src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_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_ElecRFCut_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_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              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
  94     real             *vftab;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     krf              = _mm_set1_ps(fr->ic->k_rf);
 113     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 114     crf              = _mm_set1_ps(fr->ic->c_rf);
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_vdw->data;
 120     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 121
 122     /* Setup water-specific parameters */
 123     inr              = nlist->iinr[0];
 124     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 125     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 126     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 127     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 128
 129     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 130     rcutoff_scalar   = fr->rcoulomb;
 131     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 132     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = jnrC = jnrD = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138     j_coord_offsetC = 0;
 139     j_coord_offsetD = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 165                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 166
 167         fix0             = _mm_setzero_ps();
 168         fiy0             = _mm_setzero_ps();
 169         fiz0             = _mm_setzero_ps();
 170         fix1             = _mm_setzero_ps();
 171         fiy1             = _mm_setzero_ps();
 172         fiz1             = _mm_setzero_ps();
 173         fix2             = _mm_setzero_ps();
 174         fiy2             = _mm_setzero_ps();
 175         fiz2             = _mm_setzero_ps();
 176         fix3             = _mm_setzero_ps();
 177         fiy3             = _mm_setzero_ps();
 178         fiz3             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               x+j_coord_offsetC,x+j_coord_offsetD,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_ps(ix0,jx0);
 205             dy00             = _mm_sub_ps(iy0,jy0);
 206             dz00             = _mm_sub_ps(iz0,jz0);
 207             dx10             = _mm_sub_ps(ix1,jx0);
 208             dy10             = _mm_sub_ps(iy1,jy0);
 209             dz10             = _mm_sub_ps(iz1,jz0);
 210             dx20             = _mm_sub_ps(ix2,jx0);
 211             dy20             = _mm_sub_ps(iy2,jy0);
 212             dz20             = _mm_sub_ps(iz2,jz0);
 213             dx30             = _mm_sub_ps(ix3,jx0);
 214             dy30             = _mm_sub_ps(iy3,jy0);
 215             dz30             = _mm_sub_ps(iz3,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 219             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 220             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 221             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 222
 223             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 224             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 226             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 227
 228             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 230             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 234                                                               charge+jnrC+0,charge+jnrD+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237             vdwjidx0C        = 2*vdwtype[jnrC+0];
 238             vdwjidx0D        = 2*vdwtype[jnrD+0];
 239
 240             fjx0             = _mm_setzero_ps();
 241             fjy0             = _mm_setzero_ps();
 242             fjz0             = _mm_setzero_ps();
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             r00              = _mm_mul_ps(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,
 253                                          vdwparam+vdwioffset0+vdwjidx0C,
 254                                          vdwparam+vdwioffset0+vdwjidx0D,
 255                                          &c6_00,&c12_00);
 256
 257             /* Calculate table index by multiplying r with table scale and truncate to integer */
 258             rt               = _mm_mul_ps(r00,vftabscale);
 259             vfitab           = _mm_cvttps_epi32(rt);
 260 #ifdef __XOP__
 261             vfeps            = _mm_frcz_ps(rt);
 262 #else
 263             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 264 #endif
 265             twovfeps         = _mm_add_ps(vfeps,vfeps);
 266             vfitab           = _mm_slli_epi32(vfitab,3);
 267
 268             /* CUBIC SPLINE TABLE DISPERSION */
 269             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 270             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 271             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 272             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 273             _MM_TRANSPOSE4_PS(Y,F,G,H);
 274             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 275             VV               = _mm_macc_ps(vfeps,Fp,Y);
 276             vvdw6            = _mm_mul_ps(c6_00,VV);
 277             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 278             fvdw6            = _mm_mul_ps(c6_00,FF);
 279
