src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
  91     real             *vftab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm_set1_ps(fr->ic->k_rf);
 110     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_ps(fr->ic->c_rf);
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 122     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 123     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 129     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 163
 164         fix0             = _mm_setzero_ps();
 165         fiy0             = _mm_setzero_ps();
 166         fiz0             = _mm_setzero_ps();
 167         fix1             = _mm_setzero_ps();
 168         fiy1             = _mm_setzero_ps();
 169         fiz1             = _mm_setzero_ps();
 170         fix2             = _mm_setzero_ps();
 171         fiy2             = _mm_setzero_ps();
 172         fiz2             = _mm_setzero_ps();
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_ps();
 176         vvdwsum          = _mm_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               x+j_coord_offsetC,x+j_coord_offsetD,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_ps(ix0,jx0);
 199             dy00             = _mm_sub_ps(iy0,jy0);
 200             dz00             = _mm_sub_ps(iz0,jz0);
 201             dx10             = _mm_sub_ps(ix1,jx0);
 202             dy10             = _mm_sub_ps(iy1,jy0);
 203             dz10             = _mm_sub_ps(iz1,jz0);
 204             dx20             = _mm_sub_ps(ix2,jx0);
 205             dy20             = _mm_sub_ps(iy2,jy0);
 206             dz20             = _mm_sub_ps(iz2,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 210             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 211             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 212
 213             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 214             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 215             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 216
 217             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 218             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 219             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 223                                                               charge+jnrC+0,charge+jnrD+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228
 229             fjx0             = _mm_setzero_ps();
 230             fjy0             = _mm_setzero_ps();
 231             fjz0             = _mm_setzero_ps();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             if (gmx_mm_any_lt(rsq00,rcutoff2))
 238             {
 239
 240             r00              = _mm_mul_ps(rsq00,rinv00);
 241
 242             /* Compute parameters for interactions between i and j atoms */
 243             qq00             = _mm_mul_ps(iq0,jq0);
 244             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 245                                          vdwparam+vdwioffset0+vdwjidx0B,
 246                                          vdwparam+vdwioffset0+vdwjidx0C,
 247                                          vdwparam+vdwioffset0+vdwjidx0D,
 248                                          &c6_00,&c12_00);
 249
 250             /* Calculate table index by multiplying r with table scale and truncate to integer */
 251             rt               = _mm_mul_ps(r00,vftabscale);
 252             vfitab           = _mm_cvttps_epi32(rt);
 253 #ifdef __XOP__
 254             vfeps            = _mm_frcz_ps(rt);
 255 #else
 256             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 257 #endif
 258             twovfeps         = _mm_add_ps(vfeps,vfeps);
 259             vfitab           = _mm_slli_epi32(vfitab,3);
 260
 261             /* REACTION-FIELD ELECTROSTATICS */
 262             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 263             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 264
 265             /* CUBIC SPLINE TABLE DISPERSION */
 266             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 267             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 268             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 269             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 270             _MM_TRANSPOSE4_PS(Y,F,G,H);
 271             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 272             VV               = _mm_macc_ps(vfeps,Fp,Y);
 273             vvdw6            = _mm_mul_ps(c6_00,VV);
 274             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 275             fvdw6            = _mm_mul_ps(c6_00,FF);
 276
 277             /* CUBIC SPLINE TABLE REPULSION */
 278             vfitab           = _mm_add_epi32(vfitab,ifour);
 279             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 280             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 281             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 282             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 283             _MM_TRANSPOSE4_PS(Y,F,G,H);
 284             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 285             VV               = _mm_macc_ps(vfeps,Fp,Y);
 286             vvdw12           = _mm_mul_ps(c12_00,VV);
 287             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 288             fvdw12           = _mm_mul_ps(c12_00,FF);
 289             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 290             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 291
 292             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velec            = _mm_and_ps(velec,cutoff_mask);
 296             velecsum         = _mm_add_ps(velecsum,velec);
 297             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 298             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 299
 300             fscal            = _mm_add_ps(felec,fvdw);
 301
 302             fscal            = _mm_and_ps(fscal,cutoff_mask);
 303
 304              /* Update vectorial force */
