src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 107     real             *vftab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     krf              = _mm_set1_ps(fr->ic->k_rf);
 126     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 127     crf              = _mm_set1_ps(fr->ic->c_rf);
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 138     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 139     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 143     rcutoff_scalar   = fr->rcoulomb;
 144     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 145     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = jnrC = jnrD = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151     j_coord_offsetC = 0;
 152     j_coord_offsetD = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 179
 180         fix0             = _mm_setzero_ps();
 181         fiy0             = _mm_setzero_ps();
 182         fiz0             = _mm_setzero_ps();
 183         fix1             = _mm_setzero_ps();
 184         fiy1             = _mm_setzero_ps();
 185         fiz1             = _mm_setzero_ps();
 186         fix2             = _mm_setzero_ps();
 187         fiy2             = _mm_setzero_ps();
 188         fiz2             = _mm_setzero_ps();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm_setzero_ps();
 192         vvdwsum          = _mm_setzero_ps();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                               x+j_coord_offsetC,x+j_coord_offsetD,
 211                                               &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm_sub_ps(ix0,jx0);
 215             dy00             = _mm_sub_ps(iy0,jy0);
 216             dz00             = _mm_sub_ps(iz0,jz0);
 217             dx10             = _mm_sub_ps(ix1,jx0);
 218             dy10             = _mm_sub_ps(iy1,jy0);
 219             dz10             = _mm_sub_ps(iz1,jz0);
 220             dx20             = _mm_sub_ps(ix2,jx0);
 221             dy20             = _mm_sub_ps(iy2,jy0);
 222             dz20             = _mm_sub_ps(iz2,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 226             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 227             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 228
 229             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 230             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 231             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 232
 233             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 234             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 235             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 239                                                               charge+jnrC+0,charge+jnrD+0);
 240             vdwjidx0A        = 2*vdwtype[jnrA+0];
 241             vdwjidx0B        = 2*vdwtype[jnrB+0];
 242             vdwjidx0C        = 2*vdwtype[jnrC+0];
 243             vdwjidx0D        = 2*vdwtype[jnrD+0];
 244
 245             fjx0             = _mm_setzero_ps();
 246             fjy0             = _mm_setzero_ps();
 247             fjz0             = _mm_setzero_ps();
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             if (gmx_mm_any_lt(rsq00,rcutoff2))
 254             {
 255
 256             r00              = _mm_mul_ps(rsq00,rinv00);
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             qq00             = _mm_mul_ps(iq0,jq0);
 260             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 261                                          vdwparam+vdwioffset0+vdwjidx0B,
 262                                          vdwparam+vdwioffset0+vdwjidx0C,
 263                                          vdwparam+vdwioffset0+vdwjidx0D,
 264                                          &c6_00,&c12_00);
 265
 266             /* Calculate table index by multiplying r with table scale and truncate to integer */
 267             rt               = _mm_mul_ps(r00,vftabscale);
 268             vfitab           = _mm_cvttps_epi32(rt);
 269 #ifdef __XOP__
 270             vfeps            = _mm_frcz_ps(rt);
 271 #else
 272             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 273 #endif
 274             twovfeps         = _mm_add_ps(vfeps,vfeps);
 275             vfitab           = _mm_slli_epi32(vfitab,3);
 276
 277             /* REACTION-FIELD ELECTROSTATICS */
 278             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 279             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 280
 281             /* CUBIC SPLINE TABLE DISPERSION */
 282             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 283             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 284             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 285             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 286             _MM_TRANSPOSE4_PS(Y,F,G,H);
 287             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 288             VV               = _mm_macc_ps(vfeps,Fp,Y);
 289             vvdw6            = _mm_mul_ps(c6_00,VV);
 290             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 291             fvdw6            = _mm_mul_ps(c6_00,FF);