 280             /* CUBIC SPLINE TABLE REPULSION */
 281             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm_mul_ps(c12_00,VV);
 290             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 291             fvdw12           = _mm_mul_ps(c12_00,FF);
 292             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 293             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 297
 298             fscal            = fvdw;
 299
 300              /* Update vectorial force */
 301             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 302             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 303             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 304
 305             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 306             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 307             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 308
 309             /**************************
 310              * CALCULATE INTERACTIONS *
 311              **************************/
 312
 313             if (gmx_mm_any_lt(rsq10,rcutoff2))
 314             {
 315
 316             /* Compute parameters for interactions between i and j atoms */
 317             qq10             = _mm_mul_ps(iq1,jq0);
 318
 319             /* REACTION-FIELD ELECTROSTATICS */
 320             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 321             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 322
 323             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velec            = _mm_and_ps(velec,cutoff_mask);
 327             velecsum         = _mm_add_ps(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331             fscal            = _mm_and_ps(fscal,cutoff_mask);
 332
 333              /* Update vectorial force */
 334             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 335             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 336             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 337
 338             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 339             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 340             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 341
 342             }
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             if (gmx_mm_any_lt(rsq20,rcutoff2))
 349             {
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq20             = _mm_mul_ps(iq2,jq0);
 353
 354             /* REACTION-FIELD ELECTROSTATICS */
 355             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 356             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 357
 358             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velec            = _mm_and_ps(velec,cutoff_mask);
 362             velecsum         = _mm_add_ps(velecsum,velec);
 363
 364             fscal            = felec;
 365
 366             fscal            = _mm_and_ps(fscal,cutoff_mask);
 367
 368              /* Update vectorial force */
 369             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 370             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 371             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 372
 373             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 374             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 375             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 376
 377             }
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             if (gmx_mm_any_lt(rsq30,rcutoff2))
 384             {
 385
 386             /* Compute parameters for interactions between i and j atoms */
 387             qq30             = _mm_mul_ps(iq3,jq0);
 388
 389             /* REACTION-FIELD ELECTROSTATICS */
 390             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 391             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 392
 393             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velec            = _mm_and_ps(velec,cutoff_mask);
 397             velecsum         = _mm_add_ps(velecsum,velec);
 398
 399             fscal            = felec;
 400
 401             fscal            = _mm_and_ps(fscal,cutoff_mask);
 402
 403              /* Update vectorial force */
 404             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 405             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 406             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 407
 408             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 409             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 410             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 411
 412             }
 413
 414             fjptrA             = f+j_coord_offsetA;
 415             fjptrB             = f+j_coord_offsetB;
 416             fjptrC             = f+j_coord_offsetC;
 417             fjptrD             = f+j_coord_offsetD;
 418
 419             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 420
 421             /* Inner loop uses 176 flops */
 422         }
 423
 424         if(jidx<j_index_end)
 425         {
 426
 427             /* Get j neighbor index, and coordinate index */
 428             jnrlistA         = jjnr[jidx];
 429             jnrlistB         = jjnr[jidx+1];
 430             jnrlistC         = jjnr[jidx+2];
 431             jnrlistD         = jjnr[jidx+3];
 432             /* Sign of each element will be negative for non-real atoms.
