src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_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_ElecRFCut_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_ElecRFCut_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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwioffset3;
  73     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  75     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm_set1_ps(fr->ic->k_rf);
 110     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_ps(fr->ic->c_rf);
 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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 129     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 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         shX              = shiftvec[i_shift_offset+XX];
 147         shY              = shiftvec[i_shift_offset+YY];
 148         shZ              = shiftvec[i_shift_offset+ZZ];
 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         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 160         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 161         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 162         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 163         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 164         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 165         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 166         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 167         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 168         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 169         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 170         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 171
 172         fix0             = _mm_setzero_ps();
 173         fiy0             = _mm_setzero_ps();
 174         fiz0             = _mm_setzero_ps();
 175         fix1             = _mm_setzero_ps();
 176         fiy1             = _mm_setzero_ps();
 177         fiz1             = _mm_setzero_ps();
 178         fix2             = _mm_setzero_ps();
 179         fiy2             = _mm_setzero_ps();
 180         fiz2             = _mm_setzero_ps();
 181         fix3             = _mm_setzero_ps();
 182         fiy3             = _mm_setzero_ps();
 183         fiz3             = _mm_setzero_ps();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm_setzero_ps();
 187         vvdwsum          = _mm_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               x+j_coord_offsetC,x+j_coord_offsetD,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_ps(ix0,jx0);
 211             dy00             = _mm_sub_ps(iy0,jy0);
 212             dz00             = _mm_sub_ps(iz0,jz0);
 213             dx10             = _mm_sub_ps(ix1,jx0);
 214             dy10             = _mm_sub_ps(iy1,jy0);
 215             dz10             = _mm_sub_ps(iz1,jz0);
 216             dx20             = _mm_sub_ps(ix2,jx0);
 217             dy20             = _mm_sub_ps(iy2,jy0);
 218             dz20             = _mm_sub_ps(iz2,jz0);
 219             dx30             = _mm_sub_ps(ix3,jx0);
 220             dy30             = _mm_sub_ps(iy3,jy0);
 221             dz30             = _mm_sub_ps(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 225             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 226             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 227             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 228
 229             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 230             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 231             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 232             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 233
 234             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 235             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 236             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 237
 238             /* Load parameters for j particles */
 239             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 240                                                               charge+jnrC+0,charge+jnrD+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243             vdwjidx0C        = 2*vdwtype[jnrC+0];
 244             vdwjidx0D        = 2*vdwtype[jnrD+0];
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             r00              = _mm_mul_ps(rsq00,rinv00);
 251
 252             /* Compute parameters for interactions between i and j atoms */
 253             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 254                                          vdwparam+vdwioffset0+vdwjidx0B,
 255                                          vdwparam+vdwioffset0+vdwjidx0C,
 256                                          vdwparam+vdwioffset0+vdwjidx0D,
 257                                          &c6_00,&c12_00);
 258
 259             /* Calculate table index by multiplying r with table scale and truncate to integer */
 260             rt               = _mm_mul_ps(r00,vftabscale);
 261             vfitab           = _mm_cvttps_epi32(rt);
 262             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 263             vfitab           = _mm_slli_epi32(vfitab,3);
 264
 265             /* CUBIC SPLINE TABLE DISPERSION */
 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             vvdw6            = _mm_mul_ps(c6_00,VV);
 275             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 276             fvdw6            = _mm_mul_ps(c6_00,FF);
 277
 278             /* CUBIC SPLINE TABLE REPULSION */
 279             vfitab           = _mm_add_epi32(vfitab,ifour);
 280             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 281             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 282             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 283             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 284             _MM_TRANSPOSE4_PS(Y,F,G,H);
 285             Heps             = _mm_mul_ps(vfeps,H);
