src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwNone_GeomW3P1_avx_128_fma_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 avx_128_fma_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_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwNone_GeomW3P1_VF_avx_128_fma_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 AVX_128, 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           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  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,twovfeps,vftabscale,Y,F,G,H,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     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 124     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 125     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 126
 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,
 158                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 159
 160         fix0             = _mm_setzero_ps();
 161         fiy0             = _mm_setzero_ps();
 162         fiz0             = _mm_setzero_ps();
 163         fix1             = _mm_setzero_ps();
 164         fiy1             = _mm_setzero_ps();
 165         fiz1             = _mm_setzero_ps();
 166         fix2             = _mm_setzero_ps();
 167         fiy2             = _mm_setzero_ps();
 168         fiz2             = _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             dx00             = _mm_sub_ps(ix0,jx0);
 194             dy00             = _mm_sub_ps(iy0,jy0);
 195             dz00             = _mm_sub_ps(iz0,jz0);
 196             dx10             = _mm_sub_ps(ix1,jx0);
 197             dy10             = _mm_sub_ps(iy1,jy0);
 198             dz10             = _mm_sub_ps(iz1,jz0);
 199             dx20             = _mm_sub_ps(ix2,jx0);
 200             dy20             = _mm_sub_ps(iy2,jy0);
 201             dz20             = _mm_sub_ps(iz2,jz0);
 202
 203             /* Calculate squared distance and things based on it */
 204             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 205             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 206             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 207
 208             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 209             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 210             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 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             r00              = _mm_mul_ps(rsq00,rinv00);
 225
 226             /* Compute parameters for interactions between i and j atoms */
 227             qq00             = _mm_mul_ps(iq0,jq0);
 228
 229             /* Calculate table index by multiplying r with table scale and truncate to integer */
 230             rt               = _mm_mul_ps(r00,vftabscale);
 231             vfitab           = _mm_cvttps_epi32(rt);
 232 #ifdef __XOP__
 233             vfeps            = _mm_frcz_ps(rt);
 234 #else
 235             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 236 #endif
 237             twovfeps         = _mm_add_ps(vfeps,vfeps);
 238             vfitab           = _mm_slli_epi32(vfitab,2);
 239
 240             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 241             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 242             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 243             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 244             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 245             _MM_TRANSPOSE4_PS(Y,F,G,H);
 246             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 247             VV               = _mm_macc_ps(vfeps,Fp,Y);
 248             velec            = _mm_mul_ps(qq00,VV);
 249             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 250             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 251
 252             /* Update potential sum for this i atom from the interaction with this j atom. */
 253             velecsum         = _mm_add_ps(velecsum,velec);
 254
 255             fscal            = felec;
 256
 257              /* Update vectorial force */
 258             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 259             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 260             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 261
 262             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 263             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 264             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             r10              = _mm_mul_ps(rsq10,rinv10);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq10             = _mm_mul_ps(iq1,jq0);
 274
 275             /* Calculate table index by multiplying r with table scale and truncate to integer */
 276             rt               = _mm_mul_ps(r10,vftabscale);
 277             vfitab           = _mm_cvttps_epi32(rt);
 278 #ifdef __XOP__
 279             vfeps            = _mm_frcz_ps(rt);
 280 #else
 281             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 282 #endif
 283             twovfeps         = _mm_add_ps(vfeps,vfeps);
 284             vfitab           = _mm_slli_epi32(vfitab,2);
 285
 286             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 287             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 288             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 289             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 290             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 291             _MM_TRANSPOSE4_PS(Y,F,G,H);
 292             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 293             VV               = _mm_macc_ps(vfeps,Fp,Y);
 294             velec            = _mm_mul_ps(qq10,VV);
 295             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 296             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velecsum         = _mm_add_ps(velecsum,velec);
 300
 301             fscal            = felec;
 302
 303              /* Update vectorial force */
 304             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 305             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 306             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 307
 308             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 309             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 310             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             r20              = _mm_mul_ps(rsq20,rinv20);
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq20             = _mm_mul_ps(iq2,jq0);
 320
