src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecGB_VdwNone_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_ElecGB_VdwNone_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: GeneralizedBorn
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecGB_VdwNone_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              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     __m128i          gbitab;
  77     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  78     __m128           minushalf = _mm_set1_ps(-0.5);
  79     real             *invsqrta,*dvda,*gbtab;
  80     __m128i          vfitab;
  81     __m128i          ifour       = _mm_set1_epi32(4);
  82     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  83     real             *vftab;
  84     __m128           dummy_mask,cutoff_mask;
  85     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  86     __m128           one     = _mm_set1_ps(1.0);
  87     __m128           two     = _mm_set1_ps(2.0);
  88     x                = xx[0];
  89     f                = ff[0];
  90
  91     nri              = nlist->nri;
  92     iinr             = nlist->iinr;
  93     jindex           = nlist->jindex;
  94     jjnr             = nlist->jjnr;
  95     shiftidx         = nlist->shift;
  96     gid              = nlist->gid;
  97     shiftvec         = fr->shift_vec[0];
  98     fshift           = fr->fshift[0];
  99     facel            = _mm_set1_ps(fr->epsfac);
 100     charge           = mdatoms->chargeA;
 101
 102     invsqrta         = fr->invsqrta;
 103     dvda             = fr->dvda;
 104     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 105     gbtab            = fr->gbtab.data;
 106     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 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     for(iidx=0;iidx<4*DIM;iidx++)
 119     {
 120         scratch[iidx] = 0.0;
 121     }
 122
 123     /* Start outer loop over neighborlists */
 124     for(iidx=0; iidx<nri; iidx++)
 125     {
 126         /* Load shift vector for this list */
 127         i_shift_offset   = DIM*shiftidx[iidx];
 128
 129         /* Load limits for loop over neighbors */
 130         j_index_start    = jindex[iidx];
 131         j_index_end      = jindex[iidx+1];
 132
 133         /* Get outer coordinate index */
 134         inr              = iinr[iidx];
 135         i_coord_offset   = DIM*inr;
 136
 137         /* Load i particle coords and add shift vector */
 138         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 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         isai0            = _mm_load1_ps(invsqrta+inr+0);
 147
 148         /* Reset potential sums */
 149         velecsum         = _mm_setzero_ps();
 150         vgbsum           = _mm_setzero_ps();
 151         dvdasum          = _mm_setzero_ps();
 152
 153         /* Start inner kernel loop */
 154         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 155         {
 156
 157             /* Get j neighbor index, and coordinate index */
 158             jnrA             = jjnr[jidx];
 159             jnrB             = jjnr[jidx+1];
 160             jnrC             = jjnr[jidx+2];
 161             jnrD             = jjnr[jidx+3];
 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             /* Load parameters for j particles */
 183             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 184                                                               charge+jnrC+0,charge+jnrD+0);
 185             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 186                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 187
 188             /**************************
 189              * CALCULATE INTERACTIONS *
 190              **************************/
 191
 192             r00              = _mm_mul_ps(rsq00,rinv00);
 193
 194             /* Compute parameters for interactions between i and j atoms */
 195             qq00             = _mm_mul_ps(iq0,jq0);
 196
 197             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 198             isaprod          = _mm_mul_ps(isai0,isaj0);
 199             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 200             gbscale          = _mm_mul_ps(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_ps(r00,gbscale);
 206             gbitab           = _mm_cvttps_epi32(rt);
 207             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 208             gbitab           = _mm_slli_epi32(gbitab,2);
 209
 210             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 211             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 212             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 213             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 214             _MM_TRANSPOSE4_PS(Y,F,G,H);
 215             Heps             = _mm_mul_ps(gbeps,H);
 216             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 217             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 218             vgb              = _mm_mul_ps(gbqqfactor,VV);
 219
 220             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 221             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 222             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 223             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 224             fjptrA           = dvda+jnrA;
 225             fjptrB           = dvda+jnrB;
 226             fjptrC           = dvda+jnrC;
 227             fjptrD           = dvda+jnrD;
 228             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 229             velec            = _mm_mul_ps(qq00,rinv00);
 230             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 231
 232             /* Update potential sum for this i atom from the interaction with this j atom. */
 233             velecsum         = _mm_add_ps(velecsum,velec);
