src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwLJ_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_ElecCSTab_VdwLJ_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_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_elec->data;
 114     vftabscale       = _mm_set1_ps(kernel_data->table_elec->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
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 213                                                               charge+jnrC+0,charge+jnrD+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216             vdwjidx0C        = 2*vdwtype[jnrC+0];
 217             vdwjidx0D        = 2*vdwtype[jnrD+0];
 218
 219             fjx0             = _mm_setzero_ps();
 220             fjy0             = _mm_setzero_ps();
 221             fjz0             = _mm_setzero_ps();
 222
 223             /**************************
 224              * CALCULATE INTERACTIONS *
 225              **************************/
 226
 227             r00              = _mm_mul_ps(rsq00,rinv00);
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq00             = _mm_mul_ps(iq0,jq0);
 231             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 232                                          vdwparam+vdwioffset0+vdwjidx0B,
 233                                          vdwparam+vdwioffset0+vdwjidx0C,
 234                                          vdwparam+vdwioffset0+vdwjidx0D,
 235                                          &c6_00,&c12_00);
 236
 237             /* Calculate table index by multiplying r with table scale and truncate to integer */
 238             rt               = _mm_mul_ps(r00,vftabscale);
 239             vfitab           = _mm_cvttps_epi32(rt);
 240 #ifdef __XOP__
 241             vfeps            = _mm_frcz_ps(rt);
 242 #else
 243             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 244 #endif
 245             twovfeps         = _mm_add_ps(vfeps,vfeps);
 246             vfitab           = _mm_slli_epi32(vfitab,2);
 247
 248             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 249             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 250             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 251             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 252             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 253             _MM_TRANSPOSE4_PS(Y,F,G,H);
 254             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 255             VV               = _mm_macc_ps(vfeps,Fp,Y);
 256             velec            = _mm_mul_ps(qq00,VV);
 257             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 258             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 259
 260             /* LENNARD-JONES DISPERSION/REPULSION */
 261
 262             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 263             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 264             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 265             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 266             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             velecsum         = _mm_add_ps(velecsum,velec);
 270             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 271
 272             fscal            = _mm_add_ps(felec,fvdw);
 273
 274              /* Update vectorial force */
 275             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 276             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 277             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 278
 279             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 280             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 281             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             r10              = _mm_mul_ps(rsq10,rinv10);
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq10             = _mm_mul_ps(iq1,jq0);
 291
 292             /* Calculate table index by multiplying r with table scale and truncate to integer */
 293             rt               = _mm_mul_ps(r10,vftabscale);
 294             vfitab           = _mm_cvttps_epi32(rt);
 295 #ifdef __XOP__
 296             vfeps            = _mm_frcz_ps(rt);
 297 #else
 298             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 299 #endif
 300             twovfeps         = _mm_add_ps(vfeps,vfeps);
 301             vfitab           = _mm_slli_epi32(vfitab,2);
 302
 303             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 304             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 305             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 306             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 307             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 308             _MM_TRANSPOSE4_PS(Y,F,G,H);
 309             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 310             VV               = _mm_macc_ps(vfeps,Fp,Y);
 311             velec            = _mm_mul_ps(qq10,VV);
 312             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 313             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velecsum         = _mm_add_ps(velecsum,velec);
 317
 318             fscal            = felec;
 319
 320              /* Update vectorial force */
 321             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 322             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 323             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 324
 325             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 326             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 327             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             r20              = _mm_mul_ps(rsq20,rinv20);
 334
 335             /* Compute parameters for interactions between i and j atoms */
 336             qq20             = _mm_mul_ps(iq2,jq0);
 337
 338             /* Calculate table index by multiplying r with table scale and truncate to integer */
 339             rt               = _mm_mul_ps(r20,vftabscale);
 340             vfitab           = _mm_cvttps_epi32(rt);
 341 #ifdef __XOP__
 342             vfeps            = _mm_frcz_ps(rt);
 343 #else
 344             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 345 #endif
 346             twovfeps         = _mm_add_ps(vfeps,vfeps);
 347             vfitab           = _mm_slli_epi32(vfitab,2);
 348
 349             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 350             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 351             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 352             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 353             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 354             _MM_TRANSPOSE4_PS(Y,F,G,H);
 355             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 356             VV               = _mm_macc_ps(vfeps,Fp,Y);
 357             velec            = _mm_mul_ps(qq20,VV);
 358             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 359             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velecsum         = _mm_add_ps(velecsum,velec);
 363
 364             fscal            = felec;
 365
 366              /* Update vectorial force */
 367             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 368             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 369             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 370
 371             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 372             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 373             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 374
 375             fjptrA             = f+j_coord_offsetA;
 376             fjptrB             = f+j_coord_offsetB;
 377             fjptrC             = f+j_coord_offsetC;
 378             fjptrD             = f+j_coord_offsetD;
 379
 380             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 381
 382             /* Inner loop uses 151 flops */
 383         }
 384
 385         if(jidx<j_index_end)
 386         {
 387
 388             /* Get j neighbor index, and coordinate index */
 389             jnrlistA         = jjnr[jidx];
 390             jnrlistB         = jjnr[jidx+1];
 391             jnrlistC         = jjnr[jidx+2];
 392             jnrlistD         = jjnr[jidx+3];
 393             /* Sign of each element will be negative for non-real atoms.
