src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwNone_GeomW4P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_sse4_1_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            None
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_sse4_1_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset1;
  84     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  85     int              vdwioffset2;
  86     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  87     int              vdwioffset3;
  88     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     __m128i          vfitab;
  97     __m128i          ifour       = _mm_set1_epi32(4);
  98     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  99     real             *vftab;
 100     __m128           dummy_mask,cutoff_mask;
 101     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 102     __m128           one     = _mm_set1_ps(1.0);
 103     __m128           two     = _mm_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_ps(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117
 118     vftab            = kernel_data->table_elec->data;
 119     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 124     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 125     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = jnrC = jnrD = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131     j_coord_offsetC = 0;
 132     j_coord_offsetD = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     for(iidx=0;iidx<4*DIM;iidx++)
 138     {
 139         scratch[iidx] = 0.0;
 140     }
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 158                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 159
 160         fix1             = _mm_setzero_ps();
 161         fiy1             = _mm_setzero_ps();
 162         fiz1             = _mm_setzero_ps();
 163         fix2             = _mm_setzero_ps();
 164         fiy2             = _mm_setzero_ps();
 165         fiz2             = _mm_setzero_ps();
 166         fix3             = _mm_setzero_ps();
 167         fiy3             = _mm_setzero_ps();
 168         fiz3             = _mm_setzero_ps();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184             j_coord_offsetC  = DIM*jnrC;
 185             j_coord_offsetD  = DIM*jnrD;
 186
 187             /* load j atom coordinates */
 188             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                               x+j_coord_offsetC,x+j_coord_offsetD,
 190                                               &jx0,&jy0,&jz0);
 191
 192             /* Calculate displacement vector */
 193             dx10             = _mm_sub_ps(ix1,jx0);
 194             dy10             = _mm_sub_ps(iy1,jy0);
 195             dz10             = _mm_sub_ps(iz1,jz0);
 196             dx20             = _mm_sub_ps(ix2,jx0);
 197             dy20             = _mm_sub_ps(iy2,jy0);
 198             dz20             = _mm_sub_ps(iz2,jz0);
 199             dx30             = _mm_sub_ps(ix3,jx0);
 200             dy30             = _mm_sub_ps(iy3,jy0);
 201             dz30             = _mm_sub_ps(iz3,jz0);
 202
 203             /* Calculate squared distance and things based on it */
 204             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 205             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 206             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 207
 208             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 209             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 210             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 214                                                               charge+jnrC+0,charge+jnrD+0);
 215
 216             fjx0             = _mm_setzero_ps();
 217             fjy0             = _mm_setzero_ps();
 218             fjz0             = _mm_setzero_ps();
 219
 220             /**************************
 221              * CALCULATE INTERACTIONS *
 222              **************************/
 223
 224             r10              = _mm_mul_ps(rsq10,rinv10);
 225
 226             /* Compute parameters for interactions between i and j atoms */
 227             qq10             = _mm_mul_ps(iq1,jq0);
 228
 229             /* Calculate table index by multiplying r with table scale and truncate to integer */
 230             rt               = _mm_mul_ps(r10,vftabscale);
 231             vfitab           = _mm_cvttps_epi32(rt);
 232             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 233             vfitab           = _mm_slli_epi32(vfitab,2);
 234
 235             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 236             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 237             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 238             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 239             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 240             _MM_TRANSPOSE4_PS(Y,F,G,H);
 241             Heps             = _mm_mul_ps(vfeps,H);
 242             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 243             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 244             velec            = _mm_mul_ps(qq10,VV);
 245             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 246             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 247
 248             /* Update potential sum for this i atom from the interaction with this j atom. */
 249             velecsum         = _mm_add_ps(velecsum,velec);
 250
 251             fscal            = felec;
 252