 234             vgbsum           = _mm_add_ps(vgbsum,vgb);
 235
 236             fscal            = felec;
 237
 238             /* Calculate temporary vectorial force */
 239             tx               = _mm_mul_ps(fscal,dx00);
 240             ty               = _mm_mul_ps(fscal,dy00);
 241             tz               = _mm_mul_ps(fscal,dz00);
 242
 243             /* Update vectorial force */
 244             fix0             = _mm_add_ps(fix0,tx);
 245             fiy0             = _mm_add_ps(fiy0,ty);
 246             fiz0             = _mm_add_ps(fiz0,tz);
 247
 248             fjptrA             = f+j_coord_offsetA;
 249             fjptrB             = f+j_coord_offsetB;
 250             fjptrC             = f+j_coord_offsetC;
 251             fjptrD             = f+j_coord_offsetD;
 252             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 253
 254             /* Inner loop uses 58 flops */
 255         }
 256
 257         if(jidx<j_index_end)
 258         {
 259
 260             /* Get j neighbor index, and coordinate index */
 261             jnrlistA         = jjnr[jidx];
 262             jnrlistB         = jjnr[jidx+1];
 263             jnrlistC         = jjnr[jidx+2];
 264             jnrlistD         = jjnr[jidx+3];
 265             /* Sign of each element will be negative for non-real atoms.
 266              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 267              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 268              */
 269             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 270             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 271             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 272             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 273             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 274             j_coord_offsetA  = DIM*jnrA;
 275             j_coord_offsetB  = DIM*jnrB;
 276             j_coord_offsetC  = DIM*jnrC;
 277             j_coord_offsetD  = DIM*jnrD;
 278
 279             /* load j atom coordinates */
 280             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 281                                               x+j_coord_offsetC,x+j_coord_offsetD,
 282                                               &jx0,&jy0,&jz0);
 283
 284             /* Calculate displacement vector */
 285             dx00             = _mm_sub_ps(ix0,jx0);
 286             dy00             = _mm_sub_ps(iy0,jy0);
 287             dz00             = _mm_sub_ps(iz0,jz0);
 288
 289             /* Calculate squared distance and things based on it */
 290             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 291
 292             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 293
 294             /* Load parameters for j particles */
 295             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 296                                                               charge+jnrC+0,charge+jnrD+0);
 297             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 298                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r00              = _mm_mul_ps(rsq00,rinv00);
 305             r00              = _mm_andnot_ps(dummy_mask,r00);
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq00             = _mm_mul_ps(iq0,jq0);
 309
 310             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 311             isaprod          = _mm_mul_ps(isai0,isaj0);
 312             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 313             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 314
 315             /* Calculate generalized born table index - this is a separate table from the normal one,
 316              * but we use the same procedure by multiplying r with scale and truncating to integer.
 317              */
 318             rt               = _mm_mul_ps(r00,gbscale);
 319             gbitab           = _mm_cvttps_epi32(rt);
 320             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 321             gbitab           = _mm_slli_epi32(gbitab,2);
 322
 323             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 324             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 325             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 326             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 327             _MM_TRANSPOSE4_PS(Y,F,G,H);
 328             Heps             = _mm_mul_ps(gbeps,H);
 329             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 330             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 331             vgb              = _mm_mul_ps(gbqqfactor,VV);
 332
 333             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 334             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 335             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 336             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 337             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 338             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 339             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 340             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 341             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 342             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 343             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 344             velec            = _mm_mul_ps(qq00,rinv00);
 345             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velec            = _mm_andnot_ps(dummy_mask,velec);
 349             velecsum         = _mm_add_ps(velecsum,velec);
 350             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 351             vgbsum           = _mm_add_ps(vgbsum,vgb);
 352
 353             fscal            = felec;
 354
 355             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 356