 394              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 395              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 396              */
 397             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 398             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 399             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 400             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 401             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 402             j_coord_offsetA  = DIM*jnrA;
 403             j_coord_offsetB  = DIM*jnrB;
 404             j_coord_offsetC  = DIM*jnrC;
 405             j_coord_offsetD  = DIM*jnrD;
 406
 407             /* load j atom coordinates */
 408             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 409                                               x+j_coord_offsetC,x+j_coord_offsetD,
 410                                               &jx0,&jy0,&jz0);
 411
 412             /* Calculate displacement vector */
 413             dx00             = _mm_sub_ps(ix0,jx0);
 414             dy00             = _mm_sub_ps(iy0,jy0);
 415             dz00             = _mm_sub_ps(iz0,jz0);
 416             dx10             = _mm_sub_ps(ix1,jx0);
 417             dy10             = _mm_sub_ps(iy1,jy0);
 418             dz10             = _mm_sub_ps(iz1,jz0);
 419             dx20             = _mm_sub_ps(ix2,jx0);
 420             dy20             = _mm_sub_ps(iy2,jy0);
 421             dz20             = _mm_sub_ps(iz2,jz0);
 422
 423             /* Calculate squared distance and things based on it */
 424             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 425             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 426             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 427
 428             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 429             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 430             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 431
 432             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 433
 434             /* Load parameters for j particles */
 435             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 436                                                               charge+jnrC+0,charge+jnrD+0);
 437             vdwjidx0A        = 2*vdwtype[jnrA+0];
 438             vdwjidx0B        = 2*vdwtype[jnrB+0];
 439             vdwjidx0C        = 2*vdwtype[jnrC+0];
 440             vdwjidx0D        = 2*vdwtype[jnrD+0];
 441
 442             fjx0             = _mm_setzero_ps();
 443             fjy0             = _mm_setzero_ps();
 444             fjz0             = _mm_setzero_ps();
 445
 446             /**************************
 447              * CALCULATE INTERACTIONS *
 448              **************************/
 449
 450             r00              = _mm_mul_ps(rsq00,rinv00);
 451             r00              = _mm_andnot_ps(dummy_mask,r00);
 452
 453             /* Compute parameters for interactions between i and j atoms */
 454             qq00             = _mm_mul_ps(iq0,jq0);
 455             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 456                                          vdwparam+vdwioffset0+vdwjidx0B,
 457                                          vdwparam+vdwioffset0+vdwjidx0C,
 458                                          vdwparam+vdwioffset0+vdwjidx0D,
 459                                          &c6_00,&c12_00);
 460
 461             /* Calculate table index by multiplying r with table scale and truncate to integer */
 462             rt               = _mm_mul_ps(r00,vftabscale);
 463             vfitab           = _mm_cvttps_epi32(rt);
 464 #ifdef __XOP__
 465             vfeps            = _mm_frcz_ps(rt);
 466 #else
 467             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 468 #endif
 469             twovfeps         = _mm_add_ps(vfeps,vfeps);
 470             vfitab           = _mm_slli_epi32(vfitab,2);
 471
 472             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 473             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 474             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 475             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 476             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 477             _MM_TRANSPOSE4_PS(Y,F,G,H);
 478             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 479             VV               = _mm_macc_ps(vfeps,Fp,Y);
 480             velec            = _mm_mul_ps(qq00,VV);
 481             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 482             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 483
 484             /* LENNARD-JONES DISPERSION/REPULSION */
 485
 486             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 487             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 488             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 489             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 490             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 491
 492             /* Update potential sum for this i atom from the interaction with this j atom. */
 493             velec            = _mm_andnot_ps(dummy_mask,velec);
 494             velecsum         = _mm_add_ps(velecsum,velec);
 495             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 496             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 497
 498             fscal            = _mm_add_ps(felec,fvdw);
 499
 500             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 501
 502              /* Update vectorial force */
 503             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 504             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 505             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 506
 507             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 508             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 509             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             r10              = _mm_mul_ps(rsq10,rinv10);
 516             r10              = _mm_andnot_ps(dummy_mask,r10);
 517
 518             /* Compute parameters for interactions between i and j atoms */
 519             qq10             = _mm_mul_ps(iq1,jq0);
