src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
  91     real             *vftab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_vdw->data;
 114     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 119     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 120     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 154                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 155
 156         fix0             = _mm_setzero_ps();
 157         fiy0             = _mm_setzero_ps();
 158         fiz0             = _mm_setzero_ps();
 159         fix1             = _mm_setzero_ps();
 160         fiy1             = _mm_setzero_ps();
 161         fiz1             = _mm_setzero_ps();
 162         fix2             = _mm_setzero_ps();
 163         fiy2             = _mm_setzero_ps();
 164         fiz2             = _mm_setzero_ps();
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_ps();
 168         vvdwsum          = _mm_setzero_ps();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             jnrC             = jjnr[jidx+2];
 178             jnrD             = jjnr[jidx+3];
 179             j_coord_offsetA  = DIM*jnrA;
 180             j_coord_offsetB  = DIM*jnrB;
 181             j_coord_offsetC  = DIM*jnrC;
 182             j_coord_offsetD  = DIM*jnrD;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               x+j_coord_offsetC,x+j_coord_offsetD,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_ps(ix0,jx0);
 191             dy00             = _mm_sub_ps(iy0,jy0);
 192             dz00             = _mm_sub_ps(iz0,jz0);
 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
 200             /* Calculate squared distance and things based on it */
 201             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 202             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 203             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 204
 205             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 206             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 207             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 208
 209             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 210             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 211             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 212
 213             /* Load parameters for j particles */
 214             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 215                                                               charge+jnrC+0,charge+jnrD+0);
 216             vdwjidx0A        = 2*vdwtype[jnrA+0];
 217             vdwjidx0B        = 2*vdwtype[jnrB+0];
 218             vdwjidx0C        = 2*vdwtype[jnrC+0];
 219             vdwjidx0D        = 2*vdwtype[jnrD+0];
 220
 221             fjx0             = _mm_setzero_ps();
 222             fjy0             = _mm_setzero_ps();
 223             fjz0             = _mm_setzero_ps();
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             r00              = _mm_mul_ps(rsq00,rinv00);
 230
 231             /* Compute parameters for interactions between i and j atoms */
 232             qq00             = _mm_mul_ps(iq0,jq0);
 233             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 234                                          vdwparam+vdwioffset0+vdwjidx0B,
 235                                          vdwparam+vdwioffset0+vdwjidx0C,
 236                                          vdwparam+vdwioffset0+vdwjidx0D,
 237                                          &c6_00,&c12_00);
 238
 239             /* Calculate table index by multiplying r with table scale and truncate to integer */
 240             rt               = _mm_mul_ps(r00,vftabscale);
 241             vfitab           = _mm_cvttps_epi32(rt);
 242 #ifdef __XOP__
 243             vfeps            = _mm_frcz_ps(rt);
 244 #else
 245             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 246 #endif
 247             twovfeps         = _mm_add_ps(vfeps,vfeps);
 248             vfitab           = _mm_slli_epi32(vfitab,3);
 249
 250             /* COULOMB ELECTROSTATICS */
 251             velec            = _mm_mul_ps(qq00,rinv00);
 252             felec            = _mm_mul_ps(velec,rinvsq00);
 253
 254             /* CUBIC SPLINE TABLE DISPERSION */
 255             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 256             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 257             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 258             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 259             _MM_TRANSPOSE4_PS(Y,F,G,H);
 260             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 261             VV               = _mm_macc_ps(vfeps,Fp,Y);
 262             vvdw6            = _mm_mul_ps(c6_00,VV);
 263             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 264             fvdw6            = _mm_mul_ps(c6_00,FF);
 265
 266             /* CUBIC SPLINE TABLE REPULSION */
 267             vfitab           = _mm_add_epi32(vfitab,ifour);
 268             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 269             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 270             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 271             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 272             _MM_TRANSPOSE4_PS(Y,F,G,H);
 273             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 274             VV               = _mm_macc_ps(vfeps,Fp,Y);
 275             vvdw12           = _mm_mul_ps(c12_00,VV);
 276             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 277             fvdw12           = _mm_mul_ps(c12_00,FF);
 278             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 279             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_ps(velecsum,velec);
 283             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 284
 285             fscal            = _mm_add_ps(felec,fvdw);
 286
 287              /* Update vectorial force */
 288             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 289             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 290             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 291
 292             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 293             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 294             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq10             = _mm_mul_ps(iq1,jq0);
