src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_sse4_1_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_sse4_1_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              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             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             r00              = _mm_mul_ps(rsq00,rinv00);
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 240                                          vdwparam+vdwioffset0+vdwjidx0B,
 241                                          vdwparam+vdwioffset0+vdwjidx0C,
 242                                          vdwparam+vdwioffset0+vdwjidx0D,
 243                                          &c6_00,&c12_00);
 244
 245             /* Calculate table index by multiplying r with table scale and truncate to integer */
 246             rt               = _mm_mul_ps(r00,vftabscale);
 247             vfitab           = _mm_cvttps_epi32(rt);
 248             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 249             vfitab           = _mm_slli_epi32(vfitab,3);
 250
 251             /* CUBIC SPLINE TABLE DISPERSION */
 252             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 253             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 254             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 255             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 256             _MM_TRANSPOSE4_PS(Y,F,G,H);
 257             Heps             = _mm_mul_ps(vfeps,H);
 258             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 259             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 260             vvdw6            = _mm_mul_ps(c6_00,VV);
 261             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 262             fvdw6            = _mm_mul_ps(c6_00,FF);
 263
 264             /* CUBIC SPLINE TABLE REPULSION */
 265             vfitab           = _mm_add_epi32(vfitab,ifour);
 266             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 267             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 268             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 269             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 270             _MM_TRANSPOSE4_PS(Y,F,G,H);
 271             Heps             = _mm_mul_ps(vfeps,H);
 272             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 273             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 274             vvdw12           = _mm_mul_ps(c12_00,VV);
 275             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 276             fvdw12           = _mm_mul_ps(c12_00,FF);
 277             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 278             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 282
 283             fscal            = fvdw;
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm_mul_ps(fscal,dx00);
 287             ty               = _mm_mul_ps(fscal,dy00);
 288             tz               = _mm_mul_ps(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_add_ps(fix0,tx);
 292             fiy0             = _mm_add_ps(fiy0,ty);
 293             fiz0             = _mm_add_ps(fiz0,tz);
 294
 295             fjptrA             = f+j_coord_offsetA;
 296             fjptrB             = f+j_coord_offsetB;
 297             fjptrC             = f+j_coord_offsetC;
 298             fjptrD             = f+j_coord_offsetD;
 299             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_ps(iq1,jq0);
 307
 308             /* COULOMB ELECTROSTATICS */
 309             velec            = _mm_mul_ps(qq10,rinv10);
 310             felec            = _mm_mul_ps(velec,rinvsq10);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm_add_ps(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm_mul_ps(fscal,dx10);
 319             ty               = _mm_mul_ps(fscal,dy10);
 320             tz               = _mm_mul_ps(fscal,dz10);
 321
 322             /* Update vectorial force */
 323             fix1             = _mm_add_ps(fix1,tx);
 324             fiy1             = _mm_add_ps(fiy1,ty);
 325             fiz1             = _mm_add_ps(fiz1,tz);
 326
 327             fjptrA             = f+j_coord_offsetA;
 328             fjptrB             = f+j_coord_offsetB;
 329             fjptrC             = f+j_coord_offsetC;
 330             fjptrD             = f+j_coord_offsetD;
 331             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,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             fjptrA             = f+j_coord_offsetA;
 360             fjptrB             = f+j_coord_offsetB;
 361             fjptrC             = f+j_coord_offsetC;
 362             fjptrD             = f+j_coord_offsetD;
 363             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq30             = _mm_mul_ps(iq3,jq0);
 371
 372             /* COULOMB ELECTROSTATICS */
 373             velec            = _mm_mul_ps(qq30,rinv30);
 374             felec            = _mm_mul_ps(velec,rinvsq30);
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velecsum         = _mm_add_ps(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             /* Calculate temporary vectorial force */
 382             tx               = _mm_mul_ps(fscal,dx30);
 383             ty               = _mm_mul_ps(fscal,dy30);
 384             tz               = _mm_mul_ps(fscal,dz30);
 385
 386             /* Update vectorial force */
 387             fix3             = _mm_add_ps(fix3,tx);
 388             fiy3             = _mm_add_ps(fiy3,ty);
 389             fiz3             = _mm_add_ps(fiz3,tz);
 390
 391             fjptrA             = f+j_coord_offsetA;
 392             fjptrB             = f+j_coord_offsetB;
 393             fjptrC             = f+j_coord_offsetC;
 394             fjptrD             = f+j_coord_offsetD;
 395             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 396
 397             /* Inner loop uses 140 flops */
 398         }
 399
 400         if(jidx<j_index_end)
 401         {
 402
 403             /* Get j neighbor index, and coordinate index */
 404             jnrlistA         = jjnr[jidx];
 405             jnrlistB         = jjnr[jidx+1];
 406             jnrlistC         = jjnr[jidx+2];
 407             jnrlistD         = jjnr[jidx+3];
 408             /* Sign of each element will be negative for non-real atoms.
