /* Load parameters for j particles */
 185             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 186                                                               charge+jnrC+0,charge+jnrD+0);
 187             vdwjidx0A        = 2*vdwtype[jnrA+0];
 188             vdwjidx0B        = 2*vdwtype[jnrB+0];
 189             vdwjidx0C        = 2*vdwtype[jnrC+0];
 190             vdwjidx0D        = 2*vdwtype[jnrD+0];
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             r00              = _mm_mul_ps(rsq00,rinv00);
 197
 198             /* Compute parameters for interactions between i and j atoms */
 199             qq00             = _mm_mul_ps(iq0,jq0);
 200             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 201                                          vdwparam+vdwioffset0+vdwjidx0B,
 202                                          vdwparam+vdwioffset0+vdwjidx0C,
 203                                          vdwparam+vdwioffset0+vdwjidx0D,
 204                                          &c6_00,&c12_00);
 205
 206             /* EWALD ELECTROSTATICS */
 207
 208             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 209             ewrt             = _mm_mul_ps(r00,ewtabscale);
 210             ewitab           = _mm_cvttps_epi32(ewrt);
 211             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 212             ewitab           = _mm_slli_epi32(ewitab,2);
 213             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 214             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 215             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 216             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 217             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 218             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 219             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 220             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 221             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 222
 223             /* LENNARD-JONES DISPERSION/REPULSION */
 224
 225             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 226             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 227             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 228             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 229             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 230
 231             /* Update potential sum for this i atom from the interaction with this j atom. */
 232             velecsum         = _mm_add_ps(velecsum,velec);
 233             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 234
 235             fscal            = _mm_add_ps(felec,fvdw);
 236
 237             /* Calculate temporary vectorial force */
 238             tx               = _mm_mul_ps(fscal,dx00);
 239             ty               = _mm_mul_ps(fscal,dy00);
 240             tz               = _mm_mul_ps(fscal,dz00);
 241
 242             /* Update vectorial force */
 243             fix0             = _mm_add_ps(fix0,tx);
 244             fiy0             = _mm_add_ps(fiy0,ty);
 245             fiz0             = _mm_add_ps(fiz0,tz);
 246
 247             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 248                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 249                                                    tx,ty,tz);
 250
 251             /* Inner loop uses 53 flops */
 252         }
 253
 254         if(jidx<j_index_end)
 255         {
 256
 257             /* Get j neighbor index, and coordinate index */
 258             jnrA             = jjnr[jidx];
 259             jnrB             = jjnr[jidx+1];
 260             jnrC             = jjnr[jidx+2];
 261             jnrD             = jjnr[jidx+3];
 262
 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       = (jnrA>=0) ? jnrA : 0;
 269             jnrB       = (jnrB>=0) ? jnrB : 0;
 270             jnrC       = (jnrC>=0) ? jnrC : 0;
 271             jnrD       = (jnrD>=0) ? jnrD : 0;
 272
 273             j_coord_offsetA  = DIM*jnrA;
 274             j_coord_offsetB  = DIM*jnrB;
 275             j_coord_offsetC  = DIM*jnrC;
 276             j_coord_offsetD  = DIM*jnrD;
 277
 278             /* load j atom coordinates */
 279             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 280                                               x+j_coord_offsetC,x+j_coord_offsetD,
 281                                               &jx0,&jy0,&jz0);
 282
 283             /* Calculate displacement vector */
 284             dx00             = _mm_sub_ps(ix0,jx0);
 285             dy00             = _mm_sub_ps(iy0,jy0);
 286             dz00             = _mm_sub_ps(iz0,jz0);
 287
 288             /* Calculate squared distance and things based on it */
 289             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 290
 291             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 292
 293             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 294
 295             /* Load parameters for j particles */
 296             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 297                                                               charge+jnrC+0,charge+jnrD+0);
