src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_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_ElecRF_VdwCSTab_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_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     krf              = _mm_set1_ps(fr->ic->k_rf);
 113     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 114     crf              = _mm_set1_ps(fr->ic->c_rf);
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_vdw->data;
 120     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 121
 122     /* Setup water-specific parameters */
 123     inr              = nlist->iinr[0];
 124     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 125     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 126     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 127     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 161
 162         fix0             = _mm_setzero_ps();
 163         fiy0             = _mm_setzero_ps();
 164         fiz0             = _mm_setzero_ps();
 165         fix1             = _mm_setzero_ps();
 166         fiy1             = _mm_setzero_ps();
 167         fiz1             = _mm_setzero_ps();
 168         fix2             = _mm_setzero_ps();
 169         fiy2             = _mm_setzero_ps();
 170         fiz2             = _mm_setzero_ps();
 171         fix3             = _mm_setzero_ps();
 172         fiy3             = _mm_setzero_ps();
 173         fiz3             = _mm_setzero_ps();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_ps();
 177         vvdwsum          = _mm_setzero_ps();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             jnrC             = jjnr[jidx+2];
 187             jnrD             = jjnr[jidx+3];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190             j_coord_offsetC  = DIM*jnrC;
 191             j_coord_offsetD  = DIM*jnrD;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               x+j_coord_offsetC,x+j_coord_offsetD,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_ps(ix0,jx0);
 200             dy00             = _mm_sub_ps(iy0,jy0);
 201             dz00             = _mm_sub_ps(iz0,jz0);
 202             dx10             = _mm_sub_ps(ix1,jx0);
 203             dy10             = _mm_sub_ps(iy1,jy0);
 204             dz10             = _mm_sub_ps(iz1,jz0);
 205             dx20             = _mm_sub_ps(ix2,jx0);
 206             dy20             = _mm_sub_ps(iy2,jy0);
 207             dz20             = _mm_sub_ps(iz2,jz0);
 208             dx30             = _mm_sub_ps(ix3,jx0);
 209             dy30             = _mm_sub_ps(iy3,jy0);
 210             dz30             = _mm_sub_ps(iz3,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 214             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 215             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 216             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 217
 218             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 219             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 220             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 221             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 222
 223             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 224             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 225             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 229                                                               charge+jnrC+0,charge+jnrD+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234
 235             fjx0             = _mm_setzero_ps();
 236             fjy0             = _mm_setzero_ps();
 237             fjz0             = _mm_setzero_ps();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             r00              = _mm_mul_ps(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 247                                          vdwparam+vdwioffset0+vdwjidx0B,
 248                                          vdwparam+vdwioffset0+vdwjidx0C,
 249                                          vdwparam+vdwioffset0+vdwjidx0D,
 250                                          &c6_00,&c12_00);
 251
 252             /* Calculate table index by multiplying r with table scale and truncate to integer */
 253             rt               = _mm_mul_ps(r00,vftabscale);
 254             vfitab           = _mm_cvttps_epi32(rt);
 255             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 256             vfitab           = _mm_slli_epi32(vfitab,3);
 257
 258             /* CUBIC SPLINE TABLE DISPERSION */
 259             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 260             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 261             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 262             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 263             _MM_TRANSPOSE4_PS(Y,F,G,H);
 264             Heps             = _mm_mul_ps(vfeps,H);
 265             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 266             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 267             vvdw6            = _mm_mul_ps(c6_00,VV);
 268             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 269             fvdw6            = _mm_mul_ps(c6_00,FF);
 270
 271             /* CUBIC SPLINE TABLE REPULSION */
 272             vfitab           = _mm_add_epi32(vfitab,ifour);