 302
 303             /* COULOMB ELECTROSTATICS */
 304             velec            = _mm_mul_ps(qq10,rinv10);
 305             felec            = _mm_mul_ps(velec,rinvsq10);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velecsum         = _mm_add_ps(velecsum,velec);
 309
 310             fscal            = felec;
 311
 312              /* Update vectorial force */
 313             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 314             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 315             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 316
 317             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 318             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 319             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             /* Compute parameters for interactions between i and j atoms */
 326             qq20             = _mm_mul_ps(iq2,jq0);
 327
 328             /* COULOMB ELECTROSTATICS */
 329             velec            = _mm_mul_ps(qq20,rinv20);
 330             felec            = _mm_mul_ps(velec,rinvsq20);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velecsum         = _mm_add_ps(velecsum,velec);
 334
 335             fscal            = felec;
 336
 337              /* Update vectorial force */
 338             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 339             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 340             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 341
 342             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 343             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 344             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 345
 346             fjptrA             = f+j_coord_offsetA;
 347             fjptrB             = f+j_coord_offsetB;
 348             fjptrC             = f+j_coord_offsetC;
 349             fjptrD             = f+j_coord_offsetD;
 350
 351             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 352
 353             /* Inner loop uses 128 flops */
 354         }
 355
 356         if(jidx<j_index_end)
 357         {
 358
 359             /* Get j neighbor index, and coordinate index */
 360             jnrlistA         = jjnr[jidx];
 361             jnrlistB         = jjnr[jidx+1];
 362             jnrlistC         = jjnr[jidx+2];
 363             jnrlistD         = jjnr[jidx+3];
 364             /* Sign of each element will be negative for non-real atoms.
 365              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 366              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 367              */
 368             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 369             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 370             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 371             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 372             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 373             j_coord_offsetA  = DIM*jnrA;
 374             j_coord_offsetB  = DIM*jnrB;
 375             j_coord_offsetC  = DIM*jnrC;
 376             j_coord_offsetD  = DIM*jnrD;
 377
 378             /* load j atom coordinates */
 379             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 380                                               x+j_coord_offsetC,x+j_coord_offsetD,
 381                                               &jx0,&jy0,&jz0);
 382
 383             /* Calculate displacement vector */
 384             dx00             = _mm_sub_ps(ix0,jx0);
 385             dy00             = _mm_sub_ps(iy0,jy0);
 386             dz00             = _mm_sub_ps(iz0,jz0);
 387             dx10             = _mm_sub_ps(ix1,jx0);
 388             dy10             = _mm_sub_ps(iy1,jy0);
 389             dz10             = _mm_sub_ps(iz1,jz0);
 390             dx20             = _mm_sub_ps(ix2,jx0);
 391             dy20             = _mm_sub_ps(iy2,jy0);
 392             dz20             = _mm_sub_ps(iz2,jz0);
 393
 394             /* Calculate squared distance and things based on it */
 395             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 396             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 397             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 398
 399             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 400             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 401             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 402
 403             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 404             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 405             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 406
 407             /* Load parameters for j particles */
 408             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 409                                                               charge+jnrC+0,charge+jnrD+0);
 410             vdwjidx0A        = 2*vdwtype[jnrA+0];
 411             vdwjidx0B        = 2*vdwtype[jnrB+0];
 412             vdwjidx0C        = 2*vdwtype[jnrC+0];
 413             vdwjidx0D        = 2*vdwtype[jnrD+0];
 414
 415             fjx0             = _mm_setzero_ps();
 416             fjy0             = _mm_setzero_ps();
 417             fjz0             = _mm_setzero_ps();
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             r00              = _mm_mul_ps(rsq00,rinv00);
 424             r00              = _mm_andnot_ps(dummy_mask,r00);
 425
 426             /* Compute parameters for interactions between i and j atoms */
 427             qq00             = _mm_mul_ps(iq0,jq0);
 428             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 429                                          vdwparam+vdwioffset0+vdwjidx0B,
 430                                          vdwparam+vdwioffset0+vdwjidx0C,
 431                                          vdwparam+vdwioffset0+vdwjidx0D,
 432                                          &c6_00,&c12_00);
 433
 434             /* Calculate table index by multiplying r with table scale and truncate to integer */
 435             rt               = _mm_mul_ps(r00,vftabscale);
 436             vfitab           = _mm_cvttps_epi32(rt);
 437 #ifdef __XOP__
 438             vfeps            = _mm_frcz_ps(rt);
 439 #else
 440             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 441 #endif