 409              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 410              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 411              */
 412             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 413             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 414             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 415             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 416             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 417             j_coord_offsetA  = DIM*jnrA;
 418             j_coord_offsetB  = DIM*jnrB;
 419             j_coord_offsetC  = DIM*jnrC;
 420             j_coord_offsetD  = DIM*jnrD;
 421
 422             /* load j atom coordinates */
 423             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 424                                               x+j_coord_offsetC,x+j_coord_offsetD,
 425                                               &jx0,&jy0,&jz0);
 426
 427             /* Calculate displacement vector */
 428             dx00             = _mm_sub_ps(ix0,jx0);
 429             dy00             = _mm_sub_ps(iy0,jy0);
 430             dz00             = _mm_sub_ps(iz0,jz0);
 431             dx10             = _mm_sub_ps(ix1,jx0);
 432             dy10             = _mm_sub_ps(iy1,jy0);
 433             dz10             = _mm_sub_ps(iz1,jz0);
 434             dx20             = _mm_sub_ps(ix2,jx0);
 435             dy20             = _mm_sub_ps(iy2,jy0);
 436             dz20             = _mm_sub_ps(iz2,jz0);
 437             dx30             = _mm_sub_ps(ix3,jx0);
 438             dy30             = _mm_sub_ps(iy3,jy0);
 439             dz30             = _mm_sub_ps(iz3,jz0);
 440
 441             /* Calculate squared distance and things based on it */
 442             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 443             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 444             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 445             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 446
 447             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 448             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 449             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 450             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 451
 452             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 453             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 454             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 455
 456             /* Load parameters for j particles */
 457             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 458                                                               charge+jnrC+0,charge+jnrD+0);
 459             vdwjidx0A        = 2*vdwtype[jnrA+0];
 460             vdwjidx0B        = 2*vdwtype[jnrB+0];
 461             vdwjidx0C        = 2*vdwtype[jnrC+0];
 462             vdwjidx0D        = 2*vdwtype[jnrD+0];
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             r00              = _mm_mul_ps(rsq00,rinv00);
 469             r00              = _mm_andnot_ps(dummy_mask,r00);
 470
 471             /* Compute parameters for interactions between i and j atoms */
 472             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 473                                          vdwparam+vdwioffset0+vdwjidx0B,
 474                                          vdwparam+vdwioffset0+vdwjidx0C,
 475                                          vdwparam+vdwioffset0+vdwjidx0D,
 476                                          &c6_00,&c12_00);
 477
 478             /* Calculate table index by multiplying r with table scale and truncate to integer */
 479             rt               = _mm_mul_ps(r00,vftabscale);
 480             vfitab           = _mm_cvttps_epi32(rt);
 481             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 482             vfitab           = _mm_slli_epi32(vfitab,3);
 483
 484             /* CUBIC SPLINE TABLE DISPERSION */
 485             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 486             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 487             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 488             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 489             _MM_TRANSPOSE4_PS(Y,F,G,H);
 490             Heps             = _mm_mul_ps(vfeps,H);
 491             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 492             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 493             vvdw6            = _mm_mul_ps(c6_00,VV);
 494             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 495             fvdw6            = _mm_mul_ps(c6_00,FF);
 496
 497             /* CUBIC SPLINE TABLE REPULSION */
 498             vfitab           = _mm_add_epi32(vfitab,ifour);
 499             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 500             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 501             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 502             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 503             _MM_TRANSPOSE4_PS(Y,F,G,H);
 504             Heps             = _mm_mul_ps(vfeps,H);
 505             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 506             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 507             vvdw12           = _mm_mul_ps(c12_00,VV);
 508             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 509             fvdw12           = _mm_mul_ps(c12_00,FF);
 510             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 511             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 512
 513             /* Update potential sum for this i atom from the interaction with this j atom. */
 514             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 515             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 516
 517             fscal            = fvdw;
 518
 519             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm_mul_ps(fscal,dx00);
 523             ty               = _mm_mul_ps(fscal,dy00);
 524             tz               = _mm_mul_ps(fscal,dz00);
 525
 526             /* Update vectorial force */
 527             fix0             = _mm_add_ps(fix0,tx);
 528             fiy0             = _mm_add_ps(fiy0,ty);
