src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_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_ElecCoul_VdwCSTab_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_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              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  94     real             *vftab;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 122     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 123     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = jnrC = jnrD = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130     j_coord_offsetC = 0;
 131     j_coord_offsetD = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 158
 159         fix0             = _mm_setzero_ps();
 160         fiy0             = _mm_setzero_ps();
 161         fiz0             = _mm_setzero_ps();
 162         fix1             = _mm_setzero_ps();
 163         fiy1             = _mm_setzero_ps();
 164         fiz1             = _mm_setzero_ps();
 165         fix2             = _mm_setzero_ps();
 166         fiy2             = _mm_setzero_ps();
 167         fiz2             = _mm_setzero_ps();
 168         fix3             = _mm_setzero_ps();
 169         fiy3             = _mm_setzero_ps();
 170         fiz3             = _mm_setzero_ps();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_ps();
 174         vvdwsum          = _mm_setzero_ps();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187             j_coord_offsetC  = DIM*jnrC;
 188             j_coord_offsetD  = DIM*jnrD;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               x+j_coord_offsetC,x+j_coord_offsetD,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm_sub_ps(ix0,jx0);
 197             dy00             = _mm_sub_ps(iy0,jy0);
 198             dz00             = _mm_sub_ps(iz0,jz0);
 199             dx10             = _mm_sub_ps(ix1,jx0);
 200             dy10             = _mm_sub_ps(iy1,jy0);
 201             dz10             = _mm_sub_ps(iz1,jz0);
 202             dx20             = _mm_sub_ps(ix2,jx0);
 203             dy20             = _mm_sub_ps(iy2,jy0);
 204             dz20             = _mm_sub_ps(iz2,jz0);
 205             dx30             = _mm_sub_ps(ix3,jx0);
 206             dy30             = _mm_sub_ps(iy3,jy0);
 207             dz30             = _mm_sub_ps(iz3,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 213             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 214
 215             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 216             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 217             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 218             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 219
 220             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 221             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 222             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 226                                                               charge+jnrC+0,charge+jnrD+0);
 227             vdwjidx0A        = 2*vdwtype[jnrA+0];
 228             vdwjidx0B        = 2*vdwtype[jnrB+0];
 229             vdwjidx0C        = 2*vdwtype[jnrC+0];
 230             vdwjidx0D        = 2*vdwtype[jnrD+0];
 231
 232             fjx0             = _mm_setzero_ps();
 233             fjy0             = _mm_setzero_ps();
 234             fjz0             = _mm_setzero_ps();
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             r00              = _mm_mul_ps(rsq00,rinv00);
 241
 242             /* Compute parameters for interactions between i and j atoms */
 243             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 244                                          vdwparam+vdwioffset0+vdwjidx0B,
 245                                          vdwparam+vdwioffset0+vdwjidx0C,
 246                                          vdwparam+vdwioffset0+vdwjidx0D,
 247                                          &c6_00,&c12_00);
 248
 249             /* Calculate table index by multiplying r with table scale and truncate to integer */
 250             rt               = _mm_mul_ps(r00,vftabscale);
 251             vfitab           = _mm_cvttps_epi32(rt);
 252             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 253             vfitab           = _mm_slli_epi32(vfitab,3);
 254
 255             /* CUBIC SPLINE TABLE DISPERSION */
 256             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 257             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 258             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 259             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 260             _MM_TRANSPOSE4_PS(Y,F,G,H);
 261             Heps             = _mm_mul_ps(vfeps,H);
 262             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 263             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 264             vvdw6            = _mm_mul_ps(c6_00,VV);
 265             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 266             fvdw6            = _mm_mul_ps(c6_00,FF);
 267
 268             /* CUBIC SPLINE TABLE REPULSION */
 269             vfitab           = _mm_add_epi32(vfitab,ifour);