 442             twovfeps         = _mm_add_ps(vfeps,vfeps);
 443             vfitab           = _mm_slli_epi32(vfitab,3);
 444
 445             /* COULOMB ELECTROSTATICS */
 446             velec            = _mm_mul_ps(qq00,rinv00);
 447             felec            = _mm_mul_ps(velec,rinvsq00);
 448
 449             /* CUBIC SPLINE TABLE DISPERSION */
 450             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 451             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 452             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 453             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 454             _MM_TRANSPOSE4_PS(Y,F,G,H);
 455             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 456             VV               = _mm_macc_ps(vfeps,Fp,Y);
 457             vvdw6            = _mm_mul_ps(c6_00,VV);
 458             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 459             fvdw6            = _mm_mul_ps(c6_00,FF);
 460
 461             /* CUBIC SPLINE TABLE REPULSION */
 462             vfitab           = _mm_add_epi32(vfitab,ifour);
 463             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 464             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 465             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 466             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 467             _MM_TRANSPOSE4_PS(Y,F,G,H);
 468             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 469             VV               = _mm_macc_ps(vfeps,Fp,Y);
 470             vvdw12           = _mm_mul_ps(c12_00,VV);
 471             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 472             fvdw12           = _mm_mul_ps(c12_00,FF);
 473             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 474             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 475
 476             /* Update potential sum for this i atom from the interaction with this j atom. */
 477             velec            = _mm_andnot_ps(dummy_mask,velec);
 478             velecsum         = _mm_add_ps(velecsum,velec);
 479             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 480             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 481
 482             fscal            = _mm_add_ps(felec,fvdw);
 483
 484             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 485
 486              /* Update vectorial force */
 487             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 488             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 489             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 490
 491             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 492             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 493             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq10             = _mm_mul_ps(iq1,jq0);
 501
 502             /* COULOMB ELECTROSTATICS */
 503             velec            = _mm_mul_ps(qq10,rinv10);
 504             felec            = _mm_mul_ps(velec,rinvsq10);
 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             /* Compute parameters for interactions between i and j atoms */
 528             qq20             = _mm_mul_ps(iq2,jq0);
 529
 530             /* COULOMB ELECTROSTATICS */
 531             velec            = _mm_mul_ps(qq20,rinv20);
 532             felec            = _mm_mul_ps(velec,rinvsq20);
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velec            = _mm_andnot_ps(dummy_mask,velec);
 536             velecsum         = _mm_add_ps(velecsum,velec);
 537
 538             fscal            = felec;
 539
 540             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 541
 542              /* Update vectorial force */
 543             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 544             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 545             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 546
 547             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 548             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 549             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 550
 551             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 552             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 553             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 554             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 555
 556             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 557
 558             /* Inner loop uses 129 flops */
 559         }
 560
 561         /* End of innermost loop */
 562
 563         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 564                                               f+i_coord_offset,fshift+i_shift_offset);
 565
 566         ggid                        = gid[iidx];
 567         /* Update potential energies */
 568         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 569         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 570
 571         /* Increment number of inner iterations */
 572         inneriter                  += j_index_end - j_index_start;
 573
 574         /* Outer loop uses 20 flops */
 575     }
 576
 577     /* Increment number of outer iterations */
 578     outeriter        += nri;
 579
 580     /* Update outer/inner flops */
 581
 582     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
 583 }
 584 /*
 585  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 586  * Electrostatics interaction: Coulomb
 587  * VdW interaction:            CubicSplineTable
 588  * Geometry:                   Water3-Particle
 589  * Calculate force/pot:        Force
 590  */
 591 void
 592 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 593                     (t_nblist * gmx_restrict                nlist,
 594                      rvec * gmx_restrict                    xx,
 595                      rvec * gmx_restrict                    ff,
 596                      t_forcerec * gmx_restrict              fr,
 597                      t_mdatoms * gmx_restrict               mdatoms,
 598                      nb_kernel_data_t * gmx_restrict        kernel_data,
 599                      t_nrnb * gmx_restrict                  nrnb)
 600 {
 601     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 602      * just 0 for non-waters.