 357             /* Calculate temporary vectorial force */
 358             tx               = _mm_mul_ps(fscal,dx00);
 359             ty               = _mm_mul_ps(fscal,dy00);
 360             tz               = _mm_mul_ps(fscal,dz00);
 361
 362             /* Update vectorial force */
 363             fix0             = _mm_add_ps(fix0,tx);
 364             fiy0             = _mm_add_ps(fiy0,ty);
 365             fiz0             = _mm_add_ps(fiz0,tz);
 366
 367             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 368             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 369             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 370             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 371             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 372
 373             /* Inner loop uses 59 flops */
 374         }
 375
 376         /* End of innermost loop */
 377
 378         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 379                                               f+i_coord_offset,fshift+i_shift_offset);
 380
 381         ggid                        = gid[iidx];
 382         /* Update potential energies */
 383         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 384         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 385         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 386         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 387
 388         /* Increment number of inner iterations */
 389         inneriter                  += j_index_end - j_index_start;
 390
 391         /* Outer loop uses 9 flops */
 392     }
 393
 394     /* Increment number of outer iterations */
 395     outeriter        += nri;
 396
 397     /* Update outer/inner flops */
 398
 399     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*9 + inneriter*59);
 400 }
 401 /*
 402  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse2_single
 403  * Electrostatics interaction: GeneralizedBorn
 404  * VdW interaction:            None
 405  * Geometry:                   Particle-Particle
 406  * Calculate force/pot:        Force
 407  */
 408 void
 409 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse2_single
 410                     (t_nblist * gmx_restrict                nlist,
 411                      rvec * gmx_restrict                    xx,
 412                      rvec * gmx_restrict                    ff,
 413                      t_forcerec * gmx_restrict              fr,
 414                      t_mdatoms * gmx_restrict               mdatoms,
 415                      nb_kernel_data_t * gmx_restrict        kernel_data,
 416                      t_nrnb * gmx_restrict                  nrnb)
 417 {
 418     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 419      * just 0 for non-waters.
 420      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 421      * jnr indices corresponding to data put in the four positions in the SIMD register.
 422      */
 423     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 424     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 425     int              jnrA,jnrB,jnrC,jnrD;
 426     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 427     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 428     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 429     real             rcutoff_scalar;
 430     real             *shiftvec,*fshift,*x,*f;
 431     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 432     real             scratch[4*DIM];
 433     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 434     int              vdwioffset0;
 435     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 436     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 437     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 438     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 439     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 440     real             *charge;
 441     __m128i          gbitab;
 442     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 443     __m128           minushalf = _mm_set1_ps(-0.5);
 444     real             *invsqrta,*dvda,*gbtab;
 445     __m128i          vfitab;
 446     __m128i          ifour       = _mm_set1_epi32(4);
 447     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 448     real             *vftab;
 449     __m128           dummy_mask,cutoff_mask;
 450     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 451     __m128           one     = _mm_set1_ps(1.0);
 452     __m128           two     = _mm_set1_ps(2.0);
 453     x                = xx[0];
 454     f                = ff[0];
 455
 456     nri              = nlist->nri;
 457     iinr             = nlist->iinr;
 458     jindex           = nlist->jindex;
 459     jjnr             = nlist->jjnr;
 460     shiftidx         = nlist->shift;
 461     gid              = nlist->gid;
 462     shiftvec         = fr->shift_vec[0];
 463     fshift           = fr->fshift[0];
 464     facel            = _mm_set1_ps(fr->epsfac);
 465     charge           = mdatoms->chargeA;
 466
 467     invsqrta         = fr->invsqrta;
 468     dvda             = fr->dvda;
 469     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 470     gbtab            = fr->gbtab.data;
 471     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 472
 473     /* Avoid stupid compiler warnings */
 474     jnrA = jnrB = jnrC = jnrD = 0;
 475     j_coord_offsetA = 0;
 476     j_coord_offsetB = 0;
 477     j_coord_offsetC = 0;
 478     j_coord_offsetD = 0;
 479
 480     outeriter        = 0;
 481     inneriter        = 0;
 482
 483     for(iidx=0;iidx<4*DIM;iidx++)
 484     {
 485         scratch[iidx] = 0.0;
 486     }
 487
 488     /* Start outer loop over neighborlists */
 489     for(iidx=0; iidx<nri; iidx++)
 490     {
 491         /* Load shift vector for this list */
 492         i_shift_offset   = DIM*shiftidx[iidx];