 273             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 274             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 275             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 276             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 277             _MM_TRANSPOSE4_PS(Y,F,G,H);
 278             Heps             = _mm_mul_ps(vfeps,H);
 279             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 280             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 281             vvdw12           = _mm_mul_ps(c12_00,VV);
 282             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 283             fvdw12           = _mm_mul_ps(c12_00,FF);
 284             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 285             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 289
 290             fscal            = fvdw;
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm_mul_ps(fscal,dx00);
 294             ty               = _mm_mul_ps(fscal,dy00);
 295             tz               = _mm_mul_ps(fscal,dz00);
 296
 297             /* Update vectorial force */
 298             fix0             = _mm_add_ps(fix0,tx);
 299             fiy0             = _mm_add_ps(fiy0,ty);
 300             fiz0             = _mm_add_ps(fiz0,tz);
 301
 302             fjx0             = _mm_add_ps(fjx0,tx);
 303             fjy0             = _mm_add_ps(fjy0,ty);
 304             fjz0             = _mm_add_ps(fjz0,tz);
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq10             = _mm_mul_ps(iq1,jq0);
 312
 313             /* REACTION-FIELD ELECTROSTATICS */
 314             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 315             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velecsum         = _mm_add_ps(velecsum,velec);
 319
 320             fscal            = felec;
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm_mul_ps(fscal,dx10);
 324             ty               = _mm_mul_ps(fscal,dy10);
 325             tz               = _mm_mul_ps(fscal,dz10);
 326
 327             /* Update vectorial force */
 328             fix1             = _mm_add_ps(fix1,tx);
 329             fiy1             = _mm_add_ps(fiy1,ty);
 330             fiz1             = _mm_add_ps(fiz1,tz);
 331
 332             fjx0             = _mm_add_ps(fjx0,tx);
 333             fjy0             = _mm_add_ps(fjy0,ty);
 334             fjz0             = _mm_add_ps(fjz0,tz);
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq20             = _mm_mul_ps(iq2,jq0);
 342
 343             /* REACTION-FIELD ELECTROSTATICS */
 344             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 345             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velecsum         = _mm_add_ps(velecsum,velec);
 349
 350             fscal            = felec;
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm_mul_ps(fscal,dx20);
 354             ty               = _mm_mul_ps(fscal,dy20);
 355             tz               = _mm_mul_ps(fscal,dz20);
 356
 357             /* Update vectorial force */
 358             fix2             = _mm_add_ps(fix2,tx);
 359             fiy2             = _mm_add_ps(fiy2,ty);
 360             fiz2             = _mm_add_ps(fiz2,tz);
 361
 362             fjx0             = _mm_add_ps(fjx0,tx);
 363             fjy0             = _mm_add_ps(fjy0,ty);
 364             fjz0             = _mm_add_ps(fjz0,tz);
 365
 366             /**************************
 367              * CALCULATE INTERACTIONS *
 368              **************************/
 369
 370             /* Compute parameters for interactions between i and j atoms */
 371             qq30             = _mm_mul_ps(iq3,jq0);
 372
 373             /* REACTION-FIELD ELECTROSTATICS */
 374             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 375             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velecsum         = _mm_add_ps(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382             /* Calculate temporary vectorial force */
 383             tx               = _mm_mul_ps(fscal,dx30);
 384             ty               = _mm_mul_ps(fscal,dy30);
 385             tz               = _mm_mul_ps(fscal,dz30);
 386
 387             /* Update vectorial force */
 388             fix3             = _mm_add_ps(fix3,tx);
 389             fiy3             = _mm_add_ps(fiy3,ty);
 390             fiz3             = _mm_add_ps(fiz3,tz);
 391
 392             fjx0             = _mm_add_ps(fjx0,tx);
 393             fjy0             = _mm_add_ps(fjy0,ty);
 394             fjz0             = _mm_add_ps(fjz0,tz);
 395
 396             fjptrA             = f+j_coord_offsetA;
 397             fjptrB             = f+j_coord_offsetB;
 398             fjptrC             = f+j_coord_offsetC;
 399             fjptrD             = f+j_coord_offsetD;
 400
 401             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 402
 403             /* Inner loop uses 152 flops */
 404         }
 405
 406         if(jidx<j_index_end)
 407         {
 408
 409             /* Get j neighbor index, and coordinate index */
 410             jnrlistA         = jjnr[jidx];
 411             jnrlistB         = jjnr[jidx+1];
 412             jnrlistC         = jjnr[jidx+2];
 413             jnrlistD         = jjnr[jidx+3];
 414             /* Sign of each element will be negative for non-real atoms.