 433              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 434              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 435              */
 436             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 437             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 438             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 439             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 440             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 441             j_coord_offsetA  = DIM*jnrA;
 442             j_coord_offsetB  = DIM*jnrB;
 443             j_coord_offsetC  = DIM*jnrC;
 444             j_coord_offsetD  = DIM*jnrD;
 445
 446             /* load j atom coordinates */
 447             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 448                                               x+j_coord_offsetC,x+j_coord_offsetD,
 449                                               &jx0,&jy0,&jz0);
 450
 451             /* Calculate displacement vector */
 452             dx00             = _mm_sub_ps(ix0,jx0);
 453             dy00             = _mm_sub_ps(iy0,jy0);
 454             dz00             = _mm_sub_ps(iz0,jz0);
 455             dx10             = _mm_sub_ps(ix1,jx0);
 456             dy10             = _mm_sub_ps(iy1,jy0);
 457             dz10             = _mm_sub_ps(iz1,jz0);
 458             dx20             = _mm_sub_ps(ix2,jx0);
 459             dy20             = _mm_sub_ps(iy2,jy0);
 460             dz20             = _mm_sub_ps(iz2,jz0);
 461             dx30             = _mm_sub_ps(ix3,jx0);
 462             dy30             = _mm_sub_ps(iy3,jy0);
 463             dz30             = _mm_sub_ps(iz3,jz0);
 464
 465             /* Calculate squared distance and things based on it */
 466             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 467             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 468             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 469             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 470
 471             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 472             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 473             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 474             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 475
 476             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 477             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 478             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 479
 480             /* Load parameters for j particles */
 481             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 482                                                               charge+jnrC+0,charge+jnrD+0);
 483             vdwjidx0A        = 2*vdwtype[jnrA+0];
 484             vdwjidx0B        = 2*vdwtype[jnrB+0];
 485             vdwjidx0C        = 2*vdwtype[jnrC+0];
 486             vdwjidx0D        = 2*vdwtype[jnrD+0];
 487
 488             fjx0             = _mm_setzero_ps();
 489             fjy0             = _mm_setzero_ps();
 490             fjz0             = _mm_setzero_ps();
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             r00              = _mm_mul_ps(rsq00,rinv00);
 497             r00              = _mm_andnot_ps(dummy_mask,r00);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 501                                          vdwparam+vdwioffset0+vdwjidx0B,
 502                                          vdwparam+vdwioffset0+vdwjidx0C,
 503                                          vdwparam+vdwioffset0+vdwjidx0D,
 504                                          &c6_00,&c12_00);
 505
 506             /* Calculate table index by multiplying r with table scale and truncate to integer */
 507             rt               = _mm_mul_ps(r00,vftabscale);
 508             vfitab           = _mm_cvttps_epi32(rt);
 509 #ifdef __XOP__
 510             vfeps            = _mm_frcz_ps(rt);
 511 #else
 512             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 513 #endif
 514             twovfeps         = _mm_add_ps(vfeps,vfeps);
 515             vfitab           = _mm_slli_epi32(vfitab,3);
 516
 517             /* CUBIC SPLINE TABLE DISPERSION */
 518             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 519             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 520             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 521             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 522             _MM_TRANSPOSE4_PS(Y,F,G,H);
 523             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 524             VV               = _mm_macc_ps(vfeps,Fp,Y);
 525             vvdw6            = _mm_mul_ps(c6_00,VV);
 526             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 527             fvdw6            = _mm_mul_ps(c6_00,FF);
 528
 529             /* CUBIC SPLINE TABLE REPULSION */
 530             vfitab           = _mm_add_epi32(vfitab,ifour);
 531             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 532             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 533             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 534             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 535             _MM_TRANSPOSE4_PS(Y,F,G,H);
 536             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 537             VV               = _mm_macc_ps(vfeps,Fp,Y);
 538             vvdw12           = _mm_mul_ps(c12_00,VV);
 539             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 540             fvdw12           = _mm_mul_ps(c12_00,FF);
 541             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 542             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 546             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 547
 548             fscal            = fvdw;
 549
 550             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 551
 552              /* Update vectorial force */
 553             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 554             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 555             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 556
 557             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 558             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 559             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 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             /**************************