 529             fiz0             = _mm_add_ps(fiz0,tz);
 530
 531             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 532             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 533             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 534             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 535             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq10             = _mm_mul_ps(iq1,jq0);
 543
 544             /* COULOMB ELECTROSTATICS */
 545             velec            = _mm_mul_ps(qq10,rinv10);
 546             felec            = _mm_mul_ps(velec,rinvsq10);
 547
 548             /* Update potential sum for this i atom from the interaction with this j atom. */
 549             velec            = _mm_andnot_ps(dummy_mask,velec);
 550             velecsum         = _mm_add_ps(velecsum,velec);
 551
 552             fscal            = felec;
 553
 554             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 555
 556             /* Calculate temporary vectorial force */
 557             tx               = _mm_mul_ps(fscal,dx10);
 558             ty               = _mm_mul_ps(fscal,dy10);
 559             tz               = _mm_mul_ps(fscal,dz10);
 560
 561             /* Update vectorial force */
 562             fix1             = _mm_add_ps(fix1,tx);
 563             fiy1             = _mm_add_ps(fiy1,ty);
 564             fiz1             = _mm_add_ps(fiz1,tz);
 565
 566             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 567             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 568             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 569             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 570             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             /* Compute parameters for interactions between i and j atoms */
 577             qq20             = _mm_mul_ps(iq2,jq0);
 578
 579             /* COULOMB ELECTROSTATICS */
 580             velec            = _mm_mul_ps(qq20,rinv20);
 581             felec            = _mm_mul_ps(velec,rinvsq20);
 582
 583             /* Update potential sum for this i atom from the interaction with this j atom. */
 584             velec            = _mm_andnot_ps(dummy_mask,velec);
 585             velecsum         = _mm_add_ps(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 590
 591             /* Calculate temporary vectorial force */
 592             tx               = _mm_mul_ps(fscal,dx20);
 593             ty               = _mm_mul_ps(fscal,dy20);
 594             tz               = _mm_mul_ps(fscal,dz20);
 595
 596             /* Update vectorial force */
 597             fix2             = _mm_add_ps(fix2,tx);
 598             fiy2             = _mm_add_ps(fiy2,ty);
 599             fiz2             = _mm_add_ps(fiz2,tz);
 600
 601             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 602             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 603             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 604             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 605             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq30             = _mm_mul_ps(iq3,jq0);
 613
 614             /* COULOMB ELECTROSTATICS */
 615             velec            = _mm_mul_ps(qq30,rinv30);
 616             felec            = _mm_mul_ps(velec,rinvsq30);
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             velec            = _mm_andnot_ps(dummy_mask,velec);
 620             velecsum         = _mm_add_ps(velecsum,velec);
 621
 622             fscal            = felec;
 623
 624             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 625
 626             /* Calculate temporary vectorial force */
 627             tx               = _mm_mul_ps(fscal,dx30);
 628             ty               = _mm_mul_ps(fscal,dy30);
 629             tz               = _mm_mul_ps(fscal,dz30);
 630
 631             /* Update vectorial force */
 632             fix3             = _mm_add_ps(fix3,tx);
 633             fiy3             = _mm_add_ps(fiy3,ty);
 634             fiz3             = _mm_add_ps(fiz3,tz);
 635
 636             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 637             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 638             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 639             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 640             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 641
 642             /* Inner loop uses 141 flops */
 643         }
 644
 645         /* End of innermost loop */
 646
 647         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 648                                               f+i_coord_offset,fshift+i_shift_offset);
 649
 650         ggid                        = gid[iidx];
 651         /* Update potential energies */
 652         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 653         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 654
 655         /* Increment number of inner iterations */
 656         inneriter                  += j_index_end - j_index_start;
 657
 658         /* Outer loop uses 26 flops */
 659     }
 660
 661     /* Increment number of outer iterations */
 662     outeriter        += nri;
 663
 664     /* Update outer/inner flops */
 665
 666     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 667 }
 668 /*
 669  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_single
 670  * Electrostatics interaction: Coulomb
 671  * VdW interaction:            CubicSplineTable
 672  * Geometry:                   Water4-Particle
 673  * Calculate force/pot:        Force
 674  */
 675 void
 676 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_single
 677                     (t_nblist * gmx_restrict                nlist,
 678                      rvec * gmx_restrict                    xx,
 679                      rvec * gmx_restrict                    ff,
 680                      t_forcerec * gmx_restrict              fr,
 681                      t_mdatoms * gmx_restrict               mdatoms,
 682                      nb_kernel_data_t * gmx_restrict        kernel_data,
 683                      t_nrnb * gmx_restrict                  nrnb)
 684 {
 685     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 686      * just 0 for non-waters.