 415              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 416              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 417              */
 418             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 419             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 420             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 421             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 422             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 423             j_coord_offsetA  = DIM*jnrA;
 424             j_coord_offsetB  = DIM*jnrB;
 425             j_coord_offsetC  = DIM*jnrC;
 426             j_coord_offsetD  = DIM*jnrD;
 427
 428             /* load j atom coordinates */
 429             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 430                                               x+j_coord_offsetC,x+j_coord_offsetD,
 431                                               &jx0,&jy0,&jz0);
 432
 433             /* Calculate displacement vector */
 434             dx00             = _mm_sub_ps(ix0,jx0);
 435             dy00             = _mm_sub_ps(iy0,jy0);
 436             dz00             = _mm_sub_ps(iz0,jz0);
 437             dx10             = _mm_sub_ps(ix1,jx0);
 438             dy10             = _mm_sub_ps(iy1,jy0);
 439             dz10             = _mm_sub_ps(iz1,jz0);
 440             dx20             = _mm_sub_ps(ix2,jx0);
 441             dy20             = _mm_sub_ps(iy2,jy0);
 442             dz20             = _mm_sub_ps(iz2,jz0);
 443             dx30             = _mm_sub_ps(ix3,jx0);
 444             dy30             = _mm_sub_ps(iy3,jy0);
 445             dz30             = _mm_sub_ps(iz3,jz0);
 446
 447             /* Calculate squared distance and things based on it */
 448             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 449             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 450             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 451             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 452
 453             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 454             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 455             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 456             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 457
 458             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 459             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 460             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 461
 462             /* Load parameters for j particles */
 463             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 464                                                               charge+jnrC+0,charge+jnrD+0);
 465             vdwjidx0A        = 2*vdwtype[jnrA+0];
 466             vdwjidx0B        = 2*vdwtype[jnrB+0];
 467             vdwjidx0C        = 2*vdwtype[jnrC+0];
 468             vdwjidx0D        = 2*vdwtype[jnrD+0];
 469
 470             fjx0             = _mm_setzero_ps();
 471             fjy0             = _mm_setzero_ps();
 472             fjz0             = _mm_setzero_ps();
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             r00              = _mm_mul_ps(rsq00,rinv00);
 479             r00              = _mm_andnot_ps(dummy_mask,r00);
 480
 481             /* Compute parameters for interactions between i and j atoms */
 482             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 483                                          vdwparam+vdwioffset0+vdwjidx0B,
 484                                          vdwparam+vdwioffset0+vdwjidx0C,
 485                                          vdwparam+vdwioffset0+vdwjidx0D,
 486                                          &c6_00,&c12_00);
 487
 488             /* Calculate table index by multiplying r with table scale and truncate to integer */
 489             rt               = _mm_mul_ps(r00,vftabscale);
 490             vfitab           = _mm_cvttps_epi32(rt);
 491             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 492             vfitab           = _mm_slli_epi32(vfitab,3);
 493
 494             /* CUBIC SPLINE TABLE DISPERSION */
 495             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 496             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 497             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 498             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 499             _MM_TRANSPOSE4_PS(Y,F,G,H);
 500             Heps             = _mm_mul_ps(vfeps,H);
 501             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 502             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 503             vvdw6            = _mm_mul_ps(c6_00,VV);
 504             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 505             fvdw6            = _mm_mul_ps(c6_00,FF);
 506
 507             /* CUBIC SPLINE TABLE REPULSION */
 508             vfitab           = _mm_add_epi32(vfitab,ifour);
 509             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 510             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 511             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 512             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 513             _MM_TRANSPOSE4_PS(Y,F,G,H);
 514             Heps             = _mm_mul_ps(vfeps,H);
 515             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 516             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 517             vvdw12           = _mm_mul_ps(c12_00,VV);
 518             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 519             fvdw12           = _mm_mul_ps(c12_00,FF);
 520             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 521             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 525             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 526
 527             fscal            = fvdw;
 528
 529             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_ps(fscal,dx00);
 533             ty               = _mm_mul_ps(fscal,dy00);
 534             tz               = _mm_mul_ps(fscal,dz00);
 535
 536             /* Update vectorial force */
 537             fix0             = _mm_add_ps(fix0,tx);