 638              * CALCULATE INTERACTIONS *
 639              **************************/
 640
 641             if (gmx_mm_any_lt(rsq30,rcutoff2))
 642             {
 643
 644             /* Compute parameters for interactions between i and j atoms */
 645             qq30             = _mm_mul_ps(iq3,jq0);
 646
 647             /* REACTION-FIELD ELECTROSTATICS */
 648             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 649             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 650
 651             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 652
 653             /* Update potential sum for this i atom from the interaction with this j atom. */
 654             velec            = _mm_and_ps(velec,cutoff_mask);
 655             velec            = _mm_andnot_ps(dummy_mask,velec);
 656             velecsum         = _mm_add_ps(velecsum,velec);
 657
 658             fscal            = felec;
 659
 660             fscal            = _mm_and_ps(fscal,cutoff_mask);
 661
 662             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 663
 664              /* Update vectorial force */
 665             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 666             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 667             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 668
 669             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 670             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 671             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 672
 673             }
 674
 675             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 676             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 677             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 678             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 679
 680             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 681
 682             /* Inner loop uses 177 flops */
 683         }
 684
 685         /* End of innermost loop */
 686
 687         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 688                                               f+i_coord_offset,fshift+i_shift_offset);
 689
 690         ggid                        = gid[iidx];
 691         /* Update potential energies */
 692         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 693         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 694
 695         /* Increment number of inner iterations */
 696         inneriter                  += j_index_end - j_index_start;
 697
 698         /* Outer loop uses 26 flops */
 699     }
 700
 701     /* Increment number of outer iterations */
 702     outeriter        += nri;
 703
 704     /* Update outer/inner flops */
 705
 706     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*177);
 707 }
 708 /*
 709  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 710  * Electrostatics interaction: ReactionField
 711  * VdW interaction:            CubicSplineTable
 712  * Geometry:                   Water4-Particle
 713  * Calculate force/pot:        Force
 714  */
 715 void
 716 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 717                     (t_nblist * gmx_restrict                nlist,
 718                      rvec * gmx_restrict                    xx,
 719                      rvec * gmx_restrict                    ff,
 720                      t_forcerec * gmx_restrict              fr,
 721                      t_mdatoms * gmx_restrict               mdatoms,
 722                      nb_kernel_data_t * gmx_restrict        kernel_data,
 723                      t_nrnb * gmx_restrict                  nrnb)
 724 {
 725     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 726      * just 0 for non-waters.
 727      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 728      * jnr indices corresponding to data put in the four positions in the SIMD register.
 729      */
 730     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 731     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 732     int              jnrA,jnrB,jnrC,jnrD;
 733     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 734     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 735     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 736     real             rcutoff_scalar;
 737     real             *shiftvec,*fshift,*x,*f;
 738     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 739     real             scratch[4*DIM];
 740     __m128           fscal,rcutoff,rcutoff2,jidxall;
 741     int              vdwioffset0;
 742     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 743     int              vdwioffset1;
 744     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 745     int              vdwioffset2;
 746     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 747     int              vdwioffset3;
 748     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 749     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 750     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 751     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 752     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 753     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 754     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 755     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 756     real             *charge;
 757     int              nvdwtype;
 758     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 759     int              *vdwtype;
 760     real             *vdwparam;
 761     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 762     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 763     __m128i          vfitab;
 764     __m128i          ifour       = _mm_set1_epi32(4);
 765     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 766     real             *vftab;
 767     __m128           dummy_mask,cutoff_mask;
 768     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 769     __m128           one     = _mm_set1_ps(1.0);
 770     __m128           two     = _mm_set1_ps(2.0);
 771     x                = xx[0];
 772     f                = ff[0];
 773
 774     nri              = nlist->nri;
 775     iinr             = nlist->iinr;
 776     jindex           = nlist->jindex;
 777     jjnr             = nlist->jjnr;
 778     shiftidx         = nlist->shift;