 286             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 287             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 288             vvdw12           = _mm_mul_ps(c12_00,VV);
 289             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 290             fvdw12           = _mm_mul_ps(c12_00,FF);
 291             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 292             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 296
 297             fscal            = fvdw;
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm_mul_ps(fscal,dx00);
 301             ty               = _mm_mul_ps(fscal,dy00);
 302             tz               = _mm_mul_ps(fscal,dz00);
 303
 304             /* Update vectorial force */
 305             fix0             = _mm_add_ps(fix0,tx);
 306             fiy0             = _mm_add_ps(fiy0,ty);
 307             fiz0             = _mm_add_ps(fiz0,tz);
 308
 309             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 310                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 311                                                    tx,ty,tz);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             if (gmx_mm_any_lt(rsq10,rcutoff2))
 318             {
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq10             = _mm_mul_ps(iq1,jq0);
 322
 323             /* REACTION-FIELD ELECTROSTATICS */
 324             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 325             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 326
 327             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velec            = _mm_and_ps(velec,cutoff_mask);
 331             velecsum         = _mm_add_ps(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             fscal            = _mm_and_ps(fscal,cutoff_mask);
 336
 337             /* Calculate temporary vectorial force */
 338             tx               = _mm_mul_ps(fscal,dx10);
 339             ty               = _mm_mul_ps(fscal,dy10);
 340             tz               = _mm_mul_ps(fscal,dz10);
 341
 342             /* Update vectorial force */
 343             fix1             = _mm_add_ps(fix1,tx);
 344             fiy1             = _mm_add_ps(fiy1,ty);
 345             fiz1             = _mm_add_ps(fiz1,tz);
 346
 347             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 348                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 349                                                    tx,ty,tz);
 350
 351             }
 352
 353             /**************************
 354              * CALCULATE INTERACTIONS *
 355              **************************/
 356
 357             if (gmx_mm_any_lt(rsq20,rcutoff2))
 358             {
 359
 360             /* Compute parameters for interactions between i and j atoms */
 361             qq20             = _mm_mul_ps(iq2,jq0);
 362
 363             /* REACTION-FIELD ELECTROSTATICS */
 364             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 365             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 366
 367             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 368
 369             /* Update potential sum for this i atom from the interaction with this j atom. */
 370             velec            = _mm_and_ps(velec,cutoff_mask);
 371             velecsum         = _mm_add_ps(velecsum,velec);
 372
 373             fscal            = felec;
 374
 375             fscal            = _mm_and_ps(fscal,cutoff_mask);
 376
 377             /* Calculate temporary vectorial force */
 378             tx               = _mm_mul_ps(fscal,dx20);
 379             ty               = _mm_mul_ps(fscal,dy20);
 380             tz               = _mm_mul_ps(fscal,dz20);
 381
 382             /* Update vectorial force */
 383             fix2             = _mm_add_ps(fix2,tx);
 384             fiy2             = _mm_add_ps(fiy2,ty);
 385             fiz2             = _mm_add_ps(fiz2,tz);
 386
 387             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 388                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 389                                                    tx,ty,tz);
 390
 391             }
 392
 393             /**************************
 394              * CALCULATE INTERACTIONS *
 395              **************************/
 396
 397             if (gmx_mm_any_lt(rsq30,rcutoff2))
 398             {
 399
 400             /* Compute parameters for interactions between i and j atoms */
 401             qq30             = _mm_mul_ps(iq3,jq0);
 402
 403             /* REACTION-FIELD ELECTROSTATICS */
 404             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 405             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 406
 407             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velec            = _mm_and_ps(velec,cutoff_mask);
 411             velecsum         = _mm_add_ps(velecsum,velec);
 412
 413             fscal            = felec;
 414
 415             fscal            = _mm_and_ps(fscal,cutoff_mask);
 416
 417             /* Calculate temporary vectorial force */
 418             tx               = _mm_mul_ps(fscal,dx30);
 419             ty               = _mm_mul_ps(fscal,dy30);
 420             tz               = _mm_mul_ps(fscal,dz30);
 421
 422             /* Update vectorial force */
 423             fix3             = _mm_add_ps(fix3,tx);
 424             fiy3             = _mm_add_ps(fiy3,ty);
 425             fiz3             = _mm_add_ps(fiz3,tz);
 426
 427             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 428                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 429                                                    tx,ty,tz);
 430
 431             }
 432
 433             /* Inner loop uses 164 flops */
 434         }
 435
 436         if(jidx<j_index_end)
 437         {
 438
 439             /* Get j neighbor index, and coordinate index */
 440             jnrA             = jjnr[jidx];
 441             jnrB             = jjnr[jidx+1];
 442             jnrC             = jjnr[jidx+2];
 443             jnrD             = jjnr[jidx+3];
 444
 445             /* Sign of each element will be negative for non-real atoms.