 253             /* Calculate temporary vectorial force */
 254             tx               = _mm_mul_ps(fscal,dx10);
 255             ty               = _mm_mul_ps(fscal,dy10);
 256             tz               = _mm_mul_ps(fscal,dz10);
 257
 258             /* Update vectorial force */
 259             fix1             = _mm_add_ps(fix1,tx);
 260             fiy1             = _mm_add_ps(fiy1,ty);
 261             fiz1             = _mm_add_ps(fiz1,tz);
 262
 263             fjx0             = _mm_add_ps(fjx0,tx);
 264             fjy0             = _mm_add_ps(fjy0,ty);
 265             fjz0             = _mm_add_ps(fjz0,tz);
 266
 267             /**************************
 268              * CALCULATE INTERACTIONS *
 269              **************************/
 270
 271             r20              = _mm_mul_ps(rsq20,rinv20);
 272
 273             /* Compute parameters for interactions between i and j atoms */
 274             qq20             = _mm_mul_ps(iq2,jq0);
 275
 276             /* Calculate table index by multiplying r with table scale and truncate to integer */
 277             rt               = _mm_mul_ps(r20,vftabscale);
 278             vfitab           = _mm_cvttps_epi32(rt);
 279             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 280             vfitab           = _mm_slli_epi32(vfitab,2);
 281
 282             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 283             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 284             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 285             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 286             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 287             _MM_TRANSPOSE4_PS(Y,F,G,H);
 288             Heps             = _mm_mul_ps(vfeps,H);
 289             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 290             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 291             velec            = _mm_mul_ps(qq20,VV);
 292             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 293             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velecsum         = _mm_add_ps(velecsum,velec);
 297
 298             fscal            = felec;
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm_mul_ps(fscal,dx20);
 302             ty               = _mm_mul_ps(fscal,dy20);
 303             tz               = _mm_mul_ps(fscal,dz20);
 304
 305             /* Update vectorial force */
 306             fix2             = _mm_add_ps(fix2,tx);
 307             fiy2             = _mm_add_ps(fiy2,ty);
 308             fiz2             = _mm_add_ps(fiz2,tz);
 309
 310             fjx0             = _mm_add_ps(fjx0,tx);
 311             fjy0             = _mm_add_ps(fjy0,ty);
 312             fjz0             = _mm_add_ps(fjz0,tz);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r30              = _mm_mul_ps(rsq30,rinv30);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq30             = _mm_mul_ps(iq3,jq0);
 322
 323             /* Calculate table index by multiplying r with table scale and truncate to integer */
 324             rt               = _mm_mul_ps(r30,vftabscale);
 325             vfitab           = _mm_cvttps_epi32(rt);
 326             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 327             vfitab           = _mm_slli_epi32(vfitab,2);
 328
 329             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 330             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 331             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 332             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 333             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 334             _MM_TRANSPOSE4_PS(Y,F,G,H);
 335             Heps             = _mm_mul_ps(vfeps,H);
 336             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 337             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 338             velec            = _mm_mul_ps(qq30,VV);
 339             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 340             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velecsum         = _mm_add_ps(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             /* Calculate temporary vectorial force */
 348             tx               = _mm_mul_ps(fscal,dx30);
 349             ty               = _mm_mul_ps(fscal,dy30);
 350             tz               = _mm_mul_ps(fscal,dz30);
 351
 352             /* Update vectorial force */
 353             fix3             = _mm_add_ps(fix3,tx);
 354             fiy3             = _mm_add_ps(fiy3,ty);
 355             fiz3             = _mm_add_ps(fiz3,tz);
 356
 357             fjx0             = _mm_add_ps(fjx0,tx);
 358             fjy0             = _mm_add_ps(fjy0,ty);
 359             fjz0             = _mm_add_ps(fjz0,tz);
 360
 361             fjptrA             = f+j_coord_offsetA;
 362             fjptrB             = f+j_coord_offsetB;
 363             fjptrC             = f+j_coord_offsetC;
 364             fjptrD             = f+j_coord_offsetD;
 365
 366             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 367
 368             /* Inner loop uses 129 flops */
 369         }
 370
 371         if(jidx<j_index_end)
 372         {
 373
 374             /* Get j neighbor index, and coordinate index */
 375             jnrlistA         = jjnr[jidx];
 376             jnrlistB         = jjnr[jidx+1];
 377             jnrlistC         = jjnr[jidx+2];
 378             jnrlistD         = jjnr[jidx+3];
 379             /* Sign of each element will be negative for non-real atoms.