 321             /* Calculate table index by multiplying r with table scale and truncate to integer */
 322             rt               = _mm_mul_ps(r20,vftabscale);
 323             vfitab           = _mm_cvttps_epi32(rt);
 324 #ifdef __XOP__
 325             vfeps            = _mm_frcz_ps(rt);
 326 #else
 327             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 328 #endif
 329             twovfeps         = _mm_add_ps(vfeps,vfeps);
 330             vfitab           = _mm_slli_epi32(vfitab,2);
 331
 332             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 333             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 334             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 335             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 336             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 337             _MM_TRANSPOSE4_PS(Y,F,G,H);
 338             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 339             VV               = _mm_macc_ps(vfeps,Fp,Y);
 340             velec            = _mm_mul_ps(qq20,VV);
 341             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 342             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm_add_ps(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349              /* Update vectorial force */
 350             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 351             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 352             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 353
 354             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 355             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 356             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 357
 358             fjptrA             = f+j_coord_offsetA;
 359             fjptrB             = f+j_coord_offsetB;
 360             fjptrC             = f+j_coord_offsetC;
 361             fjptrD             = f+j_coord_offsetD;
 362
 363             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 364
 365             /* Inner loop uses 138 flops */
 366         }
 367
 368         if(jidx<j_index_end)
 369         {
 370
 371             /* Get j neighbor index, and coordinate index */
 372             jnrlistA         = jjnr[jidx];
 373             jnrlistB         = jjnr[jidx+1];
 374             jnrlistC         = jjnr[jidx+2];
 375             jnrlistD         = jjnr[jidx+3];
 376             /* Sign of each element will be negative for non-real atoms.
 377              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 378              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 379              */
 380             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 381             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 382             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 383             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 384             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 385             j_coord_offsetA  = DIM*jnrA;
 386             j_coord_offsetB  = DIM*jnrB;
 387             j_coord_offsetC  = DIM*jnrC;
 388             j_coord_offsetD  = DIM*jnrD;
 389
 390             /* load j atom coordinates */
 391             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 392                                               x+j_coord_offsetC,x+j_coord_offsetD,
 393                                               &jx0,&jy0,&jz0);
 394
 395             /* Calculate displacement vector */
 396             dx00             = _mm_sub_ps(ix0,jx0);
 397             dy00             = _mm_sub_ps(iy0,jy0);
 398             dz00             = _mm_sub_ps(iz0,jz0);
 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
 406             /* Calculate squared distance and things based on it */
 407             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 408             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 409             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 410
 411             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 412             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 413             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 414
 415             /* Load parameters for j particles */
 416             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 417                                                               charge+jnrC+0,charge+jnrD+0);
 418
 419             fjx0             = _mm_setzero_ps();
 420             fjy0             = _mm_setzero_ps();
 421             fjz0             = _mm_setzero_ps();
 422
 423             /**************************
 424              * CALCULATE INTERACTIONS *
 425              **************************/
 426
 427             r00              = _mm_mul_ps(rsq00,rinv00);
 428             r00              = _mm_andnot_ps(dummy_mask,r00);
 429
 430             /* Compute parameters for interactions between i and j atoms */
 431             qq00             = _mm_mul_ps(iq0,jq0);
 432
 433             /* Calculate table index by multiplying r with table scale and truncate to integer */
 434             rt               = _mm_mul_ps(r00,vftabscale);
 435             vfitab           = _mm_cvttps_epi32(rt);
 436 #ifdef __XOP__
 437             vfeps            = _mm_frcz_ps(rt);
 438 #else
 439             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 440 #endif
 441             twovfeps         = _mm_add_ps(vfeps,vfeps);
 442             vfitab           = _mm_slli_epi32(vfitab,2);
 443
 444             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 445             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 446             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 447             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 448             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 449             _MM_TRANSPOSE4_PS(Y,F,G,H);
 450             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 451             VV               = _mm_macc_ps(vfeps,Fp,Y);
 452             velec            = _mm_mul_ps(qq00,VV);
 453             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 454             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velec            = _mm_andnot_ps(dummy_mask,velec);
 458             velecsum         = _mm_add_ps(velecsum,velec);
 459
 460             fscal            = felec;
 461
 462             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 463
 464              /* Update vectorial force */
 465             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 466             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 467             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 468