 305             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 306             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 307             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 308
 309             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 310             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 311             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 312
 313             }
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             if (gmx_mm_any_lt(rsq10,rcutoff2))
 320             {
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq10             = _mm_mul_ps(iq1,jq0);
 324
 325             /* REACTION-FIELD ELECTROSTATICS */
 326             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 327             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 328
 329             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 330
 331             /* Update potential sum for this i atom from the interaction with this j atom. */
 332             velec            = _mm_and_ps(velec,cutoff_mask);
 333             velecsum         = _mm_add_ps(velecsum,velec);
 334
 335             fscal            = felec;
 336
 337             fscal            = _mm_and_ps(fscal,cutoff_mask);
 338
 339              /* Update vectorial force */
 340             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 341             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 342             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 343
 344             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 345             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 346             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 347
 348             }
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             if (gmx_mm_any_lt(rsq20,rcutoff2))
 355             {
 356
 357             /* Compute parameters for interactions between i and j atoms */
 358             qq20             = _mm_mul_ps(iq2,jq0);
 359
 360             /* REACTION-FIELD ELECTROSTATICS */
 361             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 362             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 363
 364             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 365
 366             /* Update potential sum for this i atom from the interaction with this j atom. */
 367             velec            = _mm_and_ps(velec,cutoff_mask);
 368             velecsum         = _mm_add_ps(velecsum,velec);
 369
 370             fscal            = felec;
 371
 372             fscal            = _mm_and_ps(fscal,cutoff_mask);
 373
 374              /* Update vectorial force */
 375             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 376             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 377             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 378
 379             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 380             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 381             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 382
 383             }
 384
 385             fjptrA             = f+j_coord_offsetA;
 386             fjptrB             = f+j_coord_offsetB;
 387             fjptrC             = f+j_coord_offsetC;
 388             fjptrD             = f+j_coord_offsetD;
 389
 390             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 391
 392             /* Inner loop uses 153 flops */
 393         }
 394
 395         if(jidx<j_index_end)
 396         {
 397
 398             /* Get j neighbor index, and coordinate index */
 399             jnrlistA         = jjnr[jidx];
 400             jnrlistB         = jjnr[jidx+1];
 401             jnrlistC         = jjnr[jidx+2];
 402             jnrlistD         = jjnr[jidx+3];
 403             /* Sign of each element will be negative for non-real atoms.
 404              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 405              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 406              */
 407             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 408             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 409             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 410             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 411             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 412             j_coord_offsetA  = DIM*jnrA;
 413             j_coord_offsetB  = DIM*jnrB;
 414             j_coord_offsetC  = DIM*jnrC;
 415             j_coord_offsetD  = DIM*jnrD;
 416
 417             /* load j atom coordinates */
 418             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 419                                               x+j_coord_offsetC,x+j_coord_offsetD,
 420                                               &jx0,&jy0,&jz0);
 421
 422             /* Calculate displacement vector */
 423             dx00             = _mm_sub_ps(ix0,jx0);
 424             dy00             = _mm_sub_ps(iy0,jy0);
 425             dz00             = _mm_sub_ps(iz0,jz0);
 426             dx10             = _mm_sub_ps(ix1,jx0);
 427             dy10             = _mm_sub_ps(iy1,jy0);
 428             dz10             = _mm_sub_ps(iz1,jz0);
 429             dx20             = _mm_sub_ps(ix2,jx0);
 430             dy20             = _mm_sub_ps(iy2,jy0);
 431             dz20             = _mm_sub_ps(iz2,jz0);
 432
 433             /* Calculate squared distance and things based on it */
 434             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 435             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 436             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 437
 438             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 439             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 440             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 441
 442             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 443             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 444             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 445
 446             /* Load parameters for j particles */