 380              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 381              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 382              */
 383             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 384             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 385             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 386             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 387             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 388             j_coord_offsetA  = DIM*jnrA;
 389             j_coord_offsetB  = DIM*jnrB;
 390             j_coord_offsetC  = DIM*jnrC;
 391             j_coord_offsetD  = DIM*jnrD;
 392
 393             /* load j atom coordinates */
 394             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 395                                               x+j_coord_offsetC,x+j_coord_offsetD,
 396                                               &jx0,&jy0,&jz0);
 397
 398             /* Calculate displacement vector */
 399             dx10             = _mm_sub_ps(ix1,jx0);
 400             dy10             = _mm_sub_ps(iy1,jy0);
 401             dz10             = _mm_sub_ps(iz1,jz0);
 402             dx20             = _mm_sub_ps(ix2,jx0);
 403             dy20             = _mm_sub_ps(iy2,jy0);
 404             dz20             = _mm_sub_ps(iz2,jz0);
 405             dx30             = _mm_sub_ps(ix3,jx0);
 406             dy30             = _mm_sub_ps(iy3,jy0);
 407             dz30             = _mm_sub_ps(iz3,jz0);
 408
 409             /* Calculate squared distance and things based on it */
 410             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 411             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 412             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 413
 414             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 415             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 416             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 417
 418             /* Load parameters for j particles */
 419             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 420                                                               charge+jnrC+0,charge+jnrD+0);
 421
 422             fjx0             = _mm_setzero_ps();
 423             fjy0             = _mm_setzero_ps();
 424             fjz0             = _mm_setzero_ps();
 425
 426             /**************************
 427              * CALCULATE INTERACTIONS *
 428              **************************/
 429
 430             r10              = _mm_mul_ps(rsq10,rinv10);
 431             r10              = _mm_andnot_ps(dummy_mask,r10);
 432
 433             /* Compute parameters for interactions between i and j atoms */
 434             qq10             = _mm_mul_ps(iq1,jq0);
 435
 436             /* Calculate table index by multiplying r with table scale and truncate to integer */
 437             rt               = _mm_mul_ps(r10,vftabscale);
 438             vfitab           = _mm_cvttps_epi32(rt);
 439             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 440             vfitab           = _mm_slli_epi32(vfitab,2);
 441
 442             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 443             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 444             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 445             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 446             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 447             _MM_TRANSPOSE4_PS(Y,F,G,H);
 448             Heps             = _mm_mul_ps(vfeps,H);
 449             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 450             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 451             velec            = _mm_mul_ps(qq10,VV);
 452             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 453             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 454
 455             /* Update potential sum for this i atom from the interaction with this j atom. */
 456             velec            = _mm_andnot_ps(dummy_mask,velec);
 457             velecsum         = _mm_add_ps(velecsum,velec);
 458
 459             fscal            = felec;
 460
 461             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 462
 463             /* Calculate temporary vectorial force */
 464             tx               = _mm_mul_ps(fscal,dx10);
 465             ty               = _mm_mul_ps(fscal,dy10);
 466             tz               = _mm_mul_ps(fscal,dz10);
 467
 468             /* Update vectorial force */
 469             fix1             = _mm_add_ps(fix1,tx);
 470             fiy1             = _mm_add_ps(fiy1,ty);
 471             fiz1             = _mm_add_ps(fiz1,tz);
 472
 473             fjx0             = _mm_add_ps(fjx0,tx);
 474             fjy0             = _mm_add_ps(fjy0,ty);
 475             fjz0             = _mm_add_ps(fjz0,tz);
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             r20              = _mm_mul_ps(rsq20,rinv20);
 482             r20              = _mm_andnot_ps(dummy_mask,r20);
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             qq20             = _mm_mul_ps(iq2,jq0);
 486
 487             /* Calculate table index by multiplying r with table scale and truncate to integer */
 488             rt               = _mm_mul_ps(r20,vftabscale);
 489             vfitab           = _mm_cvttps_epi32(rt);
 490             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 491             vfitab           = _mm_slli_epi32(vfitab,2);
 492
 493             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 494             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 495             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 496             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 497             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 498             _MM_TRANSPOSE4_PS(Y,F,G,H);