 469             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 470             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 471             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 472
 473             /**************************
 474              * CALCULATE INTERACTIONS *
 475              **************************/
 476
 477             r10              = _mm_mul_ps(rsq10,rinv10);
 478             r10              = _mm_andnot_ps(dummy_mask,r10);
 479
 480             /* Compute parameters for interactions between i and j atoms */
 481             qq10             = _mm_mul_ps(iq1,jq0);
 482
 483             /* Calculate table index by multiplying r with table scale and truncate to integer */
 484             rt               = _mm_mul_ps(r10,vftabscale);
 485             vfitab           = _mm_cvttps_epi32(rt);
 486 #ifdef __XOP__
 487             vfeps            = _mm_frcz_ps(rt);
 488 #else
 489             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 490 #endif
 491             twovfeps         = _mm_add_ps(vfeps,vfeps);
 492             vfitab           = _mm_slli_epi32(vfitab,2);
 493
 494             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 495             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 496             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 497             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 498             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 499             _MM_TRANSPOSE4_PS(Y,F,G,H);
 500             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 501             VV               = _mm_macc_ps(vfeps,Fp,Y);
 502             velec            = _mm_mul_ps(qq10,VV);
 503             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 504             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 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              /* Update vectorial force */
 515             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 516             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 517             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 518
 519             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 520             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 521             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             r20              = _mm_mul_ps(rsq20,rinv20);
 528             r20              = _mm_andnot_ps(dummy_mask,r20);
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq20             = _mm_mul_ps(iq2,jq0);
 532
 533             /* Calculate table index by multiplying r with table scale and truncate to integer */
 534             rt               = _mm_mul_ps(r20,vftabscale);
 535             vfitab           = _mm_cvttps_epi32(rt);
 536 #ifdef __XOP__
 537             vfeps            = _mm_frcz_ps(rt);
 538 #else
 539             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 540 #endif
 541             twovfeps         = _mm_add_ps(vfeps,vfeps);
 542             vfitab           = _mm_slli_epi32(vfitab,2);
 543
 544             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 545             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 546             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 547             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 548             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 549             _MM_TRANSPOSE4_PS(Y,F,G,H);
 550             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 551             VV               = _mm_macc_ps(vfeps,Fp,Y);
 552             velec            = _mm_mul_ps(qq20,VV);
 553             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 554             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velec            = _mm_andnot_ps(dummy_mask,velec);
 558             velecsum         = _mm_add_ps(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 563
 564              /* Update vectorial force */
 565             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 566             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 567             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 568
 569             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 570             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 571             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 572
 573             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 574             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 575             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 576             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 577
 578             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 579
 580             /* Inner loop uses 141 flops */
 581         }
 582
 583         /* End of innermost loop */
 584
 585         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 586                                               f+i_coord_offset,fshift+i_shift_offset);
 587
 588         ggid                        = gid[iidx];
 589         /* Update potential energies */
 590         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 591
 592         /* Increment number of inner iterations */
 593         inneriter                  += j_index_end - j_index_start;
 594
 595         /* Outer loop uses 19 flops */
 596     }
 597
 598     /* Increment number of outer iterations */
 599     outeriter        += nri;
 600
 601     /* Update outer/inner flops */
 602
 603     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*141);
 604 }
 605 /*
 606  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_avx_128_fma_single
 607  * Electrostatics interaction: CubicSplineTable
 608  * VdW interaction:            None
 609  * Geometry:                   Water3-Particle
 610  * Calculate force/pot:        Force
 611  */
 612 void
 613 nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_avx_128_fma_single
 614                     (t_nblist                    * gmx_restrict       nlist,
 615                      rvec                        * gmx_restrict          xx,
 616                      rvec                        * gmx_restrict          ff,
 617                      t_forcerec                  * gmx_restrict          fr,
 618                      t_mdatoms                   * gmx_restrict     mdatoms,
 619                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 620                      t_nrnb                      * gmx_restrict        nrnb)
 621 {
 622     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 623      * just 0 for non-waters.