 292
 293             /* CUBIC SPLINE TABLE REPULSION */
 294             vfitab           = _mm_add_epi32(vfitab,ifour);
 295             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 296             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 297             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 298             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 299             _MM_TRANSPOSE4_PS(Y,F,G,H);
 300             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 301             VV               = _mm_macc_ps(vfeps,Fp,Y);
 302             vvdw12           = _mm_mul_ps(c12_00,VV);
 303             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 304             fvdw12           = _mm_mul_ps(c12_00,FF);
 305             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 306             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 307
 308             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             velec            = _mm_and_ps(velec,cutoff_mask);
 312             velecsum         = _mm_add_ps(velecsum,velec);
 313             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 314             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 315
 316             fscal            = _mm_add_ps(felec,fvdw);
 317
 318             fscal            = _mm_and_ps(fscal,cutoff_mask);
 319
 320              /* Update vectorial force */
 321             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 322             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 323             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 324
 325             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 326             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 327             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 328
 329             }
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             if (gmx_mm_any_lt(rsq10,rcutoff2))
 336             {
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq10             = _mm_mul_ps(iq1,jq0);
 340
 341             /* REACTION-FIELD ELECTROSTATICS */
 342             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 343             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 344
 345             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velec            = _mm_and_ps(velec,cutoff_mask);
 349             velecsum         = _mm_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             fscal            = _mm_and_ps(fscal,cutoff_mask);
 354
 355              /* Update vectorial force */
 356             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 357             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 358             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 359
 360             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 361             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 362             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 363
 364             }
 365
 366             /**************************
 367              * CALCULATE INTERACTIONS *
 368              **************************/
 369
 370             if (gmx_mm_any_lt(rsq20,rcutoff2))
 371             {
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq20             = _mm_mul_ps(iq2,jq0);
 375
 376             /* REACTION-FIELD ELECTROSTATICS */
 377             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 378             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 379
 380             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velec            = _mm_and_ps(velec,cutoff_mask);
 384             velecsum         = _mm_add_ps(velecsum,velec);
 385
 386             fscal            = felec;
 387
 388             fscal            = _mm_and_ps(fscal,cutoff_mask);
 389
 390              /* Update vectorial force */
 391             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 392             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 393             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 394
 395             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 396             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 397             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 398
 399             }
 400
 401             fjptrA             = f+j_coord_offsetA;
 402             fjptrB             = f+j_coord_offsetB;
 403             fjptrC             = f+j_coord_offsetC;
 404             fjptrD             = f+j_coord_offsetD;
 405
 406             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 407
 408             /* Inner loop uses 153 flops */
 409         }
 410
 411         if(jidx<j_index_end)
 412         {
 413
 414             /* Get j neighbor index, and coordinate index */
 415             jnrlistA         = jjnr[jidx];
 416             jnrlistB         = jjnr[jidx+1];
 417             jnrlistC         = jjnr[jidx+2];
 418             jnrlistD         = jjnr[jidx+3];
 419             /* Sign of each element will be negative for non-real atoms.