 687      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 688      * jnr indices corresponding to data put in the four positions in the SIMD register.
 689      */
 690     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 691     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 692     int              jnrA,jnrB,jnrC,jnrD;
 693     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 694     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 695     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 696     real             rcutoff_scalar;
 697     real             *shiftvec,*fshift,*x,*f;
 698     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 699     real             scratch[4*DIM];
 700     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 701     int              vdwioffset0;
 702     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 703     int              vdwioffset1;
 704     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 705     int              vdwioffset2;
 706     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 707     int              vdwioffset3;
 708     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 709     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 710     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 711     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 712     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 713     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 714     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 715     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 716     real             *charge;
 717     int              nvdwtype;
 718     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 719     int              *vdwtype;
 720     real             *vdwparam;
 721     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 722     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 723     __m128i          vfitab;
 724     __m128i          ifour       = _mm_set1_epi32(4);
 725     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 726     real             *vftab;
 727     __m128           dummy_mask,cutoff_mask;
 728     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 729     __m128           one     = _mm_set1_ps(1.0);
 730     __m128           two     = _mm_set1_ps(2.0);
 731     x                = xx[0];
 732     f                = ff[0];
 733
 734     nri              = nlist->nri;
 735     iinr             = nlist->iinr;
 736     jindex           = nlist->jindex;
 737     jjnr             = nlist->jjnr;
 738     shiftidx         = nlist->shift;
 739     gid              = nlist->gid;
 740     shiftvec         = fr->shift_vec[0];
 741     fshift           = fr->fshift[0];
 742     facel            = _mm_set1_ps(fr->epsfac);
 743     charge           = mdatoms->chargeA;
 744     nvdwtype         = fr->ntype;
 745     vdwparam         = fr->nbfp;
 746     vdwtype          = mdatoms->typeA;
 747
 748     vftab            = kernel_data->table_vdw->data;
 749     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 750
 751     /* Setup water-specific parameters */
 752     inr              = nlist->iinr[0];
 753     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 754     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 755     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 756     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 757
 758     /* Avoid stupid compiler warnings */
 759     jnrA = jnrB = jnrC = jnrD = 0;
 760     j_coord_offsetA = 0;
 761     j_coord_offsetB = 0;
 762     j_coord_offsetC = 0;
 763     j_coord_offsetD = 0;
 764
 765     outeriter        = 0;
 766     inneriter        = 0;
 767
 768     for(iidx=0;iidx<4*DIM;iidx++)
 769     {
 770         scratch[iidx] = 0.0;
 771     }
 772
 773     /* Start outer loop over neighborlists */
 774     for(iidx=0; iidx<nri; iidx++)
 775     {
 776         /* Load shift vector for this list */
 777         i_shift_offset   = DIM*shiftidx[iidx];
 778
 779         /* Load limits for loop over neighbors */
 780         j_index_start    = jindex[iidx];
 781         j_index_end      = jindex[iidx+1];
 782
 783         /* Get outer coordinate index */
 784         inr              = iinr[iidx];
 785         i_coord_offset   = DIM*inr;
 786
 787         /* Load i particle coords and add shift vector */
 788         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 789                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 790
 791         fix0             = _mm_setzero_ps();
 792         fiy0             = _mm_setzero_ps();
 793         fiz0             = _mm_setzero_ps();
 794         fix1             = _mm_setzero_ps();
 795         fiy1             = _mm_setzero_ps();
 796         fiz1             = _mm_setzero_ps();
 797         fix2             = _mm_setzero_ps();
 798         fiy2             = _mm_setzero_ps();
 799         fiz2             = _mm_setzero_ps();
 800         fix3             = _mm_setzero_ps();
 801         fiy3             = _mm_setzero_ps();
 802         fiz3             = _mm_setzero_ps();
 803
 804         /* Start inner kernel loop */
