src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecGB_VdwLJ_GeomP1P1_sse4_1_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_double 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_double.h"
  34 #include "kernelutil_x86_sse4_1_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse4_1_double
  38  * Electrostatics interaction: GeneralizedBorn
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_sse4_1_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     __m128i          gbitab;
  74     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
  75     __m128d          minushalf = _mm_set1_pd(-0.5);
  76     real             *invsqrta,*dvda,*gbtab;
  77     int              nvdwtype;
  78     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  79     int              *vdwtype;
  80     real             *vdwparam;
  81     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  82     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  83     __m128i          vfitab;
  84     __m128i          ifour       = _mm_set1_epi32(4);
  85     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  86     real             *vftab;
  87     __m128d          dummy_mask,cutoff_mask;
  88     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  89     __m128d          one     = _mm_set1_pd(1.0);
  90     __m128d          two     = _mm_set1_pd(2.0);
  91     x                = xx[0];
  92     f                = ff[0];
  93
  94     nri              = nlist->nri;
  95     iinr             = nlist->iinr;
  96     jindex           = nlist->jindex;
  97     jjnr             = nlist->jjnr;
  98     shiftidx         = nlist->shift;
  99     gid              = nlist->gid;
 100     shiftvec         = fr->shift_vec[0];
 101     fshift           = fr->fshift[0];
 102     facel            = _mm_set1_pd(fr->epsfac);
 103     charge           = mdatoms->chargeA;
 104     nvdwtype         = fr->ntype;
 105     vdwparam         = fr->nbfp;
 106     vdwtype          = mdatoms->typeA;
 107
 108     invsqrta         = fr->invsqrta;
 109     dvda             = fr->dvda;
 110     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 111     gbtab            = fr->gbtab.data;
 112     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 113
 114     /* Avoid stupid compiler warnings */
 115     jnrA = jnrB = 0;
 116     j_coord_offsetA = 0;
 117     j_coord_offsetB = 0;
 118
 119     outeriter        = 0;
 120     inneriter        = 0;
 121
 122     /* Start outer loop over neighborlists */
 123     for(iidx=0; iidx<nri; iidx++)
 124     {
 125         /* Load shift vector for this list */
 126         i_shift_offset   = DIM*shiftidx[iidx];
 127
 128         /* Load limits for loop over neighbors */
 129         j_index_start    = jindex[iidx];
 130         j_index_end      = jindex[iidx+1];
 131
 132         /* Get outer coordinate index */
 133         inr              = iinr[iidx];
 134         i_coord_offset   = DIM*inr;
 135
 136         /* Load i particle coords and add shift vector */
 137         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 138
 139         fix0             = _mm_setzero_pd();
 140         fiy0             = _mm_setzero_pd();
 141         fiz0             = _mm_setzero_pd();
 142
 143         /* Load parameters for i particles */
 144         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 145         isai0            = _mm_load1_pd(invsqrta+inr+0);
 146         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148         /* Reset potential sums */
 149         velecsum         = _mm_setzero_pd();
 150         vgbsum           = _mm_setzero_pd();
 151         vvdwsum          = _mm_setzero_pd();
 152         dvdasum          = _mm_setzero_pd();
 153
 154         /* Start inner kernel loop */
 155         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 156         {
 157
 158             /* Get j neighbor index, and coordinate index */
 159             jnrA             = jjnr[jidx];
 160             jnrB             = jjnr[jidx+1];
 161             j_coord_offsetA  = DIM*jnrA;
 162             j_coord_offsetB  = DIM*jnrB;
 163
 164             /* load j atom coordinates */
 165             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 166                                               &jx0,&jy0,&jz0);
 167
 168             /* Calculate displacement vector */
 169             dx00             = _mm_sub_pd(ix0,jx0);
 170             dy00             = _mm_sub_pd(iy0,jy0);
 171             dz00             = _mm_sub_pd(iz0,jz0);
 172
 173             /* Calculate squared distance and things based on it */
 174             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 175
 176             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 177
 178             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 179
 180             /* Load parameters for j particles */
 181             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 182             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 183             vdwjidx0A        = 2*vdwtype[jnrA+0];
 184             vdwjidx0B        = 2*vdwtype[jnrB+0];
 185
 186             /**************************
 187              * CALCULATE INTERACTIONS *
 188              **************************/
 189
 190             r00              = _mm_mul_pd(rsq00,rinv00);
 191
 192             /* Compute parameters for interactions between i and j atoms */
 193             qq00             = _mm_mul_pd(iq0,jq0);
 194             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 195                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 196
 197             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 198             isaprod          = _mm_mul_pd(isai0,isaj0);
 199             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 200             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 201
 202             /* Calculate generalized born table index - this is a separate table from the normal one,
 203              * but we use the same procedure by multiplying r with scale and truncating to integer.