 420              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 421              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 422              */
 423             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 424             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 425             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 426             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 427             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 428             j_coord_offsetA  = DIM*jnrA;
 429             j_coord_offsetB  = DIM*jnrB;
 430             j_coord_offsetC  = DIM*jnrC;
 431             j_coord_offsetD  = DIM*jnrD;
 432
 433             /* load j atom coordinates */
 434             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 435                                               x+j_coord_offsetC,x+j_coord_offsetD,
 436                                               &jx0,&jy0,&jz0);
 437
 438             /* Calculate displacement vector */
 439             dx00             = _mm_sub_ps(ix0,jx0);
 440             dy00             = _mm_sub_ps(iy0,jy0);
 441             dz00             = _mm_sub_ps(iz0,jz0);
 442             dx10             = _mm_sub_ps(ix1,jx0);
 443             dy10             = _mm_sub_ps(iy1,jy0);
 444             dz10             = _mm_sub_ps(iz1,jz0);
 445             dx20             = _mm_sub_ps(ix2,jx0);
 446             dy20             = _mm_sub_ps(iy2,jy0);
 447             dz20             = _mm_sub_ps(iz2,jz0);
 448
 449             /* Calculate squared distance and things based on it */
 450             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 451             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 452             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 453
 454             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 455             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 456             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 457
 458             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 459             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 460             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 461
 462             /* Load parameters for j particles */
 463             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 464                                                               charge+jnrC+0,charge+jnrD+0);
 465             vdwjidx0A        = 2*vdwtype[jnrA+0];
 466             vdwjidx0B        = 2*vdwtype[jnrB+0];
 467             vdwjidx0C        = 2*vdwtype[jnrC+0];
 468             vdwjidx0D        = 2*vdwtype[jnrD+0];
 469
 470             fjx0             = _mm_setzero_ps();
 471             fjy0             = _mm_setzero_ps();
 472             fjz0             = _mm_setzero_ps();
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             if (gmx_mm_any_lt(rsq00,rcutoff2))
 479             {
 480
 481             r00              = _mm_mul_ps(rsq00,rinv00);
 482             r00              = _mm_andnot_ps(dummy_mask,r00);
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             qq00             = _mm_mul_ps(iq0,jq0);
 486             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 487                                          vdwparam+vdwioffset0+vdwjidx0B,
 488                                          vdwparam+vdwioffset0+vdwjidx0C,
 489                                          vdwparam+vdwioffset0+vdwjidx0D,
 490                                          &c6_00,&c12_00);
 491
 492             /* Calculate table index by multiplying r with table scale and truncate to integer */
 493             rt               = _mm_mul_ps(r00,vftabscale);
 494             vfitab           = _mm_cvttps_epi32(rt);
 495 #ifdef __XOP__
 496             vfeps            = _mm_frcz_ps(rt);
 497 #else
 498             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 499 #endif
 500             twovfeps         = _mm_add_ps(vfeps,vfeps);
 501             vfitab           = _mm_slli_epi32(vfitab,3);
 502
 503             /* REACTION-FIELD ELECTROSTATICS */
 504             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 505             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 506
 507             /* CUBIC SPLINE TABLE DISPERSION */
 508             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 509             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 510             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 511             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 512             _MM_TRANSPOSE4_PS(Y,F,G,H);
 513             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 514             VV               = _mm_macc_ps(vfeps,Fp,Y);
 515             vvdw6            = _mm_mul_ps(c6_00,VV);
 516             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 517             fvdw6            = _mm_mul_ps(c6_00,FF);
 518
 519             /* CUBIC SPLINE TABLE REPULSION */
 520             vfitab           = _mm_add_epi32(vfitab,ifour);
 521             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 522             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 523             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 524             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 525             _MM_TRANSPOSE4_PS(Y,F,G,H);
 526             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 527             VV               = _mm_macc_ps(vfeps,Fp,Y);
 528             vvdw12           = _mm_mul_ps(c12_00,VV);
 529             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 530             fvdw12           = _mm_mul_ps(c12_00,FF);
 531             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 532             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 533
 534             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 535
 536             /* Update potential sum for this i atom from the interaction with this j atom. */
 537             velec            = _mm_and_ps(velec,cutoff_mask);
 538             velec            = _mm_andnot_ps(dummy_mask,velec);
 539             velecsum         = _mm_add_ps(velecsum,velec);
