src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_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_ElecRF_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_ElecRF_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     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 161
 162         fix0             = _mm_setzero_ps();
 163         fiy0             = _mm_setzero_ps();
 164         fiz0             = _mm_setzero_ps();
 165         fix1             = _mm_setzero_ps();
 166         fiy1             = _mm_setzero_ps();
 167         fiz1             = _mm_setzero_ps();
 168         fix2             = _mm_setzero_ps();
 169         fiy2             = _mm_setzero_ps();
 170         fiz2             = _mm_setzero_ps();
 171         fix3             = _mm_setzero_ps();
 172         fiy3             = _mm_setzero_ps();
 173         fiz3             = _mm_setzero_ps();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_ps();
 177         vvdwsum          = _mm_setzero_ps();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             jnrC             = jjnr[jidx+2];
 187             jnrD             = jjnr[jidx+3];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190             j_coord_offsetC  = DIM*jnrC;
 191             j_coord_offsetD  = DIM*jnrD;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               x+j_coord_offsetC,x+j_coord_offsetD,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_ps(ix0,jx0);
 200             dy00             = _mm_sub_ps(iy0,jy0);
 201             dz00             = _mm_sub_ps(iz0,jz0);
 202             dx10             = _mm_sub_ps(ix1,jx0);
 203             dy10             = _mm_sub_ps(iy1,jy0);
 204             dz10             = _mm_sub_ps(iz1,jz0);
 205             dx20             = _mm_sub_ps(ix2,jx0);
 206             dy20             = _mm_sub_ps(iy2,jy0);
 207             dz20             = _mm_sub_ps(iz2,jz0);
 208             dx30             = _mm_sub_ps(ix3,jx0);
 209             dy30             = _mm_sub_ps(iy3,jy0);
 210             dz30             = _mm_sub_ps(iz3,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 214             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 215             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 216             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 217
 218             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 219             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 220             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 221             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 222
 223             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 224             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 225             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 229                                                               charge+jnrC+0,charge+jnrD+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234
 235             fjx0             = _mm_setzero_ps();
 236             fjy0             = _mm_setzero_ps();
 237             fjz0             = _mm_setzero_ps();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             r00              = _mm_mul_ps(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 247                                          vdwparam+vdwioffset0+vdwjidx0B,
 248                                          vdwparam+vdwioffset0+vdwjidx0C,
 249                                          vdwparam+vdwioffset0+vdwjidx0D,
 250                                          &c6_00,&c12_00);
 251
 252             /* Calculate table index by multiplying r with table scale and truncate to integer */
 253             rt               = _mm_mul_ps(r00,vftabscale);
 254             vfitab           = _mm_cvttps_epi32(rt);
 255 #ifdef __XOP__
 256             vfeps            = _mm_frcz_ps(rt);
 257 #else
 258             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 259 #endif
 260             twovfeps         = _mm_add_ps(vfeps,vfeps);
 261             vfitab           = _mm_slli_epi32(vfitab,3);
 262
 263             /* CUBIC SPLINE TABLE DISPERSION */
 264             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 265             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 266             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 267             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 268             _MM_TRANSPOSE4_PS(Y,F,G,H);
 269             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 270             VV               = _mm_macc_ps(vfeps,Fp,Y);
 271             vvdw6            = _mm_mul_ps(c6_00,VV);
 272             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 273             fvdw6            = _mm_mul_ps(c6_00,FF);
 274
 275             /* CUBIC SPLINE TABLE REPULSION */
 276             vfitab           = _mm_add_epi32(vfitab,ifour);
 277             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 278             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 279             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 280             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 281             _MM_TRANSPOSE4_PS(Y,F,G,H);
 282             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 283             VV               = _mm_macc_ps(vfeps,Fp,Y);
 284             vvdw12           = _mm_mul_ps(c12_00,VV);
 285             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 286             fvdw12           = _mm_mul_ps(c12_00,FF);
 287             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 288             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 292
 293             fscal            = fvdw;
 294
 295              /* Update vectorial force */
 296             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 297             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 298             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 299
 300             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 301             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 302             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq10             = _mm_mul_ps(iq1,jq0);
 310
 311             /* REACTION-FIELD ELECTROSTATICS */
 312             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 313             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velecsum         = _mm_add_ps(velecsum,velec);
 317
 318             fscal            = felec;
 319
 320              /* Update vectorial force */
 321             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 322             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 323             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 324
 325             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 326             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 327             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             /* Compute parameters for interactions between i and j atoms */
 334             qq20             = _mm_mul_ps(iq2,jq0);
 335
 336             /* REACTION-FIELD ELECTROSTATICS */
 337             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 338             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velecsum         = _mm_add_ps(velecsum,velec);