 805         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 806         {
 807
 808             /* Get j neighbor index, and coordinate index */
 809             jnrA             = jjnr[jidx];
 810             jnrB             = jjnr[jidx+1];
 811             jnrC             = jjnr[jidx+2];
 812             jnrD             = jjnr[jidx+3];
 813             j_coord_offsetA  = DIM*jnrA;
 814             j_coord_offsetB  = DIM*jnrB;
 815             j_coord_offsetC  = DIM*jnrC;
 816             j_coord_offsetD  = DIM*jnrD;
 817
 818             /* load j atom coordinates */
 819             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 820                                               x+j_coord_offsetC,x+j_coord_offsetD,
 821                                               &jx0,&jy0,&jz0);
 822
 823             /* Calculate displacement vector */
 824             dx00             = _mm_sub_ps(ix0,jx0);
 825             dy00             = _mm_sub_ps(iy0,jy0);
 826             dz00             = _mm_sub_ps(iz0,jz0);
 827             dx10             = _mm_sub_ps(ix1,jx0);
 828             dy10             = _mm_sub_ps(iy1,jy0);
 829             dz10             = _mm_sub_ps(iz1,jz0);
 830             dx20             = _mm_sub_ps(ix2,jx0);
 831             dy20             = _mm_sub_ps(iy2,jy0);
 832             dz20             = _mm_sub_ps(iz2,jz0);
 833             dx30             = _mm_sub_ps(ix3,jx0);
 834             dy30             = _mm_sub_ps(iy3,jy0);
 835             dz30             = _mm_sub_ps(iz3,jz0);
 836
 837             /* Calculate squared distance and things based on it */
 838             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 839             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 840             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 841             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 842
 843             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 844             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 845             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 846             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 847
 848             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 849             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 850             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 851
 852             /* Load parameters for j particles */
 853             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 854                                                               charge+jnrC+0,charge+jnrD+0);
 855             vdwjidx0A        = 2*vdwtype[jnrA+0];
 856             vdwjidx0B        = 2*vdwtype[jnrB+0];
 857             vdwjidx0C        = 2*vdwtype[jnrC+0];
 858             vdwjidx0D        = 2*vdwtype[jnrD+0];
 859
 860             /**************************
 861              * CALCULATE INTERACTIONS *
 862              **************************/
 863
 864             r00              = _mm_mul_ps(rsq00,rinv00);
 865
 866             /* Compute parameters for interactions between i and j atoms */
 867             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 868                                          vdwparam+vdwioffset0+vdwjidx0B,
 869                                          vdwparam+vdwioffset0+vdwjidx0C,
 870                                          vdwparam+vdwioffset0+vdwjidx0D,
 871                                          &c6_00,&c12_00);
 872
 873             /* Calculate table index by multiplying r with table scale and truncate to integer */
 874             rt               = _mm_mul_ps(r00,vftabscale);
 875             vfitab           = _mm_cvttps_epi32(rt);
 876             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 877             vfitab           = _mm_slli_epi32(vfitab,3);
 878
 879             /* CUBIC SPLINE TABLE DISPERSION */
 880             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 881             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 882             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 883             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 884             _MM_TRANSPOSE4_PS(Y,F,G,H);
 885             Heps             = _mm_mul_ps(vfeps,H);
 886             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 887             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 888             fvdw6            = _mm_mul_ps(c6_00,FF);
 889
 890             /* CUBIC SPLINE TABLE REPULSION */
 891             vfitab           = _mm_add_epi32(vfitab,ifour);
 892             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 893             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 894             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 895             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 896             _MM_TRANSPOSE4_PS(Y,F,G,H);
 897             Heps             = _mm_mul_ps(vfeps,H);
 898             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 899             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 900             fvdw12           = _mm_mul_ps(c12_00,FF);