 540             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 541             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 542             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 543
 544             fscal            = _mm_add_ps(felec,fvdw);
 545
 546             fscal            = _mm_and_ps(fscal,cutoff_mask);
 547
 548             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 549
 550              /* Update vectorial force */
 551             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 552             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 553             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 554
 555             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 556             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 557             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 558
 559             }
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             if (gmx_mm_any_lt(rsq10,rcutoff2))
 566             {
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             qq10             = _mm_mul_ps(iq1,jq0);
 570
 571             /* REACTION-FIELD ELECTROSTATICS */
 572             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 573             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 574
 575             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm_and_ps(velec,cutoff_mask);
 579             velec            = _mm_andnot_ps(dummy_mask,velec);
 580             velecsum         = _mm_add_ps(velecsum,velec);
 581
 582             fscal            = felec;
 583
 584             fscal            = _mm_and_ps(fscal,cutoff_mask);
 585
 586             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 587
 588              /* Update vectorial force */
 589             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 590             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 591             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 592
 593             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 594             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 595             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 596
 597             }
 598
 599             /**************************
 600              * CALCULATE INTERACTIONS *
 601              **************************/
 602
 603             if (gmx_mm_any_lt(rsq20,rcutoff2))
 604             {
 605
 606             /* Compute parameters for interactions between i and j atoms */
 607             qq20             = _mm_mul_ps(iq2,jq0);
 608
 609             /* REACTION-FIELD ELECTROSTATICS */
 610             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 611             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 612
 613             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 614
 615             /* Update potential sum for this i atom from the interaction with this j atom. */
 616             velec            = _mm_and_ps(velec,cutoff_mask);
 617             velec            = _mm_andnot_ps(dummy_mask,velec);
 618             velecsum         = _mm_add_ps(velecsum,velec);
 619
 620             fscal            = felec;
 621
 622             fscal            = _mm_and_ps(fscal,cutoff_mask);
 623
 624             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 625
 626              /* Update vectorial force */
 627             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 628             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 629             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 630
 631             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 632             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 633             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 634
 635             }
 636
 637             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 638             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 639             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 640             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 641
 642             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 643
 644             /* Inner loop uses 154 flops */
 645         }
 646
 647         /* End of innermost loop */
 648
 649         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 650                                               f+i_coord_offset,fshift+i_shift_offset);
 651
 652         ggid                        = gid[iidx];
 653         /* Update potential energies */
 654         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 655         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 656
 657         /* Increment number of inner iterations */
 658         inneriter                  += j_index_end - j_index_start;
 659
 660         /* Outer loop uses 20 flops */
 661     }
 662
 663     /* Increment number of outer iterations */
 664     outeriter        += nri;
 665
 666     /* Update outer/inner flops */
 667
 668     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 669 }
 670 /*
 671  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 672  * Electrostatics interaction: ReactionField
 673  * VdW interaction:            CubicSplineTable
 674  * Geometry:                   Water3-Particle
 675  * Calculate force/pot:        Force
 676  */
 677 void
 678 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 679                     (t_nblist                    * gmx_restrict       nlist,
 680                      rvec                        * gmx_restrict          xx,
 681                      rvec                        * gmx_restrict          ff,
 682                      t_forcerec                  * gmx_restrict          fr,
 683                      t_mdatoms                   * gmx_restrict     mdatoms,
 684                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 685                      t_nrnb                      * gmx_restrict        nrnb)
 686 {
 687     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 688      * just 0 for non-waters.
 689      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 690      * jnr indices corresponding to data put in the four positions in the SIMD register.