 624      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 625      * jnr indices corresponding to data put in the four positions in the SIMD register.
 626      */
 627     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 628     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 629     int              jnrA,jnrB,jnrC,jnrD;
 630     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 631     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 632     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 633     real             rcutoff_scalar;
 634     real             *shiftvec,*fshift,*x,*f;
 635     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 636     real             scratch[4*DIM];
 637     __m128           fscal,rcutoff,rcutoff2,jidxall;
 638     int              vdwioffset0;
 639     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 640     int              vdwioffset1;
 641     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 642     int              vdwioffset2;
 643     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 644     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 645     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 646     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 647     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 648     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 649     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 650     real             *charge;
 651     __m128i          vfitab;
 652     __m128i          ifour       = _mm_set1_epi32(4);
 653     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 654     real             *vftab;
 655     __m128           dummy_mask,cutoff_mask;
 656     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 657     __m128           one     = _mm_set1_ps(1.0);
 658     __m128           two     = _mm_set1_ps(2.0);
 659     x                = xx[0];
 660     f                = ff[0];
 661
 662     nri              = nlist->nri;
 663     iinr             = nlist->iinr;
 664     jindex           = nlist->jindex;
 665     jjnr             = nlist->jjnr;
 666     shiftidx         = nlist->shift;
 667     gid              = nlist->gid;
 668     shiftvec         = fr->shift_vec[0];
 669     fshift           = fr->fshift[0];
 670     facel            = _mm_set1_ps(fr->epsfac);
 671     charge           = mdatoms->chargeA;
 672
 673     vftab            = kernel_data->table_elec->data;
 674     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 675
 676     /* Setup water-specific parameters */
 677     inr              = nlist->iinr[0];
 678     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 679     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 680     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 681
 682     /* Avoid stupid compiler warnings */
 683     jnrA = jnrB = jnrC = jnrD = 0;
 684     j_coord_offsetA = 0;
 685     j_coord_offsetB = 0;
 686     j_coord_offsetC = 0;
 687     j_coord_offsetD = 0;
 688
 689     outeriter        = 0;
 690     inneriter        = 0;
 691
 692     for(iidx=0;iidx<4*DIM;iidx++)
 693     {
 694         scratch[iidx] = 0.0;
 695     }
 696
 697     /* Start outer loop over neighborlists */
 698     for(iidx=0; iidx<nri; iidx++)
 699     {
 700         /* Load shift vector for this list */
 701         i_shift_offset   = DIM*shiftidx[iidx];
 702
 703         /* Load limits for loop over neighbors */
 704         j_index_start    = jindex[iidx];
 705         j_index_end      = jindex[iidx+1];
 706
 707         /* Get outer coordinate index */
 708         inr              = iinr[iidx];
 709         i_coord_offset   = DIM*inr;
 710
 711         /* Load i particle coords and add shift vector */
 712         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 713                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 714
 715         fix0             = _mm_setzero_ps();
 716         fiy0             = _mm_setzero_ps();
 717         fiz0             = _mm_setzero_ps();
 718         fix1             = _mm_setzero_ps();
 719         fiy1             = _mm_setzero_ps();
 720         fiz1             = _mm_setzero_ps();
 721         fix2             = _mm_setzero_ps();
 722         fiy2             = _mm_setzero_ps();
 723         fiz2             = _mm_setzero_ps();
 724
 725         /* Start inner kernel loop */
 726         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 727         {
 728
 729             /* Get j neighbor index, and coordinate index */
 730             jnrA             = jjnr[jidx];
 731             jnrB             = jjnr[jidx+1];
 732             jnrC             = jjnr[jidx+2];
 733             jnrD             = jjnr[jidx+3];
 734             j_coord_offsetA  = DIM*jnrA;
 735             j_coord_offsetB  = DIM*jnrB;
 736             j_coord_offsetC  = DIM*jnrC;
 737             j_coord_offsetD  = DIM*jnrD;
 738
 739             /* load j atom coordinates */
 740             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 741                                               x+j_coord_offsetC,x+j_coord_offsetD,
 742                                               &jx0,&jy0,&jz0);
 743
 744             /* Calculate displacement vector */
 745             dx00             = _mm_sub_ps(ix0,jx0);
 746             dy00             = _mm_sub_ps(iy0,jy0);
 747             dz00             = _mm_sub_ps(iz0,jz0);
 748             dx10             = _mm_sub_ps(ix1,jx0);
 749             dy10             = _mm_sub_ps(iy1,jy0);
 750             dz10             = _mm_sub_ps(iz1,jz0);
 751             dx20             = _mm_sub_ps(ix2,jx0);
 752             dy20             = _mm_sub_ps(iy2,jy0);
 753             dz20             = _mm_sub_ps(iz2,jz0);