 446              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 447              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 448              */
 449             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 450             jnrA       = (jnrA>=0) ? jnrA : 0;
 451             jnrB       = (jnrB>=0) ? jnrB : 0;
 452             jnrC       = (jnrC>=0) ? jnrC : 0;
 453             jnrD       = (jnrD>=0) ? jnrD : 0;
 454
 455             j_coord_offsetA  = DIM*jnrA;
 456             j_coord_offsetB  = DIM*jnrB;
 457             j_coord_offsetC  = DIM*jnrC;
 458             j_coord_offsetD  = DIM*jnrD;
 459
 460             /* load j atom coordinates */
 461             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 462                                               x+j_coord_offsetC,x+j_coord_offsetD,
 463                                               &jx0,&jy0,&jz0);
 464
 465             /* Calculate displacement vector */
 466             dx00             = _mm_sub_ps(ix0,jx0);
 467             dy00             = _mm_sub_ps(iy0,jy0);
 468             dz00             = _mm_sub_ps(iz0,jz0);
 469             dx10             = _mm_sub_ps(ix1,jx0);
 470             dy10             = _mm_sub_ps(iy1,jy0);
 471             dz10             = _mm_sub_ps(iz1,jz0);
 472             dx20             = _mm_sub_ps(ix2,jx0);
 473             dy20             = _mm_sub_ps(iy2,jy0);
 474             dz20             = _mm_sub_ps(iz2,jz0);
 475             dx30             = _mm_sub_ps(ix3,jx0);
 476             dy30             = _mm_sub_ps(iy3,jy0);
 477             dz30             = _mm_sub_ps(iz3,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 481             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 482             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 483             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 484
 485             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 486             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 487             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 488             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 489
 490             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 491             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 492             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 493
 494             /* Load parameters for j particles */
 495             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 496                                                               charge+jnrC+0,charge+jnrD+0);
 497             vdwjidx0A        = 2*vdwtype[jnrA+0];
 498             vdwjidx0B        = 2*vdwtype[jnrB+0];
 499             vdwjidx0C        = 2*vdwtype[jnrC+0];
 500             vdwjidx0D        = 2*vdwtype[jnrD+0];
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             r00              = _mm_mul_ps(rsq00,rinv00);
 507             r00              = _mm_andnot_ps(dummy_mask,r00);
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 511                                          vdwparam+vdwioffset0+vdwjidx0B,
 512                                          vdwparam+vdwioffset0+vdwjidx0C,
 513                                          vdwparam+vdwioffset0+vdwjidx0D,
 514                                          &c6_00,&c12_00);
 515
 516             /* Calculate table index by multiplying r with table scale and truncate to integer */
 517             rt               = _mm_mul_ps(r00,vftabscale);
 518             vfitab           = _mm_cvttps_epi32(rt);
 519             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 520             vfitab           = _mm_slli_epi32(vfitab,3);
 521
 522             /* CUBIC SPLINE TABLE DISPERSION */
 523             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 524             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 525             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 526             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 527             _MM_TRANSPOSE4_PS(Y,F,G,H);
 528             Heps             = _mm_mul_ps(vfeps,H);
 529             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 530             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 531             vvdw6            = _mm_mul_ps(c6_00,VV);
 532             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 533             fvdw6            = _mm_mul_ps(c6_00,FF);
 534
 535             /* CUBIC SPLINE TABLE REPULSION */
 536             vfitab           = _mm_add_epi32(vfitab,ifour);
 537             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 538             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 539             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 540             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 541             _MM_TRANSPOSE4_PS(Y,F,G,H);
 542             Heps             = _mm_mul_ps(vfeps,H);
 543             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 544             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 545             vvdw12           = _mm_mul_ps(c12_00,VV);
 546             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 547             fvdw12           = _mm_mul_ps(c12_00,FF);
 548             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 549             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 553             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 554
 555             fscal            = fvdw;
 556
 557             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm_mul_ps(fscal,dx00);
 561             ty               = _mm_mul_ps(fscal,dy00);
 562             tz               = _mm_mul_ps(fscal,dz00);
 563
 564             /* Update vectorial force */
 565             fix0             = _mm_add_ps(fix0,tx);
 566             fiy0             = _mm_add_ps(fiy0,ty);
 567             fiz0             = _mm_add_ps(fiz0,tz);
 568
 569             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 570                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 571                                                    tx,ty,tz);
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             if (gmx_mm_any_lt(rsq10,rcutoff2))
 578             {
 579
 580             /* Compute parameters for interactions between i and j atoms */
 581             qq10             = _mm_mul_ps(iq1,jq0);
 582
 583             /* REACTION-FIELD ELECTROSTATICS */
 584             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 585             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 586
 587             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 588