 204              */
 205             rt               = _mm_mul_pd(r00,gbscale);
 206             gbitab           = _mm_cvttpd_epi32(rt);
 207             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 208             gbitab           = _mm_slli_epi32(gbitab,2);
 209
 210             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 211             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 212             GMX_MM_TRANSPOSE2_PD(Y,F);
 213             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 214             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 215             GMX_MM_TRANSPOSE2_PD(G,H);
 216             Heps             = _mm_mul_pd(gbeps,H);
 217             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 218             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 219             vgb              = _mm_mul_pd(gbqqfactor,VV);
 220
 221             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 222             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 223             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 224             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 225             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 226             velec            = _mm_mul_pd(qq00,rinv00);
 227             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 228
 229             /* LENNARD-JONES DISPERSION/REPULSION */
 230
 231             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 232             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 233             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 234             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 235             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 236
 237             /* Update potential sum for this i atom from the interaction with this j atom. */
 238             velecsum         = _mm_add_pd(velecsum,velec);
 239             vgbsum           = _mm_add_pd(vgbsum,vgb);
 240             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 241
 242             fscal            = _mm_add_pd(felec,fvdw);
 243
 244             /* Calculate temporary vectorial force */
 245             tx               = _mm_mul_pd(fscal,dx00);
 246             ty               = _mm_mul_pd(fscal,dy00);
 247             tz               = _mm_mul_pd(fscal,dz00);
 248
 249             /* Update vectorial force */
 250             fix0             = _mm_add_pd(fix0,tx);
 251             fiy0             = _mm_add_pd(fiy0,ty);
 252             fiz0             = _mm_add_pd(fiz0,tz);
 253
 254             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 255
 256             /* Inner loop uses 71 flops */
 257         }
 258
 259         if(jidx<j_index_end)
 260         {
 261
 262             jnrA             = jjnr[jidx];
 263             j_coord_offsetA  = DIM*jnrA;
 264
 265             /* load j atom coordinates */
 266             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 267                                               &jx0,&jy0,&jz0);
 268
 269             /* Calculate displacement vector */
 270             dx00             = _mm_sub_pd(ix0,jx0);
 271             dy00             = _mm_sub_pd(iy0,jy0);
 272             dz00             = _mm_sub_pd(iz0,jz0);
 273
 274             /* Calculate squared distance and things based on it */
 275             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 276
 277             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 278
 279             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 280
 281             /* Load parameters for j particles */
 282             jq0              = _mm_load_sd(charge+jnrA+0);
 283             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 284             vdwjidx0A        = 2*vdwtype[jnrA+0];
 285
 286             /**************************
 287              * CALCULATE INTERACTIONS *
 288              **************************/
 289
 290             r00              = _mm_mul_pd(rsq00,rinv00);
 291
 292             /* Compute parameters for interactions between i and j atoms */
 293             qq00             = _mm_mul_pd(iq0,jq0);
 294             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 295
 296             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 297             isaprod          = _mm_mul_pd(isai0,isaj0);
 298             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 299             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 300
 301             /* Calculate generalized born table index - this is a separate table from the normal one,
 302              * but we use the same procedure by multiplying r with scale and truncating to integer.