 691      */
 692     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 693     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 694     int              jnrA,jnrB,jnrC,jnrD;
 695     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 696     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 697     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 698     real             rcutoff_scalar;
 699     real             *shiftvec,*fshift,*x,*f;
 700     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 701     real             scratch[4*DIM];
 702     __m128           fscal,rcutoff,rcutoff2,jidxall;
 703     int              vdwioffset0;
 704     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 705     int              vdwioffset1;
 706     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 707     int              vdwioffset2;
 708     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 709     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 710     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 711     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 712     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 713     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 714     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 715     real             *charge;
 716     int              nvdwtype;
 717     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 718     int              *vdwtype;
 719     real             *vdwparam;
 720     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 721     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 722     __m128i          vfitab;
 723     __m128i          ifour       = _mm_set1_epi32(4);
 724     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 725     real             *vftab;
 726     __m128           dummy_mask,cutoff_mask;
 727     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 728     __m128           one     = _mm_set1_ps(1.0);
 729     __m128           two     = _mm_set1_ps(2.0);
 730     x                = xx[0];
 731     f                = ff[0];
 732
 733     nri              = nlist->nri;
 734     iinr             = nlist->iinr;
 735     jindex           = nlist->jindex;
 736     jjnr             = nlist->jjnr;
 737     shiftidx         = nlist->shift;
 738     gid              = nlist->gid;
 739     shiftvec         = fr->shift_vec[0];
 740     fshift           = fr->fshift[0];
 741     facel            = _mm_set1_ps(fr->epsfac);
 742     charge           = mdatoms->chargeA;
 743     krf              = _mm_set1_ps(fr->ic->k_rf);
 744     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 745     crf              = _mm_set1_ps(fr->ic->c_rf);
 746     nvdwtype         = fr->ntype;
 747     vdwparam         = fr->nbfp;
 748     vdwtype          = mdatoms->typeA;
 749
 750     vftab            = kernel_data->table_vdw->data;
 751     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 752
 753     /* Setup water-specific parameters */
 754     inr              = nlist->iinr[0];
 755     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 756     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 757     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 758     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 759
 760     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 761     rcutoff_scalar   = fr->rcoulomb;
 762     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 763     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 764
 765     /* Avoid stupid compiler warnings */
 766     jnrA = jnrB = jnrC = jnrD = 0;
 767     j_coord_offsetA = 0;
 768     j_coord_offsetB = 0;
 769     j_coord_offsetC = 0;
 770     j_coord_offsetD = 0;
 771
 772     outeriter        = 0;
 773     inneriter        = 0;
 774
 775     for(iidx=0;iidx<4*DIM;iidx++)
 776     {
 777         scratch[iidx] = 0.0;
 778     }
 779
 780     /* Start outer loop over neighborlists */
 781     for(iidx=0; iidx<nri; iidx++)
 782     {
 783         /* Load shift vector for this list */
 784         i_shift_offset   = DIM*shiftidx[iidx];
 785
 786         /* Load limits for loop over neighbors */
 787         j_index_start    = jindex[iidx];
 788         j_index_end      = jindex[iidx+1];
 789
 790         /* Get outer coordinate index */
 791         inr              = iinr[iidx];
 792         i_coord_offset   = DIM*inr;
 793
 794         /* Load i particle coords and add shift vector */
 795         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 796                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 797
 798         fix0             = _mm_setzero_ps();
 799         fiy0             = _mm_setzero_ps();
 800         fiz0             = _mm_setzero_ps();
 801         fix1             = _mm_setzero_ps();
 802         fiy1             = _mm_setzero_ps();
 803         fiz1             = _mm_setzero_ps();
 804         fix2             = _mm_setzero_ps();
 805         fiy2             = _mm_setzero_ps();
 806         fiz2             = _mm_setzero_ps();
 807
 808         /* Start inner kernel loop */
 809         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 810         {
 811
 812             /* Get j neighbor index, and coordinate index */
 813             jnrA             = jjnr[jidx];
 814             jnrB             = jjnr[jidx+1];
 815             jnrC             = jjnr[jidx+2];
 816             jnrD             = jjnr[jidx+3];
 817             j_coord_offsetA  = DIM*jnrA;
 818             j_coord_offsetB  = DIM*jnrB;
 819             j_coord_offsetC  = DIM*jnrC;
 820             j_coord_offsetD  = DIM*jnrD;
 821
 822             /* load j atom coordinates */
 823             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 824                                               x+j_coord_offsetC,x+j_coord_offsetD,
 825                                               &jx0,&jy0,&jz0);
 826
 827             /* Calculate displacement vector */
 828             dx00             = _mm_sub_ps(ix0,jx0);
 829             dy00             = _mm_sub_ps(iy0,jy0);