 589             /* Update potential sum for this i atom from the interaction with this j atom. */
 590             velec            = _mm_and_ps(velec,cutoff_mask);
 591             velec            = _mm_andnot_ps(dummy_mask,velec);
 592             velecsum         = _mm_add_ps(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             fscal            = _mm_and_ps(fscal,cutoff_mask);
 597
 598             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 599
 600             /* Calculate temporary vectorial force */
 601             tx               = _mm_mul_ps(fscal,dx10);
 602             ty               = _mm_mul_ps(fscal,dy10);
 603             tz               = _mm_mul_ps(fscal,dz10);
 604
 605             /* Update vectorial force */
 606             fix1             = _mm_add_ps(fix1,tx);
 607             fiy1             = _mm_add_ps(fiy1,ty);
 608             fiz1             = _mm_add_ps(fiz1,tz);
 609
 610             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 611                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 612                                                    tx,ty,tz);
 613
 614             }
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             if (gmx_mm_any_lt(rsq20,rcutoff2))
 621             {
 622
 623             /* Compute parameters for interactions between i and j atoms */
 624             qq20             = _mm_mul_ps(iq2,jq0);
 625
 626             /* REACTION-FIELD ELECTROSTATICS */
 627             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 628             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 629
 630             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 631
 632             /* Update potential sum for this i atom from the interaction with this j atom. */
 633             velec            = _mm_and_ps(velec,cutoff_mask);
 634             velec            = _mm_andnot_ps(dummy_mask,velec);
 635             velecsum         = _mm_add_ps(velecsum,velec);
 636
 637             fscal            = felec;
 638
 639             fscal            = _mm_and_ps(fscal,cutoff_mask);
 640
 641             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 642
 643             /* Calculate temporary vectorial force */
 644             tx               = _mm_mul_ps(fscal,dx20);
 645             ty               = _mm_mul_ps(fscal,dy20);
 646             tz               = _mm_mul_ps(fscal,dz20);
 647
 648             /* Update vectorial force */
 649             fix2             = _mm_add_ps(fix2,tx);
 650             fiy2             = _mm_add_ps(fiy2,ty);
 651             fiz2             = _mm_add_ps(fiz2,tz);
 652
 653             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 654                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 655                                                    tx,ty,tz);
 656
 657             }
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             if (gmx_mm_any_lt(rsq30,rcutoff2))
 664             {
 665
 666             /* Compute parameters for interactions between i and j atoms */
 667             qq30             = _mm_mul_ps(iq3,jq0);
 668
 669             /* REACTION-FIELD ELECTROSTATICS */
 670             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 671             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 672
 673             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 674
 675             /* Update potential sum for this i atom from the interaction with this j atom. */
 676             velec            = _mm_and_ps(velec,cutoff_mask);
 677             velec            = _mm_andnot_ps(dummy_mask,velec);
 678             velecsum         = _mm_add_ps(velecsum,velec);
 679
 680             fscal            = felec;
 681
 682             fscal            = _mm_and_ps(fscal,cutoff_mask);
 683
 684             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 685
 686             /* Calculate temporary vectorial force */
 687             tx               = _mm_mul_ps(fscal,dx30);
 688             ty               = _mm_mul_ps(fscal,dy30);
 689             tz               = _mm_mul_ps(fscal,dz30);
 690
 691             /* Update vectorial force */
 692             fix3             = _mm_add_ps(fix3,tx);
 693             fiy3             = _mm_add_ps(fiy3,ty);
 694             fiz3             = _mm_add_ps(fiz3,tz);
 695
 696             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 697                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 698                                                    tx,ty,tz);
 699
 700             }
 701
 702             /* Inner loop uses 165 flops */
 703         }
 704
 705         /* End of innermost loop */
 706
 707         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 708                                               f+i_coord_offset,fshift+i_shift_offset);
 709
 710         ggid                        = gid[iidx];
 711         /* Update potential energies */
 712         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 713         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 714
 715         /* Increment number of inner iterations */
 716         inneriter                  += j_index_end - j_index_start;
 717
 718         /* Outer loop uses 38 flops */
 719     }
 720
 721     /* Increment number of outer iterations */
 722     outeriter        += nri;
 723
 724     /* Update outer/inner flops */
 725
 726     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*38 + inneriter*165);
 727 }
 728 /*
 729  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse2_single
 730  * Electrostatics interaction: ReactionField
 731  * VdW interaction:            CubicSplineTable
 732  * Geometry:                   Water4-Particle
 733  * Calculate force/pot:        Force
 734  */
 735 void
 736 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse2_single
 737                     (t_nblist * gmx_restrict                nlist,
 738                      rvec * gmx_restrict                    xx,
 739                      rvec * gmx_restrict                    ff,
 740                      t_forcerec * gmx_restrict              fr,
 741                      t_mdatoms * gmx_restrict               mdatoms,
 742                      nb_kernel_data_t * gmx_restrict        kernel_data,
 743                      t_nrnb * gmx_restrict                  nrnb)
 744 {
 745     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 746      * just 0 for non-waters.
 747      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 748      * jnr indices corresponding to data put in the four positions in the SIMD register.