 303              */
 304             rt               = _mm_mul_pd(r00,gbscale);
 305             gbitab           = _mm_cvttpd_epi32(rt);
 306             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 307             gbitab           = _mm_slli_epi32(gbitab,2);
 308
 309             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 310             F                = _mm_setzero_pd();
 311             GMX_MM_TRANSPOSE2_PD(Y,F);
 312             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 313             H                = _mm_setzero_pd();
 314             GMX_MM_TRANSPOSE2_PD(G,H);
 315             Heps             = _mm_mul_pd(gbeps,H);
 316             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 317             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 318             vgb              = _mm_mul_pd(gbqqfactor,VV);
 319
 320             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 321             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 322             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 323             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 324             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 325             velec            = _mm_mul_pd(qq00,rinv00);
 326             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 327
 328             /* LENNARD-JONES DISPERSION/REPULSION */
 329
 330             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 331             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 332             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 333             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 334             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 338             velecsum         = _mm_add_pd(velecsum,velec);
 339             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 340             vgbsum           = _mm_add_pd(vgbsum,vgb);
 341             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 342             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 343
 344             fscal            = _mm_add_pd(felec,fvdw);
 345
 346             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm_mul_pd(fscal,dx00);
 350             ty               = _mm_mul_pd(fscal,dy00);
 351             tz               = _mm_mul_pd(fscal,dz00);
 352
 353             /* Update vectorial force */
 354             fix0             = _mm_add_pd(fix0,tx);
 355             fiy0             = _mm_add_pd(fiy0,ty);
 356             fiz0             = _mm_add_pd(fiz0,tz);
 357
 358             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 359
 360             /* Inner loop uses 71 flops */
 361         }
 362
 363         /* End of innermost loop */
 364
 365         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 366                                               f+i_coord_offset,fshift+i_shift_offset);
 367
 368         ggid                        = gid[iidx];
 369         /* Update potential energies */
 370         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 371         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 372         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 373         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 374         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 375
 376         /* Increment number of inner iterations */
 377         inneriter                  += j_index_end - j_index_start;
 378
 379         /* Outer loop uses 10 flops */
 380     }
 381
 382     /* Increment number of outer iterations */
 383     outeriter        += nri;
 384
 385     /* Update outer/inner flops */
 386
 387     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*71);
 388 }
 389 /*
 390  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_double
 391  * Electrostatics interaction: GeneralizedBorn
 392  * VdW interaction:            LennardJones
 393  * Geometry:                   Particle-Particle
 394  * Calculate force/pot:        Force
 395  */
 396 void
 397 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_sse4_1_double
 398                     (t_nblist * gmx_restrict                nlist,
 399                      rvec * gmx_restrict                    xx,
 400                      rvec * gmx_restrict                    ff,
 401                      t_forcerec * gmx_restrict              fr,
 402                      t_mdatoms * gmx_restrict               mdatoms,
 403                      nb_kernel_data_t * gmx_restrict        kernel_data,
 404                      t_nrnb * gmx_restrict                  nrnb)
 405 {
 406     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 407      * just 0 for non-waters.
 408      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 409      * jnr indices corresponding to data put in the four positions in the SIMD register.