 342
 343             fscal            = felec;
 344
 345              /* Update vectorial force */
 346             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 347             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 348             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 349
 350             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 351             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 352             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 353
 354             /**************************
 355              * CALCULATE INTERACTIONS *
 356              **************************/
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq30             = _mm_mul_ps(iq3,jq0);
 360
 361             /* REACTION-FIELD ELECTROSTATICS */
 362             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 363             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 364
 365             /* Update potential sum for this i atom from the interaction with this j atom. */
 366             velecsum         = _mm_add_ps(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370              /* Update vectorial force */
 371             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 372             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 373             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 374
 375             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 376             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 377             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 378
 379             fjptrA             = f+j_coord_offsetA;
 380             fjptrB             = f+j_coord_offsetB;
 381             fjptrC             = f+j_coord_offsetC;
 382             fjptrD             = f+j_coord_offsetD;
 383
 384             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 385
 386             /* Inner loop uses 164 flops */
 387         }
 388
 389         if(jidx<j_index_end)
 390         {
 391
 392             /* Get j neighbor index, and coordinate index */
 393             jnrlistA         = jjnr[jidx];
 394             jnrlistB         = jjnr[jidx+1];
 395             jnrlistC         = jjnr[jidx+2];
 396             jnrlistD         = jjnr[jidx+3];
 397             /* Sign of each element will be negative for non-real atoms.
 398              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 399              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 400              */
 401             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 402             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 403             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 404             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 405             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 406             j_coord_offsetA  = DIM*jnrA;
 407             j_coord_offsetB  = DIM*jnrB;
 408             j_coord_offsetC  = DIM*jnrC;
 409             j_coord_offsetD  = DIM*jnrD;
 410
 411             /* load j atom coordinates */
 412             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 413                                               x+j_coord_offsetC,x+j_coord_offsetD,
 414                                               &jx0,&jy0,&jz0);
 415
 416             /* Calculate displacement vector */
 417             dx00             = _mm_sub_ps(ix0,jx0);
 418             dy00             = _mm_sub_ps(iy0,jy0);
 419             dz00             = _mm_sub_ps(iz0,jz0);
 420             dx10             = _mm_sub_ps(ix1,jx0);
 421             dy10             = _mm_sub_ps(iy1,jy0);
 422             dz10             = _mm_sub_ps(iz1,jz0);
 423             dx20             = _mm_sub_ps(ix2,jx0);
 424             dy20             = _mm_sub_ps(iy2,jy0);
 425             dz20             = _mm_sub_ps(iz2,jz0);
 426             dx30             = _mm_sub_ps(ix3,jx0);
 427             dy30             = _mm_sub_ps(iy3,jy0);
 428             dz30             = _mm_sub_ps(iz3,jz0);
 429
 430             /* Calculate squared distance and things based on it */
 431             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 432             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 433             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 434             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 435
 436             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 437             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 438             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 439             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 440
 441             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 442             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 443             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 444
 445             /* Load parameters for j particles */
 446             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 447                                                               charge+jnrC+0,charge+jnrD+0);
 448             vdwjidx0A        = 2*vdwtype[jnrA+0];
 449             vdwjidx0B        = 2*vdwtype[jnrB+0];
 450             vdwjidx0C        = 2*vdwtype[jnrC+0];
 451             vdwjidx0D        = 2*vdwtype[jnrD+0];
 452
 453             fjx0             = _mm_setzero_ps();
 454             fjy0             = _mm_setzero_ps();
 455             fjz0             = _mm_setzero_ps();
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             r00              = _mm_mul_ps(rsq00,rinv00);
 462             r00              = _mm_andnot_ps(dummy_mask,r00);
 463
 464             /* Compute parameters for interactions between i and j atoms */
 465             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 466                                          vdwparam+vdwioffset0+vdwjidx0B,
 467                                          vdwparam+vdwioffset0+vdwjidx0C,
 468                                          vdwparam+vdwioffset0+vdwjidx0D,
 469                                          &c6_00,&c12_00);
 470
 471             /* Calculate table index by multiplying r with table scale and truncate to integer */
 472             rt               = _mm_mul_ps(r00,vftabscale);
 473             vfitab           = _mm_cvttps_epi32(rt);
 474 #ifdef __XOP__
 475             vfeps            = _mm_frcz_ps(rt);
 476 #else
 477             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 478 #endif
 479             twovfeps         = _mm_add_ps(vfeps,vfeps);
 480             vfitab           = _mm_slli_epi32(vfitab,3);
 481
 482             /* CUBIC SPLINE TABLE DISPERSION */
 483             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 484             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 485             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 486             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 487             _MM_TRANSPOSE4_PS(Y,F,G,H);
 488             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 489             VV               = _mm_macc_ps(vfeps,Fp,Y);
 490             vvdw6            = _mm_mul_ps(c6_00,VV);