 754
 755             /* Calculate squared distance and things based on it */
 756             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 757             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 758             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 759
 760             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 761             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 762             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 763
 764             /* Load parameters for j particles */
 765             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 766                                                               charge+jnrC+0,charge+jnrD+0);
 767
 768             fjx0             = _mm_setzero_ps();
 769             fjy0             = _mm_setzero_ps();
 770             fjz0             = _mm_setzero_ps();
 771
 772             /**************************
 773              * CALCULATE INTERACTIONS *
 774              **************************/
 775
 776             r00              = _mm_mul_ps(rsq00,rinv00);
 777
 778             /* Compute parameters for interactions between i and j atoms */
 779             qq00             = _mm_mul_ps(iq0,jq0);
 780
 781             /* Calculate table index by multiplying r with table scale and truncate to integer */
 782             rt               = _mm_mul_ps(r00,vftabscale);
 783             vfitab           = _mm_cvttps_epi32(rt);
 784 #ifdef __XOP__
 785             vfeps            = _mm_frcz_ps(rt);
 786 #else
 787             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 788 #endif
 789             twovfeps         = _mm_add_ps(vfeps,vfeps);
 790             vfitab           = _mm_slli_epi32(vfitab,2);
 791
 792             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 793             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 794             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 795             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 796             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 797             _MM_TRANSPOSE4_PS(Y,F,G,H);
 798             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 799             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 800             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 801
 802             fscal            = felec;
 803
 804              /* Update vectorial force */
 805             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 806             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 807             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 808
 809             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 810             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 811             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 812
 813             /**************************
 814              * CALCULATE INTERACTIONS *
 815              **************************/
 816
 817             r10              = _mm_mul_ps(rsq10,rinv10);
 818
 819             /* Compute parameters for interactions between i and j atoms */
 820             qq10             = _mm_mul_ps(iq1,jq0);
 821
 822             /* Calculate table index by multiplying r with table scale and truncate to integer */
 823             rt               = _mm_mul_ps(r10,vftabscale);
 824             vfitab           = _mm_cvttps_epi32(rt);
 825 #ifdef __XOP__
 826             vfeps            = _mm_frcz_ps(rt);
 827 #else
 828             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 829 #endif
 830             twovfeps         = _mm_add_ps(vfeps,vfeps);
 831             vfitab           = _mm_slli_epi32(vfitab,2);
 832
 833             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 834             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 835             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 836             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 837             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 838             _MM_TRANSPOSE4_PS(Y,F,G,H);
 839             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 840             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 841             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 842
 843             fscal            = felec;
 844
 845              /* Update vectorial force */
 846             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 847             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 848             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 849
 850             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 851             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 852             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 853
 854             /**************************
 855              * CALCULATE INTERACTIONS *
 856              **************************/
 857
 858             r20              = _mm_mul_ps(rsq20,rinv20);
 859
 860             /* Compute parameters for interactions between i and j atoms */
 861             qq20             = _mm_mul_ps(iq2,jq0);
 862
 863             /* Calculate table index by multiplying r with table scale and truncate to integer */
 864             rt               = _mm_mul_ps(r20,vftabscale);
 865             vfitab           = _mm_cvttps_epi32(rt);
 866 #ifdef __XOP__
 867             vfeps            = _mm_frcz_ps(rt);
 868 #else
 869             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 870 #endif
 871             twovfeps         = _mm_add_ps(vfeps,vfeps);
 872             vfitab           = _mm_slli_epi32(vfitab,2);
 873