 538             fiy0             = _mm_add_ps(fiy0,ty);
 539             fiz0             = _mm_add_ps(fiz0,tz);
 540
 541             fjx0             = _mm_add_ps(fjx0,tx);
 542             fjy0             = _mm_add_ps(fjy0,ty);
 543             fjz0             = _mm_add_ps(fjz0,tz);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             /* Compute parameters for interactions between i and j atoms */
 550             qq10             = _mm_mul_ps(iq1,jq0);
 551
 552             /* REACTION-FIELD ELECTROSTATICS */
 553             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 554             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velec            = _mm_andnot_ps(dummy_mask,velec);
 558             velecsum         = _mm_add_ps(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm_mul_ps(fscal,dx10);
 566             ty               = _mm_mul_ps(fscal,dy10);
 567             tz               = _mm_mul_ps(fscal,dz10);
 568
 569             /* Update vectorial force */
 570             fix1             = _mm_add_ps(fix1,tx);
 571             fiy1             = _mm_add_ps(fiy1,ty);
 572             fiz1             = _mm_add_ps(fiz1,tz);
 573
 574             fjx0             = _mm_add_ps(fjx0,tx);
 575             fjy0             = _mm_add_ps(fjy0,ty);
 576             fjz0             = _mm_add_ps(fjz0,tz);
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq20             = _mm_mul_ps(iq2,jq0);
 584
 585             /* REACTION-FIELD ELECTROSTATICS */
 586             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 587             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 588
 589             /* Update potential sum for this i atom from the interaction with this j atom. */
 590             velec            = _mm_andnot_ps(dummy_mask,velec);
 591             velecsum         = _mm_add_ps(velecsum,velec);
 592
 593             fscal            = felec;
 594
 595             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 596
 597             /* Calculate temporary vectorial force */
 598             tx               = _mm_mul_ps(fscal,dx20);
 599             ty               = _mm_mul_ps(fscal,dy20);
 600             tz               = _mm_mul_ps(fscal,dz20);
 601
 602             /* Update vectorial force */
 603             fix2             = _mm_add_ps(fix2,tx);
 604             fiy2             = _mm_add_ps(fiy2,ty);
 605             fiz2             = _mm_add_ps(fiz2,tz);
 606
 607             fjx0             = _mm_add_ps(fjx0,tx);
 608             fjy0             = _mm_add_ps(fjy0,ty);
 609             fjz0             = _mm_add_ps(fjz0,tz);
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             qq30             = _mm_mul_ps(iq3,jq0);
 617
 618             /* REACTION-FIELD ELECTROSTATICS */
 619             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 620             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 621
 622             /* Update potential sum for this i atom from the interaction with this j atom. */
 623             velec            = _mm_andnot_ps(dummy_mask,velec);
 624             velecsum         = _mm_add_ps(velecsum,velec);
 625
 626             fscal            = felec;
 627
 628             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 629
 630             /* Calculate temporary vectorial force */
 631             tx               = _mm_mul_ps(fscal,dx30);
 632             ty               = _mm_mul_ps(fscal,dy30);
 633             tz               = _mm_mul_ps(fscal,dz30);
 634
 635             /* Update vectorial force */
 636             fix3             = _mm_add_ps(fix3,tx);
 637             fiy3             = _mm_add_ps(fiy3,ty);
 638             fiz3             = _mm_add_ps(fiz3,tz);
 639
 640             fjx0             = _mm_add_ps(fjx0,tx);
 641             fjy0             = _mm_add_ps(fjy0,ty);
 642             fjz0             = _mm_add_ps(fjz0,tz);
 643
 644             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 645             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 646             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 647             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 648
 649             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 650
 651             /* Inner loop uses 153 flops */
 652         }
 653
 654         /* End of innermost loop */
 655
 656         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 657                                               f+i_coord_offset,fshift+i_shift_offset);
 658
 659         ggid                        = gid[iidx];
 660         /* Update potential energies */
 661         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 662         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 663
 664         /* Increment number of inner iterations */
 665         inneriter                  += j_index_end - j_index_start;
 666
 667         /* Outer loop uses 26 flops */
 668     }
 669
 670     /* Increment number of outer iterations */
 671     outeriter        += nri;
 672
 673     /* Update outer/inner flops */
 674
 675     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*153);
 676 }
 677 /*
 678  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse2_single
 679  * Electrostatics interaction: ReactionField
 680  * VdW interaction:            CubicSplineTable
 681  * Geometry:                   Water4-Particle
 682  * Calculate force/pot:        Force
 683  */
 684 void
 685 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse2_single
 686                     (t_nblist * gmx_restrict                nlist,
 687                      rvec * gmx_restrict                    xx,
 688                      rvec * gmx_restrict                    ff,
 689                      t_forcerec * gmx_restrict              fr,
 690                      t_mdatoms * gmx_restrict               mdatoms,
 691                      nb_kernel_data_t * gmx_restrict        kernel_data,
 692                      t_nrnb * gmx_restrict                  nrnb)
 693 {
 694     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 695      * just 0 for non-waters.
 696      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 697      * jnr indices corresponding to data put in the four positions in the SIMD register.