 749      */
 750     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 751     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 752     int              jnrA,jnrB,jnrC,jnrD;
 753     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 754     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 755     real             shX,shY,shZ,rcutoff_scalar;
 756     real             *shiftvec,*fshift,*x,*f;
 757     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 758     int              vdwioffset0;
 759     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 760     int              vdwioffset1;
 761     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 762     int              vdwioffset2;
 763     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 764     int              vdwioffset3;
 765     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 766     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 767     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 768     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 769     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 770     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 771     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 772     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 773     real             *charge;
 774     int              nvdwtype;
 775     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 776     int              *vdwtype;
 777     real             *vdwparam;
 778     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 779     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 780     __m128i          vfitab;
 781     __m128i          ifour       = _mm_set1_epi32(4);
 782     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 783     real             *vftab;
 784     __m128           dummy_mask,cutoff_mask;
 785     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 786     __m128           one     = _mm_set1_ps(1.0);
 787     __m128           two     = _mm_set1_ps(2.0);
 788     x                = xx[0];
 789     f                = ff[0];
 790
 791     nri              = nlist->nri;
 792     iinr             = nlist->iinr;
 793     jindex           = nlist->jindex;
 794     jjnr             = nlist->jjnr;
 795     shiftidx         = nlist->shift;
 796     gid              = nlist->gid;
 797     shiftvec         = fr->shift_vec[0];
 798     fshift           = fr->fshift[0];
 799     facel            = _mm_set1_ps(fr->epsfac);
 800     charge           = mdatoms->chargeA;
 801     krf              = _mm_set1_ps(fr->ic->k_rf);
 802     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 803     crf              = _mm_set1_ps(fr->ic->c_rf);
 804     nvdwtype         = fr->ntype;
 805     vdwparam         = fr->nbfp;
 806     vdwtype          = mdatoms->typeA;
 807
 808     vftab            = kernel_data->table_vdw->data;
 809     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 810
 811     /* Setup water-specific parameters */
 812     inr              = nlist->iinr[0];
 813     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 814     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 815     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 816     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 817
 818     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 819     rcutoff_scalar   = fr->rcoulomb;
 820     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 821     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 822
 823     /* Avoid stupid compiler warnings */
 824     jnrA = jnrB = jnrC = jnrD = 0;
 825     j_coord_offsetA = 0;
 826     j_coord_offsetB = 0;
 827     j_coord_offsetC = 0;
 828     j_coord_offsetD = 0;
 829
 830     outeriter        = 0;
 831     inneriter        = 0;
 832
 833     /* Start outer loop over neighborlists */
 834     for(iidx=0; iidx<nri; iidx++)
 835     {
 836         /* Load shift vector for this list */
 837         i_shift_offset   = DIM*shiftidx[iidx];
 838         shX              = shiftvec[i_shift_offset+XX];
 839         shY              = shiftvec[i_shift_offset+YY];
 840         shZ              = shiftvec[i_shift_offset+ZZ];
 841
 842         /* Load limits for loop over neighbors */
 843         j_index_start    = jindex[iidx];
 844         j_index_end      = jindex[iidx+1];
 845
 846         /* Get outer coordinate index */
 847         inr              = iinr[iidx];
 848         i_coord_offset   = DIM*inr;
 849
 850         /* Load i particle coords and add shift vector */
 851         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 852         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 853         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 854         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 855         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 856         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 857         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 858         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 859         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 860         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 861         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 862         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 863
 864         fix0             = _mm_setzero_ps();
 865         fiy0             = _mm_setzero_ps();
 866         fiz0             = _mm_setzero_ps();
 867         fix1             = _mm_setzero_ps();
 868         fiy1             = _mm_setzero_ps();
 869         fiz1             = _mm_setzero_ps();
 870         fix2             = _mm_setzero_ps();
 871         fiy2             = _mm_setzero_ps();
 872         fiz2             = _mm_setzero_ps();
 873         fix3             = _mm_setzero_ps();
 874         fiy3             = _mm_setzero_ps();
 875         fiz3             = _mm_setzero_ps();
 876
 877         /* Start inner kernel loop */
 878         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 879         {
 880
 881             /* Get j neighbor index, and coordinate index */
 882             jnrA             = jjnr[jidx];
 883             jnrB             = jjnr[jidx+1];
 884             jnrC             = jjnr[jidx+2];
 885             jnrD             = jjnr[jidx+3];
 886
 887             j_coord_offsetA  = DIM*jnrA;
 888             j_coord_offsetB  = DIM*jnrB;
 889             j_coord_offsetC  = DIM*jnrC;
 890             j_coord_offsetD  = DIM*jnrD;
 891
 892             /* load j atom coordinates */
 893             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 894                                               x+j_coord_offsetC,x+j_coord_offsetD,
 895                                               &jx0,&jy0,&jz0);
 896
 897             /* Calculate displacement vector */
 898             dx00             = _mm_sub_ps(ix0,jx0);
 899             dy00             = _mm_sub_ps(iy0,jy0);
 900             dz00             = _mm_sub_ps(iz0,jz0);
 901             dx10             = _mm_sub_ps(ix1,jx0);
 902             dy10             = _mm_sub_ps(iy1,jy0);
 903             dz10             = _mm_sub_ps(iz1,jz0);
 904             dx20             = _mm_sub_ps(ix2,jx0);
 905             dy20             = _mm_sub_ps(iy2,jy0);
 906             dz20             = _mm_sub_ps(iz2,jz0);
 907             dx30             = _mm_sub_ps(ix3,jx0);
 908             dy30             = _mm_sub_ps(iy3,jy0);
 909             dz30             = _mm_sub_ps(iz3,jz0);
 910
 911             /* Calculate squared distance and things based on it */
 912             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 913             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 914             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 915             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 916
 917             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 918             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 919             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 920             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 921
 922             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 923             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 924             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 925