 520
 521             /* Calculate table index by multiplying r with table scale and truncate to integer */
 522             rt               = _mm_mul_ps(r10,vftabscale);
 523             vfitab           = _mm_cvttps_epi32(rt);
 524 #ifdef __XOP__
 525             vfeps            = _mm_frcz_ps(rt);
 526 #else
 527             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 528 #endif
 529             twovfeps         = _mm_add_ps(vfeps,vfeps);
 530             vfitab           = _mm_slli_epi32(vfitab,2);
 531
 532             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 533             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 534             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 535             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 536             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 537             _MM_TRANSPOSE4_PS(Y,F,G,H);
 538             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 539             VV               = _mm_macc_ps(vfeps,Fp,Y);
 540             velec            = _mm_mul_ps(qq10,VV);
 541             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 542             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             velec            = _mm_andnot_ps(dummy_mask,velec);
 546             velecsum         = _mm_add_ps(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 551
 552              /* Update vectorial force */
 553             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 554             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 555             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 556
 557             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 558             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 559             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             r20              = _mm_mul_ps(rsq20,rinv20);
 566             r20              = _mm_andnot_ps(dummy_mask,r20);
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             qq20             = _mm_mul_ps(iq2,jq0);
 570
 571             /* Calculate table index by multiplying r with table scale and truncate to integer */
 572             rt               = _mm_mul_ps(r20,vftabscale);
 573             vfitab           = _mm_cvttps_epi32(rt);
 574 #ifdef __XOP__
 575             vfeps            = _mm_frcz_ps(rt);
 576 #else
 577             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 578 #endif
 579             twovfeps         = _mm_add_ps(vfeps,vfeps);
 580             vfitab           = _mm_slli_epi32(vfitab,2);
 581
 582             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 583             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 584             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 585             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 586             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 587             _MM_TRANSPOSE4_PS(Y,F,G,H);
 588             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 589             VV               = _mm_macc_ps(vfeps,Fp,Y);
 590             velec            = _mm_mul_ps(qq20,VV);
 591             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 592             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 593
 594             /* Update potential sum for this i atom from the interaction with this j atom. */
 595             velec            = _mm_andnot_ps(dummy_mask,velec);
 596             velecsum         = _mm_add_ps(velecsum,velec);
 597
 598             fscal            = felec;
 599
 600             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 601
 602              /* Update vectorial force */
 603             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 604             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 605             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 606
 607             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 608             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 609             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 610
 611             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 612             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 613             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 614             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 615
 616             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 617
 618             /* Inner loop uses 154 flops */
 619         }
 620
 621         /* End of innermost loop */
 622
 623         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 624                                               f+i_coord_offset,fshift+i_shift_offset);
 625
 626         ggid                        = gid[iidx];
 627         /* Update potential energies */
 628         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 629         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 630
 631         /* Increment number of inner iterations */
 632         inneriter                  += j_index_end - j_index_start;
 633
 634         /* Outer loop uses 20 flops */
 635     }
 636
 637     /* Increment number of outer iterations */
 638     outeriter        += nri;
 639
 640     /* Update outer/inner flops */
 641
 642     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 643 }
 644 /*
 645  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_single
 646  * Electrostatics interaction: CubicSplineTable
 647  * VdW interaction:            LennardJones
 648  * Geometry:                   Water3-Particle
 649  * Calculate force/pot:        Force
 650  */
 651 void
 652 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_avx_128_fma_single
 653                     (t_nblist * gmx_restrict                nlist,
 654                      rvec * gmx_restrict                    xx,
 655                      rvec * gmx_restrict                    ff,
 656                      t_forcerec * gmx_restrict              fr,
 657                      t_mdatoms * gmx_restrict               mdatoms,
 658                      nb_kernel_data_t * gmx_restrict        kernel_data,
 659                      t_nrnb * gmx_restrict                  nrnb)
 660 {
 661     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 662      * just 0 for non-waters.