 830             dz00             = _mm_sub_ps(iz0,jz0);
 831             dx10             = _mm_sub_ps(ix1,jx0);
 832             dy10             = _mm_sub_ps(iy1,jy0);
 833             dz10             = _mm_sub_ps(iz1,jz0);
 834             dx20             = _mm_sub_ps(ix2,jx0);
 835             dy20             = _mm_sub_ps(iy2,jy0);
 836             dz20             = _mm_sub_ps(iz2,jz0);
 837
 838             /* Calculate squared distance and things based on it */
 839             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 840             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 841             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 842
 843             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 844             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 845             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 846
 847             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 848             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 849             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 850
 851             /* Load parameters for j particles */
 852             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 853                                                               charge+jnrC+0,charge+jnrD+0);
 854             vdwjidx0A        = 2*vdwtype[jnrA+0];
 855             vdwjidx0B        = 2*vdwtype[jnrB+0];
 856             vdwjidx0C        = 2*vdwtype[jnrC+0];
 857             vdwjidx0D        = 2*vdwtype[jnrD+0];
 858
 859             fjx0             = _mm_setzero_ps();
 860             fjy0             = _mm_setzero_ps();
 861             fjz0             = _mm_setzero_ps();
 862
 863             /**************************
 864              * CALCULATE INTERACTIONS *
 865              **************************/
 866
 867             if (gmx_mm_any_lt(rsq00,rcutoff2))
 868             {
 869
 870             r00              = _mm_mul_ps(rsq00,rinv00);
 871
 872             /* Compute parameters for interactions between i and j atoms */
 873             qq00             = _mm_mul_ps(iq0,jq0);
 874             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 875                                          vdwparam+vdwioffset0+vdwjidx0B,
 876                                          vdwparam+vdwioffset0+vdwjidx0C,
 877                                          vdwparam+vdwioffset0+vdwjidx0D,
 878                                          &c6_00,&c12_00);
 879
 880             /* Calculate table index by multiplying r with table scale and truncate to integer */
 881             rt               = _mm_mul_ps(r00,vftabscale);
 882             vfitab           = _mm_cvttps_epi32(rt);
 883 #ifdef __XOP__
 884             vfeps            = _mm_frcz_ps(rt);
 885 #else
 886             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 887 #endif
 888             twovfeps         = _mm_add_ps(vfeps,vfeps);
 889             vfitab           = _mm_slli_epi32(vfitab,3);
 890
 891             /* REACTION-FIELD ELECTROSTATICS */
 892             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 893
 894             /* CUBIC SPLINE TABLE DISPERSION */
 895             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 896             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 897             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 898             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 899             _MM_TRANSPOSE4_PS(Y,F,G,H);
 900             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 901             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 902             fvdw6            = _mm_mul_ps(c6_00,FF);
 903
 904             /* CUBIC SPLINE TABLE REPULSION */
 905             vfitab           = _mm_add_epi32(vfitab,ifour);
 906             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 907             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 908             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 909             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 910             _MM_TRANSPOSE4_PS(Y,F,G,H);
 911             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 912             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 913             fvdw12           = _mm_mul_ps(c12_00,FF);
 914             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 915
 916             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 917
 918             fscal            = _mm_add_ps(felec,fvdw);
 919
 920             fscal            = _mm_and_ps(fscal,cutoff_mask);
 921
 922              /* Update vectorial force */
 923             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 924             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 925             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 926
 927             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 928             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 929             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 930
 931             }
 932
 933             /**************************
 934              * CALCULATE INTERACTIONS *
 935              **************************/
 936
 937             if (gmx_mm_any_lt(rsq10,rcutoff2))
 938             {
 939
 940             /* Compute parameters for interactions between i and j atoms */
 941             qq10             = _mm_mul_ps(iq1,jq0);
 942
 943             /* REACTION-FIELD ELECTROSTATICS */
 944             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 945
 946             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 947
 948             fscal            = felec;
 949
 950             fscal            = _mm_and_ps(fscal,cutoff_mask);
 951
 952              /* Update vectorial force */
 953             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 954             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 955             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 956