 901             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 902
 903             fscal            = fvdw;
 904
 905             /* Calculate temporary vectorial force */
 906             tx               = _mm_mul_ps(fscal,dx00);
 907             ty               = _mm_mul_ps(fscal,dy00);
 908             tz               = _mm_mul_ps(fscal,dz00);
 909
 910             /* Update vectorial force */
 911             fix0             = _mm_add_ps(fix0,tx);
 912             fiy0             = _mm_add_ps(fiy0,ty);
 913             fiz0             = _mm_add_ps(fiz0,tz);
 914
 915             fjptrA             = f+j_coord_offsetA;
 916             fjptrB             = f+j_coord_offsetB;
 917             fjptrC             = f+j_coord_offsetC;
 918             fjptrD             = f+j_coord_offsetD;
 919             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 920
 921             /**************************
 922              * CALCULATE INTERACTIONS *
 923              **************************/
 924
 925             /* Compute parameters for interactions between i and j atoms */
 926             qq10             = _mm_mul_ps(iq1,jq0);
 927
 928             /* COULOMB ELECTROSTATICS */
 929             velec            = _mm_mul_ps(qq10,rinv10);
 930             felec            = _mm_mul_ps(velec,rinvsq10);
 931
 932             fscal            = felec;
 933
 934             /* Calculate temporary vectorial force */
 935             tx               = _mm_mul_ps(fscal,dx10);
 936             ty               = _mm_mul_ps(fscal,dy10);
 937             tz               = _mm_mul_ps(fscal,dz10);
 938
 939             /* Update vectorial force */
 940             fix1             = _mm_add_ps(fix1,tx);
 941             fiy1             = _mm_add_ps(fiy1,ty);
 942             fiz1             = _mm_add_ps(fiz1,tz);
 943
 944             fjptrA             = f+j_coord_offsetA;
 945             fjptrB             = f+j_coord_offsetB;
 946             fjptrC             = f+j_coord_offsetC;
 947             fjptrD             = f+j_coord_offsetD;
 948             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 949
 950             /**************************
 951              * CALCULATE INTERACTIONS *
 952              **************************/
 953
 954             /* Compute parameters for interactions between i and j atoms */
 955             qq20             = _mm_mul_ps(iq2,jq0);
 956
 957             /* COULOMB ELECTROSTATICS */
 958             velec            = _mm_mul_ps(qq20,rinv20);
 959             felec            = _mm_mul_ps(velec,rinvsq20);
 960
 961             fscal            = felec;
 962
 963             /* Calculate temporary vectorial force */
 964             tx               = _mm_mul_ps(fscal,dx20);
 965             ty               = _mm_mul_ps(fscal,dy20);
 966             tz               = _mm_mul_ps(fscal,dz20);
 967
 968             /* Update vectorial force */
 969             fix2             = _mm_add_ps(fix2,tx);
 970             fiy2             = _mm_add_ps(fiy2,ty);
 971             fiz2             = _mm_add_ps(fiz2,tz);
 972
 973             fjptrA             = f+j_coord_offsetA;
 974             fjptrB             = f+j_coord_offsetB;
 975             fjptrC             = f+j_coord_offsetC;
 976             fjptrD             = f+j_coord_offsetD;
 977             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 978
 979             /**************************
 980              * CALCULATE INTERACTIONS *
 981              **************************/
 982
 983             /* Compute parameters for interactions between i and j atoms */
 984             qq30             = _mm_mul_ps(iq3,jq0);
 985
 986             /* COULOMB ELECTROSTATICS */
 987             velec            = _mm_mul_ps(qq30,rinv30);
 988             felec            = _mm_mul_ps(velec,rinvsq30);
 989
 990             fscal            = felec;
 991
 992             /* Calculate temporary vectorial force */
 993             tx               = _mm_mul_ps(fscal,dx30);
 994             ty               = _mm_mul_ps(fscal,dy30);
 995             tz               = _mm_mul_ps(fscal,dz30);
 996
 997             /* Update vectorial force */
 998             fix3             = _mm_add_ps(fix3,tx);
 999             fiy3             = _mm_add_ps(fiy3,ty);
1000             fiz3             = _mm_add_ps(fiz3,tz);
1001
1002             fjptrA             = f+j_coord_offsetA;
1003             fjptrB             = f+j_coord_offsetB;
1004             fjptrC             = f+j_coord_offsetC;
1005             fjptrD             = f+j_coord_offsetD;
1006             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1007
1008             /* Inner loop uses 129 flops */
1009         }
1010
1011         if(jidx<j_index_end)
1012         {
1013
1014             /* Get j neighbor index, and coordinate index */
1015             jnrlistA         = jjnr[jidx];
1016             jnrlistB         = jjnr[jidx+1];
1017             jnrlistC         = jjnr[jidx+2];
1018             jnrlistD         = jjnr[jidx+3];
1019             /* Sign of each element will be negative for non-real atoms.