 603      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 604      * jnr indices corresponding to data put in the four positions in the SIMD register.
 605      */
 606     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 607     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 608     int              jnrA,jnrB,jnrC,jnrD;
 609     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 610     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 611     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 612     real             rcutoff_scalar;
 613     real             *shiftvec,*fshift,*x,*f;
 614     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 615     real             scratch[4*DIM];
 616     __m128           fscal,rcutoff,rcutoff2,jidxall;
 617     int              vdwioffset0;
 618     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 619     int              vdwioffset1;
 620     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 621     int              vdwioffset2;
 622     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 623     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 624     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 625     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 626     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 627     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 628     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 629     real             *charge;
 630     int              nvdwtype;
 631     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 632     int              *vdwtype;
 633     real             *vdwparam;
 634     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 635     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 636     __m128i          vfitab;
 637     __m128i          ifour       = _mm_set1_epi32(4);
 638     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 639     real             *vftab;
 640     __m128           dummy_mask,cutoff_mask;
 641     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 642     __m128           one     = _mm_set1_ps(1.0);
 643     __m128           two     = _mm_set1_ps(2.0);
 644     x                = xx[0];
 645     f                = ff[0];
 646
 647     nri              = nlist->nri;
 648     iinr             = nlist->iinr;
 649     jindex           = nlist->jindex;
 650     jjnr             = nlist->jjnr;
 651     shiftidx         = nlist->shift;
 652     gid              = nlist->gid;
 653     shiftvec         = fr->shift_vec[0];
 654     fshift           = fr->fshift[0];
 655     facel            = _mm_set1_ps(fr->epsfac);
 656     charge           = mdatoms->chargeA;
 657     nvdwtype         = fr->ntype;
 658     vdwparam         = fr->nbfp;
 659     vdwtype          = mdatoms->typeA;
 660
 661     vftab            = kernel_data->table_vdw->data;
 662     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 663
 664     /* Setup water-specific parameters */
 665     inr              = nlist->iinr[0];
 666     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 667     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 668     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 669     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 670
 671     /* Avoid stupid compiler warnings */
 672     jnrA = jnrB = jnrC = jnrD = 0;
 673     j_coord_offsetA = 0;
 674     j_coord_offsetB = 0;
 675     j_coord_offsetC = 0;
 676     j_coord_offsetD = 0;
 677
 678     outeriter        = 0;
 679     inneriter        = 0;
 680
 681     for(iidx=0;iidx<4*DIM;iidx++)
 682     {
 683         scratch[iidx] = 0.0;
 684     }
 685
 686     /* Start outer loop over neighborlists */
 687     for(iidx=0; iidx<nri; iidx++)
 688     {
 689         /* Load shift vector for this list */
 690         i_shift_offset   = DIM*shiftidx[iidx];
 691
 692         /* Load limits for loop over neighbors */
 693         j_index_start    = jindex[iidx];
 694         j_index_end      = jindex[iidx+1];
 695
 696         /* Get outer coordinate index */
 697         inr              = iinr[iidx];
 698         i_coord_offset   = DIM*inr;
 699
 700         /* Load i particle coords and add shift vector */
 701         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 702                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 703
 704         fix0             = _mm_setzero_ps();
 705         fiy0             = _mm_setzero_ps();
 706         fiz0             = _mm_setzero_ps();
 707         fix1             = _mm_setzero_ps();
 708         fiy1             = _mm_setzero_ps();
 709         fiz1             = _mm_setzero_ps();
 710         fix2             = _mm_setzero_ps();
 711         fiy2             = _mm_setzero_ps();
 712         fiz2             = _mm_setzero_ps();
 713
 714         /* Start inner kernel loop */
 715         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 716         {
 717
 718             /* Get j neighbor index, and coordinate index */
 719             jnrA             = jjnr[jidx];
 720             jnrB             = jjnr[jidx+1];
 721             jnrC             = jjnr[jidx+2];
 722             jnrD             = jjnr[jidx+3];
 723             j_coord_offsetA  = DIM*jnrA;
 724             j_coord_offsetB  = DIM*jnrB;
 725             j_coord_offsetC  = DIM*jnrC;
 726             j_coord_offsetD  = DIM*jnrD;
 727
 728             /* load j atom coordinates */
 729             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 730                                               x+j_coord_offsetC,x+j_coord_offsetD,
 731                                               &jx0,&jy0,&jz0);