 779     gid              = nlist->gid;
 780     shiftvec         = fr->shift_vec[0];
 781     fshift           = fr->fshift[0];
 782     facel            = _mm_set1_ps(fr->epsfac);
 783     charge           = mdatoms->chargeA;
 784     krf              = _mm_set1_ps(fr->ic->k_rf);
 785     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 786     crf              = _mm_set1_ps(fr->ic->c_rf);
 787     nvdwtype         = fr->ntype;
 788     vdwparam         = fr->nbfp;
 789     vdwtype          = mdatoms->typeA;
 790
 791     vftab            = kernel_data->table_vdw->data;
 792     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 793
 794     /* Setup water-specific parameters */
 795     inr              = nlist->iinr[0];
 796     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 797     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 798     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 799     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 800
 801     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 802     rcutoff_scalar   = fr->rcoulomb;
 803     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 804     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 805
 806     /* Avoid stupid compiler warnings */
 807     jnrA = jnrB = jnrC = jnrD = 0;
 808     j_coord_offsetA = 0;
 809     j_coord_offsetB = 0;
 810     j_coord_offsetC = 0;
 811     j_coord_offsetD = 0;
 812
 813     outeriter        = 0;
 814     inneriter        = 0;
 815
 816     for(iidx=0;iidx<4*DIM;iidx++)
 817     {
 818         scratch[iidx] = 0.0;
 819     }
 820
 821     /* Start outer loop over neighborlists */
 822     for(iidx=0; iidx<nri; iidx++)
 823     {
 824         /* Load shift vector for this list */
 825         i_shift_offset   = DIM*shiftidx[iidx];
 826
 827         /* Load limits for loop over neighbors */
 828         j_index_start    = jindex[iidx];
 829         j_index_end      = jindex[iidx+1];
 830
 831         /* Get outer coordinate index */
 832         inr              = iinr[iidx];
 833         i_coord_offset   = DIM*inr;
 834
 835         /* Load i particle coords and add shift vector */
 836         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 837                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 838
 839         fix0             = _mm_setzero_ps();
 840         fiy0             = _mm_setzero_ps();
 841         fiz0             = _mm_setzero_ps();
 842         fix1             = _mm_setzero_ps();
 843         fiy1             = _mm_setzero_ps();
 844         fiz1             = _mm_setzero_ps();
 845         fix2             = _mm_setzero_ps();
 846         fiy2             = _mm_setzero_ps();
 847         fiz2             = _mm_setzero_ps();
 848         fix3             = _mm_setzero_ps();
 849         fiy3             = _mm_setzero_ps();
 850         fiz3             = _mm_setzero_ps();
 851
 852         /* Start inner kernel loop */
 853         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 854         {
 855
 856             /* Get j neighbor index, and coordinate index */
 857             jnrA             = jjnr[jidx];
 858             jnrB             = jjnr[jidx+1];
 859             jnrC             = jjnr[jidx+2];
 860             jnrD             = jjnr[jidx+3];
 861             j_coord_offsetA  = DIM*jnrA;
 862             j_coord_offsetB  = DIM*jnrB;
 863             j_coord_offsetC  = DIM*jnrC;
 864             j_coord_offsetD  = DIM*jnrD;
 865
 866             /* load j atom coordinates */
 867             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 868                                               x+j_coord_offsetC,x+j_coord_offsetD,
 869                                               &jx0,&jy0,&jz0);
 870
 871             /* Calculate displacement vector */
 872             dx00             = _mm_sub_ps(ix0,jx0);
 873             dy00             = _mm_sub_ps(iy0,jy0);
 874             dz00             = _mm_sub_ps(iz0,jz0);
 875             dx10             = _mm_sub_ps(ix1,jx0);
 876             dy10             = _mm_sub_ps(iy1,jy0);
 877             dz10             = _mm_sub_ps(iz1,jz0);
 878             dx20             = _mm_sub_ps(ix2,jx0);
 879             dy20             = _mm_sub_ps(iy2,jy0);
 880             dz20             = _mm_sub_ps(iz2,jz0);
 881             dx30             = _mm_sub_ps(ix3,jx0);
 882             dy30             = _mm_sub_ps(iy3,jy0);
 883             dz30             = _mm_sub_ps(iz3,jz0);
 884
 885             /* Calculate squared distance and things based on it */
 886             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 887             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 888             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 889             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 890
 891             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 892             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 893             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 894             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 895
 896             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 897             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 898             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 899
 900             /* Load parameters for j particles */
 901             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 902                                                               charge+jnrC+0,charge+jnrD+0);
 903             vdwjidx0A        = 2*vdwtype[jnrA+0];
 904             vdwjidx0B        = 2*vdwtype[jnrB+0];
 905             vdwjidx0C        = 2*vdwtype[jnrC+0];
 906             vdwjidx0D        = 2*vdwtype[jnrD+0];
 907
 908             fjx0             = _mm_setzero_ps();
 909             fjy0             = _mm_setzero_ps();
 910             fjz0             = _mm_setzero_ps();
 911
 912             /**************************
 913              * CALCULATE INTERACTIONS *
 914              **************************/
 915
 916             r00              = _mm_mul_ps(rsq00,rinv00);
 917
 918             /* Compute parameters for interactions between i and j atoms */
 919             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 920                                          vdwparam+vdwioffset0+vdwjidx0B,
 921                                          vdwparam+vdwioffset0+vdwjidx0C,