 957             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 958             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 959             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 960
 961             }
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             if (gmx_mm_any_lt(rsq20,rcutoff2))
 968             {
 969
 970             /* Compute parameters for interactions between i and j atoms */
 971             qq20             = _mm_mul_ps(iq2,jq0);
 972
 973             /* REACTION-FIELD ELECTROSTATICS */
 974             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 975
 976             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 977
 978             fscal            = felec;
 979
 980             fscal            = _mm_and_ps(fscal,cutoff_mask);
 981
 982              /* Update vectorial force */
 983             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 984             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 985             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 986
 987             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 988             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 989             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 990
 991             }
 992
 993             fjptrA             = f+j_coord_offsetA;
 994             fjptrB             = f+j_coord_offsetB;
 995             fjptrC             = f+j_coord_offsetC;
 996             fjptrD             = f+j_coord_offsetD;
 997
 998             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 999
1000             /* Inner loop uses 126 flops */
1001         }
1002
1003         if(jidx<j_index_end)
1004         {
1005
1006             /* Get j neighbor index, and coordinate index */
1007             jnrlistA         = jjnr[jidx];
1008             jnrlistB         = jjnr[jidx+1];
1009             jnrlistC         = jjnr[jidx+2];
1010             jnrlistD         = jjnr[jidx+3];
1011             /* Sign of each element will be negative for non-real atoms.
1012              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1013              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1014              */
1015             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1016             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1017             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1018             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1019             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1020             j_coord_offsetA  = DIM*jnrA;
1021             j_coord_offsetB  = DIM*jnrB;
1022             j_coord_offsetC  = DIM*jnrC;
1023             j_coord_offsetD  = DIM*jnrD;
1024
1025             /* load j atom coordinates */
1026             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1027                                               x+j_coord_offsetC,x+j_coord_offsetD,
1028                                               &jx0,&jy0,&jz0);
1029
1030             /* Calculate displacement vector */
1031             dx00             = _mm_sub_ps(ix0,jx0);
1032             dy00             = _mm_sub_ps(iy0,jy0);
1033             dz00             = _mm_sub_ps(iz0,jz0);
1034             dx10             = _mm_sub_ps(ix1,jx0);
1035             dy10             = _mm_sub_ps(iy1,jy0);
1036             dz10             = _mm_sub_ps(iz1,jz0);
1037             dx20             = _mm_sub_ps(ix2,jx0);
1038             dy20             = _mm_sub_ps(iy2,jy0);
1039             dz20             = _mm_sub_ps(iz2,jz0);
1040
1041             /* Calculate squared distance and things based on it */
1042             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1043             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1044             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1045
1046             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1047             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1048             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1049
1050             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1051             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1052             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1053
1054             /* Load parameters for j particles */
1055             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1056                                                               charge+jnrC+0,charge+jnrD+0);
1057             vdwjidx0A        = 2*vdwtype[jnrA+0];
1058             vdwjidx0B        = 2*vdwtype[jnrB+0];
1059             vdwjidx0C        = 2*vdwtype[jnrC+0];
1060             vdwjidx0D        = 2*vdwtype[jnrD+0];
1061
1062             fjx0             = _mm_setzero_ps();
1063             fjy0             = _mm_setzero_ps();
1064             fjz0             = _mm_setzero_ps();
1065
1066             /**************************
1067              * CALCULATE INTERACTIONS *
1068              **************************/
1069
1070             if (gmx_mm_any_lt(rsq00,rcutoff2))
1071             {
1072
1073             r00              = _mm_mul_ps(rsq00,rinv00);
1074             r00              = _mm_andnot_ps(dummy_mask,r00);
1075
1076             /* Compute parameters for interactions between i and j atoms */
1077             qq00             = _mm_mul_ps(iq0,jq0);
1078             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1079                                          vdwparam+vdwioffset0+vdwjidx0B,
1080                                          vdwparam+vdwioffset0+vdwjidx0C,
1081                                          vdwparam+vdwioffset0+vdwjidx0D,
1082                                          &c6_00,&c12_00);
1083
1084             /* Calculate table index by multiplying r with table scale and truncate to integer */
1085             rt               = _mm_mul_ps(r00,vftabscale);