 493
 494         /* Load limits for loop over neighbors */
 495         j_index_start    = jindex[iidx];
 496         j_index_end      = jindex[iidx+1];
 497
 498         /* Get outer coordinate index */
 499         inr              = iinr[iidx];
 500         i_coord_offset   = DIM*inr;
 501
 502         /* Load i particle coords and add shift vector */
 503         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 504
 505         fix0             = _mm_setzero_ps();
 506         fiy0             = _mm_setzero_ps();
 507         fiz0             = _mm_setzero_ps();
 508
 509         /* Load parameters for i particles */
 510         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 511         isai0            = _mm_load1_ps(invsqrta+inr+0);
 512
 513         dvdasum          = _mm_setzero_ps();
 514
 515         /* Start inner kernel loop */
 516         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 517         {
 518
 519             /* Get j neighbor index, and coordinate index */
 520             jnrA             = jjnr[jidx];
 521             jnrB             = jjnr[jidx+1];
 522             jnrC             = jjnr[jidx+2];
 523             jnrD             = jjnr[jidx+3];
 524             j_coord_offsetA  = DIM*jnrA;
 525             j_coord_offsetB  = DIM*jnrB;
 526             j_coord_offsetC  = DIM*jnrC;
 527             j_coord_offsetD  = DIM*jnrD;
 528
 529             /* load j atom coordinates */
 530             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 531                                               x+j_coord_offsetC,x+j_coord_offsetD,
 532                                               &jx0,&jy0,&jz0);
 533
 534             /* Calculate displacement vector */
 535             dx00             = _mm_sub_ps(ix0,jx0);
 536             dy00             = _mm_sub_ps(iy0,jy0);
 537             dz00             = _mm_sub_ps(iz0,jz0);
 538
 539             /* Calculate squared distance and things based on it */
 540             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 541
 542             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 543
 544             /* Load parameters for j particles */
 545             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 546                                                               charge+jnrC+0,charge+jnrD+0);
 547             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 548                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             r00              = _mm_mul_ps(rsq00,rinv00);
 555
 556             /* Compute parameters for interactions between i and j atoms */
 557             qq00             = _mm_mul_ps(iq0,jq0);
 558
 559             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 560             isaprod          = _mm_mul_ps(isai0,isaj0);
 561             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 562             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 563
 564             /* Calculate generalized born table index - this is a separate table from the normal one,
 565              * but we use the same procedure by multiplying r with scale and truncating to integer.
 566              */
 567             rt               = _mm_mul_ps(r00,gbscale);
 568             gbitab           = _mm_cvttps_epi32(rt);
 569             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 570             gbitab           = _mm_slli_epi32(gbitab,2);
 571
 572             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 573             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 574             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 575             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 576             _MM_TRANSPOSE4_PS(Y,F,G,H);
 577             Heps             = _mm_mul_ps(gbeps,H);
 578             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 579             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 580             vgb              = _mm_mul_ps(gbqqfactor,VV);
 581
 582             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 583             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 584             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 585             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 586             fjptrA           = dvda+jnrA;
 587             fjptrB           = dvda+jnrB;
 588             fjptrC           = dvda+jnrC;
 589             fjptrD           = dvda+jnrD;
 590             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 591             velec            = _mm_mul_ps(qq00,rinv00);
 592             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 593
 594             fscal            = felec;
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm_mul_ps(fscal,dx00);
 598             ty               = _mm_mul_ps(fscal,dy00);
 599             tz               = _mm_mul_ps(fscal,dz00);
 600
 601             /* Update vectorial force */
 602             fix0             = _mm_add_ps(fix0,tx);
 603             fiy0             = _mm_add_ps(fiy0,ty);
 604             fiz0             = _mm_add_ps(fiz0,tz);
 605
 606             fjptrA             = f+j_coord_offsetA;
 607             fjptrB             = f+j_coord_offsetB;
 608             fjptrC             = f+j_coord_offsetC;
 609             fjptrD             = f+j_coord_offsetD;
 610             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 611
 612             /* Inner loop uses 56 flops */
 613         }
 614
 615         if(jidx<j_index_end)
 616         {
 617
 618             /* Get j neighbor index, and coordinate index */
 619             jnrlistA         = jjnr[jidx];
 620             jnrlistB         = jjnr[jidx+1];
 621             jnrlistC         = jjnr[jidx+2];
 622             jnrlistD         = jjnr[jidx+3];
 623             /* Sign of each element will be negative for non-real atoms.