 922                                          vdwparam+vdwioffset0+vdwjidx0D,
 923                                          &c6_00,&c12_00);
 924
 925             /* Calculate table index by multiplying r with table scale and truncate to integer */
 926             rt               = _mm_mul_ps(r00,vftabscale);
 927             vfitab           = _mm_cvttps_epi32(rt);
 928 #ifdef __XOP__
 929             vfeps            = _mm_frcz_ps(rt);
 930 #else
 931             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 932 #endif
 933             twovfeps         = _mm_add_ps(vfeps,vfeps);
 934             vfitab           = _mm_slli_epi32(vfitab,3);
 935
 936             /* CUBIC SPLINE TABLE DISPERSION */
 937             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 938             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 939             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 940             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 941             _MM_TRANSPOSE4_PS(Y,F,G,H);
 942             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 943             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 944             fvdw6            = _mm_mul_ps(c6_00,FF);
 945
 946             /* CUBIC SPLINE TABLE REPULSION */
 947             vfitab           = _mm_add_epi32(vfitab,ifour);
 948             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 949             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 950             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 951             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 952             _MM_TRANSPOSE4_PS(Y,F,G,H);
 953             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 954             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 955             fvdw12           = _mm_mul_ps(c12_00,FF);
 956             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 957
 958             fscal            = fvdw;
 959
 960              /* Update vectorial force */
 961             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 962             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 963             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 964
 965             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 966             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 967             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             if (gmx_mm_any_lt(rsq10,rcutoff2))
 974             {
 975
 976             /* Compute parameters for interactions between i and j atoms */
 977             qq10             = _mm_mul_ps(iq1,jq0);
 978
 979             /* REACTION-FIELD ELECTROSTATICS */
 980             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 981
 982             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 983
 984             fscal            = felec;
 985
 986             fscal            = _mm_and_ps(fscal,cutoff_mask);
 987
 988              /* Update vectorial force */
 989             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 990             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 991             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 992
 993             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 994             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 995             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 996
 997             }
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             if (gmx_mm_any_lt(rsq20,rcutoff2))
1004             {
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             qq20             = _mm_mul_ps(iq2,jq0);
1008
1009             /* REACTION-FIELD ELECTROSTATICS */
1010             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1011
1012             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1013
1014             fscal            = felec;
1015
1016             fscal            = _mm_and_ps(fscal,cutoff_mask);
1017
1018              /* Update vectorial force */
1019             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1020             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1021             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1022
1023             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1024             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1025             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1026
1027             }
1028
1029             /**************************
1030              * CALCULATE INTERACTIONS *
1031              **************************/
1032
1033             if (gmx_mm_any_lt(rsq30,rcutoff2))
1034             {
1035
1036             /* Compute parameters for interactions between i and j atoms */
1037             qq30             = _mm_mul_ps(iq3,jq0);
1038
1039             /* REACTION-FIELD ELECTROSTATICS */
1040             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1041
1042             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1043
1044             fscal            = felec;
1045
1046             fscal            = _mm_and_ps(fscal,cutoff_mask);
1047
1048              /* Update vectorial force */
1049             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1050             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1051             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1052
1053             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1054             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1055             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1056
1057             }
1058
1059             fjptrA             = f+j_coord_offsetA;
1060             fjptrB             = f+j_coord_offsetB;
1061             fjptrC             = f+j_coord_offsetC;
1062             fjptrD             = f+j_coord_offsetD;
1063
1064             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1065
1066             /* Inner loop uses 150 flops */
1067         }
1068
1069         if(jidx<j_index_end)
1070         {
1071
1072             /* Get j neighbor index, and coordinate index */
1073             jnrlistA         = jjnr[jidx];
1074             jnrlistB         = jjnr[jidx+1];
1075             jnrlistC         = jjnr[jidx+2];
1076             jnrlistD         = jjnr[jidx+3];
1077             /* Sign of each element will be negative for non-real atoms.