 698      */
 699     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 700     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 701     int              jnrA,jnrB,jnrC,jnrD;
 702     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 703     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 704     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 705     real             rcutoff_scalar;
 706     real             *shiftvec,*fshift,*x,*f;
 707     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 708     real             scratch[4*DIM];
 709     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 710     int              vdwioffset0;
 711     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 712     int              vdwioffset1;
 713     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 714     int              vdwioffset2;
 715     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 716     int              vdwioffset3;
 717     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 718     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 719     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 720     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 721     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 722     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 723     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 724     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 725     real             *charge;
 726     int              nvdwtype;
 727     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 728     int              *vdwtype;
 729     real             *vdwparam;
 730     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 731     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 732     __m128i          vfitab;
 733     __m128i          ifour       = _mm_set1_epi32(4);
 734     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 735     real             *vftab;
 736     __m128           dummy_mask,cutoff_mask;
 737     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 738     __m128           one     = _mm_set1_ps(1.0);
 739     __m128           two     = _mm_set1_ps(2.0);
 740     x                = xx[0];
 741     f                = ff[0];
 742
 743     nri              = nlist->nri;
 744     iinr             = nlist->iinr;
 745     jindex           = nlist->jindex;
 746     jjnr             = nlist->jjnr;
 747     shiftidx         = nlist->shift;
 748     gid              = nlist->gid;
 749     shiftvec         = fr->shift_vec[0];
 750     fshift           = fr->fshift[0];
 751     facel            = _mm_set1_ps(fr->epsfac);
 752     charge           = mdatoms->chargeA;
 753     krf              = _mm_set1_ps(fr->ic->k_rf);
 754     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 755     crf              = _mm_set1_ps(fr->ic->c_rf);
 756     nvdwtype         = fr->ntype;
 757     vdwparam         = fr->nbfp;
 758     vdwtype          = mdatoms->typeA;
 759
 760     vftab            = kernel_data->table_vdw->data;
 761     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 762
 763     /* Setup water-specific parameters */
 764     inr              = nlist->iinr[0];
 765     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 766     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 767     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 768     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 769
 770     /* Avoid stupid compiler warnings */
 771     jnrA = jnrB = jnrC = jnrD = 0;
 772     j_coord_offsetA = 0;
 773     j_coord_offsetB = 0;
 774     j_coord_offsetC = 0;
 775     j_coord_offsetD = 0;
 776
 777     outeriter        = 0;
 778     inneriter        = 0;
 779
 780     for(iidx=0;iidx<4*DIM;iidx++)
 781     {
 782         scratch[iidx] = 0.0;
 783     }
 784
 785     /* Start outer loop over neighborlists */
 786     for(iidx=0; iidx<nri; iidx++)
 787     {
 788         /* Load shift vector for this list */
 789         i_shift_offset   = DIM*shiftidx[iidx];
 790
 791         /* Load limits for loop over neighbors */
 792         j_index_start    = jindex[iidx];
 793         j_index_end      = jindex[iidx+1];
 794
 795         /* Get outer coordinate index */
 796         inr              = iinr[iidx];
 797         i_coord_offset   = DIM*inr;
 798
 799         /* Load i particle coords and add shift vector */
 800         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 801                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 802
 803         fix0             = _mm_setzero_ps();
 804         fiy0             = _mm_setzero_ps();
 805         fiz0             = _mm_setzero_ps();
 806         fix1             = _mm_setzero_ps();
 807         fiy1             = _mm_setzero_ps();
 808         fiz1             = _mm_setzero_ps();
 809         fix2             = _mm_setzero_ps();
 810         fiy2             = _mm_setzero_ps();
 811         fiz2             = _mm_setzero_ps();
 812         fix3             = _mm_setzero_ps();
 813         fiy3             = _mm_setzero_ps();
 814         fiz3             = _mm_setzero_ps();
 815
 816         /* Start inner kernel loop */
 817         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 818         {
 819
 820             /* Get j neighbor index, and coordinate index */
 821             jnrA             = jjnr[jidx];
 822             jnrB             = jjnr[jidx+1];
 823             jnrC             = jjnr[jidx+2];
 824             jnrD             = jjnr[jidx+3];
 825             j_coord_offsetA  = DIM*jnrA;
 826             j_coord_offsetB  = DIM*jnrB;
 827             j_coord_offsetC  = DIM*jnrC;
 828             j_coord_offsetD  = DIM*jnrD;
 829
 830             /* load j atom coordinates */
 831             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 832                                               x+j_coord_offsetC,x+j_coord_offsetD,
 833                                               &jx0,&jy0,&jz0);
 834