 663      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 664      * jnr indices corresponding to data put in the four positions in the SIMD register.
 665      */
 666     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 667     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 668     int              jnrA,jnrB,jnrC,jnrD;
 669     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 670     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 671     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 672     real             rcutoff_scalar;
 673     real             *shiftvec,*fshift,*x,*f;
 674     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 675     real             scratch[4*DIM];
 676     __m128           fscal,rcutoff,rcutoff2,jidxall;
 677     int              vdwioffset0;
 678     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 679     int              vdwioffset1;
 680     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 681     int              vdwioffset2;
 682     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 683     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 684     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 685     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 686     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 687     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 688     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 689     real             *charge;
 690     int              nvdwtype;
 691     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 692     int              *vdwtype;
 693     real             *vdwparam;
 694     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 695     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 696     __m128i          vfitab;
 697     __m128i          ifour       = _mm_set1_epi32(4);
 698     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 699     real             *vftab;
 700     __m128           dummy_mask,cutoff_mask;
 701     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 702     __m128           one     = _mm_set1_ps(1.0);
 703     __m128           two     = _mm_set1_ps(2.0);
 704     x                = xx[0];
 705     f                = ff[0];
 706
 707     nri              = nlist->nri;
 708     iinr             = nlist->iinr;
 709     jindex           = nlist->jindex;
 710     jjnr             = nlist->jjnr;
 711     shiftidx         = nlist->shift;
 712     gid              = nlist->gid;
 713     shiftvec         = fr->shift_vec[0];
 714     fshift           = fr->fshift[0];
 715     facel            = _mm_set1_ps(fr->epsfac);
 716     charge           = mdatoms->chargeA;
 717     nvdwtype         = fr->ntype;
 718     vdwparam         = fr->nbfp;
 719     vdwtype          = mdatoms->typeA;
 720
 721     vftab            = kernel_data->table_elec->data;
 722     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 723
 724     /* Setup water-specific parameters */
 725     inr              = nlist->iinr[0];
 726     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 727     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 728     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 729     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 730
 731     /* Avoid stupid compiler warnings */
 732     jnrA = jnrB = jnrC = jnrD = 0;
 733     j_coord_offsetA = 0;
 734     j_coord_offsetB = 0;
 735     j_coord_offsetC = 0;
 736     j_coord_offsetD = 0;
 737
 738     outeriter        = 0;
 739     inneriter        = 0;
 740
 741     for(iidx=0;iidx<4*DIM;iidx++)
 742     {
 743         scratch[iidx] = 0.0;
 744     }
 745
 746     /* Start outer loop over neighborlists */
 747     for(iidx=0; iidx<nri; iidx++)
 748     {
 749         /* Load shift vector for this list */
 750         i_shift_offset   = DIM*shiftidx[iidx];
 751
 752         /* Load limits for loop over neighbors */
 753         j_index_start    = jindex[iidx];
 754         j_index_end      = jindex[iidx+1];
 755
 756         /* Get outer coordinate index */
 757         inr              = iinr[iidx];
 758         i_coord_offset   = DIM*inr;
 759
 760         /* Load i particle coords and add shift vector */
 761         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 762                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 763
 764         fix0             = _mm_setzero_ps();
 765         fiy0             = _mm_setzero_ps();
 766         fiz0             = _mm_setzero_ps();
 767         fix1             = _mm_setzero_ps();
 768         fiy1             = _mm_setzero_ps();
 769         fiz1             = _mm_setzero_ps();
 770         fix2             = _mm_setzero_ps();
 771         fiy2             = _mm_setzero_ps();
 772         fiz2             = _mm_setzero_ps();
 773
 774         /* Start inner kernel loop */
 775         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 776         {
 777
 778             /* Get j neighbor index, and coordinate index */
 779             jnrA             = jjnr[jidx];
 780             jnrB             = jjnr[jidx+1];
 781             jnrC             = jjnr[jidx+2];
 782             jnrD             = jjnr[jidx+3];
 783             j_coord_offsetA  = DIM*jnrA;
 784             j_coord_offsetB  = DIM*jnrB;
 785             j_coord_offsetC  = DIM*jnrC;
 786             j_coord_offsetD  = DIM*jnrD;
 787
 788             /* load j atom coordinates */
 789             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 790                                               x+j_coord_offsetC,x+j_coord_offsetD,
 791                                               &jx0,&jy0,&jz0);
 792
 793             /* Calculate displacement vector */
 794             dx00             = _mm_sub_ps(ix0,jx0);
 795             dy00             = _mm_sub_ps(iy0,jy0);
 796             dz00             = _mm_sub_ps(iz0,jz0);
 797             dx10             = _mm_sub_ps(ix1,jx0);
 798             dy10             = _mm_sub_ps(iy1,jy0);
 799             dz10             = _mm_sub_ps(iz1,jz0);
 800             dx20             = _mm_sub_ps(ix2,jx0);
 801             dy20             = _mm_sub_ps(iy2,jy0);
 802             dz20             = _mm_sub_ps(iz2,jz0);
 803
 804             /* Calculate squared distance and things based on it */