 499             Heps             = _mm_mul_ps(vfeps,H);
 500             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 501             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 502             velec            = _mm_mul_ps(qq20,VV);
 503             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 504             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 505
 506             /* Update potential sum for this i atom from the interaction with this j atom. */
 507             velec            = _mm_andnot_ps(dummy_mask,velec);
 508             velecsum         = _mm_add_ps(velecsum,velec);
 509
 510             fscal            = felec;
 511
 512             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 513
 514             /* Calculate temporary vectorial force */
 515             tx               = _mm_mul_ps(fscal,dx20);
 516             ty               = _mm_mul_ps(fscal,dy20);
 517             tz               = _mm_mul_ps(fscal,dz20);
 518
 519             /* Update vectorial force */
 520             fix2             = _mm_add_ps(fix2,tx);
 521             fiy2             = _mm_add_ps(fiy2,ty);
 522             fiz2             = _mm_add_ps(fiz2,tz);
 523
 524             fjx0             = _mm_add_ps(fjx0,tx);
 525             fjy0             = _mm_add_ps(fjy0,ty);
 526             fjz0             = _mm_add_ps(fjz0,tz);
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             r30              = _mm_mul_ps(rsq30,rinv30);
 533             r30              = _mm_andnot_ps(dummy_mask,r30);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq30             = _mm_mul_ps(iq3,jq0);
 537
 538             /* Calculate table index by multiplying r with table scale and truncate to integer */
 539             rt               = _mm_mul_ps(r30,vftabscale);
 540             vfitab           = _mm_cvttps_epi32(rt);
 541             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 542             vfitab           = _mm_slli_epi32(vfitab,2);
 543
 544             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 545             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 546             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 547             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 548             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 549             _MM_TRANSPOSE4_PS(Y,F,G,H);
 550             Heps             = _mm_mul_ps(vfeps,H);
 551             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 552             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 553             velec            = _mm_mul_ps(qq30,VV);
 554             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 555             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             velec            = _mm_andnot_ps(dummy_mask,velec);
 559             velecsum         = _mm_add_ps(velecsum,velec);
 560
 561             fscal            = felec;
 562
 563             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm_mul_ps(fscal,dx30);
 567             ty               = _mm_mul_ps(fscal,dy30);
 568             tz               = _mm_mul_ps(fscal,dz30);
 569
 570             /* Update vectorial force */
 571             fix3             = _mm_add_ps(fix3,tx);
 572             fiy3             = _mm_add_ps(fiy3,ty);
 573             fiz3             = _mm_add_ps(fiz3,tz);
 574
 575             fjx0             = _mm_add_ps(fjx0,tx);
 576             fjy0             = _mm_add_ps(fjy0,ty);
 577             fjz0             = _mm_add_ps(fjz0,tz);
 578
 579             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 580             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 581             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 582             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 583
 584             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 585
 586             /* Inner loop uses 132 flops */
 587         }
 588
 589         /* End of innermost loop */
 590
 591         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 592                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 593
 594         ggid                        = gid[iidx];
 595         /* Update potential energies */
 596         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 597
 598         /* Increment number of inner iterations */
 599         inneriter                  += j_index_end - j_index_start;
 600
 601         /* Outer loop uses 19 flops */
 602     }
 603
 604     /* Increment number of outer iterations */
 605     outeriter        += nri;
 606
 607     /* Update outer/inner flops */
 608
 609     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*132);
 610 }
 611 /*
 612  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse4_1_single
 613  * Electrostatics interaction: CubicSplineTable
 614  * VdW interaction:            None
 615  * Geometry:                   Water4-Particle
 616  * Calculate force/pot:        Force
 617  */
 618 void
 619 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse4_1_single
 620                     (t_nblist                    * gmx_restrict       nlist,
 621                      rvec                        * gmx_restrict          xx,
 622                      rvec                        * gmx_restrict          ff,
 623                      t_forcerec                  * gmx_restrict          fr,
 624                      t_mdatoms                   * gmx_restrict     mdatoms,
 625                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 626                      t_nrnb                      * gmx_restrict        nrnb)
 627 {
 628     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 629      * just 0 for non-waters.
 630      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 631      * jnr indices corresponding to data put in the four positions in the SIMD register.