 410      */
 411     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 412     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 413     int              jnrA,jnrB;
 414     int              j_coord_offsetA,j_coord_offsetB;
 415     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 416     real             rcutoff_scalar;
 417     real             *shiftvec,*fshift,*x,*f;
 418     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 419     int              vdwioffset0;
 420     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 421     int              vdwjidx0A,vdwjidx0B;
 422     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 423     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 424     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 425     real             *charge;
 426     __m128i          gbitab;
 427     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 428     __m128d          minushalf = _mm_set1_pd(-0.5);
 429     real             *invsqrta,*dvda,*gbtab;
 430     int              nvdwtype;
 431     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 432     int              *vdwtype;
 433     real             *vdwparam;
 434     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 435     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 436     __m128i          vfitab;
 437     __m128i          ifour       = _mm_set1_epi32(4);
 438     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 439     real             *vftab;
 440     __m128d          dummy_mask,cutoff_mask;
 441     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 442     __m128d          one     = _mm_set1_pd(1.0);
 443     __m128d          two     = _mm_set1_pd(2.0);
 444     x                = xx[0];
 445     f                = ff[0];
 446
 447     nri              = nlist->nri;
 448     iinr             = nlist->iinr;
 449     jindex           = nlist->jindex;
 450     jjnr             = nlist->jjnr;
 451     shiftidx         = nlist->shift;
 452     gid              = nlist->gid;
 453     shiftvec         = fr->shift_vec[0];
 454     fshift           = fr->fshift[0];
 455     facel            = _mm_set1_pd(fr->epsfac);
 456     charge           = mdatoms->chargeA;
 457     nvdwtype         = fr->ntype;
 458     vdwparam         = fr->nbfp;
 459     vdwtype          = mdatoms->typeA;
 460
 461     invsqrta         = fr->invsqrta;
 462     dvda             = fr->dvda;
 463     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 464     gbtab            = fr->gbtab.data;
 465     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 466
 467     /* Avoid stupid compiler warnings */
 468     jnrA = jnrB = 0;
 469     j_coord_offsetA = 0;
 470     j_coord_offsetB = 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
 481         /* Load limits for loop over neighbors */
 482         j_index_start    = jindex[iidx];
 483         j_index_end      = jindex[iidx+1];
 484
 485         /* Get outer coordinate index */
 486         inr              = iinr[iidx];
 487         i_coord_offset   = DIM*inr;
 488
 489         /* Load i particle coords and add shift vector */
 490         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 491
 492         fix0             = _mm_setzero_pd();
 493         fiy0             = _mm_setzero_pd();
 494         fiz0             = _mm_setzero_pd();
 495
 496         /* Load parameters for i particles */
 497         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 498         isai0            = _mm_load1_pd(invsqrta+inr+0);
 499         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 500
 501         dvdasum          = _mm_setzero_pd();
 502
 503         /* Start inner kernel loop */
 504         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 505         {
 506
 507             /* Get j neighbor index, and coordinate index */
 508             jnrA             = jjnr[jidx];
 509             jnrB             = jjnr[jidx+1];
 510             j_coord_offsetA  = DIM*jnrA;
 511             j_coord_offsetB  = DIM*jnrB;
 512
 513             /* load j atom coordinates */
 514             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 515                                               &jx0,&jy0,&jz0);
 516
 517             /* Calculate displacement vector */
 518             dx00             = _mm_sub_pd(ix0,jx0);
 519             dy00             = _mm_sub_pd(iy0,jy0);
 520             dz00             = _mm_sub_pd(iz0,jz0);
 521
 522             /* Calculate squared distance and things based on it */
 523             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 524
 525             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 526
 527             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 528
 529             /* Load parameters for j particles */
 530             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 531             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 532             vdwjidx0A        = 2*vdwtype[jnrA+0];
 533             vdwjidx0B        = 2*vdwtype[jnrB+0];
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             r00              = _mm_mul_pd(rsq00,rinv00);
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq00             = _mm_mul_pd(iq0,jq0);
 543             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 544                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 545
 546             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 547             isaprod          = _mm_mul_pd(isai0,isaj0);
 548             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 549             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 550
 551             /* Calculate generalized born table index - this is a separate table from the normal one,
 552              * but we use the same procedure by multiplying r with scale and truncating to integer.