1078              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1079              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1080              */
1081             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1082             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1083             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1084             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1085             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1086             j_coord_offsetA  = DIM*jnrA;
1087             j_coord_offsetB  = DIM*jnrB;
1088             j_coord_offsetC  = DIM*jnrC;
1089             j_coord_offsetD  = DIM*jnrD;
1090
1091             /* load j atom coordinates */
1092             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1093                                               x+j_coord_offsetC,x+j_coord_offsetD,
1094                                               &jx0,&jy0,&jz0);
1095
1096             /* Calculate displacement vector */
1097             dx00             = _mm_sub_ps(ix0,jx0);
1098             dy00             = _mm_sub_ps(iy0,jy0);
1099             dz00             = _mm_sub_ps(iz0,jz0);
1100             dx10             = _mm_sub_ps(ix1,jx0);
1101             dy10             = _mm_sub_ps(iy1,jy0);
1102             dz10             = _mm_sub_ps(iz1,jz0);
1103             dx20             = _mm_sub_ps(ix2,jx0);
1104             dy20             = _mm_sub_ps(iy2,jy0);
1105             dz20             = _mm_sub_ps(iz2,jz0);
1106             dx30             = _mm_sub_ps(ix3,jx0);
1107             dy30             = _mm_sub_ps(iy3,jy0);
1108             dz30             = _mm_sub_ps(iz3,jz0);
1109
1110             /* Calculate squared distance and things based on it */
1111             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1112             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1113             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1114             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1115
1116             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1117             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1118             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1119             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1120
1121             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1122             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1123             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1124
1125             /* Load parameters for j particles */
1126             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1127                                                               charge+jnrC+0,charge+jnrD+0);
1128             vdwjidx0A        = 2*vdwtype[jnrA+0];
1129             vdwjidx0B        = 2*vdwtype[jnrB+0];
1130             vdwjidx0C        = 2*vdwtype[jnrC+0];
1131             vdwjidx0D        = 2*vdwtype[jnrD+0];
1132
1133             fjx0             = _mm_setzero_ps();
1134             fjy0             = _mm_setzero_ps();
1135             fjz0             = _mm_setzero_ps();
1136
1137             /**************************
1138              * CALCULATE INTERACTIONS *
1139              **************************/
1140
1141             r00              = _mm_mul_ps(rsq00,rinv00);
1142             r00              = _mm_andnot_ps(dummy_mask,r00);
1143
1144             /* Compute parameters for interactions between i and j atoms */
1145             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1146                                          vdwparam+vdwioffset0+vdwjidx0B,
1147                                          vdwparam+vdwioffset0+vdwjidx0C,
1148                                          vdwparam+vdwioffset0+vdwjidx0D,
1149                                          &c6_00,&c12_00);
1150
1151             /* Calculate table index by multiplying r with table scale and truncate to integer */
1152             rt               = _mm_mul_ps(r00,vftabscale);
1153             vfitab           = _mm_cvttps_epi32(rt);
1154 #ifdef __XOP__
1155             vfeps            = _mm_frcz_ps(rt);
1156 #else
1157             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1158 #endif
1159             twovfeps         = _mm_add_ps(vfeps,vfeps);
1160             vfitab           = _mm_slli_epi32(vfitab,3);
1161
1162             /* CUBIC SPLINE TABLE DISPERSION */
1163             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1164             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1165             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1166             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1167             _MM_TRANSPOSE4_PS(Y,F,G,H);
1168             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1169             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1170             fvdw6            = _mm_mul_ps(c6_00,FF);
1171
1172             /* CUBIC SPLINE TABLE REPULSION */
1173             vfitab           = _mm_add_epi32(vfitab,ifour);
1174             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1175             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1176             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1177             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1178             _MM_TRANSPOSE4_PS(Y,F,G,H);
1179             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1180             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1181             fvdw12           = _mm_mul_ps(c12_00,FF);