 805             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 806             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 807             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 808
 809             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 810             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 811             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 812
 813             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 814
 815             /* Load parameters for j particles */
 816             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 817                                                               charge+jnrC+0,charge+jnrD+0);
 818             vdwjidx0A        = 2*vdwtype[jnrA+0];
 819             vdwjidx0B        = 2*vdwtype[jnrB+0];
 820             vdwjidx0C        = 2*vdwtype[jnrC+0];
 821             vdwjidx0D        = 2*vdwtype[jnrD+0];
 822
 823             fjx0             = _mm_setzero_ps();
 824             fjy0             = _mm_setzero_ps();
 825             fjz0             = _mm_setzero_ps();
 826
 827             /**************************
 828              * CALCULATE INTERACTIONS *
 829              **************************/
 830
 831             r00              = _mm_mul_ps(rsq00,rinv00);
 832
 833             /* Compute parameters for interactions between i and j atoms */
 834             qq00             = _mm_mul_ps(iq0,jq0);
 835             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 836                                          vdwparam+vdwioffset0+vdwjidx0B,
 837                                          vdwparam+vdwioffset0+vdwjidx0C,
 838                                          vdwparam+vdwioffset0+vdwjidx0D,
 839                                          &c6_00,&c12_00);
 840
 841             /* Calculate table index by multiplying r with table scale and truncate to integer */
 842             rt               = _mm_mul_ps(r00,vftabscale);
 843             vfitab           = _mm_cvttps_epi32(rt);
 844 #ifdef __XOP__
 845             vfeps            = _mm_frcz_ps(rt);
 846 #else
 847             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 848 #endif
 849             twovfeps         = _mm_add_ps(vfeps,vfeps);
 850             vfitab           = _mm_slli_epi32(vfitab,2);
 851
 852             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 853             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 854             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 855             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 856             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 857             _MM_TRANSPOSE4_PS(Y,F,G,H);
 858             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 859             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 860             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 861
 862             /* LENNARD-JONES DISPERSION/REPULSION */
 863
 864             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 865             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 866
 867             fscal            = _mm_add_ps(felec,fvdw);
 868
 869              /* Update vectorial force */
 870             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 871             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 872             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 873
 874             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 875             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 876             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             r10              = _mm_mul_ps(rsq10,rinv10);
 883
 884             /* Compute parameters for interactions between i and j atoms */
 885             qq10             = _mm_mul_ps(iq1,jq0);
 886
 887             /* Calculate table index by multiplying r with table scale and truncate to integer */
 888             rt               = _mm_mul_ps(r10,vftabscale);
 889             vfitab           = _mm_cvttps_epi32(rt);
 890 #ifdef __XOP__
 891             vfeps            = _mm_frcz_ps(rt);
 892 #else
 893             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 894 #endif
 895             twovfeps         = _mm_add_ps(vfeps,vfeps);
 896             vfitab           = _mm_slli_epi32(vfitab,2);
 897
 898             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 899             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 900             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 901             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 902             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 903             _MM_TRANSPOSE4_PS(Y,F,G,H);
 904             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 905             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 906             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 907
 908             fscal            = felec;
 909
 910              /* Update vectorial force */
 911             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 912             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 913             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 914
 915             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 916             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 917             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 918
 919             /**************************
 920              * CALCULATE INTERACTIONS *
 921              **************************/
 922
 923             r20              = _mm_mul_ps(rsq20,rinv20);
 924
 925             /* Compute parameters for interactions between i and j atoms */
 926             qq20             = _mm_mul_ps(iq2,jq0);
 927
 928             /* Calculate table index by multiplying r with table scale and truncate to integer */
 929             rt               = _mm_mul_ps(r20,vftabscale);
 930             vfitab           = _mm_cvttps_epi32(rt);
 931 #ifdef __XOP__
 932             vfeps            = _mm_frcz_ps(rt);
 933 #else
 934             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 935 #endif