 732
 733             /* Calculate displacement vector */
 734             dx00             = _mm_sub_ps(ix0,jx0);
 735             dy00             = _mm_sub_ps(iy0,jy0);
 736             dz00             = _mm_sub_ps(iz0,jz0);
 737             dx10             = _mm_sub_ps(ix1,jx0);
 738             dy10             = _mm_sub_ps(iy1,jy0);
 739             dz10             = _mm_sub_ps(iz1,jz0);
 740             dx20             = _mm_sub_ps(ix2,jx0);
 741             dy20             = _mm_sub_ps(iy2,jy0);
 742             dz20             = _mm_sub_ps(iz2,jz0);
 743
 744             /* Calculate squared distance and things based on it */
 745             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 746             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 747             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 748
 749             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 750             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 751             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 752
 753             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 754             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 755             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 756
 757             /* Load parameters for j particles */
 758             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 759                                                               charge+jnrC+0,charge+jnrD+0);
 760             vdwjidx0A        = 2*vdwtype[jnrA+0];
 761             vdwjidx0B        = 2*vdwtype[jnrB+0];
 762             vdwjidx0C        = 2*vdwtype[jnrC+0];
 763             vdwjidx0D        = 2*vdwtype[jnrD+0];
 764
 765             fjx0             = _mm_setzero_ps();
 766             fjy0             = _mm_setzero_ps();
 767             fjz0             = _mm_setzero_ps();
 768
 769             /**************************
 770              * CALCULATE INTERACTIONS *
 771              **************************/
 772
 773             r00              = _mm_mul_ps(rsq00,rinv00);
 774
 775             /* Compute parameters for interactions between i and j atoms */
 776             qq00             = _mm_mul_ps(iq0,jq0);
 777             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 778                                          vdwparam+vdwioffset0+vdwjidx0B,
 779                                          vdwparam+vdwioffset0+vdwjidx0C,
 780                                          vdwparam+vdwioffset0+vdwjidx0D,
 781                                          &c6_00,&c12_00);
 782
 783             /* Calculate table index by multiplying r with table scale and truncate to integer */
 784             rt               = _mm_mul_ps(r00,vftabscale);
 785             vfitab           = _mm_cvttps_epi32(rt);
 786 #ifdef __XOP__
 787             vfeps            = _mm_frcz_ps(rt);
 788 #else
 789             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 790 #endif
 791             twovfeps         = _mm_add_ps(vfeps,vfeps);
 792             vfitab           = _mm_slli_epi32(vfitab,3);
 793
 794             /* COULOMB ELECTROSTATICS */
 795             velec            = _mm_mul_ps(qq00,rinv00);
 796             felec            = _mm_mul_ps(velec,rinvsq00);
 797
 798             /* CUBIC SPLINE TABLE DISPERSION */
 799             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 800             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 801             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 802             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 803             _MM_TRANSPOSE4_PS(Y,F,G,H);
 804             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 805             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 806             fvdw6            = _mm_mul_ps(c6_00,FF);
 807
 808             /* CUBIC SPLINE TABLE REPULSION */
 809             vfitab           = _mm_add_epi32(vfitab,ifour);
 810             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 811             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 812             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 813             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 814             _MM_TRANSPOSE4_PS(Y,F,G,H);
 815             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 816             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 817             fvdw12           = _mm_mul_ps(c12_00,FF);
 818             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 819
 820             fscal            = _mm_add_ps(felec,fvdw);
 821
 822              /* Update vectorial force */
 823             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 824             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 825             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 826
 827             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 828             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 829             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 830
 831             /**************************
 832              * CALCULATE INTERACTIONS *
 833              **************************/
 834
 835             /* Compute parameters for interactions between i and j atoms */
 836             qq10             = _mm_mul_ps(iq1,jq0);
 837
 838             /* COULOMB ELECTROSTATICS */
 839             velec            = _mm_mul_ps(qq10,rinv10);
 840             felec            = _mm_mul_ps(velec,rinvsq10);
 841
 842             fscal            = felec;