 270             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 271             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 272             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 273             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 274             _MM_TRANSPOSE4_PS(Y,F,G,H);
 275             Heps             = _mm_mul_ps(vfeps,H);
 276             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 277             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 278             vvdw12           = _mm_mul_ps(c12_00,VV);
 279             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 280             fvdw12           = _mm_mul_ps(c12_00,FF);
 281             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 282             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 286
 287             fscal            = fvdw;
 288
 289             /* Calculate temporary vectorial force */
 290             tx               = _mm_mul_ps(fscal,dx00);
 291             ty               = _mm_mul_ps(fscal,dy00);
 292             tz               = _mm_mul_ps(fscal,dz00);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm_add_ps(fix0,tx);
 296             fiy0             = _mm_add_ps(fiy0,ty);
 297             fiz0             = _mm_add_ps(fiz0,tz);
 298
 299             fjx0             = _mm_add_ps(fjx0,tx);
 300             fjy0             = _mm_add_ps(fjy0,ty);
 301             fjz0             = _mm_add_ps(fjz0,tz);
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq10             = _mm_mul_ps(iq1,jq0);
 309
 310             /* COULOMB ELECTROSTATICS */
 311             velec            = _mm_mul_ps(qq10,rinv10);
 312             felec            = _mm_mul_ps(velec,rinvsq10);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velecsum         = _mm_add_ps(velecsum,velec);
 316
 317             fscal            = felec;
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm_mul_ps(fscal,dx10);
 321             ty               = _mm_mul_ps(fscal,dy10);
 322             tz               = _mm_mul_ps(fscal,dz10);
 323
 324             /* Update vectorial force */
 325             fix1             = _mm_add_ps(fix1,tx);
 326             fiy1             = _mm_add_ps(fiy1,ty);
 327             fiz1             = _mm_add_ps(fiz1,tz);
 328
 329             fjx0             = _mm_add_ps(fjx0,tx);
 330             fjy0             = _mm_add_ps(fjy0,ty);
 331             fjz0             = _mm_add_ps(fjz0,tz);
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq20             = _mm_mul_ps(iq2,jq0);
 339
 340             /* COULOMB ELECTROSTATICS */
 341             velec            = _mm_mul_ps(qq20,rinv20);
 342             felec            = _mm_mul_ps(velec,rinvsq20);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm_add_ps(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm_mul_ps(fscal,dx20);
 351             ty               = _mm_mul_ps(fscal,dy20);
 352             tz               = _mm_mul_ps(fscal,dz20);
 353
 354             /* Update vectorial force */
 355             fix2             = _mm_add_ps(fix2,tx);
 356             fiy2             = _mm_add_ps(fiy2,ty);
 357             fiz2             = _mm_add_ps(fiz2,tz);
 358
 359             fjx0             = _mm_add_ps(fjx0,tx);
 360             fjy0             = _mm_add_ps(fjy0,ty);
 361             fjz0             = _mm_add_ps(fjz0,tz);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             /* Compute parameters for interactions between i and j atoms */
 368             qq30             = _mm_mul_ps(iq3,jq0);
 369
 370             /* COULOMB ELECTROSTATICS */
 371             velec            = _mm_mul_ps(qq30,rinv30);
 372             felec            = _mm_mul_ps(velec,rinvsq30);
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velecsum         = _mm_add_ps(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             /* Calculate temporary vectorial force */
 380             tx               = _mm_mul_ps(fscal,dx30);
 381             ty               = _mm_mul_ps(fscal,dy30);
 382             tz               = _mm_mul_ps(fscal,dz30);
 383
 384             /* Update vectorial force */
 385             fix3             = _mm_add_ps(fix3,tx);
 386             fiy3             = _mm_add_ps(fiy3,ty);
 387             fiz3             = _mm_add_ps(fiz3,tz);
 388
 389             fjx0             = _mm_add_ps(fjx0,tx);
 390             fjy0             = _mm_add_ps(fjy0,ty);
 391             fjz0             = _mm_add_ps(fjz0,tz);
 392
 393             fjptrA             = f+j_coord_offsetA;
 394             fjptrB             = f+j_coord_offsetB;
 395             fjptrC             = f+j_coord_offsetC;
 396             fjptrD             = f+j_coord_offsetD;
 397
 398             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 399
 400             /* Inner loop uses 140 flops */
 401         }
 402
 403         if(jidx<j_index_end)
 404         {
 405
 406             /* Get j neighbor index, and coordinate index */
 407             jnrlistA         = jjnr[jidx];
 408             jnrlistB         = jjnr[jidx+1];
 409             jnrlistC         = jjnr[jidx+2];
 410             jnrlistD         = jjnr[jidx+3];
 411             /* Sign of each element will be negative for non-real atoms.