 491             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 492             fvdw6            = _mm_mul_ps(c6_00,FF);
 493
 494             /* CUBIC SPLINE TABLE REPULSION */
 495             vfitab           = _mm_add_epi32(vfitab,ifour);
 496             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 497             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 498             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 499             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 500             _MM_TRANSPOSE4_PS(Y,F,G,H);
 501             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 502             VV               = _mm_macc_ps(vfeps,Fp,Y);
 503             vvdw12           = _mm_mul_ps(c12_00,VV);
 504             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 505             fvdw12           = _mm_mul_ps(c12_00,FF);
 506             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 507             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 508
 509             /* Update potential sum for this i atom from the interaction with this j atom. */
 510             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 511             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 512
 513             fscal            = fvdw;
 514
 515             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 516
 517              /* Update vectorial force */
 518             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 519             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 520             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 521
 522             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 523             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 524             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq10             = _mm_mul_ps(iq1,jq0);
 532
 533             /* REACTION-FIELD ELECTROSTATICS */
 534             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 535             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velec            = _mm_andnot_ps(dummy_mask,velec);
 539             velecsum         = _mm_add_ps(velecsum,velec);
 540
 541             fscal            = felec;
 542
 543             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 544
 545              /* Update vectorial force */
 546             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 547             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 548             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 549
 550             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 551             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 552             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             /* Compute parameters for interactions between i and j atoms */
 559             qq20             = _mm_mul_ps(iq2,jq0);
 560
 561             /* REACTION-FIELD ELECTROSTATICS */
 562             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 563             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 564
 565             /* Update potential sum for this i atom from the interaction with this j atom. */
 566             velec            = _mm_andnot_ps(dummy_mask,velec);
 567             velecsum         = _mm_add_ps(velecsum,velec);
 568
 569             fscal            = felec;
 570
 571             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 572
 573              /* Update vectorial force */
 574             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 575             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 576             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 577
 578             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 579             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 580             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 581
 582             /**************************
 583              * CALCULATE INTERACTIONS *
 584              **************************/
 585
 586             /* Compute parameters for interactions between i and j atoms */
 587             qq30             = _mm_mul_ps(iq3,jq0);
 588
 589             /* REACTION-FIELD ELECTROSTATICS */
 590             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 591             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velec            = _mm_andnot_ps(dummy_mask,velec);
 595             velecsum         = _mm_add_ps(velecsum,velec);
 596
 597             fscal            = felec;
 598
 599             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 600
 601              /* Update vectorial force */
 602             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 603             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 604             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 605
 606             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 607             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 608             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 609
 610             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 611             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 612             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 613             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 614
 615             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 616
 617             /* Inner loop uses 165 flops */
 618         }
 619
 620         /* End of innermost loop */
 621
 622         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 623                                               f+i_coord_offset,fshift+i_shift_offset);
 624
 625         ggid                        = gid[iidx];
 626         /* Update potential energies */
 627         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 628         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 629
 630         /* Increment number of inner iterations */
 631         inneriter                  += j_index_end - j_index_start;
 632
 633         /* Outer loop uses 26 flops */
 634     }
 635
 636     /* Increment number of outer iterations */
 637     outeriter        += nri;
 638
 639     /* Update outer/inner flops */
 640
 641     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*165);
 642 }
 643 /*
 644  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 645  * Electrostatics interaction: ReactionField
 646  * VdW interaction:            CubicSplineTable
 647  * Geometry:                   Water4-Particle
 648  * Calculate force/pot:        Force
 649  */
 650 void
 651 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 652                     (t_nblist * gmx_restrict                nlist,
 653                      rvec * gmx_restrict                    xx,
 654                      rvec * gmx_restrict                    ff,
 655                      t_forcerec * gmx_restrict              fr,
 656                      t_mdatoms * gmx_restrict               mdatoms,
 657                      nb_kernel_data_t * gmx_restrict        kernel_data,
 658                      t_nrnb * gmx_restrict                  nrnb)
 659 {
 660     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 661      * just 0 for non-waters.