 298             vdwjidx0A        = 2*vdwtype[jnrA+0];
 299             vdwjidx0B        = 2*vdwtype[jnrB+0];
 300             vdwjidx0C        = 2*vdwtype[jnrC+0];
 301             vdwjidx0D        = 2*vdwtype[jnrD+0];
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             r00              = _mm_mul_ps(rsq00,rinv00);
 308             r00              = _mm_andnot_ps(dummy_mask,r00);
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq00             = _mm_mul_ps(iq0,jq0);
 312             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 313                                          vdwparam+vdwioffset0+vdwjidx0B,
 314                                          vdwparam+vdwioffset0+vdwjidx0C,
 315                                          vdwparam+vdwioffset0+vdwjidx0D,
 316                                          &c6_00,&c12_00);
 317
 318             /* EWALD ELECTROSTATICS */
 319
 320             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 321             ewrt             = _mm_mul_ps(r00,ewtabscale);
 322             ewitab           = _mm_cvttps_epi32(ewrt);
 323             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 324             ewitab           = _mm_slli_epi32(ewitab,2);
 325             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 326             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 327             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 328             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 329             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 330             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 331             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 332             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 333             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 334
 335             /* LENNARD-JONES DISPERSION/REPULSION */
 336
 337             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 338             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 339             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 340             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 341             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velec            = _mm_andnot_ps(dummy_mask,velec);
 345             velecsum         = _mm_add_ps(velecsum,velec);
 346             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 347             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 348
 349             fscal            = _mm_add_ps(felec,fvdw);
 350
 351             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm_mul_ps(fscal,dx00);
 355             ty               = _mm_mul_ps(fscal,dy00);
 356             tz               = _mm_mul_ps(fscal,dz00);
 357
 358             /* Update vectorial force */
 359             fix0             = _mm_add_ps(fix0,tx);
 360             fiy0             = _mm_add_ps(fiy0,ty);
 361             fiz0             = _mm_add_ps(fiz0,tz);
 362
 363             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 364                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 365                                                    tx,ty,tz);
 366
 367             /* Inner loop uses 54 flops */
 368         }
 369
 370         /* End of innermost loop */
 371
 372         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 373                                               f+i_coord_offset,fshift+i_shift_offset);
 374
 375         ggid                        = gid[iidx];
 376         /* Update potential energies */
 377         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 378         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 379
 380         /* Increment number of inner iterations */
 381         inneriter                  += j_index_end - j_index_start;
 382
 383         /* Outer loop uses 12 flops */
 384     }
 385
 386     /* Increment number of outer iterations */
 387     outeriter        += nri;
 388
 389     /* Update outer/inner flops */
 390
 391     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*12 + inneriter*54);
 392 }
 393 /*
 394  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse2_single
 395  * Electrostatics interaction: Ewald
 396  * VdW interaction:            LennardJones
 397  * Geometry:                   Particle-Particle
 398  * Calculate force/pot:        Force
 399  */
 400 void
 401 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse2_single
 402                     (t_nblist * gmx_restrict                nlist,
 403                      rvec * gmx_restrict                    xx,
 404                      rvec * gmx_restrict                    ff,
 405                      t_forcerec * gmx_restrict              fr,
 406                      t_mdatoms * gmx_restrict               mdatoms,
 407                      nb_kernel_data_t * gmx_restrict        kernel_data,
 408                      t_nrnb * gmx_restrict                  nrnb)
 409 {
 410     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 411      * just 0 for non-waters.
 412      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 413      * jnr indices corresponding to data put in the four positions in the SIMD register.