1086             vfitab           = _mm_cvttps_epi32(rt);
1087 #ifdef __XOP__
1088             vfeps            = _mm_frcz_ps(rt);
1089 #else
1090             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1091 #endif
1092             twovfeps         = _mm_add_ps(vfeps,vfeps);
1093             vfitab           = _mm_slli_epi32(vfitab,3);
1094
1095             /* REACTION-FIELD ELECTROSTATICS */
1096             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1097
1098             /* CUBIC SPLINE TABLE DISPERSION */
1099             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1100             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1101             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1102             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1103             _MM_TRANSPOSE4_PS(Y,F,G,H);
1104             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1105             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1106             fvdw6            = _mm_mul_ps(c6_00,FF);
1107
1108             /* CUBIC SPLINE TABLE REPULSION */
1109             vfitab           = _mm_add_epi32(vfitab,ifour);
1110             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1111             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1112             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1113             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1114             _MM_TRANSPOSE4_PS(Y,F,G,H);
1115             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1116             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1117             fvdw12           = _mm_mul_ps(c12_00,FF);
1118             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1119
1120             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1121
1122             fscal            = _mm_add_ps(felec,fvdw);
1123
1124             fscal            = _mm_and_ps(fscal,cutoff_mask);
1125
1126             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1127
1128              /* Update vectorial force */
1129             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1130             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1131             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1132
1133             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1134             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1135             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1136
1137             }
1138
1139             /**************************
1140              * CALCULATE INTERACTIONS *
1141              **************************/
1142
1143             if (gmx_mm_any_lt(rsq10,rcutoff2))
1144             {
1145
1146             /* Compute parameters for interactions between i and j atoms */
1147             qq10             = _mm_mul_ps(iq1,jq0);
1148
1149             /* REACTION-FIELD ELECTROSTATICS */
1150             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1151
1152             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1153
1154             fscal            = felec;
1155
1156             fscal            = _mm_and_ps(fscal,cutoff_mask);
1157
1158             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1159
1160              /* Update vectorial force */
1161             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1162             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1163             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1164
1165             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1166             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1167             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1168
1169             }
1170
1171             /**************************
1172              * CALCULATE INTERACTIONS *
1173              **************************/
1174
1175             if (gmx_mm_any_lt(rsq20,rcutoff2))
1176             {
1177
1178             /* Compute parameters for interactions between i and j atoms */
1179             qq20             = _mm_mul_ps(iq2,jq0);
1180
1181             /* REACTION-FIELD ELECTROSTATICS */
1182             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1183
1184             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1185
1186             fscal            = felec;
1187
1188             fscal            = _mm_and_ps(fscal,cutoff_mask);
1189
1190             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1191
1192              /* Update vectorial force */
1193             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1194             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1195             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1196
1197             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1198             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1199             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1200
1201             }
1202
1203             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1204             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1205             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1206             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1207
1208             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1209
1210             /* Inner loop uses 127 flops */
1211         }
1212
1213         /* End of innermost loop */
1214
1215         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1216                                               f+i_coord_offset,fshift+i_shift_offset);
1217
1218         /* Increment number of inner iterations */
1219         inneriter                  += j_index_end - j_index_start;
1220
1221         /* Outer loop uses 18 flops */
1222     }
1223
1224     /* Increment number of outer iterations */
1225     outeriter        += nri;
1226
1227     /* Update outer/inner flops */
1228
1229     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);
1230 }