 412              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 413              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 414              */
 415             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 416             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 417             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 418             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 419             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 420             j_coord_offsetA  = DIM*jnrA;
 421             j_coord_offsetB  = DIM*jnrB;
 422             j_coord_offsetC  = DIM*jnrC;
 423             j_coord_offsetD  = DIM*jnrD;
 424
 425             /* load j atom coordinates */
 426             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 427                                               x+j_coord_offsetC,x+j_coord_offsetD,
 428                                               &jx0,&jy0,&jz0);
 429
 430             /* Calculate displacement vector */
 431             dx00             = _mm_sub_ps(ix0,jx0);
 432             dy00             = _mm_sub_ps(iy0,jy0);
 433             dz00             = _mm_sub_ps(iz0,jz0);
 434             dx10             = _mm_sub_ps(ix1,jx0);
 435             dy10             = _mm_sub_ps(iy1,jy0);
 436             dz10             = _mm_sub_ps(iz1,jz0);
 437             dx20             = _mm_sub_ps(ix2,jx0);
 438             dy20             = _mm_sub_ps(iy2,jy0);
 439             dz20             = _mm_sub_ps(iz2,jz0);
 440             dx30             = _mm_sub_ps(ix3,jx0);
 441             dy30             = _mm_sub_ps(iy3,jy0);
 442             dz30             = _mm_sub_ps(iz3,jz0);
 443
 444             /* Calculate squared distance and things based on it */
 445             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 446             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 447             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 448             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 449
 450             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 451             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 452             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 453             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 454
 455             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 456             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 457             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 458
 459             /* Load parameters for j particles */
 460             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 461                                                               charge+jnrC+0,charge+jnrD+0);
 462             vdwjidx0A        = 2*vdwtype[jnrA+0];
 463             vdwjidx0B        = 2*vdwtype[jnrB+0];
 464             vdwjidx0C        = 2*vdwtype[jnrC+0];
 465             vdwjidx0D        = 2*vdwtype[jnrD+0];
 466
 467             fjx0             = _mm_setzero_ps();
 468             fjy0             = _mm_setzero_ps();
 469             fjz0             = _mm_setzero_ps();
 470
 471             /**************************
 472              * CALCULATE INTERACTIONS *
 473              **************************/
 474
 475             r00              = _mm_mul_ps(rsq00,rinv00);
 476             r00              = _mm_andnot_ps(dummy_mask,r00);
 477
 478             /* Compute parameters for interactions between i and j atoms */
 479             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 480                                          vdwparam+vdwioffset0+vdwjidx0B,
 481                                          vdwparam+vdwioffset0+vdwjidx0C,
 482                                          vdwparam+vdwioffset0+vdwjidx0D,
 483                                          &c6_00,&c12_00);
 484
 485             /* Calculate table index by multiplying r with table scale and truncate to integer */
 486             rt               = _mm_mul_ps(r00,vftabscale);
 487             vfitab           = _mm_cvttps_epi32(rt);
 488             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 489             vfitab           = _mm_slli_epi32(vfitab,3);
 490
 491             /* CUBIC SPLINE TABLE DISPERSION */
 492             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 493             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 494             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 495             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 496             _MM_TRANSPOSE4_PS(Y,F,G,H);
 497             Heps             = _mm_mul_ps(vfeps,H);
 498             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 499             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 500             vvdw6            = _mm_mul_ps(c6_00,VV);
 501             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 502             fvdw6            = _mm_mul_ps(c6_00,FF);
 503
 504             /* CUBIC SPLINE TABLE REPULSION */
 505             vfitab           = _mm_add_epi32(vfitab,ifour);
 506             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 507             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 508             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 509             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 510             _MM_TRANSPOSE4_PS(Y,F,G,H);
 511             Heps             = _mm_mul_ps(vfeps,H);
 512             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 513             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 514             vvdw12           = _mm_mul_ps(c12_00,VV);
 515             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 516             fvdw12           = _mm_mul_ps(c12_00,FF);
 517             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 518             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 522             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 523
 524             fscal            = fvdw;
 525
 526             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 527
 528             /* Calculate temporary vectorial force */
 529             tx               = _mm_mul_ps(fscal,dx00);
 530             ty               = _mm_mul_ps(fscal,dy00);
 531             tz               = _mm_mul_ps(fscal,dz00);
 532
 533             /* Update vectorial force */
 534             fix0             = _mm_add_ps(fix0,tx);