 414      */
 415     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 416     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 417     int              jnrA,jnrB,jnrC,jnrD;
 418     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 419     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 420     real             shX,shY,shZ,rcutoff_scalar;
 421     real             *shiftvec,*fshift,*x,*f;
 422     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 423     int              vdwioffset0;
 424     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 425     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 426     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 427     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 428     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 429     real             *charge;
 430     int              nvdwtype;
 431     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 432     int              *vdwtype;
 433     real             *vdwparam;
 434     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 435     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 436     __m128i          ewitab;
 437     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 438     real             *ewtab;
 439     __m128           dummy_mask,cutoff_mask;
 440     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 441     __m128           one     = _mm_set1_ps(1.0);
 442     __m128           two     = _mm_set1_ps(2.0);
 443     x                = xx[0];
 444     f                = ff[0];
 445
 446     nri              = nlist->nri;
 447     iinr             = nlist->iinr;
 448     jindex           = nlist->jindex;
 449     jjnr             = nlist->jjnr;
 450     shiftidx         = nlist->shift;
 451     gid              = nlist->gid;
 452     shiftvec         = fr->shift_vec[0];
 453     fshift           = fr->fshift[0];
 454     facel            = _mm_set1_ps(fr->epsfac);
 455     charge           = mdatoms->chargeA;
 456     nvdwtype         = fr->ntype;
 457     vdwparam         = fr->nbfp;
 458     vdwtype          = mdatoms->typeA;
 459
 460     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 461     ewtab            = fr->ic->tabq_coul_F;
 462     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 463     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 464
 465     /* Avoid stupid compiler warnings */
 466     jnrA = jnrB = jnrC = jnrD = 0;
 467     j_coord_offsetA = 0;
 468     j_coord_offsetB = 0;
 469     j_coord_offsetC = 0;
 470     j_coord_offsetD = 0;
 471
 472     outeriter        = 0;
 473     inneriter        = 0;
 474
 475     /* Start outer loop over neighborlists */
 476     for(iidx=0; iidx<nri; iidx++)
 477     {
 478         /* Load shift vector for this list */
 479         i_shift_offset   = DIM*shiftidx[iidx];
 480         shX              = shiftvec[i_shift_offset+XX];
 481         shY              = shiftvec[i_shift_offset+YY];
 482         shZ              = shiftvec[i_shift_offset+ZZ];
 483
 484         /* Load limits for loop over neighbors */
 485         j_index_start    = jindex[iidx];
 486         j_index_end      = jindex[iidx+1];
 487
 488         /* Get outer coordinate index */
 489         inr              = iinr[iidx];
 490         i_coord_offset   = DIM*inr;
 491
 492         /* Load i particle coords and add shift vector */
 493         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 494         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 495         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 496
 497         fix0             = _mm_setzero_ps();
 498         fiy0             = _mm_setzero_ps();
 499         fiz0             = _mm_setzero_ps();
 500
 501         /* Load parameters for i particles */
 502         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 503         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 504
 505         /* Start inner kernel loop */
 506         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 507         {
 508
 509             /* Get j neighbor index, and coordinate index */
 510             jnrA             = jjnr[jidx];
 511             jnrB             = jjnr[jidx+1];
 512             jnrC             = jjnr[jidx+2];
 513             jnrD             = jjnr[jidx+3];
 514
 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             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 562             ewrt             = _mm_mul_ps(r00,ewtabscale);
 563             ewitab           = _mm_cvttps_epi32(ewrt);
 564             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 565             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 566                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 567                                          &ewtabF,&ewtabFn);
 568             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 569             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 570
 571             /* LENNARD-JONES DISPERSION/REPULSION */
 572
 573             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 574             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 575
 576             fscal            = _mm_add_ps(felec,fvdw);
 577
 578             /* Calculate temporary vectorial force */
 579             tx               = _mm_mul_ps(fscal,dx00);
 580             ty               = _mm_mul_ps(fscal,dy00);
 581             tz               = _mm_mul_ps(fscal,dz00);
 582
 583             /* Update vectorial force */
 584             fix0             = _mm_add_ps(fix0,tx);
 585             fiy0             = _mm_add_ps(fiy0,ty);
 586             fiz0             = _mm_add_ps(fiz0,tz);
 587
 588             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 589                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 590                                                    tx,ty,tz);
 591
 592             /* Inner loop uses 43 flops */
 593         }
 594
 595         if(jidx<j_index_end)
 596         {
 597
 598             /* Get j neighbor index, and coordinate index */
 599             jnrA             = jjnr[jidx];
 600             jnrB             = jjnr[jidx+1];
 601             jnrC             = jjnr[jidx+2];
 602             jnrD             = jjnr[jidx+3];
 603
 604             /* Sign of each element will be negative for non-real atoms.