1182             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1183
1184             fscal            = fvdw;
1185
1186             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1187
1188              /* Update vectorial force */
1189             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1190             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1191             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1192
1193             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1194             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1195             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1196
1197             /**************************
1198              * CALCULATE INTERACTIONS *
1199              **************************/
1200
1201             if (gmx_mm_any_lt(rsq10,rcutoff2))
1202             {
1203
1204             /* Compute parameters for interactions between i and j atoms */
1205             qq10             = _mm_mul_ps(iq1,jq0);
1206
1207             /* REACTION-FIELD ELECTROSTATICS */
1208             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1209
1210             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1211
1212             fscal            = felec;
1213
1214             fscal            = _mm_and_ps(fscal,cutoff_mask);
1215
1216             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1217
1218              /* Update vectorial force */
1219             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1220             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1221             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1222
1223             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1224             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1225             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1226
1227             }
1228
1229             /**************************
1230              * CALCULATE INTERACTIONS *
1231              **************************/
1232
1233             if (gmx_mm_any_lt(rsq20,rcutoff2))
1234             {
1235
1236             /* Compute parameters for interactions between i and j atoms */
1237             qq20             = _mm_mul_ps(iq2,jq0);
1238
1239             /* REACTION-FIELD ELECTROSTATICS */
1240             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1241
1242             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1243
1244             fscal            = felec;
1245
1246             fscal            = _mm_and_ps(fscal,cutoff_mask);
1247
1248             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1249
1250              /* Update vectorial force */
1251             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1252             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1253             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1254
1255             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1256             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1257             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1258
1259             }
1260
1261             /**************************
1262              * CALCULATE INTERACTIONS *
1263              **************************/
1264
1265             if (gmx_mm_any_lt(rsq30,rcutoff2))
1266             {
1267
1268             /* Compute parameters for interactions between i and j atoms */
1269             qq30             = _mm_mul_ps(iq3,jq0);
1270
1271             /* REACTION-FIELD ELECTROSTATICS */
1272             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1273
1274             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1275
1276             fscal            = felec;
1277
1278             fscal            = _mm_and_ps(fscal,cutoff_mask);
1279
1280             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1281
1282              /* Update vectorial force */
1283             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1284             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1285             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1286
1287             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1288             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1289             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1290
1291             }
1292
1293             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1294             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1295             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1296             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1297
1298             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1299
1300             /* Inner loop uses 151 flops */
1301         }
1302
1303         /* End of innermost loop */
1304
1305         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1306                                               f+i_coord_offset,fshift+i_shift_offset);
1307
1308         /* Increment number of inner iterations */
1309         inneriter                  += j_index_end - j_index_start;
1310
1311         /* Outer loop uses 24 flops */
1312     }
1313
1314     /* Increment number of outer iterations */
1315     outeriter        += nri;
1316
1317     /* Update outer/inner flops */
1318
1319     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*151);
1320 }