 535             fiy0             = _mm_add_ps(fiy0,ty);
 536             fiz0             = _mm_add_ps(fiz0,tz);
 537
 538             fjx0             = _mm_add_ps(fjx0,tx);
 539             fjy0             = _mm_add_ps(fjy0,ty);
 540             fjz0             = _mm_add_ps(fjz0,tz);
 541
 542             /**************************
 543              * CALCULATE INTERACTIONS *
 544              **************************/
 545
 546             /* Compute parameters for interactions between i and j atoms */
 547             qq10             = _mm_mul_ps(iq1,jq0);
 548
 549             /* COULOMB ELECTROSTATICS */
 550             velec            = _mm_mul_ps(qq10,rinv10);
 551             felec            = _mm_mul_ps(velec,rinvsq10);
 552
 553             /* Update potential sum for this i atom from the interaction with this j atom. */
 554             velec            = _mm_andnot_ps(dummy_mask,velec);
 555             velecsum         = _mm_add_ps(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm_mul_ps(fscal,dx10);
 563             ty               = _mm_mul_ps(fscal,dy10);
 564             tz               = _mm_mul_ps(fscal,dz10);
 565
 566             /* Update vectorial force */
 567             fix1             = _mm_add_ps(fix1,tx);
 568             fiy1             = _mm_add_ps(fiy1,ty);
 569             fiz1             = _mm_add_ps(fiz1,tz);
 570
 571             fjx0             = _mm_add_ps(fjx0,tx);
 572             fjy0             = _mm_add_ps(fjy0,ty);
 573             fjz0             = _mm_add_ps(fjz0,tz);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             /* Compute parameters for interactions between i and j atoms */
 580             qq20             = _mm_mul_ps(iq2,jq0);
 581
 582             /* COULOMB ELECTROSTATICS */
 583             velec            = _mm_mul_ps(qq20,rinv20);
 584             felec            = _mm_mul_ps(velec,rinvsq20);
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velec            = _mm_andnot_ps(dummy_mask,velec);
 588             velecsum         = _mm_add_ps(velecsum,velec);
 589
 590             fscal            = felec;
 591
 592             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm_mul_ps(fscal,dx20);
 596             ty               = _mm_mul_ps(fscal,dy20);
 597             tz               = _mm_mul_ps(fscal,dz20);
 598
 599             /* Update vectorial force */
 600             fix2             = _mm_add_ps(fix2,tx);
 601             fiy2             = _mm_add_ps(fiy2,ty);
 602             fiz2             = _mm_add_ps(fiz2,tz);
 603
 604             fjx0             = _mm_add_ps(fjx0,tx);
 605             fjy0             = _mm_add_ps(fjy0,ty);
 606             fjz0             = _mm_add_ps(fjz0,tz);
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             /* Compute parameters for interactions between i and j atoms */
 613             qq30             = _mm_mul_ps(iq3,jq0);
 614
 615             /* COULOMB ELECTROSTATICS */
 616             velec            = _mm_mul_ps(qq30,rinv30);
 617             felec            = _mm_mul_ps(velec,rinvsq30);
 618
 619             /* Update potential sum for this i atom from the interaction with this j atom. */
 620             velec            = _mm_andnot_ps(dummy_mask,velec);
 621             velecsum         = _mm_add_ps(velecsum,velec);
 622
 623             fscal            = felec;
 624
 625             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 626
 627             /* Calculate temporary vectorial force */
 628             tx               = _mm_mul_ps(fscal,dx30);
 629             ty               = _mm_mul_ps(fscal,dy30);
 630             tz               = _mm_mul_ps(fscal,dz30);
 631
 632             /* Update vectorial force */
 633             fix3             = _mm_add_ps(fix3,tx);
 634             fiy3             = _mm_add_ps(fiy3,ty);
 635             fiz3             = _mm_add_ps(fiz3,tz);
 636
 637             fjx0             = _mm_add_ps(fjx0,tx);
 638             fjy0             = _mm_add_ps(fjy0,ty);
 639             fjz0             = _mm_add_ps(fjz0,tz);
 640
 641             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 642             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 643             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 644             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 645
 646             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 647
 648             /* Inner loop uses 141 flops */
 649         }
 650
 651         /* End of innermost loop */
 652
 653         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 654                                               f+i_coord_offset,fshift+i_shift_offset);
 655
 656         ggid                        = gid[iidx];
 657         /* Update potential energies */
 658         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 659         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 660
 661         /* Increment number of inner iterations */
 662         inneriter                  += j_index_end - j_index_start;
 663
 664         /* Outer loop uses 26 flops */
 665     }
 666
 667     /* Increment number of outer iterations */
 668     outeriter        += nri;
 669
 670     /* Update outer/inner flops */
 671
 672     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 673 }
 674 /*
 675  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse2_single
 676  * Electrostatics interaction: Coulomb
 677  * VdW interaction:            CubicSplineTable
 678  * Geometry:                   Water4-Particle
 679  * Calculate force/pot:        Force
 680  */
 681 void
 682 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse2_single
 683                     (t_nblist * gmx_restrict                nlist,
 684                      rvec * gmx_restrict                    xx,
 685                      rvec * gmx_restrict                    ff,
 686                      t_forcerec * gmx_restrict              fr,
 687                      t_mdatoms * gmx_restrict               mdatoms,
 688                      nb_kernel_data_t * gmx_restrict        kernel_data,
 689                      t_nrnb * gmx_restrict                  nrnb)
 690 {
 691     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 692      * just 0 for non-waters.
 693      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 694      * jnr indices corresponding to data put in the four positions in the SIMD register.