 662      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 663      * jnr indices corresponding to data put in the four positions in the SIMD register.
 664      */
 665     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 666     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 667     int              jnrA,jnrB,jnrC,jnrD;
 668     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 669     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 670     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 671     real             rcutoff_scalar;
 672     real             *shiftvec,*fshift,*x,*f;
 673     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 674     real             scratch[4*DIM];
 675     __m128           fscal,rcutoff,rcutoff2,jidxall;
 676     int              vdwioffset0;
 677     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 678     int              vdwioffset1;
 679     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 680     int              vdwioffset2;
 681     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 682     int              vdwioffset3;
 683     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 684     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 685     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 686     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 687     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 688     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 689     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 690     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 691     real             *charge;
 692     int              nvdwtype;
 693     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 694     int              *vdwtype;
 695     real             *vdwparam;
 696     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 697     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 698     __m128i          vfitab;
 699     __m128i          ifour       = _mm_set1_epi32(4);
 700     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 701     real             *vftab;
 702     __m128           dummy_mask,cutoff_mask;
 703     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 704     __m128           one     = _mm_set1_ps(1.0);
 705     __m128           two     = _mm_set1_ps(2.0);
 706     x                = xx[0];
 707     f                = ff[0];
 708
 709     nri              = nlist->nri;
 710     iinr             = nlist->iinr;
 711     jindex           = nlist->jindex;
 712     jjnr             = nlist->jjnr;
 713     shiftidx         = nlist->shift;
 714     gid              = nlist->gid;
 715     shiftvec         = fr->shift_vec[0];
 716     fshift           = fr->fshift[0];
 717     facel            = _mm_set1_ps(fr->epsfac);
 718     charge           = mdatoms->chargeA;
 719     krf              = _mm_set1_ps(fr->ic->k_rf);
 720     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 721     crf              = _mm_set1_ps(fr->ic->c_rf);
 722     nvdwtype         = fr->ntype;
 723     vdwparam         = fr->nbfp;
 724     vdwtype          = mdatoms->typeA;
 725
 726     vftab            = kernel_data->table_vdw->data;
 727     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 728
 729     /* Setup water-specific parameters */
 730     inr              = nlist->iinr[0];
 731     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 732     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 733     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 734     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 735
 736     /* Avoid stupid compiler warnings */
 737     jnrA = jnrB = jnrC = jnrD = 0;
 738     j_coord_offsetA = 0;
 739     j_coord_offsetB = 0;
 740     j_coord_offsetC = 0;
 741     j_coord_offsetD = 0;
 742
 743     outeriter        = 0;
 744     inneriter        = 0;
 745
 746     for(iidx=0;iidx<4*DIM;iidx++)
 747     {
 748         scratch[iidx] = 0.0;
 749     }
 750
 751     /* Start outer loop over neighborlists */
 752     for(iidx=0; iidx<nri; iidx++)
 753     {
 754         /* Load shift vector for this list */
 755         i_shift_offset   = DIM*shiftidx[iidx];
 756
 757         /* Load limits for loop over neighbors */
 758         j_index_start    = jindex[iidx];
 759         j_index_end      = jindex[iidx+1];
 760
 761         /* Get outer coordinate index */
 762         inr              = iinr[iidx];
 763         i_coord_offset   = DIM*inr;
 764
 765         /* Load i particle coords and add shift vector */
 766         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 767                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 768
 769         fix0             = _mm_setzero_ps();
 770         fiy0             = _mm_setzero_ps();
 771         fiz0             = _mm_setzero_ps();
 772         fix1             = _mm_setzero_ps();
 773         fiy1             = _mm_setzero_ps();
 774         fiz1             = _mm_setzero_ps();
 775         fix2             = _mm_setzero_ps();
 776         fiy2             = _mm_setzero_ps();
 777         fiz2             = _mm_setzero_ps();
 778         fix3             = _mm_setzero_ps();
 779         fiy3             = _mm_setzero_ps();
 780         fiz3             = _mm_setzero_ps();
 781
 782         /* Start inner kernel loop */
 783         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 784         {
 785
 786             /* Get j neighbor index, and coordinate index */
 787             jnrA             = jjnr[jidx];
 788             jnrB             = jjnr[jidx+1];
 789             jnrC             = jjnr[jidx+2];
 790             jnrD             = jjnr[jidx+3];
 791             j_coord_offsetA  = DIM*jnrA;
 792             j_coord_offsetB  = DIM*jnrB;
 793             j_coord_offsetC  = DIM*jnrC;
 794             j_coord_offsetD  = DIM*jnrD;
 795
 796             /* load j atom coordinates */
 797             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 798                                               x+j_coord_offsetC,x+j_coord_offsetD,
 799                                               &jx0,&jy0,&jz0);
 800
 801             /* Calculate displacement vector */
 802             dx00             = _mm_sub_ps(ix0,jx0);
 803             dy00             = _mm_sub_ps(iy0,jy0);