 605              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 606              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 607              */
 608             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 609             jnrA       = (jnrA>=0) ? jnrA : 0;
 610             jnrB       = (jnrB>=0) ? jnrB : 0;
 611             jnrC       = (jnrC>=0) ? jnrC : 0;
 612             jnrD       = (jnrD>=0) ? jnrD : 0;
 613
 614             j_coord_offsetA  = DIM*jnrA;
 615             j_coord_offsetB  = DIM*jnrB;
 616             j_coord_offsetC  = DIM*jnrC;
 617             j_coord_offsetD  = DIM*jnrD;
 618
 619             /* load j atom coordinates */
 620             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 621                                               x+j_coord_offsetC,x+j_coord_offsetD,
 622                                               &jx0,&jy0,&jz0);
 623
 624             /* Calculate displacement vector */
 625             dx00             = _mm_sub_ps(ix0,jx0);
 626             dy00             = _mm_sub_ps(iy0,jy0);
 627             dz00             = _mm_sub_ps(iz0,jz0);
 628
 629             /* Calculate squared distance and things based on it */
 630             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 631
 632             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 633
 634             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 635
 636             /* Load parameters for j particles */
 637             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 638                                                               charge+jnrC+0,charge+jnrD+0);
 639             vdwjidx0A        = 2*vdwtype[jnrA+0];
 640             vdwjidx0B        = 2*vdwtype[jnrB+0];
 641             vdwjidx0C        = 2*vdwtype[jnrC+0];
 642             vdwjidx0D        = 2*vdwtype[jnrD+0];
 643
 644             /**************************
 645              * CALCULATE INTERACTIONS *
 646              **************************/
 647
 648             r00              = _mm_mul_ps(rsq00,rinv00);
 649             r00              = _mm_andnot_ps(dummy_mask,r00);
 650
 651             /* Compute parameters for interactions between i and j atoms */
 652             qq00             = _mm_mul_ps(iq0,jq0);
 653             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 654                                          vdwparam+vdwioffset0+vdwjidx0B,
 655                                          vdwparam+vdwioffset0+vdwjidx0C,
 656                                          vdwparam+vdwioffset0+vdwjidx0D,
 657                                          &c6_00,&c12_00);
 658
 659             /* EWALD ELECTROSTATICS */
 660
 661             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 662             ewrt             = _mm_mul_ps(r00,ewtabscale);
 663             ewitab           = _mm_cvttps_epi32(ewrt);
 664             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 665             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 666                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 667                                          &ewtabF,&ewtabFn);
 668             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 669             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 670
 671             /* LENNARD-JONES DISPERSION/REPULSION */
 672
 673             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 674             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 675
 676             fscal            = _mm_add_ps(felec,fvdw);
 677
 678             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 679
 680             /* Calculate temporary vectorial force */
 681             tx               = _mm_mul_ps(fscal,dx00);
 682             ty               = _mm_mul_ps(fscal,dy00);
 683             tz               = _mm_mul_ps(fscal,dz00);
 684
 685             /* Update vectorial force */
 686             fix0             = _mm_add_ps(fix0,tx);
 687             fiy0             = _mm_add_ps(fiy0,ty);
 688             fiz0             = _mm_add_ps(fiz0,tz);
 689
 690             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 691                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 692                                                    tx,ty,tz);
 693
 694             /* Inner loop uses 44 flops */
 695         }
 696
 697         /* End of innermost loop */
 698
 699         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 700                                               f+i_coord_offset,fshift+i_shift_offset);
 701
 702         /* Increment number of inner iterations */
 703         inneriter                  += j_index_end - j_index_start;
 704
 705         /* Outer loop uses 10 flops */
 706     }
 707
 708     /* Increment number of outer iterations */
 709     outeriter        += nri;
 710
 711     /* Update outer/inner flops */
 712
 713     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*10 + inneriter*44);
 714 }