 695      */
 696     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 697     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 698     int              jnrA,jnrB,jnrC,jnrD;
 699     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 700     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 701     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 702     real             rcutoff_scalar;
 703     real             *shiftvec,*fshift,*x,*f;
 704     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 705     real             scratch[4*DIM];
 706     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 707     int              vdwioffset0;
 708     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 709     int              vdwioffset1;
 710     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 711     int              vdwioffset2;
 712     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 713     int              vdwioffset3;
 714     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 715     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 716     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 717     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 718     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 719     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 720     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 721     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 722     real             *charge;
 723     int              nvdwtype;
 724     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 725     int              *vdwtype;
 726     real             *vdwparam;
 727     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 728     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 729     __m128i          vfitab;
 730     __m128i          ifour       = _mm_set1_epi32(4);
 731     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 732     real             *vftab;
 733     __m128           dummy_mask,cutoff_mask;
 734     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 735     __m128           one     = _mm_set1_ps(1.0);
 736     __m128           two     = _mm_set1_ps(2.0);
 737     x                = xx[0];
 738     f                = ff[0];
 739
 740     nri              = nlist->nri;
 741     iinr             = nlist->iinr;
 742     jindex           = nlist->jindex;
 743     jjnr             = nlist->jjnr;
 744     shiftidx         = nlist->shift;
 745     gid              = nlist->gid;
 746     shiftvec         = fr->shift_vec[0];
 747     fshift           = fr->fshift[0];
 748     facel            = _mm_set1_ps(fr->epsfac);
 749     charge           = mdatoms->chargeA;
 750     nvdwtype         = fr->ntype;
 751     vdwparam         = fr->nbfp;
 752     vdwtype          = mdatoms->typeA;
 753
 754     vftab            = kernel_data->table_vdw->data;
 755     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 756
 757     /* Setup water-specific parameters */
 758     inr              = nlist->iinr[0];
 759     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 760     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 761     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 762     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 763
 764     /* Avoid stupid compiler warnings */
 765     jnrA = jnrB = jnrC = jnrD = 0;
 766     j_coord_offsetA = 0;
 767     j_coord_offsetB = 0;
 768     j_coord_offsetC = 0;
 769     j_coord_offsetD = 0;
 770
 771     outeriter        = 0;
 772     inneriter        = 0;
 773
 774     for(iidx=0;iidx<4*DIM;iidx++)
 775     {
 776         scratch[iidx] = 0.0;
 777     }
 778
 779     /* Start outer loop over neighborlists */
 780     for(iidx=0; iidx<nri; iidx++)
 781     {
 782         /* Load shift vector for this list */
 783         i_shift_offset   = DIM*shiftidx[iidx];
 784
 785         /* Load limits for loop over neighbors */
 786         j_index_start    = jindex[iidx];
 787         j_index_end      = jindex[iidx+1];
 788
 789         /* Get outer coordinate index */
 790         inr              = iinr[iidx];
 791         i_coord_offset   = DIM*inr;
 792
 793         /* Load i particle coords and add shift vector */
 794         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 795                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 796
 797         fix0             = _mm_setzero_ps();
 798         fiy0             = _mm_setzero_ps();
 799         fiz0             = _mm_setzero_ps();
 800         fix1             = _mm_setzero_ps();
 801         fiy1             = _mm_setzero_ps();
 802         fiz1             = _mm_setzero_ps();
 803         fix2             = _mm_setzero_ps();
 804         fiy2             = _mm_setzero_ps();
 805         fiz2             = _mm_setzero_ps();
 806         fix3             = _mm_setzero_ps();
 807         fiy3             = _mm_setzero_ps();
 808         fiz3             = _mm_setzero_ps();
 809
 810         /* Start inner kernel loop */
 811         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 812         {
 813
 814             /* Get j neighbor index, and coordinate index */
 815             jnrA             = jjnr[jidx];
 816             jnrB             = jjnr[jidx+1];
 817             jnrC             = jjnr[jidx+2];
 818             jnrD             = jjnr[jidx+3];
 819             j_coord_offsetA  = DIM*jnrA;
 820             j_coord_offsetB  = DIM*jnrB;
 821             j_coord_offsetC  = DIM*jnrC;
 822             j_coord_offsetD  = DIM*jnrD;
 823
 824             /* load j atom coordinates */
 825             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 826                                               x+j_coord_offsetC,x+j_coord_offsetD,
 827                                               &jx0,&jy0,&jz0);
 828
 829             /* Calculate displacement vector */
 830             dx00             = _mm_sub_ps(ix0,jx0);
 831             dy00             = _mm_sub_ps(iy0,jy0);
 832             dz00             = _mm_sub_ps(iz0,jz0);
 833             dx10             = _mm_sub_ps(ix1,jx0);
 834             dy10             = _mm_sub_ps(iy1,jy0);
 835             dz10             = _mm_sub_ps(iz1,jz0);
 836             dx20             = _mm_sub_ps(ix2,jx0);
 837             dy20             = _mm_sub_ps(iy2,jy0);
 838             dz20             = _mm_sub_ps(iz2,jz0);
 839             dx30             = _mm_sub_ps(ix3,jx0);
 840             dy30             = _mm_sub_ps(iy3,jy0);
 841             dz30             = _mm_sub_ps(iz3,jz0);
 842
 843             /* Calculate squared distance and things based on it */
 844             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 845             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 846             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 847             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 848
 849             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 850             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 851             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 852             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 853
 854             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 855             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 856             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 857
 858             /* Load parameters for j particles */
 859             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 860                                                               charge+jnrC+0,charge+jnrD+0);