 835             /* Calculate displacement vector */
 836             dx00             = _mm_sub_ps(ix0,jx0);
 837             dy00             = _mm_sub_ps(iy0,jy0);
 838             dz00             = _mm_sub_ps(iz0,jz0);
 839             dx10             = _mm_sub_ps(ix1,jx0);
 840             dy10             = _mm_sub_ps(iy1,jy0);
 841             dz10             = _mm_sub_ps(iz1,jz0);
 842             dx20             = _mm_sub_ps(ix2,jx0);
 843             dy20             = _mm_sub_ps(iy2,jy0);
 844             dz20             = _mm_sub_ps(iz2,jz0);
 845             dx30             = _mm_sub_ps(ix3,jx0);
 846             dy30             = _mm_sub_ps(iy3,jy0);
 847             dz30             = _mm_sub_ps(iz3,jz0);
 848
 849             /* Calculate squared distance and things based on it */
 850             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 851             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 852             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 853             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 854
 855             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 856             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 857             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 858             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 859
 860             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 861             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 862             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 863
 864             /* Load parameters for j particles */
 865             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 866                                                               charge+jnrC+0,charge+jnrD+0);
 867             vdwjidx0A        = 2*vdwtype[jnrA+0];
 868             vdwjidx0B        = 2*vdwtype[jnrB+0];
 869             vdwjidx0C        = 2*vdwtype[jnrC+0];
 870             vdwjidx0D        = 2*vdwtype[jnrD+0];
 871
 872             fjx0             = _mm_setzero_ps();
 873             fjy0             = _mm_setzero_ps();
 874             fjz0             = _mm_setzero_ps();
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             r00              = _mm_mul_ps(rsq00,rinv00);
 881
 882             /* Compute parameters for interactions between i and j atoms */
 883             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 884                                          vdwparam+vdwioffset0+vdwjidx0B,
 885                                          vdwparam+vdwioffset0+vdwjidx0C,
 886                                          vdwparam+vdwioffset0+vdwjidx0D,
 887                                          &c6_00,&c12_00);
 888
 889             /* Calculate table index by multiplying r with table scale and truncate to integer */
 890             rt               = _mm_mul_ps(r00,vftabscale);
 891             vfitab           = _mm_cvttps_epi32(rt);
 892             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 893             vfitab           = _mm_slli_epi32(vfitab,3);
 894
 895             /* CUBIC SPLINE TABLE DISPERSION */
 896             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 897             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 898             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 899             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 900             _MM_TRANSPOSE4_PS(Y,F,G,H);
 901             Heps             = _mm_mul_ps(vfeps,H);
 902             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 903             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 904             fvdw6            = _mm_mul_ps(c6_00,FF);
 905
 906             /* CUBIC SPLINE TABLE REPULSION */
 907             vfitab           = _mm_add_epi32(vfitab,ifour);
 908             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 909             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 910             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 911             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 912             _MM_TRANSPOSE4_PS(Y,F,G,H);
 913             Heps             = _mm_mul_ps(vfeps,H);
 914             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 915             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 916             fvdw12           = _mm_mul_ps(c12_00,FF);
 917             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 918
 919             fscal            = fvdw;
 920
 921             /* Calculate temporary vectorial force */
 922             tx               = _mm_mul_ps(fscal,dx00);
 923             ty               = _mm_mul_ps(fscal,dy00);
 924             tz               = _mm_mul_ps(fscal,dz00);
 925
 926             /* Update vectorial force */
 927             fix0             = _mm_add_ps(fix0,tx);
 928             fiy0             = _mm_add_ps(fiy0,ty);
 929             fiz0             = _mm_add_ps(fiz0,tz);
 930
 931             fjx0             = _mm_add_ps(fjx0,tx);
 932             fjy0             = _mm_add_ps(fjy0,ty);
 933             fjz0             = _mm_add_ps(fjz0,tz);
 934
 935             /**************************
 936              * CALCULATE INTERACTIONS *
 937              **************************/
 938
 939             /* Compute parameters for interactions between i and j atoms */
 940             qq10             = _mm_mul_ps(iq1,jq0);
 941
 942             /* REACTION-FIELD ELECTROSTATICS */
 943             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 944
 945             fscal            = felec;
 946
 947             /* Calculate temporary vectorial force */
 948             tx               = _mm_mul_ps(fscal,dx10);
 949             ty               = _mm_mul_ps(fscal,dy10);
 950             tz               = _mm_mul_ps(fscal,dz10);
 951
 952             /* Update vectorial force */
 953             fix1             = _mm_add_ps(fix1,tx);