 632      */
 633     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 634     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 635     int              jnrA,jnrB,jnrC,jnrD;
 636     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 637     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 638     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 639     real             rcutoff_scalar;
 640     real             *shiftvec,*fshift,*x,*f;
 641     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 642     real             scratch[4*DIM];
 643     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 644     int              vdwioffset1;
 645     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 646     int              vdwioffset2;
 647     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 648     int              vdwioffset3;
 649     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 650     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 651     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 652     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 653     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 654     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 655     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 656     real             *charge;
 657     __m128i          vfitab;
 658     __m128i          ifour       = _mm_set1_epi32(4);
 659     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 660     real             *vftab;
 661     __m128           dummy_mask,cutoff_mask;
 662     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 663     __m128           one     = _mm_set1_ps(1.0);
 664     __m128           two     = _mm_set1_ps(2.0);
 665     x                = xx[0];
 666     f                = ff[0];
 667
 668     nri              = nlist->nri;
 669     iinr             = nlist->iinr;
 670     jindex           = nlist->jindex;
 671     jjnr             = nlist->jjnr;
 672     shiftidx         = nlist->shift;
 673     gid              = nlist->gid;
 674     shiftvec         = fr->shift_vec[0];
 675     fshift           = fr->fshift[0];
 676     facel            = _mm_set1_ps(fr->epsfac);
 677     charge           = mdatoms->chargeA;
 678
 679     vftab            = kernel_data->table_elec->data;
 680     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 681
 682     /* Setup water-specific parameters */
 683     inr              = nlist->iinr[0];
 684     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 685     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 686     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 687
 688     /* Avoid stupid compiler warnings */
 689     jnrA = jnrB = jnrC = jnrD = 0;
 690     j_coord_offsetA = 0;
 691     j_coord_offsetB = 0;
 692     j_coord_offsetC = 0;
 693     j_coord_offsetD = 0;
 694
 695     outeriter        = 0;
 696     inneriter        = 0;
 697
 698     for(iidx=0;iidx<4*DIM;iidx++)
 699     {
 700         scratch[iidx] = 0.0;
 701     }
 702
 703     /* Start outer loop over neighborlists */
 704     for(iidx=0; iidx<nri; iidx++)
 705     {
 706         /* Load shift vector for this list */
 707         i_shift_offset   = DIM*shiftidx[iidx];
 708
 709         /* Load limits for loop over neighbors */
 710         j_index_start    = jindex[iidx];
 711         j_index_end      = jindex[iidx+1];
 712
 713         /* Get outer coordinate index */
 714         inr              = iinr[iidx];
 715         i_coord_offset   = DIM*inr;
 716
 717         /* Load i particle coords and add shift vector */
 718         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 719                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 720
 721         fix1             = _mm_setzero_ps();
 722         fiy1             = _mm_setzero_ps();
 723         fiz1             = _mm_setzero_ps();
 724         fix2             = _mm_setzero_ps();
 725         fiy2             = _mm_setzero_ps();
 726         fiz2             = _mm_setzero_ps();
 727         fix3             = _mm_setzero_ps();
 728         fiy3             = _mm_setzero_ps();
 729         fiz3             = _mm_setzero_ps();
 730
 731         /* Start inner kernel loop */
 732         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 733         {
 734
 735             /* Get j neighbor index, and coordinate index */
 736             jnrA             = jjnr[jidx];
 737             jnrB             = jjnr[jidx+1];
 738             jnrC             = jjnr[jidx+2];
 739             jnrD             = jjnr[jidx+3];
 740             j_coord_offsetA  = DIM*jnrA;
 741             j_coord_offsetB  = DIM*jnrB;
 742             j_coord_offsetC  = DIM*jnrC;
 743             j_coord_offsetD  = DIM*jnrD;
 744
 745             /* load j atom coordinates */
 746             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 747                                               x+j_coord_offsetC,x+j_coord_offsetD,
 748                                               &jx0,&jy0,&jz0);
 749
 750             /* Calculate displacement vector */
 751             dx10             = _mm_sub_ps(ix1,jx0);
 752             dy10             = _mm_sub_ps(iy1,jy0);
 753             dz10             = _mm_sub_ps(iz1,jz0);
 754             dx20             = _mm_sub_ps(ix2,jx0);
 755             dy20             = _mm_sub_ps(iy2,jy0);
 756             dz20             = _mm_sub_ps(iz2,jz0);
 757             dx30             = _mm_sub_ps(ix3,jx0);
 758             dy30             = _mm_sub_ps(iy3,jy0);
 759             dz30             = _mm_sub_ps(iz3,jz0);
 760
 761             /* Calculate squared distance and things based on it */
 762             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 763             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 764             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 765