 861             vdwjidx0A        = 2*vdwtype[jnrA+0];
 862             vdwjidx0B        = 2*vdwtype[jnrB+0];
 863             vdwjidx0C        = 2*vdwtype[jnrC+0];
 864             vdwjidx0D        = 2*vdwtype[jnrD+0];
 865
 866             fjx0             = _mm_setzero_ps();
 867             fjy0             = _mm_setzero_ps();
 868             fjz0             = _mm_setzero_ps();
 869
 870             /**************************
 871              * CALCULATE INTERACTIONS *
 872              **************************/
 873
 874             r00              = _mm_mul_ps(rsq00,rinv00);
 875
 876             /* Compute parameters for interactions between i and j atoms */
 877             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 878                                          vdwparam+vdwioffset0+vdwjidx0B,
 879                                          vdwparam+vdwioffset0+vdwjidx0C,
 880                                          vdwparam+vdwioffset0+vdwjidx0D,
 881                                          &c6_00,&c12_00);
 882
 883             /* Calculate table index by multiplying r with table scale and truncate to integer */
 884             rt               = _mm_mul_ps(r00,vftabscale);
 885             vfitab           = _mm_cvttps_epi32(rt);
 886             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 887             vfitab           = _mm_slli_epi32(vfitab,3);
 888
 889             /* CUBIC SPLINE TABLE DISPERSION */
 890             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 891             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 892             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 893             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 894             _MM_TRANSPOSE4_PS(Y,F,G,H);
 895             Heps             = _mm_mul_ps(vfeps,H);
 896             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 897             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 898             fvdw6            = _mm_mul_ps(c6_00,FF);
 899
 900             /* CUBIC SPLINE TABLE REPULSION */
 901             vfitab           = _mm_add_epi32(vfitab,ifour);
 902             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 903             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 904             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 905             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 906             _MM_TRANSPOSE4_PS(Y,F,G,H);
 907             Heps             = _mm_mul_ps(vfeps,H);
 908             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 909             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 910             fvdw12           = _mm_mul_ps(c12_00,FF);
 911             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 912
 913             fscal            = fvdw;
 914
 915             /* Calculate temporary vectorial force */
 916             tx               = _mm_mul_ps(fscal,dx00);
 917             ty               = _mm_mul_ps(fscal,dy00);
 918             tz               = _mm_mul_ps(fscal,dz00);
 919
 920             /* Update vectorial force */
 921             fix0             = _mm_add_ps(fix0,tx);
 922             fiy0             = _mm_add_ps(fiy0,ty);
 923             fiz0             = _mm_add_ps(fiz0,tz);
 924
 925             fjx0             = _mm_add_ps(fjx0,tx);
 926             fjy0             = _mm_add_ps(fjy0,ty);
 927             fjz0             = _mm_add_ps(fjz0,tz);
 928
 929             /**************************
 930              * CALCULATE INTERACTIONS *
 931              **************************/
 932
 933             /* Compute parameters for interactions between i and j atoms */
 934             qq10             = _mm_mul_ps(iq1,jq0);
 935
 936             /* COULOMB ELECTROSTATICS */
 937             velec            = _mm_mul_ps(qq10,rinv10);
 938             felec            = _mm_mul_ps(velec,rinvsq10);
 939
 940             fscal            = felec;
 941
 942             /* Calculate temporary vectorial force */
 943             tx               = _mm_mul_ps(fscal,dx10);
 944             ty               = _mm_mul_ps(fscal,dy10);
 945             tz               = _mm_mul_ps(fscal,dz10);
 946
 947             /* Update vectorial force */
 948             fix1             = _mm_add_ps(fix1,tx);
 949             fiy1             = _mm_add_ps(fiy1,ty);
 950             fiz1             = _mm_add_ps(fiz1,tz);
 951
 952             fjx0             = _mm_add_ps(fjx0,tx);
 953             fjy0             = _mm_add_ps(fjy0,ty);
 954             fjz0             = _mm_add_ps(fjz0,tz);
 955
 956             /**************************
 957              * CALCULATE INTERACTIONS *
 958              **************************/
 959
 960             /* Compute parameters for interactions between i and j atoms */
 961             qq20             = _mm_mul_ps(iq2,jq0);
 962
 963             /* COULOMB ELECTROSTATICS */
 964             velec            = _mm_mul_ps(qq20,rinv20);
 965             felec            = _mm_mul_ps(velec,rinvsq20);
 966
 967             fscal            = felec;
 968
 969             /* Calculate temporary vectorial force */
 970             tx               = _mm_mul_ps(fscal,dx20);
 971             ty               = _mm_mul_ps(fscal,dy20);
 972             tz               = _mm_mul_ps(fscal,dz20);
 973
 974             /* Update vectorial force */
 975             fix2             = _mm_add_ps(fix2,tx);
 976             fiy2             = _mm_add_ps(fiy2,ty);
 977             fiz2             = _mm_add_ps(fiz2,tz);
 978
 979             fjx0             = _mm_add_ps(fjx0,tx);
 980             fjy0             = _mm_add_ps(fjy0,ty);
 981             fjz0             = _mm_add_ps(fjz0,tz);
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             /* Compute parameters for interactions between i and j atoms */
 988             qq30             = _mm_mul_ps(iq3,jq0);
 989
 990             /* COULOMB ELECTROSTATICS */
 991             velec            = _mm_mul_ps(qq30,rinv30);
 992             felec            = _mm_mul_ps(velec,rinvsq30);
 993
 994             fscal            = felec;
 995
 996             /* Calculate temporary vectorial force */
 997             tx               = _mm_mul_ps(fscal,dx30);
 998             ty               = _mm_mul_ps(fscal,dy30);
 999             tz               = _mm_mul_ps(fscal,dz30);
1000
1001             /* Update vectorial force */
1002             fix3             = _mm_add_ps(fix3,tx);
1003             fiy3             = _mm_add_ps(fiy3,ty);
1004             fiz3             = _mm_add_ps(fiz3,tz);
1005
1006             fjx0             = _mm_add_ps(fjx0,tx);
1007             fjy0             = _mm_add_ps(fjy0,ty);
1008             fjz0             = _mm_add_ps(fjz0,tz);
1009
1010             fjptrA             = f+j_coord_offsetA;
1011             fjptrB             = f+j_coord_offsetB;
1012             fjptrC             = f+j_coord_offsetC;
1013             fjptrD             = f+j_coord_offsetD;
1014
1015             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1016
1017             /* Inner loop uses 129 flops */
1018         }
1019
1020         if(jidx<j_index_end)
1021         {
1022
1023             /* Get j neighbor index, and coordinate index */
1024             jnrlistA         = jjnr[jidx];
1025             jnrlistB         = jjnr[jidx+1];
1026             jnrlistC         = jjnr[jidx+2];
1027             jnrlistD         = jjnr[jidx+3];
1028             /* Sign of each element will be negative for non-real atoms.