 447             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 448                                                               charge+jnrC+0,charge+jnrD+0);
 449             vdwjidx0A        = 2*vdwtype[jnrA+0];
 450             vdwjidx0B        = 2*vdwtype[jnrB+0];
 451             vdwjidx0C        = 2*vdwtype[jnrC+0];
 452             vdwjidx0D        = 2*vdwtype[jnrD+0];
 453
 454             fjx0             = _mm_setzero_ps();
 455             fjy0             = _mm_setzero_ps();
 456             fjz0             = _mm_setzero_ps();
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             if (gmx_mm_any_lt(rsq00,rcutoff2))
 463             {
 464
 465             r00              = _mm_mul_ps(rsq00,rinv00);
 466             r00              = _mm_andnot_ps(dummy_mask,r00);
 467
 468             /* Compute parameters for interactions between i and j atoms */
 469             qq00             = _mm_mul_ps(iq0,jq0);
 470             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 471                                          vdwparam+vdwioffset0+vdwjidx0B,
 472                                          vdwparam+vdwioffset0+vdwjidx0C,
 473                                          vdwparam+vdwioffset0+vdwjidx0D,
 474                                          &c6_00,&c12_00);
 475
 476             /* Calculate table index by multiplying r with table scale and truncate to integer */
 477             rt               = _mm_mul_ps(r00,vftabscale);
 478             vfitab           = _mm_cvttps_epi32(rt);
 479 #ifdef __XOP__
 480             vfeps            = _mm_frcz_ps(rt);
 481 #else
 482             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 483 #endif
 484             twovfeps         = _mm_add_ps(vfeps,vfeps);
 485             vfitab           = _mm_slli_epi32(vfitab,3);
 486
 487             /* REACTION-FIELD ELECTROSTATICS */
 488             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 489             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 490
 491             /* CUBIC SPLINE TABLE DISPERSION */
 492             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 493             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 494             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 495             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 496             _MM_TRANSPOSE4_PS(Y,F,G,H);
 497             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 498             VV               = _mm_macc_ps(vfeps,Fp,Y);
 499             vvdw6            = _mm_mul_ps(c6_00,VV);
 500             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 501             fvdw6            = _mm_mul_ps(c6_00,FF);
 502
 503             /* CUBIC SPLINE TABLE REPULSION */
 504             vfitab           = _mm_add_epi32(vfitab,ifour);
 505             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 506             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 507             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 508             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 509             _MM_TRANSPOSE4_PS(Y,F,G,H);
 510             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 511             VV               = _mm_macc_ps(vfeps,Fp,Y);
 512             vvdw12           = _mm_mul_ps(c12_00,VV);
 513             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 514             fvdw12           = _mm_mul_ps(c12_00,FF);
 515             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 516             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 517
 518             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velec            = _mm_and_ps(velec,cutoff_mask);
 522             velec            = _mm_andnot_ps(dummy_mask,velec);
 523             velecsum         = _mm_add_ps(velecsum,velec);
 524             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 525             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 526             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 527
 528             fscal            = _mm_add_ps(felec,fvdw);
 529
 530             fscal            = _mm_and_ps(fscal,cutoff_mask);
 531
 532             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 533
 534              /* Update vectorial force */
 535             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 536             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 537             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 538
 539             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 540             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 541             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 542
 543             }
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             if (gmx_mm_any_lt(rsq10,rcutoff2))
 550             {
 551
 552             /* Compute parameters for interactions between i and j atoms */
 553             qq10             = _mm_mul_ps(iq1,jq0);
 554
 555             /* REACTION-FIELD ELECTROSTATICS */
 556             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 557             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 558
 559             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 560
 561             /* Update potential sum for this i atom from the interaction with this j atom. */
 562             velec            = _mm_and_ps(velec,cutoff_mask);
 563             velec            = _mm_andnot_ps(dummy_mask,velec);
 564             velecsum         = _mm_add_ps(velecsum,velec);
 565
 566             fscal            = felec;
 567
 568             fscal            = _mm_and_ps(fscal,cutoff_mask);
 569
 570             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 571
 572              /* Update vectorial force */
 573             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 574             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 575             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 576
 577             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 578             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 579             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 580