 766             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 767             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 768             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 769
 770             /* Load parameters for j particles */
 771             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 772                                                               charge+jnrC+0,charge+jnrD+0);
 773
 774             fjx0             = _mm_setzero_ps();
 775             fjy0             = _mm_setzero_ps();
 776             fjz0             = _mm_setzero_ps();
 777
 778             /**************************
 779              * CALCULATE INTERACTIONS *
 780              **************************/
 781
 782             r10              = _mm_mul_ps(rsq10,rinv10);
 783
 784             /* Compute parameters for interactions between i and j atoms */
 785             qq10             = _mm_mul_ps(iq1,jq0);
 786
 787             /* Calculate table index by multiplying r with table scale and truncate to integer */
 788             rt               = _mm_mul_ps(r10,vftabscale);
 789             vfitab           = _mm_cvttps_epi32(rt);
 790             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 791             vfitab           = _mm_slli_epi32(vfitab,2);
 792
 793             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 794             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 795             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 796             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 797             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 798             _MM_TRANSPOSE4_PS(Y,F,G,H);
 799             Heps             = _mm_mul_ps(vfeps,H);
 800             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 801             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 802             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 803
 804             fscal            = felec;
 805
 806             /* Calculate temporary vectorial force */
 807             tx               = _mm_mul_ps(fscal,dx10);
 808             ty               = _mm_mul_ps(fscal,dy10);
 809             tz               = _mm_mul_ps(fscal,dz10);
 810
 811             /* Update vectorial force */
 812             fix1             = _mm_add_ps(fix1,tx);
 813             fiy1             = _mm_add_ps(fiy1,ty);
 814             fiz1             = _mm_add_ps(fiz1,tz);
 815
 816             fjx0             = _mm_add_ps(fjx0,tx);
 817             fjy0             = _mm_add_ps(fjy0,ty);
 818             fjz0             = _mm_add_ps(fjz0,tz);
 819
 820             /**************************
 821              * CALCULATE INTERACTIONS *
 822              **************************/
 823
 824             r20              = _mm_mul_ps(rsq20,rinv20);
 825
 826             /* Compute parameters for interactions between i and j atoms */
 827             qq20             = _mm_mul_ps(iq2,jq0);
 828
 829             /* Calculate table index by multiplying r with table scale and truncate to integer */
 830             rt               = _mm_mul_ps(r20,vftabscale);
 831             vfitab           = _mm_cvttps_epi32(rt);
 832             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 833             vfitab           = _mm_slli_epi32(vfitab,2);
 834
 835             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 836             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 837             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 838             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 839             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 840             _MM_TRANSPOSE4_PS(Y,F,G,H);
 841             Heps             = _mm_mul_ps(vfeps,H);
 842             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 843             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 844             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 845
 846             fscal            = felec;
 847
 848             /* Calculate temporary vectorial force */
 849             tx               = _mm_mul_ps(fscal,dx20);
 850             ty               = _mm_mul_ps(fscal,dy20);
 851             tz               = _mm_mul_ps(fscal,dz20);
 852
 853             /* Update vectorial force */
 854             fix2             = _mm_add_ps(fix2,tx);
 855             fiy2             = _mm_add_ps(fiy2,ty);
 856             fiz2             = _mm_add_ps(fiz2,tz);
 857
 858             fjx0             = _mm_add_ps(fjx0,tx);
 859             fjy0             = _mm_add_ps(fjy0,ty);
 860             fjz0             = _mm_add_ps(fjz0,tz);
 861
 862             /**************************
 863              * CALCULATE INTERACTIONS *
 864              **************************/
 865
 866             r30              = _mm_mul_ps(rsq30,rinv30);
 867
 868             /* Compute parameters for interactions between i and j atoms */
 869             qq30             = _mm_mul_ps(iq3,jq0);
 870
 871             /* Calculate table index by multiplying r with table scale and truncate to integer */
 872             rt               = _mm_mul_ps(r30,vftabscale);
 873             vfitab           = _mm_cvttps_epi32(rt);
 874             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 875             vfitab           = _mm_slli_epi32(vfitab,2);
 876
 877             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 878             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 879             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 880             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 881             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 882             _MM_TRANSPOSE4_PS(Y,F,G,H);