1029              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1030              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1031              */
1032             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1033             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1034             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1035             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1036             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1037             j_coord_offsetA  = DIM*jnrA;
1038             j_coord_offsetB  = DIM*jnrB;
1039             j_coord_offsetC  = DIM*jnrC;
1040             j_coord_offsetD  = DIM*jnrD;
1041
1042             /* load j atom coordinates */
1043             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1044                                               x+j_coord_offsetC,x+j_coord_offsetD,
1045                                               &jx0,&jy0,&jz0);
1046
1047             /* Calculate displacement vector */
1048             dx00             = _mm_sub_ps(ix0,jx0);
1049             dy00             = _mm_sub_ps(iy0,jy0);
1050             dz00             = _mm_sub_ps(iz0,jz0);
1051             dx10             = _mm_sub_ps(ix1,jx0);
1052             dy10             = _mm_sub_ps(iy1,jy0);
1053             dz10             = _mm_sub_ps(iz1,jz0);
1054             dx20             = _mm_sub_ps(ix2,jx0);
1055             dy20             = _mm_sub_ps(iy2,jy0);
1056             dz20             = _mm_sub_ps(iz2,jz0);
1057             dx30             = _mm_sub_ps(ix3,jx0);
1058             dy30             = _mm_sub_ps(iy3,jy0);
1059             dz30             = _mm_sub_ps(iz3,jz0);
1060
1061             /* Calculate squared distance and things based on it */
1062             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1063             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1064             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1065             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1066
1067             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1068             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1069             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1070             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1071
1072             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1073             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1074             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1075
1076             /* Load parameters for j particles */
1077             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1078                                                               charge+jnrC+0,charge+jnrD+0);
1079             vdwjidx0A        = 2*vdwtype[jnrA+0];
1080             vdwjidx0B        = 2*vdwtype[jnrB+0];
1081             vdwjidx0C        = 2*vdwtype[jnrC+0];
1082             vdwjidx0D        = 2*vdwtype[jnrD+0];
1083
1084             fjx0             = _mm_setzero_ps();
1085             fjy0             = _mm_setzero_ps();
1086             fjz0             = _mm_setzero_ps();
1087
1088             /**************************
1089              * CALCULATE INTERACTIONS *
1090              **************************/
1091
1092             r00              = _mm_mul_ps(rsq00,rinv00);
1093             r00              = _mm_andnot_ps(dummy_mask,r00);
1094
1095             /* Compute parameters for interactions between i and j atoms */
1096             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1097                                          vdwparam+vdwioffset0+vdwjidx0B,
1098                                          vdwparam+vdwioffset0+vdwjidx0C,
1099                                          vdwparam+vdwioffset0+vdwjidx0D,
1100                                          &c6_00,&c12_00);
1101
1102             /* Calculate table index by multiplying r with table scale and truncate to integer */
1103             rt               = _mm_mul_ps(r00,vftabscale);
1104             vfitab           = _mm_cvttps_epi32(rt);
1105             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1106             vfitab           = _mm_slli_epi32(vfitab,3);
1107
1108             /* CUBIC SPLINE TABLE DISPERSION */
1109             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1110             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1111             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1112             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1113             _MM_TRANSPOSE4_PS(Y,F,G,H);
1114             Heps             = _mm_mul_ps(vfeps,H);
1115             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1116             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1117             fvdw6            = _mm_mul_ps(c6_00,FF);
1118
1119             /* CUBIC SPLINE TABLE REPULSION */
1120             vfitab           = _mm_add_epi32(vfitab,ifour);
1121             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1122             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1123             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1124             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1125             _MM_TRANSPOSE4_PS(Y,F,G,H);
1126             Heps             = _mm_mul_ps(vfeps,H);
1127             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1128             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1129             fvdw12           = _mm_mul_ps(c12_00,FF);