 954             fiy1             = _mm_add_ps(fiy1,ty);
 955             fiz1             = _mm_add_ps(fiz1,tz);
 956
 957             fjx0             = _mm_add_ps(fjx0,tx);
 958             fjy0             = _mm_add_ps(fjy0,ty);
 959             fjz0             = _mm_add_ps(fjz0,tz);
 960
 961             /**************************
 962              * CALCULATE INTERACTIONS *
 963              **************************/
 964
 965             /* Compute parameters for interactions between i and j atoms */
 966             qq20             = _mm_mul_ps(iq2,jq0);
 967
 968             /* REACTION-FIELD ELECTROSTATICS */
 969             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 970
 971             fscal            = felec;
 972
 973             /* Calculate temporary vectorial force */
 974             tx               = _mm_mul_ps(fscal,dx20);
 975             ty               = _mm_mul_ps(fscal,dy20);
 976             tz               = _mm_mul_ps(fscal,dz20);
 977
 978             /* Update vectorial force */
 979             fix2             = _mm_add_ps(fix2,tx);
 980             fiy2             = _mm_add_ps(fiy2,ty);
 981             fiz2             = _mm_add_ps(fiz2,tz);
 982
 983             fjx0             = _mm_add_ps(fjx0,tx);
 984             fjy0             = _mm_add_ps(fjy0,ty);
 985             fjz0             = _mm_add_ps(fjz0,tz);
 986
 987             /**************************
 988              * CALCULATE INTERACTIONS *
 989              **************************/
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             qq30             = _mm_mul_ps(iq3,jq0);
 993
 994             /* REACTION-FIELD ELECTROSTATICS */
 995             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 996
 997             fscal            = felec;
 998
 999             /* Calculate temporary vectorial force */
1000             tx               = _mm_mul_ps(fscal,dx30);
1001             ty               = _mm_mul_ps(fscal,dy30);
1002             tz               = _mm_mul_ps(fscal,dz30);
1003
1004             /* Update vectorial force */
1005             fix3             = _mm_add_ps(fix3,tx);
1006             fiy3             = _mm_add_ps(fiy3,ty);
1007             fiz3             = _mm_add_ps(fiz3,tz);
1008
1009             fjx0             = _mm_add_ps(fjx0,tx);
1010             fjy0             = _mm_add_ps(fjy0,ty);
1011             fjz0             = _mm_add_ps(fjz0,tz);
1012
1013             fjptrA             = f+j_coord_offsetA;
1014             fjptrB             = f+j_coord_offsetB;
1015             fjptrC             = f+j_coord_offsetC;
1016             fjptrD             = f+j_coord_offsetD;
1017
1018             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1019
1020             /* Inner loop uses 129 flops */
1021         }
1022
1023         if(jidx<j_index_end)
1024         {
1025
1026             /* Get j neighbor index, and coordinate index */
1027             jnrlistA         = jjnr[jidx];
1028             jnrlistB         = jjnr[jidx+1];
1029             jnrlistC         = jjnr[jidx+2];
1030             jnrlistD         = jjnr[jidx+3];
1031             /* Sign of each element will be negative for non-real atoms.
1032              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1033              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1034              */
1035             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1036             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1037             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1038             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1039             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1040             j_coord_offsetA  = DIM*jnrA;
1041             j_coord_offsetB  = DIM*jnrB;
1042             j_coord_offsetC  = DIM*jnrC;
1043             j_coord_offsetD  = DIM*jnrD;
1044
1045             /* load j atom coordinates */
1046             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1047                                               x+j_coord_offsetC,x+j_coord_offsetD,
1048                                               &jx0,&jy0,&jz0);
1049
1050             /* Calculate displacement vector */
1051             dx00             = _mm_sub_ps(ix0,jx0);
1052             dy00             = _mm_sub_ps(iy0,jy0);
1053             dz00             = _mm_sub_ps(iz0,jz0);
1054             dx10             = _mm_sub_ps(ix1,jx0);
1055             dy10             = _mm_sub_ps(iy1,jy0);
1056             dz10             = _mm_sub_ps(iz1,jz0);
1057             dx20             = _mm_sub_ps(ix2,jx0);
1058             dy20             = _mm_sub_ps(iy2,jy0);
1059             dz20             = _mm_sub_ps(iz2,jz0);
1060             dx30             = _mm_sub_ps(ix3,jx0);
1061             dy30             = _mm_sub_ps(iy3,jy0);
1062             dz30             = _mm_sub_ps(iz3,jz0);
1063
1064             /* Calculate squared distance and things based on it */
1065             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1066             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1067             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1068             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1069
1070             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1071             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1072             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1073             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1074
1075             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1076             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1077             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1078
1079             /* Load parameters for j particles */
1080             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1081                                                               charge+jnrC+0,charge+jnrD+0);