 883             Heps             = _mm_mul_ps(vfeps,H);
 884             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 885             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 886             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 887
 888             fscal            = felec;
 889
 890             /* Calculate temporary vectorial force */
 891             tx               = _mm_mul_ps(fscal,dx30);
 892             ty               = _mm_mul_ps(fscal,dy30);
 893             tz               = _mm_mul_ps(fscal,dz30);
 894
 895             /* Update vectorial force */
 896             fix3             = _mm_add_ps(fix3,tx);
 897             fiy3             = _mm_add_ps(fiy3,ty);
 898             fiz3             = _mm_add_ps(fiz3,tz);
 899
 900             fjx0             = _mm_add_ps(fjx0,tx);
 901             fjy0             = _mm_add_ps(fjy0,ty);
 902             fjz0             = _mm_add_ps(fjz0,tz);
 903
 904             fjptrA             = f+j_coord_offsetA;
 905             fjptrB             = f+j_coord_offsetB;
 906             fjptrC             = f+j_coord_offsetC;
 907             fjptrD             = f+j_coord_offsetD;
 908
 909             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 910
 911             /* Inner loop uses 117 flops */
 912         }
 913
 914         if(jidx<j_index_end)
 915         {
 916
 917             /* Get j neighbor index, and coordinate index */
 918             jnrlistA         = jjnr[jidx];
 919             jnrlistB         = jjnr[jidx+1];
 920             jnrlistC         = jjnr[jidx+2];
 921             jnrlistD         = jjnr[jidx+3];
 922             /* Sign of each element will be negative for non-real atoms.
 923              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 924              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 925              */
 926             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 927             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 928             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 929             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 930             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 931             j_coord_offsetA  = DIM*jnrA;
 932             j_coord_offsetB  = DIM*jnrB;
 933             j_coord_offsetC  = DIM*jnrC;
 934             j_coord_offsetD  = DIM*jnrD;
 935
 936             /* load j atom coordinates */
 937             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 938                                               x+j_coord_offsetC,x+j_coord_offsetD,
 939                                               &jx0,&jy0,&jz0);
 940
 941             /* Calculate displacement vector */
 942             dx10             = _mm_sub_ps(ix1,jx0);
 943             dy10             = _mm_sub_ps(iy1,jy0);
 944             dz10             = _mm_sub_ps(iz1,jz0);
 945             dx20             = _mm_sub_ps(ix2,jx0);
 946             dy20             = _mm_sub_ps(iy2,jy0);
 947             dz20             = _mm_sub_ps(iz2,jz0);
 948             dx30             = _mm_sub_ps(ix3,jx0);
 949             dy30             = _mm_sub_ps(iy3,jy0);
 950             dz30             = _mm_sub_ps(iz3,jz0);
 951
 952             /* Calculate squared distance and things based on it */
 953             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 954             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 955             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 956
 957             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 958             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 959             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 960
 961             /* Load parameters for j particles */
 962             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 963                                                               charge+jnrC+0,charge+jnrD+0);
 964
 965             fjx0             = _mm_setzero_ps();
 966             fjy0             = _mm_setzero_ps();
 967             fjz0             = _mm_setzero_ps();
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             r10              = _mm_mul_ps(rsq10,rinv10);
 974             r10              = _mm_andnot_ps(dummy_mask,r10);
 975
 976             /* Compute parameters for interactions between i and j atoms */
 977             qq10             = _mm_mul_ps(iq1,jq0);
 978
 979             /* Calculate table index by multiplying r with table scale and truncate to integer */
 980             rt               = _mm_mul_ps(r10,vftabscale);
 981             vfitab           = _mm_cvttps_epi32(rt);
 982             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 983             vfitab           = _mm_slli_epi32(vfitab,2);
 984
 985             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 986             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 987             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 988             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 989             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 990             _MM_TRANSPOSE4_PS(Y,F,G,H);
 991             Heps             = _mm_mul_ps(vfeps,H);
 992             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 993             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 994             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 995
 996             fscal            = felec;
 997
 998             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 999
1000             /* Calculate temporary vectorial force */
1001             tx               = _mm_mul_ps(fscal,dx10);