 936             twovfeps         = _mm_add_ps(vfeps,vfeps);
 937             vfitab           = _mm_slli_epi32(vfitab,2);
 938
 939             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 940             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 941             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 942             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 943             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 944             _MM_TRANSPOSE4_PS(Y,F,G,H);
 945             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 946             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 947             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 948
 949             fscal            = felec;
 950
 951              /* Update vectorial force */
 952             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 953             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 954             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 955
 956             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 957             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 958             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 959
 960             fjptrA             = f+j_coord_offsetA;
 961             fjptrB             = f+j_coord_offsetB;
 962             fjptrC             = f+j_coord_offsetC;
 963             fjptrD             = f+j_coord_offsetD;
 964
 965             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 966
 967             /* Inner loop uses 134 flops */
 968         }
 969
 970         if(jidx<j_index_end)
 971         {
 972
 973             /* Get j neighbor index, and coordinate index */
 974             jnrlistA         = jjnr[jidx];
 975             jnrlistB         = jjnr[jidx+1];
 976             jnrlistC         = jjnr[jidx+2];
 977             jnrlistD         = jjnr[jidx+3];
 978             /* Sign of each element will be negative for non-real atoms.
 979              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 980              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 981              */
 982             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 983             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 984             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 985             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 986             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 987             j_coord_offsetA  = DIM*jnrA;
 988             j_coord_offsetB  = DIM*jnrB;
 989             j_coord_offsetC  = DIM*jnrC;
 990             j_coord_offsetD  = DIM*jnrD;
 991
 992             /* load j atom coordinates */
 993             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 994                                               x+j_coord_offsetC,x+j_coord_offsetD,
 995                                               &jx0,&jy0,&jz0);
 996
 997             /* Calculate displacement vector */
 998             dx00             = _mm_sub_ps(ix0,jx0);
 999             dy00             = _mm_sub_ps(iy0,jy0);
1000             dz00             = _mm_sub_ps(iz0,jz0);
1001             dx10             = _mm_sub_ps(ix1,jx0);
1002             dy10             = _mm_sub_ps(iy1,jy0);
1003             dz10             = _mm_sub_ps(iz1,jz0);
1004             dx20             = _mm_sub_ps(ix2,jx0);
1005             dy20             = _mm_sub_ps(iy2,jy0);
1006             dz20             = _mm_sub_ps(iz2,jz0);
1007
1008             /* Calculate squared distance and things based on it */
1009             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1010             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1011             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1012
1013             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1014             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1015             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1016
1017             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1018
1019             /* Load parameters for j particles */
1020             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1021                                                               charge+jnrC+0,charge+jnrD+0);
1022             vdwjidx0A        = 2*vdwtype[jnrA+0];
1023             vdwjidx0B        = 2*vdwtype[jnrB+0];
1024             vdwjidx0C        = 2*vdwtype[jnrC+0];
1025             vdwjidx0D        = 2*vdwtype[jnrD+0];
1026
1027             fjx0             = _mm_setzero_ps();
1028             fjy0             = _mm_setzero_ps();
1029             fjz0             = _mm_setzero_ps();
1030
1031             /**************************
1032              * CALCULATE INTERACTIONS *
1033              **************************/
1034
1035             r00              = _mm_mul_ps(rsq00,rinv00);
1036             r00              = _mm_andnot_ps(dummy_mask,r00);
1037
1038             /* Compute parameters for interactions between i and j atoms */
1039             qq00             = _mm_mul_ps(iq0,jq0);
1040             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1041                                          vdwparam+vdwioffset0+vdwjidx0B,
1042                                          vdwparam+vdwioffset0+vdwjidx0C,
1043                                          vdwparam+vdwioffset0+vdwjidx0D,
1044                                          &c6_00,&c12_00);
1045
1046             /* Calculate table index by multiplying r with table scale and truncate to integer */
1047             rt               = _mm_mul_ps(r00,vftabscale);
1048             vfitab           = _mm_cvttps_epi32(rt);
1049 #ifdef __XOP__
1050             vfeps            = _mm_frcz_ps(rt);
1051 #else
1052             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1053 #endif
1054             twovfeps         = _mm_add_ps(vfeps,vfeps);
1055             vfitab           = _mm_slli_epi32(vfitab,2);
1056
1057             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1058             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1059             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1060             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1061             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1062             _MM_TRANSPOSE4_PS(Y,F,G,H);