1082             vdwjidx0A        = 2*vdwtype[jnrA+0];
1083             vdwjidx0B        = 2*vdwtype[jnrB+0];
1084             vdwjidx0C        = 2*vdwtype[jnrC+0];
1085             vdwjidx0D        = 2*vdwtype[jnrD+0];
1086
1087             fjx0             = _mm_setzero_ps();
1088             fjy0             = _mm_setzero_ps();
1089             fjz0             = _mm_setzero_ps();
1090
1091             /**************************
1092              * CALCULATE INTERACTIONS *
1093              **************************/
1094
1095             r00              = _mm_mul_ps(rsq00,rinv00);
1096             r00              = _mm_andnot_ps(dummy_mask,r00);
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1100                                          vdwparam+vdwioffset0+vdwjidx0B,
1101                                          vdwparam+vdwioffset0+vdwjidx0C,
1102                                          vdwparam+vdwioffset0+vdwjidx0D,
1103                                          &c6_00,&c12_00);
1104
1105             /* Calculate table index by multiplying r with table scale and truncate to integer */
1106             rt               = _mm_mul_ps(r00,vftabscale);
1107             vfitab           = _mm_cvttps_epi32(rt);
1108             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1109             vfitab           = _mm_slli_epi32(vfitab,3);
1110
1111             /* CUBIC SPLINE TABLE DISPERSION */
1112             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1113             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1114             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1115             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1116             _MM_TRANSPOSE4_PS(Y,F,G,H);
1117             Heps             = _mm_mul_ps(vfeps,H);
1118             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1119             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1120             fvdw6            = _mm_mul_ps(c6_00,FF);
1121
1122             /* CUBIC SPLINE TABLE REPULSION */
1123             vfitab           = _mm_add_epi32(vfitab,ifour);
1124             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1125             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1126             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1127             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1128             _MM_TRANSPOSE4_PS(Y,F,G,H);
1129             Heps             = _mm_mul_ps(vfeps,H);
1130             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1131             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1132             fvdw12           = _mm_mul_ps(c12_00,FF);
1133             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1134
1135             fscal            = fvdw;
1136
1137             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm_mul_ps(fscal,dx00);
1141             ty               = _mm_mul_ps(fscal,dy00);
1142             tz               = _mm_mul_ps(fscal,dz00);
1143
1144             /* Update vectorial force */
1145             fix0             = _mm_add_ps(fix0,tx);
1146             fiy0             = _mm_add_ps(fiy0,ty);
1147             fiz0             = _mm_add_ps(fiz0,tz);
1148
1149             fjx0             = _mm_add_ps(fjx0,tx);
1150             fjy0             = _mm_add_ps(fjy0,ty);
1151             fjz0             = _mm_add_ps(fjz0,tz);
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             /* Compute parameters for interactions between i and j atoms */
1158             qq10             = _mm_mul_ps(iq1,jq0);
1159
1160             /* REACTION-FIELD ELECTROSTATICS */
1161             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1166
1167             /* Calculate temporary vectorial force */
1168             tx               = _mm_mul_ps(fscal,dx10);
1169             ty               = _mm_mul_ps(fscal,dy10);
1170             tz               = _mm_mul_ps(fscal,dz10);
1171
1172             /* Update vectorial force */
1173             fix1             = _mm_add_ps(fix1,tx);
1174             fiy1             = _mm_add_ps(fiy1,ty);
1175             fiz1             = _mm_add_ps(fiz1,tz);
1176
1177             fjx0             = _mm_add_ps(fjx0,tx);
1178             fjy0             = _mm_add_ps(fjy0,ty);
1179             fjz0             = _mm_add_ps(fjz0,tz);
1180
1181             /**************************
1182              * CALCULATE INTERACTIONS *
1183              **************************/
1184
1185             /* Compute parameters for interactions between i and j atoms */
1186             qq20             = _mm_mul_ps(iq2,jq0);
1187
1188             /* REACTION-FIELD ELECTROSTATICS */
1189             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1190
1191             fscal            = felec;
1192
1193             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1194
1195             /* Calculate temporary vectorial force */
1196             tx               = _mm_mul_ps(fscal,dx20);
1197             ty               = _mm_mul_ps(fscal,dy20);
1198             tz               = _mm_mul_ps(fscal,dz20);
1199
1200             /* Update vectorial force */
1201             fix2             = _mm_add_ps(fix2,tx);
1202             fiy2             = _mm_add_ps(fiy2,ty);
1203             fiz2             = _mm_add_ps(fiz2,tz);
1204
1205             fjx0             = _mm_add_ps(fjx0,tx);
1206             fjy0             = _mm_add_ps(fjy0,ty);
1207             fjz0             = _mm_add_ps(fjz0,tz);
1208
1209             /**************************
1210              * CALCULATE INTERACTIONS *
1211              **************************/
1212
1213             /* Compute parameters for interactions between i and j atoms */
1214             qq30             = _mm_mul_ps(iq3,jq0);
1215
1216             /* REACTION-FIELD ELECTROSTATICS */
1217             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
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 }