 843
 844              /* Update vectorial force */
 845             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 846             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 847             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 848
 849             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 850             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 851             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 852
 853             /**************************
 854              * CALCULATE INTERACTIONS *
 855              **************************/
 856
 857             /* Compute parameters for interactions between i and j atoms */
 858             qq20             = _mm_mul_ps(iq2,jq0);
 859
 860             /* COULOMB ELECTROSTATICS */
 861             velec            = _mm_mul_ps(qq20,rinv20);
 862             felec            = _mm_mul_ps(velec,rinvsq20);
 863
 864             fscal            = felec;
 865
 866              /* Update vectorial force */
 867             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 868             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 869             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 870
 871             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 872             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 873             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 874
 875             fjptrA             = f+j_coord_offsetA;
 876             fjptrB             = f+j_coord_offsetB;
 877             fjptrC             = f+j_coord_offsetC;
 878             fjptrD             = f+j_coord_offsetD;
 879
 880             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 881
 882             /* Inner loop uses 117 flops */
 883         }
 884
 885         if(jidx<j_index_end)
 886         {
 887
 888             /* Get j neighbor index, and coordinate index */
 889             jnrlistA         = jjnr[jidx];
 890             jnrlistB         = jjnr[jidx+1];
 891             jnrlistC         = jjnr[jidx+2];
 892             jnrlistD         = jjnr[jidx+3];
 893             /* Sign of each element will be negative for non-real atoms.
 894              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 895              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 896              */
 897             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 898             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 899             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 900             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 901             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 902             j_coord_offsetA  = DIM*jnrA;
 903             j_coord_offsetB  = DIM*jnrB;
 904             j_coord_offsetC  = DIM*jnrC;
 905             j_coord_offsetD  = DIM*jnrD;
 906
 907             /* load j atom coordinates */
 908             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 909                                               x+j_coord_offsetC,x+j_coord_offsetD,
 910                                               &jx0,&jy0,&jz0);
 911
 912             /* Calculate displacement vector */
 913             dx00             = _mm_sub_ps(ix0,jx0);
 914             dy00             = _mm_sub_ps(iy0,jy0);
 915             dz00             = _mm_sub_ps(iz0,jz0);
 916             dx10             = _mm_sub_ps(ix1,jx0);
 917             dy10             = _mm_sub_ps(iy1,jy0);
 918             dz10             = _mm_sub_ps(iz1,jz0);
 919             dx20             = _mm_sub_ps(ix2,jx0);
 920             dy20             = _mm_sub_ps(iy2,jy0);
 921             dz20             = _mm_sub_ps(iz2,jz0);
 922
 923             /* Calculate squared distance and things based on it */
 924             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 925             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 926             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 927
 928             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 929             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 930             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 931
 932             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 933             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 934             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 935
 936             /* Load parameters for j particles */
 937             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 938                                                               charge+jnrC+0,charge+jnrD+0);
 939             vdwjidx0A        = 2*vdwtype[jnrA+0];
 940             vdwjidx0B        = 2*vdwtype[jnrB+0];
 941             vdwjidx0C        = 2*vdwtype[jnrC+0];
 942             vdwjidx0D        = 2*vdwtype[jnrD+0];
 943
 944             fjx0             = _mm_setzero_ps();
 945             fjy0             = _mm_setzero_ps();
 946             fjz0             = _mm_setzero_ps();
 947
 948             /**************************
 949              * CALCULATE INTERACTIONS *
 950              **************************/
 951
 952             r00              = _mm_mul_ps(rsq00,rinv00);
 953             r00              = _mm_andnot_ps(dummy_mask,r00);
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq00             = _mm_mul_ps(iq0,jq0);
 957             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 958                                          vdwparam+vdwioffset0+vdwjidx0B,