 804             dz00             = _mm_sub_ps(iz0,jz0);
 805             dx10             = _mm_sub_ps(ix1,jx0);
 806             dy10             = _mm_sub_ps(iy1,jy0);
 807             dz10             = _mm_sub_ps(iz1,jz0);
 808             dx20             = _mm_sub_ps(ix2,jx0);
 809             dy20             = _mm_sub_ps(iy2,jy0);
 810             dz20             = _mm_sub_ps(iz2,jz0);
 811             dx30             = _mm_sub_ps(ix3,jx0);
 812             dy30             = _mm_sub_ps(iy3,jy0);
 813             dz30             = _mm_sub_ps(iz3,jz0);
 814
 815             /* Calculate squared distance and things based on it */
 816             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 817             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 818             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 819             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 820
 821             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 822             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 823             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 824             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 825
 826             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 827             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 828             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 829
 830             /* Load parameters for j particles */
 831             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 832                                                               charge+jnrC+0,charge+jnrD+0);
 833             vdwjidx0A        = 2*vdwtype[jnrA+0];
 834             vdwjidx0B        = 2*vdwtype[jnrB+0];
 835             vdwjidx0C        = 2*vdwtype[jnrC+0];
 836             vdwjidx0D        = 2*vdwtype[jnrD+0];
 837
 838             fjx0             = _mm_setzero_ps();
 839             fjy0             = _mm_setzero_ps();
 840             fjz0             = _mm_setzero_ps();
 841
 842             /**************************
 843              * CALCULATE INTERACTIONS *
 844              **************************/
 845
 846             r00              = _mm_mul_ps(rsq00,rinv00);
 847
 848             /* Compute parameters for interactions between i and j atoms */
 849             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 850                                          vdwparam+vdwioffset0+vdwjidx0B,
 851                                          vdwparam+vdwioffset0+vdwjidx0C,
 852                                          vdwparam+vdwioffset0+vdwjidx0D,
 853                                          &c6_00,&c12_00);
 854
 855             /* Calculate table index by multiplying r with table scale and truncate to integer */
 856             rt               = _mm_mul_ps(r00,vftabscale);
 857             vfitab           = _mm_cvttps_epi32(rt);
 858 #ifdef __XOP__
 859             vfeps            = _mm_frcz_ps(rt);
 860 #else
 861             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 862 #endif
 863             twovfeps         = _mm_add_ps(vfeps,vfeps);
 864             vfitab           = _mm_slli_epi32(vfitab,3);
 865
 866             /* CUBIC SPLINE TABLE DISPERSION */
 867             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 868             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 869             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 870             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 871             _MM_TRANSPOSE4_PS(Y,F,G,H);
 872             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 873             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 874             fvdw6            = _mm_mul_ps(c6_00,FF);
 875
 876             /* CUBIC SPLINE TABLE REPULSION */
 877             vfitab           = _mm_add_epi32(vfitab,ifour);
 878             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 879             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 880             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 881             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 882             _MM_TRANSPOSE4_PS(Y,F,G,H);
 883             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 884             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 885             fvdw12           = _mm_mul_ps(c12_00,FF);
 886             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 887
 888             fscal            = fvdw;
 889
 890              /* Update vectorial force */
 891             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 892             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 893             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 894
 895             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 896             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 897             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 898
 899             /**************************
 900              * CALCULATE INTERACTIONS *
 901              **************************/
 902
 903             /* Compute parameters for interactions between i and j atoms */
 904             qq10             = _mm_mul_ps(iq1,jq0);
 905
 906             /* REACTION-FIELD ELECTROSTATICS */
 907             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 908
 909             fscal            = felec;
 910
 911              /* Update vectorial force */
 912             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 913             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 914             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 915
 916             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 917             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 918             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 919
 920             /**************************
 921              * CALCULATE INTERACTIONS *
 922              **************************/
 923
 924             /* Compute parameters for interactions between i and j atoms */
 925             qq20             = _mm_mul_ps(iq2,jq0);
 926