 874             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 875             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 876             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 877             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 878             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 879             _MM_TRANSPOSE4_PS(Y,F,G,H);
 880             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 881             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 882             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 883
 884             fscal            = felec;
 885
 886              /* Update vectorial force */
 887             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 888             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 889             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 890
 891             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 892             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 893             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 894
 895             fjptrA             = f+j_coord_offsetA;
 896             fjptrB             = f+j_coord_offsetB;
 897             fjptrC             = f+j_coord_offsetC;
 898             fjptrD             = f+j_coord_offsetD;
 899
 900             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 901
 902             /* Inner loop uses 126 flops */
 903         }
 904
 905         if(jidx<j_index_end)
 906         {
 907
 908             /* Get j neighbor index, and coordinate index */
 909             jnrlistA         = jjnr[jidx];
 910             jnrlistB         = jjnr[jidx+1];
 911             jnrlistC         = jjnr[jidx+2];
 912             jnrlistD         = jjnr[jidx+3];
 913             /* Sign of each element will be negative for non-real atoms.
 914              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 915              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 916              */
 917             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 918             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 919             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 920             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 921             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 922             j_coord_offsetA  = DIM*jnrA;
 923             j_coord_offsetB  = DIM*jnrB;
 924             j_coord_offsetC  = DIM*jnrC;
 925             j_coord_offsetD  = DIM*jnrD;
 926
 927             /* load j atom coordinates */
 928             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 929                                               x+j_coord_offsetC,x+j_coord_offsetD,
 930                                               &jx0,&jy0,&jz0);
 931
 932             /* Calculate displacement vector */
 933             dx00             = _mm_sub_ps(ix0,jx0);
 934             dy00             = _mm_sub_ps(iy0,jy0);
 935             dz00             = _mm_sub_ps(iz0,jz0);
 936             dx10             = _mm_sub_ps(ix1,jx0);
 937             dy10             = _mm_sub_ps(iy1,jy0);
 938             dz10             = _mm_sub_ps(iz1,jz0);
 939             dx20             = _mm_sub_ps(ix2,jx0);
 940             dy20             = _mm_sub_ps(iy2,jy0);
 941             dz20             = _mm_sub_ps(iz2,jz0);
 942
 943             /* Calculate squared distance and things based on it */
 944             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 945             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 946             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 947
 948             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 949             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 950             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 951
 952             /* Load parameters for j particles */
 953             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 954                                                               charge+jnrC+0,charge+jnrD+0);
 955
 956             fjx0             = _mm_setzero_ps();
 957             fjy0             = _mm_setzero_ps();
 958             fjz0             = _mm_setzero_ps();
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             r00              = _mm_mul_ps(rsq00,rinv00);
 965             r00              = _mm_andnot_ps(dummy_mask,r00);
 966
 967             /* Compute parameters for interactions between i and j atoms */
 968             qq00             = _mm_mul_ps(iq0,jq0);
 969
 970             /* Calculate table index by multiplying r with table scale and truncate to integer */
 971             rt               = _mm_mul_ps(r00,vftabscale);
 972             vfitab           = _mm_cvttps_epi32(rt);
 973 #ifdef __XOP__
 974             vfeps            = _mm_frcz_ps(rt);
 975 #else
 976             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 977 #endif
 978             twovfeps         = _mm_add_ps(vfeps,vfeps);
 979             vfitab           = _mm_slli_epi32(vfitab,2);
 980
 981             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 982             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 983             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 984             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 985             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 986             _MM_TRANSPOSE4_PS(Y,F,G,H);
 987             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 988             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 989             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 990