 959                                          vdwparam+vdwioffset0+vdwjidx0C,
 960                                          vdwparam+vdwioffset0+vdwjidx0D,
 961                                          &c6_00,&c12_00);
 962
 963             /* Calculate table index by multiplying r with table scale and truncate to integer */
 964             rt               = _mm_mul_ps(r00,vftabscale);
 965             vfitab           = _mm_cvttps_epi32(rt);
 966 #ifdef __XOP__
 967             vfeps            = _mm_frcz_ps(rt);
 968 #else
 969             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 970 #endif
 971             twovfeps         = _mm_add_ps(vfeps,vfeps);
 972             vfitab           = _mm_slli_epi32(vfitab,3);
 973
 974             /* COULOMB ELECTROSTATICS */
 975             velec            = _mm_mul_ps(qq00,rinv00);
 976             felec            = _mm_mul_ps(velec,rinvsq00);
 977
 978             /* CUBIC SPLINE TABLE DISPERSION */
 979             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 980             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 981             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 982             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 983             _MM_TRANSPOSE4_PS(Y,F,G,H);
 984             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 985             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 986             fvdw6            = _mm_mul_ps(c6_00,FF);
 987
 988             /* CUBIC SPLINE TABLE REPULSION */
 989             vfitab           = _mm_add_epi32(vfitab,ifour);
 990             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 991             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 992             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 993             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 994             _MM_TRANSPOSE4_PS(Y,F,G,H);
 995             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 996             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 997             fvdw12           = _mm_mul_ps(c12_00,FF);
 998             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 999
1000             fscal            = _mm_add_ps(felec,fvdw);
1001
1002             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1003
1004              /* Update vectorial force */
1005             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1006             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1007             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1008
1009             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1010             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1011             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             /* Compute parameters for interactions between i and j atoms */
1018             qq10             = _mm_mul_ps(iq1,jq0);
1019
1020             /* COULOMB ELECTROSTATICS */
1021             velec            = _mm_mul_ps(qq10,rinv10);
1022             felec            = _mm_mul_ps(velec,rinvsq10);
1023
1024             fscal            = felec;
1025
1026             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1027
1028              /* Update vectorial force */
1029             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1030             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1031             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1032
1033             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1034             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1035             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             /* Compute parameters for interactions between i and j atoms */
1042             qq20             = _mm_mul_ps(iq2,jq0);
1043
1044             /* COULOMB ELECTROSTATICS */
1045             velec            = _mm_mul_ps(qq20,rinv20);
1046             felec            = _mm_mul_ps(velec,rinvsq20);
1047
1048             fscal            = felec;
1049
1050             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1051
1052              /* Update vectorial force */
1053             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1054             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1055             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1056
1057             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1058             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1059             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1060
1061             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1062             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1063             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1064             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1065
1066             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1067
1068             /* Inner loop uses 118 flops */
1069         }
1070
1071         /* End of innermost loop */
1072
1073         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1074                                               f+i_coord_offset,fshift+i_shift_offset);
1075
1076         /* Increment number of inner iterations */
1077         inneriter                  += j_index_end - j_index_start;
1078
1079         /* Outer loop uses 18 flops */
1080     }
1081
1082     /* Increment number of outer iterations */
1083     outeriter        += nri;
1084
1085     /* Update outer/inner flops */
1086
1087     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1088 }