 927             /* REACTION-FIELD ELECTROSTATICS */
 928             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 929
 930             fscal            = felec;
 931
 932              /* Update vectorial force */
 933             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 934             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 935             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 936
 937             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 938             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 939             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 940
 941             /**************************
 942              * CALCULATE INTERACTIONS *
 943              **************************/
 944
 945             /* Compute parameters for interactions between i and j atoms */
 946             qq30             = _mm_mul_ps(iq3,jq0);
 947
 948             /* REACTION-FIELD ELECTROSTATICS */
 949             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 950
 951             fscal            = felec;
 952
 953              /* Update vectorial force */
 954             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 955             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 956             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 957
 958             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 959             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 960             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 961
 962             fjptrA             = f+j_coord_offsetA;
 963             fjptrB             = f+j_coord_offsetB;
 964             fjptrC             = f+j_coord_offsetC;
 965             fjptrD             = f+j_coord_offsetD;
 966
 967             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 968
 969             /* Inner loop uses 141 flops */
 970         }
 971
 972         if(jidx<j_index_end)
 973         {
 974
 975             /* Get j neighbor index, and coordinate index */
 976             jnrlistA         = jjnr[jidx];
 977             jnrlistB         = jjnr[jidx+1];
 978             jnrlistC         = jjnr[jidx+2];
 979             jnrlistD         = jjnr[jidx+3];
 980             /* Sign of each element will be negative for non-real atoms.
 981              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 982              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 983              */
 984             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 985             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 986             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 987             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 988             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 989             j_coord_offsetA  = DIM*jnrA;
 990             j_coord_offsetB  = DIM*jnrB;
 991             j_coord_offsetC  = DIM*jnrC;
 992             j_coord_offsetD  = DIM*jnrD;
 993
 994             /* load j atom coordinates */
 995             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 996                                               x+j_coord_offsetC,x+j_coord_offsetD,
 997                                               &jx0,&jy0,&jz0);
 998
 999             /* Calculate displacement vector */
1000             dx00             = _mm_sub_ps(ix0,jx0);
1001             dy00             = _mm_sub_ps(iy0,jy0);
1002             dz00             = _mm_sub_ps(iz0,jz0);
1003             dx10             = _mm_sub_ps(ix1,jx0);
1004             dy10             = _mm_sub_ps(iy1,jy0);
1005             dz10             = _mm_sub_ps(iz1,jz0);
1006             dx20             = _mm_sub_ps(ix2,jx0);
1007             dy20             = _mm_sub_ps(iy2,jy0);
1008             dz20             = _mm_sub_ps(iz2,jz0);
1009             dx30             = _mm_sub_ps(ix3,jx0);
1010             dy30             = _mm_sub_ps(iy3,jy0);
1011             dz30             = _mm_sub_ps(iz3,jz0);
1012
1013             /* Calculate squared distance and things based on it */
1014             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1015             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1016             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1017             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1018
1019             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1020             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1021             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1022             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1023
1024             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1025             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1026             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1027
1028             /* Load parameters for j particles */
1029             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1030                                                               charge+jnrC+0,charge+jnrD+0);
1031             vdwjidx0A        = 2*vdwtype[jnrA+0];
1032             vdwjidx0B        = 2*vdwtype[jnrB+0];
1033             vdwjidx0C        = 2*vdwtype[jnrC+0];
1034             vdwjidx0D        = 2*vdwtype[jnrD+0];
1035
1036             fjx0             = _mm_setzero_ps();
1037             fjy0             = _mm_setzero_ps();
1038             fjz0             = _mm_setzero_ps();
1039
1040             /**************************
1041              * CALCULATE INTERACTIONS *
1042              **************************/
1043
1044             r00              = _mm_mul_ps(rsq00,rinv00);
1045             r00              = _mm_andnot_ps(dummy_mask,r00);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1049                                          vdwparam+vdwioffset0+vdwjidx0B,
1050                                          vdwparam+vdwioffset0+vdwjidx0C,
1051                                          vdwparam+vdwioffset0+vdwjidx0D,
1052                                          &c6_00,&c12_00);
1053
1054             /* Calculate table index by multiplying r with table scale and truncate to integer */
1055             rt               = _mm_mul_ps(r00,vftabscale);