 926             /* Load parameters for j particles */
 927             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 928                                                               charge+jnrC+0,charge+jnrD+0);
 929             vdwjidx0A        = 2*vdwtype[jnrA+0];
 930             vdwjidx0B        = 2*vdwtype[jnrB+0];
 931             vdwjidx0C        = 2*vdwtype[jnrC+0];
 932             vdwjidx0D        = 2*vdwtype[jnrD+0];
 933
 934             /**************************
 935              * CALCULATE INTERACTIONS *
 936              **************************/
 937
 938             r00              = _mm_mul_ps(rsq00,rinv00);
 939
 940             /* Compute parameters for interactions between i and j atoms */
 941             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 942                                          vdwparam+vdwioffset0+vdwjidx0B,
 943                                          vdwparam+vdwioffset0+vdwjidx0C,
 944                                          vdwparam+vdwioffset0+vdwjidx0D,
 945                                          &c6_00,&c12_00);
 946
 947             /* Calculate table index by multiplying r with table scale and truncate to integer */
 948             rt               = _mm_mul_ps(r00,vftabscale);
 949             vfitab           = _mm_cvttps_epi32(rt);
 950             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 951             vfitab           = _mm_slli_epi32(vfitab,3);
 952
 953             /* CUBIC SPLINE TABLE DISPERSION */
 954             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 955             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 956             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 957             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 958             _MM_TRANSPOSE4_PS(Y,F,G,H);
 959             Heps             = _mm_mul_ps(vfeps,H);
 960             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 961             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 962             fvdw6            = _mm_mul_ps(c6_00,FF);
 963
 964             /* CUBIC SPLINE TABLE REPULSION */
 965             vfitab           = _mm_add_epi32(vfitab,ifour);
 966             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 967             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 968             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 969             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 970             _MM_TRANSPOSE4_PS(Y,F,G,H);
 971             Heps             = _mm_mul_ps(vfeps,H);
 972             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 973             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 974             fvdw12           = _mm_mul_ps(c12_00,FF);
 975             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 976
 977             fscal            = fvdw;
 978
 979             /* Calculate temporary vectorial force */
 980             tx               = _mm_mul_ps(fscal,dx00);
 981             ty               = _mm_mul_ps(fscal,dy00);
 982             tz               = _mm_mul_ps(fscal,dz00);
 983
 984             /* Update vectorial force */
 985             fix0             = _mm_add_ps(fix0,tx);
 986             fiy0             = _mm_add_ps(fiy0,ty);
 987             fiz0             = _mm_add_ps(fiz0,tz);
 988
 989             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 990                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 991                                                    tx,ty,tz);
 992
 993             /**************************
 994              * CALCULATE INTERACTIONS *
 995              **************************/
 996
 997             if (gmx_mm_any_lt(rsq10,rcutoff2))
 998             {
 999
1000             /* Compute parameters for interactions between i and j atoms */
1001             qq10             = _mm_mul_ps(iq1,jq0);
1002
1003             /* REACTION-FIELD ELECTROSTATICS */
1004             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1005
1006             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1007
1008             fscal            = felec;
1009
1010             fscal            = _mm_and_ps(fscal,cutoff_mask);
1011
1012             /* Calculate temporary vectorial force */
1013             tx               = _mm_mul_ps(fscal,dx10);
1014             ty               = _mm_mul_ps(fscal,dy10);
1015             tz               = _mm_mul_ps(fscal,dz10);
1016
1017             /* Update vectorial force */
1018             fix1             = _mm_add_ps(fix1,tx);
1019             fiy1             = _mm_add_ps(fiy1,ty);
1020             fiz1             = _mm_add_ps(fiz1,tz);
1021
1022             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1023                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1024                                                    tx,ty,tz);
1025
1026             }
1027
1028             /**************************
1029              * CALCULATE INTERACTIONS *
1030              **************************/
1031
1032             if (gmx_mm_any_lt(rsq20,rcutoff2))
1033             {
1034
1035             /* Compute parameters for interactions between i and j atoms */
1036             qq20             = _mm_mul_ps(iq2,jq0);
1037
1038             /* REACTION-FIELD ELECTROSTATICS */
1039             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1040
1041             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1042
1043             fscal            = felec;
1044
1045             fscal            = _mm_and_ps(fscal,cutoff_mask);
1046
1047             /* Calculate temporary vectorial force */
1048             tx               = _mm_mul_ps(fscal,dx20);
1049             ty               = _mm_mul_ps(fscal,dy20);
1050             tz               = _mm_mul_ps(fscal,dz20);
1051
1052             /* Update vectorial force */
1053             fix2             = _mm_add_ps(fix2,tx);
1054             fiy2             = _mm_add_ps(fiy2,ty);
1055             fiz2             = _mm_add_ps(fiz2,tz);
1056
1057             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1058                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1059                                                    tx,ty,tz);
1060
1061             }
1062
1063             /**************************
1064              * CALCULATE INTERACTIONS *
1065              **************************/
1066
1067             if (gmx_mm_any_lt(rsq30,rcutoff2))
1068             {
1069
1070             /* Compute parameters for interactions between i and j atoms */
1071             qq30             = _mm_mul_ps(iq3,jq0);
1072
1073             /* REACTION-FIELD ELECTROSTATICS */
1074             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1075
1076             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1077
1078             fscal            = felec;
1079
1080             fscal            = _mm_and_ps(fscal,cutoff_mask);
1081
1082             /* Calculate temporary vectorial force */
1083             tx               = _mm_mul_ps(fscal,dx30);
1084             ty               = _mm_mul_ps(fscal,dy30);
1085             tz               = _mm_mul_ps(fscal,dz30);
1086
1087             /* Update vectorial force */
1088             fix3             = _mm_add_ps(fix3,tx);
1089             fiy3             = _mm_add_ps(fiy3,ty);
1090             fiz3             = _mm_add_ps(fiz3,tz);
1091
1092             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1093                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1094                                                    tx,ty,tz);
1095
1096             }
1097
1098             /* Inner loop uses 138 flops */
1099         }
1100
1101         if(jidx<j_index_end)
1102         {
1103
1104             /* Get j neighbor index, and coordinate index */
1105             jnrA             = jjnr[jidx];
1106             jnrB             = jjnr[jidx+1];
1107             jnrC             = jjnr[jidx+2];
1108             jnrD             = jjnr[jidx+3];
1109
1110             /* Sign of each element will be negative for non-real atoms.