 624              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 625              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 626              */
 627             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 628             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 629             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 630             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 631             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 632             j_coord_offsetA  = DIM*jnrA;
 633             j_coord_offsetB  = DIM*jnrB;
 634             j_coord_offsetC  = DIM*jnrC;
 635             j_coord_offsetD  = DIM*jnrD;
 636
 637             /* load j atom coordinates */
 638             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 639                                               x+j_coord_offsetC,x+j_coord_offsetD,
 640                                               &jx0,&jy0,&jz0);
 641
 642             /* Calculate displacement vector */
 643             dx00             = _mm_sub_ps(ix0,jx0);
 644             dy00             = _mm_sub_ps(iy0,jy0);
 645             dz00             = _mm_sub_ps(iz0,jz0);
 646
 647             /* Calculate squared distance and things based on it */
 648             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 649
 650             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 651
 652             /* Load parameters for j particles */
 653             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 654                                                               charge+jnrC+0,charge+jnrD+0);
 655             isaj0            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 656                                                               invsqrta+jnrC+0,invsqrta+jnrD+0);
 657
 658             /**************************
 659              * CALCULATE INTERACTIONS *
 660              **************************/
 661
 662             r00              = _mm_mul_ps(rsq00,rinv00);
 663             r00              = _mm_andnot_ps(dummy_mask,r00);
 664
 665             /* Compute parameters for interactions between i and j atoms */
 666             qq00             = _mm_mul_ps(iq0,jq0);
 667
 668             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 669             isaprod          = _mm_mul_ps(isai0,isaj0);
 670             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq00,_mm_mul_ps(isaprod,gbinvepsdiff)));
 671             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 672
 673             /* Calculate generalized born table index - this is a separate table from the normal one,
 674              * but we use the same procedure by multiplying r with scale and truncating to integer.
 675              */
 676             rt               = _mm_mul_ps(r00,gbscale);
 677             gbitab           = _mm_cvttps_epi32(rt);
 678             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 679             gbitab           = _mm_slli_epi32(gbitab,2);
 680
 681             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 682             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 683             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 684             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 685             _MM_TRANSPOSE4_PS(Y,F,G,H);
 686             Heps             = _mm_mul_ps(gbeps,H);
 687             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 688             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 689             vgb              = _mm_mul_ps(gbqqfactor,VV);
 690
 691             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 692             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 693             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r00)));
 694             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 695             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 696             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 697             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 698             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 699             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 700             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 701             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj0,isaj0)));
 702             velec            = _mm_mul_ps(qq00,rinv00);
 703             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv00),fgb),rinv00);
 704
 705             fscal            = felec;
 706
 707             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 708
 709             /* Calculate temporary vectorial force */
 710             tx               = _mm_mul_ps(fscal,dx00);
 711             ty               = _mm_mul_ps(fscal,dy00);
 712             tz               = _mm_mul_ps(fscal,dz00);
 713
 714             /* Update vectorial force */
 715             fix0             = _mm_add_ps(fix0,tx);
 716             fiy0             = _mm_add_ps(fiy0,ty);
 717             fiz0             = _mm_add_ps(fiz0,tz);
 718
 719             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 720             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 721             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 722             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 723             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 724
 725             /* Inner loop uses 57 flops */
 726         }
 727
 728         /* End of innermost loop */
 729
 730         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 731                                               f+i_coord_offset,fshift+i_shift_offset);
 732
 733         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai0,isai0));
 734         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
 735
 736         /* Increment number of inner iterations */
 737         inneriter                  += j_index_end - j_index_start;
 738
 739         /* Outer loop uses 7 flops */
 740     }
 741
 742     /* Increment number of outer iterations */
 743     outeriter        += nri;
 744
 745     /* Update outer/inner flops */
 746
 747     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*57);
 748 }