1056             vfitab           = _mm_cvttps_epi32(rt);
1057 #ifdef __XOP__
1058             vfeps            = _mm_frcz_ps(rt);
1059 #else
1060             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1061 #endif
1062             twovfeps         = _mm_add_ps(vfeps,vfeps);
1063             vfitab           = _mm_slli_epi32(vfitab,3);
1064
1065             /* CUBIC SPLINE TABLE DISPERSION */
1066             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1067             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1068             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1069             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1070             _MM_TRANSPOSE4_PS(Y,F,G,H);
1071             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1072             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1073             fvdw6            = _mm_mul_ps(c6_00,FF);
1074
1075             /* CUBIC SPLINE TABLE REPULSION */
1076             vfitab           = _mm_add_epi32(vfitab,ifour);
1077             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1078             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1079             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1080             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1081             _MM_TRANSPOSE4_PS(Y,F,G,H);
1082             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1083             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1084             fvdw12           = _mm_mul_ps(c12_00,FF);
1085             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1086
1087             fscal            = fvdw;
1088
1089             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1090
1091              /* Update vectorial force */
1092             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1093             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1094             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1095
1096             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1097             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1098             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1099
1100             /**************************
1101              * CALCULATE INTERACTIONS *
1102              **************************/
1103
1104             /* Compute parameters for interactions between i and j atoms */
1105             qq10             = _mm_mul_ps(iq1,jq0);
1106
1107             /* REACTION-FIELD ELECTROSTATICS */
1108             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1109
1110             fscal            = felec;
1111
1112             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1113
1114              /* Update vectorial force */
1115             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1116             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1117             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1118
1119             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1120             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1121             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1122
1123             /**************************
1124              * CALCULATE INTERACTIONS *
1125              **************************/
1126
1127             /* Compute parameters for interactions between i and j atoms */
1128             qq20             = _mm_mul_ps(iq2,jq0);
1129
1130             /* REACTION-FIELD ELECTROSTATICS */
1131             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1132
1133             fscal            = felec;
1134
1135             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1136
1137              /* Update vectorial force */
1138             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1139             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1140             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1141
1142             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1143             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1144             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1145
1146             /**************************
1147              * CALCULATE INTERACTIONS *
1148              **************************/
1149
1150             /* Compute parameters for interactions between i and j atoms */
1151             qq30             = _mm_mul_ps(iq3,jq0);
1152
1153             /* REACTION-FIELD ELECTROSTATICS */
1154             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1155
1156             fscal            = felec;
1157
1158             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1159
1160              /* Update vectorial force */
1161             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1162             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1163             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1164
1165             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1166             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1167             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1168
1169             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1170             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1171             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1172             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1173
1174             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1175
1176             /* Inner loop uses 142 flops */
1177         }
1178
1179         /* End of innermost loop */
1180
1181         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1182                                               f+i_coord_offset,fshift+i_shift_offset);
1183
1184         /* Increment number of inner iterations */
1185         inneriter                  += j_index_end - j_index_start;
1186
1187         /* Outer loop uses 24 flops */
1188     }
1189
1190     /* Increment number of outer iterations */
1191     outeriter        += nri;
1192
1193     /* Update outer/inner flops */
1194
1195     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*142);
1196 }