1020              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1021              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1022              */
1023             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1024             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1025             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1026             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1027             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1028             j_coord_offsetA  = DIM*jnrA;
1029             j_coord_offsetB  = DIM*jnrB;
1030             j_coord_offsetC  = DIM*jnrC;
1031             j_coord_offsetD  = DIM*jnrD;
1032
1033             /* load j atom coordinates */
1034             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1035                                               x+j_coord_offsetC,x+j_coord_offsetD,
1036                                               &jx0,&jy0,&jz0);
1037
1038             /* Calculate displacement vector */
1039             dx00             = _mm_sub_ps(ix0,jx0);
1040             dy00             = _mm_sub_ps(iy0,jy0);
1041             dz00             = _mm_sub_ps(iz0,jz0);
1042             dx10             = _mm_sub_ps(ix1,jx0);
1043             dy10             = _mm_sub_ps(iy1,jy0);
1044             dz10             = _mm_sub_ps(iz1,jz0);
1045             dx20             = _mm_sub_ps(ix2,jx0);
1046             dy20             = _mm_sub_ps(iy2,jy0);
1047             dz20             = _mm_sub_ps(iz2,jz0);
1048             dx30             = _mm_sub_ps(ix3,jx0);
1049             dy30             = _mm_sub_ps(iy3,jy0);
1050             dz30             = _mm_sub_ps(iz3,jz0);
1051
1052             /* Calculate squared distance and things based on it */
1053             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1054             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1055             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1056             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1057
1058             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1059             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1060             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1061             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1062
1063             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1064             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1065             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1066
1067             /* Load parameters for j particles */
1068             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1069                                                               charge+jnrC+0,charge+jnrD+0);
1070             vdwjidx0A        = 2*vdwtype[jnrA+0];
1071             vdwjidx0B        = 2*vdwtype[jnrB+0];
1072             vdwjidx0C        = 2*vdwtype[jnrC+0];
1073             vdwjidx0D        = 2*vdwtype[jnrD+0];
1074
1075             /**************************
1076              * CALCULATE INTERACTIONS *
1077              **************************/
1078
1079             r00              = _mm_mul_ps(rsq00,rinv00);
1080             r00              = _mm_andnot_ps(dummy_mask,r00);
1081
1082             /* Compute parameters for interactions between i and j atoms */
1083             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1084                                          vdwparam+vdwioffset0+vdwjidx0B,
1085                                          vdwparam+vdwioffset0+vdwjidx0C,
1086                                          vdwparam+vdwioffset0+vdwjidx0D,
1087                                          &c6_00,&c12_00);
1088
1089             /* Calculate table index by multiplying r with table scale and truncate to integer */
1090             rt               = _mm_mul_ps(r00,vftabscale);
1091             vfitab           = _mm_cvttps_epi32(rt);
1092             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1093             vfitab           = _mm_slli_epi32(vfitab,3);
1094
1095             /* CUBIC SPLINE TABLE DISPERSION */
1096             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1097             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1098             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1099             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1100             _MM_TRANSPOSE4_PS(Y,F,G,H);
1101             Heps             = _mm_mul_ps(vfeps,H);
1102             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1103             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1104             fvdw6            = _mm_mul_ps(c6_00,FF);
1105
1106             /* CUBIC SPLINE TABLE REPULSION */
1107             vfitab           = _mm_add_epi32(vfitab,ifour);
1108             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1109             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1110             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1111             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1112             _MM_TRANSPOSE4_PS(Y,F,G,H);
1113             Heps             = _mm_mul_ps(vfeps,H);
1114             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1115             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1116             fvdw12           = _mm_mul_ps(c12_00,FF);
1117             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1118
1119             fscal            = fvdw;
1120