1111              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1112              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1113              */
1114             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1115             jnrA       = (jnrA>=0) ? jnrA : 0;
1116             jnrB       = (jnrB>=0) ? jnrB : 0;
1117             jnrC       = (jnrC>=0) ? jnrC : 0;
1118             jnrD       = (jnrD>=0) ? jnrD : 0;
1119
1120             j_coord_offsetA  = DIM*jnrA;
1121             j_coord_offsetB  = DIM*jnrB;
1122             j_coord_offsetC  = DIM*jnrC;
1123             j_coord_offsetD  = DIM*jnrD;
1124
1125             /* load j atom coordinates */
1126             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1127                                               x+j_coord_offsetC,x+j_coord_offsetD,
1128                                               &jx0,&jy0,&jz0);
1129
1130             /* Calculate displacement vector */
1131             dx00             = _mm_sub_ps(ix0,jx0);
1132             dy00             = _mm_sub_ps(iy0,jy0);
1133             dz00             = _mm_sub_ps(iz0,jz0);
1134             dx10             = _mm_sub_ps(ix1,jx0);
1135             dy10             = _mm_sub_ps(iy1,jy0);
1136             dz10             = _mm_sub_ps(iz1,jz0);
1137             dx20             = _mm_sub_ps(ix2,jx0);
1138             dy20             = _mm_sub_ps(iy2,jy0);
1139             dz20             = _mm_sub_ps(iz2,jz0);
1140             dx30             = _mm_sub_ps(ix3,jx0);
1141             dy30             = _mm_sub_ps(iy3,jy0);
1142             dz30             = _mm_sub_ps(iz3,jz0);
1143
1144             /* Calculate squared distance and things based on it */
1145             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1146             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1147             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1148             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1149
1150             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1151             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1152             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1153             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1154
1155             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1156             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1157             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1158
1159             /* Load parameters for j particles */
1160             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1161                                                               charge+jnrC+0,charge+jnrD+0);
1162             vdwjidx0A        = 2*vdwtype[jnrA+0];
1163             vdwjidx0B        = 2*vdwtype[jnrB+0];
1164             vdwjidx0C        = 2*vdwtype[jnrC+0];
1165             vdwjidx0D        = 2*vdwtype[jnrD+0];
1166
1167             /**************************
1168              * CALCULATE INTERACTIONS *
1169              **************************/
1170
1171             r00              = _mm_mul_ps(rsq00,rinv00);
1172             r00              = _mm_andnot_ps(dummy_mask,r00);
1173
1174             /* Compute parameters for interactions between i and j atoms */
1175             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1176                                          vdwparam+vdwioffset0+vdwjidx0B,
1177                                          vdwparam+vdwioffset0+vdwjidx0C,
1178                                          vdwparam+vdwioffset0+vdwjidx0D,
1179                                          &c6_00,&c12_00);
1180
1181             /* Calculate table index by multiplying r with table scale and truncate to integer */
1182             rt               = _mm_mul_ps(r00,vftabscale);
1183             vfitab           = _mm_cvttps_epi32(rt);
1184             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1185             vfitab           = _mm_slli_epi32(vfitab,3);
1186
1187             /* CUBIC SPLINE TABLE DISPERSION */
1188             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1189             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1190             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1191             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1192             _MM_TRANSPOSE4_PS(Y,F,G,H);
1193             Heps             = _mm_mul_ps(vfeps,H);
1194             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1195             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1196             fvdw6            = _mm_mul_ps(c6_00,FF);
1197
1198             /* CUBIC SPLINE TABLE REPULSION */
1199             vfitab           = _mm_add_epi32(vfitab,ifour);
1200             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1201             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1202             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1203             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1204             _MM_TRANSPOSE4_PS(Y,F,G,H);
1205             Heps             = _mm_mul_ps(vfeps,H);
1206             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1207             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1208             fvdw12           = _mm_mul_ps(c12_00,FF);
1209             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1210
1211             fscal            = fvdw;
1212
1213             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1214
1215             /* Calculate temporary vectorial force */
1216             tx               = _mm_mul_ps(fscal,dx00);