1063             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1064             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1065             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1066
1067             /* LENNARD-JONES DISPERSION/REPULSION */
1068
1069             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1070             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1071
1072             fscal            = _mm_add_ps(felec,fvdw);
1073
1074             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1075
1076              /* Update vectorial force */
1077             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1078             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1079             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1080
1081             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1082             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1083             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1084
1085             /**************************
1086              * CALCULATE INTERACTIONS *
1087              **************************/
1088
1089             r10              = _mm_mul_ps(rsq10,rinv10);
1090             r10              = _mm_andnot_ps(dummy_mask,r10);
1091
1092             /* Compute parameters for interactions between i and j atoms */
1093             qq10             = _mm_mul_ps(iq1,jq0);
1094
1095             /* Calculate table index by multiplying r with table scale and truncate to integer */
1096             rt               = _mm_mul_ps(r10,vftabscale);
1097             vfitab           = _mm_cvttps_epi32(rt);
1098 #ifdef __XOP__
1099             vfeps            = _mm_frcz_ps(rt);
1100 #else
1101             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1102 #endif
1103             twovfeps         = _mm_add_ps(vfeps,vfeps);
1104             vfitab           = _mm_slli_epi32(vfitab,2);
1105
1106             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1107             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1108             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1109             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1110             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1111             _MM_TRANSPOSE4_PS(Y,F,G,H);
1112             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1113             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1114             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1115
1116             fscal            = felec;
1117
1118             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1119
1120              /* Update vectorial force */
1121             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1122             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1123             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1124
1125             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1126             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1127             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             r20              = _mm_mul_ps(rsq20,rinv20);
1134             r20              = _mm_andnot_ps(dummy_mask,r20);
1135
1136             /* Compute parameters for interactions between i and j atoms */
1137             qq20             = _mm_mul_ps(iq2,jq0);
1138
1139             /* Calculate table index by multiplying r with table scale and truncate to integer */
1140             rt               = _mm_mul_ps(r20,vftabscale);
1141             vfitab           = _mm_cvttps_epi32(rt);
1142 #ifdef __XOP__
1143             vfeps            = _mm_frcz_ps(rt);
1144 #else
1145             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1146 #endif
1147             twovfeps         = _mm_add_ps(vfeps,vfeps);
1148             vfitab           = _mm_slli_epi32(vfitab,2);
1149
1150             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1151             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1152             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1153             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1154             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1155             _MM_TRANSPOSE4_PS(Y,F,G,H);
1156             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1157             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1158             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1159
1160             fscal            = felec;
1161
1162             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1163
1164              /* Update vectorial force */
1165             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1166             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1167             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1168
1169             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1170             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1171             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1172
1173             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1174             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1175             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1176             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1177
1178             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1179
1180             /* Inner loop uses 137 flops */
1181         }
1182
1183         /* End of innermost loop */
1184
1185         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1186                                               f+i_coord_offset,fshift+i_shift_offset);
1187
1188         /* Increment number of inner iterations */
1189         inneriter                  += j_index_end - j_index_start;
1190
1191         /* Outer loop uses 18 flops */
1192     }
1193
1194     /* Increment number of outer iterations */
1195     outeriter        += nri;
1196
1197     /* Update outer/inner flops */
1198
1199     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1200 }