1121             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1122
1123             /* Calculate temporary vectorial force */
1124             tx               = _mm_mul_ps(fscal,dx00);
1125             ty               = _mm_mul_ps(fscal,dy00);
1126             tz               = _mm_mul_ps(fscal,dz00);
1127
1128             /* Update vectorial force */
1129             fix0             = _mm_add_ps(fix0,tx);
1130             fiy0             = _mm_add_ps(fiy0,ty);
1131             fiz0             = _mm_add_ps(fiz0,tz);
1132
1133             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1134             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1135             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1136             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1137             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1138
1139             /**************************
1140              * CALCULATE INTERACTIONS *
1141              **************************/
1142
1143             /* Compute parameters for interactions between i and j atoms */
1144             qq10             = _mm_mul_ps(iq1,jq0);
1145
1146             /* COULOMB ELECTROSTATICS */
1147             velec            = _mm_mul_ps(qq10,rinv10);
1148             felec            = _mm_mul_ps(velec,rinvsq10);
1149
1150             fscal            = felec;
1151
1152             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1153
1154             /* Calculate temporary vectorial force */
1155             tx               = _mm_mul_ps(fscal,dx10);
1156             ty               = _mm_mul_ps(fscal,dy10);
1157             tz               = _mm_mul_ps(fscal,dz10);
1158
1159             /* Update vectorial force */
1160             fix1             = _mm_add_ps(fix1,tx);
1161             fiy1             = _mm_add_ps(fiy1,ty);
1162             fiz1             = _mm_add_ps(fiz1,tz);
1163
1164             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1165             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1166             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1167             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1168             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1169
1170             /**************************
1171              * CALCULATE INTERACTIONS *
1172              **************************/
1173
1174             /* Compute parameters for interactions between i and j atoms */
1175             qq20             = _mm_mul_ps(iq2,jq0);
1176
1177             /* COULOMB ELECTROSTATICS */
1178             velec            = _mm_mul_ps(qq20,rinv20);
1179             felec            = _mm_mul_ps(velec,rinvsq20);
1180
1181             fscal            = felec;
1182
1183             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1184
1185             /* Calculate temporary vectorial force */
1186             tx               = _mm_mul_ps(fscal,dx20);
1187             ty               = _mm_mul_ps(fscal,dy20);
1188             tz               = _mm_mul_ps(fscal,dz20);
1189
1190             /* Update vectorial force */
1191             fix2             = _mm_add_ps(fix2,tx);
1192             fiy2             = _mm_add_ps(fiy2,ty);
1193             fiz2             = _mm_add_ps(fiz2,tz);
1194
1195             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1196             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1197             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1198             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1199             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             /* Compute parameters for interactions between i and j atoms */
1206             qq30             = _mm_mul_ps(iq3,jq0);
1207
1208             /* COULOMB ELECTROSTATICS */
1209             velec            = _mm_mul_ps(qq30,rinv30);
1210             felec            = _mm_mul_ps(velec,rinvsq30);
1211
1212             fscal            = felec;
1213
1214             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1215
1216             /* Calculate temporary vectorial force */
1217             tx               = _mm_mul_ps(fscal,dx30);
1218             ty               = _mm_mul_ps(fscal,dy30);
1219             tz               = _mm_mul_ps(fscal,dz30);
1220
1221             /* Update vectorial force */
1222             fix3             = _mm_add_ps(fix3,tx);
1223             fiy3             = _mm_add_ps(fiy3,ty);
1224             fiz3             = _mm_add_ps(fiz3,tz);
1225
1226             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1227             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1228             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1229             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1230             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1231
1232             /* Inner loop uses 130 flops */
1233         }
1234
1235         /* End of innermost loop */
1236
1237         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1238                                               f+i_coord_offset,fshift+i_shift_offset);
1239
1240         /* Increment number of inner iterations */
1241         inneriter                  += j_index_end - j_index_start;
1242
1243         /* Outer loop uses 24 flops */
1244     }
1245
1246     /* Increment number of outer iterations */
1247     outeriter        += nri;
1248
1249     /* Update outer/inner flops */
1250
1251     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1252 }