1217             ty               = _mm_mul_ps(fscal,dy00);
1218             tz               = _mm_mul_ps(fscal,dz00);
1219
1220             /* Update vectorial force */
1221             fix0             = _mm_add_ps(fix0,tx);
1222             fiy0             = _mm_add_ps(fiy0,ty);
1223             fiz0             = _mm_add_ps(fiz0,tz);
1224
1225             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1226                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1227                                                    tx,ty,tz);
1228
1229             /**************************
1230              * CALCULATE INTERACTIONS *
1231              **************************/
1232
1233             if (gmx_mm_any_lt(rsq10,rcutoff2))
1234             {
1235
1236             /* Compute parameters for interactions between i and j atoms */
1237             qq10             = _mm_mul_ps(iq1,jq0);
1238
1239             /* REACTION-FIELD ELECTROSTATICS */
1240             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1241
1242             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1243
1244             fscal            = felec;
1245
1246             fscal            = _mm_and_ps(fscal,cutoff_mask);
1247
1248             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1249
1250             /* Calculate temporary vectorial force */
1251             tx               = _mm_mul_ps(fscal,dx10);
1252             ty               = _mm_mul_ps(fscal,dy10);
1253             tz               = _mm_mul_ps(fscal,dz10);
1254
1255             /* Update vectorial force */
1256             fix1             = _mm_add_ps(fix1,tx);
1257             fiy1             = _mm_add_ps(fiy1,ty);
1258             fiz1             = _mm_add_ps(fiz1,tz);
1259
1260             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1261                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1262                                                    tx,ty,tz);
1263
1264             }
1265
1266             /**************************
1267              * CALCULATE INTERACTIONS *
1268              **************************/
1269
1270             if (gmx_mm_any_lt(rsq20,rcutoff2))
1271             {
1272
1273             /* Compute parameters for interactions between i and j atoms */
1274             qq20             = _mm_mul_ps(iq2,jq0);
1275
1276             /* REACTION-FIELD ELECTROSTATICS */
1277             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1278
1279             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1280
1281             fscal            = felec;
1282
1283             fscal            = _mm_and_ps(fscal,cutoff_mask);
1284
1285             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1286
1287             /* Calculate temporary vectorial force */
1288             tx               = _mm_mul_ps(fscal,dx20);
1289             ty               = _mm_mul_ps(fscal,dy20);
1290             tz               = _mm_mul_ps(fscal,dz20);
1291
1292             /* Update vectorial force */
1293             fix2             = _mm_add_ps(fix2,tx);
1294             fiy2             = _mm_add_ps(fiy2,ty);
1295             fiz2             = _mm_add_ps(fiz2,tz);
1296
1297             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1298                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1299                                                    tx,ty,tz);
1300
1301             }
1302
1303             /**************************
1304              * CALCULATE INTERACTIONS *
1305              **************************/
1306
1307             if (gmx_mm_any_lt(rsq30,rcutoff2))
1308             {
1309
1310             /* Compute parameters for interactions between i and j atoms */
1311             qq30             = _mm_mul_ps(iq3,jq0);
1312
1313             /* REACTION-FIELD ELECTROSTATICS */
1314             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1315
1316             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1317
1318             fscal            = felec;
1319
1320             fscal            = _mm_and_ps(fscal,cutoff_mask);
1321
1322             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1323
1324             /* Calculate temporary vectorial force */
1325             tx               = _mm_mul_ps(fscal,dx30);
1326             ty               = _mm_mul_ps(fscal,dy30);
1327             tz               = _mm_mul_ps(fscal,dz30);
1328
1329             /* Update vectorial force */
1330             fix3             = _mm_add_ps(fix3,tx);
1331             fiy3             = _mm_add_ps(fiy3,ty);
1332             fiz3             = _mm_add_ps(fiz3,tz);
1333
1334             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1335                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1336                                                    tx,ty,tz);
1337
1338             }
1339
1340             /* Inner loop uses 139 flops */
1341         }
1342
1343         /* End of innermost loop */
1344
1345         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1346                                               f+i_coord_offset,fshift+i_shift_offset);
1347
1348         /* Increment number of inner iterations */
1349         inneriter                  += j_index_end - j_index_start;
1350
1351         /* Outer loop uses 36 flops */
1352     }
1353
1354     /* Increment number of outer iterations */
1355     outeriter        += nri;
1356
1357     /* Update outer/inner flops */
1358
1359     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*36 + inneriter*139);
1360 }