1130             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1131
1132             fscal            = fvdw;
1133
1134             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1135
1136             /* Calculate temporary vectorial force */
1137             tx               = _mm_mul_ps(fscal,dx00);
1138             ty               = _mm_mul_ps(fscal,dy00);
1139             tz               = _mm_mul_ps(fscal,dz00);
1140
1141             /* Update vectorial force */
1142             fix0             = _mm_add_ps(fix0,tx);
1143             fiy0             = _mm_add_ps(fiy0,ty);
1144             fiz0             = _mm_add_ps(fiz0,tz);
1145
1146             fjx0             = _mm_add_ps(fjx0,tx);
1147             fjy0             = _mm_add_ps(fjy0,ty);
1148             fjz0             = _mm_add_ps(fjz0,tz);
1149
1150             /**************************
1151              * CALCULATE INTERACTIONS *
1152              **************************/
1153
1154             /* Compute parameters for interactions between i and j atoms */
1155             qq10             = _mm_mul_ps(iq1,jq0);
1156
1157             /* COULOMB ELECTROSTATICS */
1158             velec            = _mm_mul_ps(qq10,rinv10);
1159             felec            = _mm_mul_ps(velec,rinvsq10);
1160
1161             fscal            = felec;
1162
1163             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1164
1165             /* Calculate temporary vectorial force */
1166             tx               = _mm_mul_ps(fscal,dx10);
1167             ty               = _mm_mul_ps(fscal,dy10);
1168             tz               = _mm_mul_ps(fscal,dz10);
1169
1170             /* Update vectorial force */
1171             fix1             = _mm_add_ps(fix1,tx);
1172             fiy1             = _mm_add_ps(fiy1,ty);
1173             fiz1             = _mm_add_ps(fiz1,tz);
1174
1175             fjx0             = _mm_add_ps(fjx0,tx);
1176             fjy0             = _mm_add_ps(fjy0,ty);
1177             fjz0             = _mm_add_ps(fjz0,tz);
1178
1179             /**************************
1180              * CALCULATE INTERACTIONS *
1181              **************************/
1182
1183             /* Compute parameters for interactions between i and j atoms */
1184             qq20             = _mm_mul_ps(iq2,jq0);
1185
1186             /* COULOMB ELECTROSTATICS */
1187             velec            = _mm_mul_ps(qq20,rinv20);
1188             felec            = _mm_mul_ps(velec,rinvsq20);
1189
1190             fscal            = felec;
1191
1192             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1193
1194             /* Calculate temporary vectorial force */
1195             tx               = _mm_mul_ps(fscal,dx20);
1196             ty               = _mm_mul_ps(fscal,dy20);
1197             tz               = _mm_mul_ps(fscal,dz20);
1198
1199             /* Update vectorial force */
1200             fix2             = _mm_add_ps(fix2,tx);
1201             fiy2             = _mm_add_ps(fiy2,ty);
1202             fiz2             = _mm_add_ps(fiz2,tz);
1203
1204             fjx0             = _mm_add_ps(fjx0,tx);
1205             fjy0             = _mm_add_ps(fjy0,ty);
1206             fjz0             = _mm_add_ps(fjz0,tz);
1207
1208             /**************************
1209              * CALCULATE INTERACTIONS *
1210              **************************/
1211
1212             /* Compute parameters for interactions between i and j atoms */
1213             qq30             = _mm_mul_ps(iq3,jq0);
1214
1215             /* COULOMB ELECTROSTATICS */
1216             velec            = _mm_mul_ps(qq30,rinv30);
1217             felec            = _mm_mul_ps(velec,rinvsq30);
1218
1219             fscal            = felec;
1220
1221             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1222
1223             /* Calculate temporary vectorial force */
1224             tx               = _mm_mul_ps(fscal,dx30);
1225             ty               = _mm_mul_ps(fscal,dy30);
1226             tz               = _mm_mul_ps(fscal,dz30);
1227
1228             /* Update vectorial force */
1229             fix3             = _mm_add_ps(fix3,tx);
1230             fiy3             = _mm_add_ps(fiy3,ty);
1231             fiz3             = _mm_add_ps(fiz3,tz);
1232
1233             fjx0             = _mm_add_ps(fjx0,tx);
1234             fjy0             = _mm_add_ps(fjy0,ty);
1235             fjz0             = _mm_add_ps(fjz0,tz);
1236
1237             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1238             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1239             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1240             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1241
1242             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1243
1244             /* Inner loop uses 130 flops */
1245         }
1246
1247         /* End of innermost loop */
1248
1249         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1250                                               f+i_coord_offset,fshift+i_shift_offset);
1251
1252         /* Increment number of inner iterations */
1253         inneriter                  += j_index_end - j_index_start;
1254
1255         /* Outer loop uses 24 flops */
1256     }
1257
1258     /* Increment number of outer iterations */
1259     outeriter        += nri;
1260
1261     /* Update outer/inner flops */
1262
1263     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1264 }