 553              */
 554             rt               = _mm_mul_pd(r00,gbscale);
 555             gbitab           = _mm_cvttpd_epi32(rt);
 556             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 557             gbitab           = _mm_slli_epi32(gbitab,2);
 558
 559             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 560             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 561             GMX_MM_TRANSPOSE2_PD(Y,F);
 562             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 563             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 564             GMX_MM_TRANSPOSE2_PD(G,H);
 565             Heps             = _mm_mul_pd(gbeps,H);
 566             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 567             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 568             vgb              = _mm_mul_pd(gbqqfactor,VV);
 569
 570             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 571             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 572             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 573             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 574             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 575             velec            = _mm_mul_pd(qq00,rinv00);
 576             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 577
 578             /* LENNARD-JONES DISPERSION/REPULSION */
 579
 580             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 581             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 582
 583             fscal            = _mm_add_pd(felec,fvdw);
 584
 585             /* Calculate temporary vectorial force */
 586             tx               = _mm_mul_pd(fscal,dx00);
 587             ty               = _mm_mul_pd(fscal,dy00);
 588             tz               = _mm_mul_pd(fscal,dz00);
 589
 590             /* Update vectorial force */
 591             fix0             = _mm_add_pd(fix0,tx);
 592             fiy0             = _mm_add_pd(fiy0,ty);
 593             fiz0             = _mm_add_pd(fiz0,tz);
 594
 595             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 596
 597             /* Inner loop uses 64 flops */
 598         }
 599
 600         if(jidx<j_index_end)
 601         {
 602
 603             jnrA             = jjnr[jidx];
 604             j_coord_offsetA  = DIM*jnrA;
 605
 606             /* load j atom coordinates */
 607             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 608                                               &jx0,&jy0,&jz0);
 609
 610             /* Calculate displacement vector */
 611             dx00             = _mm_sub_pd(ix0,jx0);
 612             dy00             = _mm_sub_pd(iy0,jy0);
 613             dz00             = _mm_sub_pd(iz0,jz0);
 614
 615             /* Calculate squared distance and things based on it */
 616             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 617
 618             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 619
 620             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 621
 622             /* Load parameters for j particles */
 623             jq0              = _mm_load_sd(charge+jnrA+0);
 624             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 625             vdwjidx0A        = 2*vdwtype[jnrA+0];
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             r00              = _mm_mul_pd(rsq00,rinv00);
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq00             = _mm_mul_pd(iq0,jq0);
 635             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 636
 637             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 638             isaprod          = _mm_mul_pd(isai0,isaj0);
 639             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 640             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 641
 642             /* Calculate generalized born table index - this is a separate table from the normal one,
 643              * but we use the same procedure by multiplying r with scale and truncating to integer.
 644              */
 645             rt               = _mm_mul_pd(r00,gbscale);
 646             gbitab           = _mm_cvttpd_epi32(rt);
 647             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 648             gbitab           = _mm_slli_epi32(gbitab,2);
 649
 650             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 651             F                = _mm_setzero_pd();
 652             GMX_MM_TRANSPOSE2_PD(Y,F);
 653             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 654             H                = _mm_setzero_pd();
 655             GMX_MM_TRANSPOSE2_PD(G,H);
 656             Heps             = _mm_mul_pd(gbeps,H);
 657             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 658             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 659             vgb              = _mm_mul_pd(gbqqfactor,VV);
 660
 661             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 662             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 663             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 664             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 665             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 666             velec            = _mm_mul_pd(qq00,rinv00);
 667             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 668
 669             /* LENNARD-JONES DISPERSION/REPULSION */
 670
 671             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 672             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 673
 674             fscal            = _mm_add_pd(felec,fvdw);
 675
 676             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 677
 678             /* Calculate temporary vectorial force */
 679             tx               = _mm_mul_pd(fscal,dx00);
 680             ty               = _mm_mul_pd(fscal,dy00);
 681             tz               = _mm_mul_pd(fscal,dz00);
 682
 683             /* Update vectorial force */
 684             fix0             = _mm_add_pd(fix0,tx);
 685             fiy0             = _mm_add_pd(fiy0,ty);
 686             fiz0             = _mm_add_pd(fiz0,tz);
 687
 688             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 689
 690             /* Inner loop uses 64 flops */
 691         }
 692
 693         /* End of innermost loop */
 694
 695         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 696                                               f+i_coord_offset,fshift+i_shift_offset);
 697
 698         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 699         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 700
 701         /* Increment number of inner iterations */
 702         inneriter                  += j_index_end - j_index_start;
 703
 704         /* Outer loop uses 7 flops */
 705     }
 706
 707     /* Increment number of outer iterations */
 708     outeriter        += nri;
 709
 710     /* Update outer/inner flops */
 711
 712     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 713 }