1002             ty               = _mm_mul_ps(fscal,dy10);
1003             tz               = _mm_mul_ps(fscal,dz10);
1004
1005             /* Update vectorial force */
1006             fix1             = _mm_add_ps(fix1,tx);
1007             fiy1             = _mm_add_ps(fiy1,ty);
1008             fiz1             = _mm_add_ps(fiz1,tz);
1009
1010             fjx0             = _mm_add_ps(fjx0,tx);
1011             fjy0             = _mm_add_ps(fjy0,ty);
1012             fjz0             = _mm_add_ps(fjz0,tz);
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             r20              = _mm_mul_ps(rsq20,rinv20);
1019             r20              = _mm_andnot_ps(dummy_mask,r20);
1020
1021             /* Compute parameters for interactions between i and j atoms */
1022             qq20             = _mm_mul_ps(iq2,jq0);
1023
1024             /* Calculate table index by multiplying r with table scale and truncate to integer */
1025             rt               = _mm_mul_ps(r20,vftabscale);
1026             vfitab           = _mm_cvttps_epi32(rt);
1027             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1028             vfitab           = _mm_slli_epi32(vfitab,2);
1029
1030             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1031             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1032             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1033             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1034             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1035             _MM_TRANSPOSE4_PS(Y,F,G,H);
1036             Heps             = _mm_mul_ps(vfeps,H);
1037             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1038             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1039             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1040
1041             fscal            = felec;
1042
1043             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1044
1045             /* Calculate temporary vectorial force */
1046             tx               = _mm_mul_ps(fscal,dx20);
1047             ty               = _mm_mul_ps(fscal,dy20);
1048             tz               = _mm_mul_ps(fscal,dz20);
1049
1050             /* Update vectorial force */
1051             fix2             = _mm_add_ps(fix2,tx);
1052             fiy2             = _mm_add_ps(fiy2,ty);
1053             fiz2             = _mm_add_ps(fiz2,tz);
1054
1055             fjx0             = _mm_add_ps(fjx0,tx);
1056             fjy0             = _mm_add_ps(fjy0,ty);
1057             fjz0             = _mm_add_ps(fjz0,tz);
1058
1059             /**************************
1060              * CALCULATE INTERACTIONS *
1061              **************************/
1062
1063             r30              = _mm_mul_ps(rsq30,rinv30);
1064             r30              = _mm_andnot_ps(dummy_mask,r30);
1065
1066             /* Compute parameters for interactions between i and j atoms */
1067             qq30             = _mm_mul_ps(iq3,jq0);
1068
1069             /* Calculate table index by multiplying r with table scale and truncate to integer */
1070             rt               = _mm_mul_ps(r30,vftabscale);
1071             vfitab           = _mm_cvttps_epi32(rt);
1072             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1073             vfitab           = _mm_slli_epi32(vfitab,2);
1074
1075             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1076             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1077             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1078             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1079             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1080             _MM_TRANSPOSE4_PS(Y,F,G,H);
1081             Heps             = _mm_mul_ps(vfeps,H);
1082             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1083             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1084             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1085
1086             fscal            = felec;
1087
1088             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1089
1090             /* Calculate temporary vectorial force */
1091             tx               = _mm_mul_ps(fscal,dx30);
1092             ty               = _mm_mul_ps(fscal,dy30);
1093             tz               = _mm_mul_ps(fscal,dz30);
1094
1095             /* Update vectorial force */
1096             fix3             = _mm_add_ps(fix3,tx);
1097             fiy3             = _mm_add_ps(fiy3,ty);
1098             fiz3             = _mm_add_ps(fiz3,tz);
1099
1100             fjx0             = _mm_add_ps(fjx0,tx);
1101             fjy0             = _mm_add_ps(fjy0,ty);
1102             fjz0             = _mm_add_ps(fjz0,tz);
1103
1104             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1105             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1106             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1107             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1108
1109             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1110
1111             /* Inner loop uses 120 flops */
1112         }
1113
1114         /* End of innermost loop */
1115
1116         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1117                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1118
1119         /* Increment number of inner iterations */
1120         inneriter                  += j_index_end - j_index_start;
1121
1122         /* Outer loop uses 18 flops */
1123     }
1124
1125     /* Increment number of outer iterations */
1126     outeriter        += nri;
1127
1128     /* Update outer/inner flops */
1129
1130     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);
1131 }