 991             fscal            = felec;
 992
 993             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 994
 995              /* Update vectorial force */
 996             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 997             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 998             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 999
1000             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1001             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1002             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             r10              = _mm_mul_ps(rsq10,rinv10);
1009             r10              = _mm_andnot_ps(dummy_mask,r10);
1010
1011             /* Compute parameters for interactions between i and j atoms */
1012             qq10             = _mm_mul_ps(iq1,jq0);
1013
1014             /* Calculate table index by multiplying r with table scale and truncate to integer */
1015             rt               = _mm_mul_ps(r10,vftabscale);
1016             vfitab           = _mm_cvttps_epi32(rt);
1017 #ifdef __XOP__
1018             vfeps            = _mm_frcz_ps(rt);
1019 #else
1020             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1021 #endif
1022             twovfeps         = _mm_add_ps(vfeps,vfeps);
1023             vfitab           = _mm_slli_epi32(vfitab,2);
1024
1025             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1026             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1027             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1028             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1029             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1030             _MM_TRANSPOSE4_PS(Y,F,G,H);
1031             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1032             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1033             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1034
1035             fscal            = felec;
1036
1037             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1038
1039              /* Update vectorial force */
1040             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1041             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1042             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1043
1044             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1045             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1046             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1047
1048             /**************************
1049              * CALCULATE INTERACTIONS *
1050              **************************/
1051
1052             r20              = _mm_mul_ps(rsq20,rinv20);
1053             r20              = _mm_andnot_ps(dummy_mask,r20);
1054
1055             /* Compute parameters for interactions between i and j atoms */
1056             qq20             = _mm_mul_ps(iq2,jq0);
1057
1058             /* Calculate table index by multiplying r with table scale and truncate to integer */
1059             rt               = _mm_mul_ps(r20,vftabscale);
1060             vfitab           = _mm_cvttps_epi32(rt);
1061 #ifdef __XOP__
1062             vfeps            = _mm_frcz_ps(rt);
1063 #else
1064             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1065 #endif
1066             twovfeps         = _mm_add_ps(vfeps,vfeps);
1067             vfitab           = _mm_slli_epi32(vfitab,2);
1068
1069             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1070             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1071             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1072             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1073             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1074             _MM_TRANSPOSE4_PS(Y,F,G,H);
1075             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1076             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1077             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1078
1079             fscal            = felec;
1080
1081             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1082
1083              /* Update vectorial force */
1084             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1085             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1086             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1087
1088             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1089             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1090             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1091
1092             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1093             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1094             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1095             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1096
1097             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1098
1099             /* Inner loop uses 129 flops */
1100         }
1101
1102         /* End of innermost loop */
1103
1104         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1105                                               f+i_coord_offset,fshift+i_shift_offset);
1106
1107         /* Increment number of inner iterations */
1108         inneriter                  += j_index_end - j_index_start;
1109
1110         /* Outer loop uses 18 flops */
1111     }
1112
1113     /* Increment number of outer iterations */
1114     outeriter        += nri;
1115
1116     /* Update outer/inner flops */
1117
1118     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*129);
1119 }