 581             }
 582
 583             /**************************
 584              * CALCULATE INTERACTIONS *
 585              **************************/
 586
 587             if (gmx_mm_any_lt(rsq20,rcutoff2))
 588             {
 589
 590             /* Compute parameters for interactions between i and j atoms */
 591             qq20             = _mm_mul_ps(iq2,jq0);
 592
 593             /* REACTION-FIELD ELECTROSTATICS */
 594             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 595             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 596
 597             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 598
 599             /* Update potential sum for this i atom from the interaction with this j atom. */
 600             velec            = _mm_and_ps(velec,cutoff_mask);
 601             velec            = _mm_andnot_ps(dummy_mask,velec);
 602             velecsum         = _mm_add_ps(velecsum,velec);
 603
 604             fscal            = felec;
 605
 606             fscal            = _mm_and_ps(fscal,cutoff_mask);
 607
 608             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 609
 610              /* Update vectorial force */
 611             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 612             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 613             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 614
 615             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 616             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 617             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 618
 619             }
 620
 621             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 622             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 623             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 624             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 625
 626             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 627
 628             /* Inner loop uses 154 flops */
 629         }
 630
 631         /* End of innermost loop */
 632
 633         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 634                                               f+i_coord_offset,fshift+i_shift_offset);
 635
 636         ggid                        = gid[iidx];
 637         /* Update potential energies */
 638         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 639         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 640
 641         /* Increment number of inner iterations */
 642         inneriter                  += j_index_end - j_index_start;
 643
 644         /* Outer loop uses 20 flops */
 645     }
 646
 647     /* Increment number of outer iterations */
 648     outeriter        += nri;
 649
 650     /* Update outer/inner flops */
 651
 652     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*154);
 653 }
 654 /*
 655  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 656  * Electrostatics interaction: ReactionField
 657  * VdW interaction:            CubicSplineTable
 658  * Geometry:                   Water3-Particle
 659  * Calculate force/pot:        Force
 660  */
 661 void
 662 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 663                     (t_nblist * gmx_restrict                nlist,
 664                      rvec * gmx_restrict                    xx,
 665                      rvec * gmx_restrict                    ff,
 666                      t_forcerec * gmx_restrict              fr,
 667                      t_mdatoms * gmx_restrict               mdatoms,
 668                      nb_kernel_data_t * gmx_restrict        kernel_data,
 669                      t_nrnb * gmx_restrict                  nrnb)
 670 {
 671     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 672      * just 0 for non-waters.
 673      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 674      * jnr indices corresponding to data put in the four positions in the SIMD register.
 675      */
 676     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 677     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 678     int              jnrA,jnrB,jnrC,jnrD;
 679     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 680     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 681     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 682     real             rcutoff_scalar;
 683     real             *shiftvec,*fshift,*x,*f;
 684     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 685     real             scratch[4*DIM];
 686     __m128           fscal,rcutoff,rcutoff2,jidxall;
 687     int              vdwioffset0;
 688     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 689     int              vdwioffset1;
 690     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 691     int              vdwioffset2;
 692     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 693     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 694     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 695     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 696     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 697     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 698     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 699     real             *charge;
 700     int              nvdwtype;
 701     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 702     int              *vdwtype;
 703     real             *vdwparam;
 704     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 705     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 706     __m128i          vfitab;
 707     __m128i          ifour       = _mm_set1_epi32(4);
 708     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 709     real             *vftab;
 710     __m128           dummy_mask,cutoff_mask;
 711     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 712     __m128           one     = _mm_set1_ps(1.0);
 713     __m128           two     = _mm_set1_ps(2.0);
 714     x                = xx[0];
 715     f                = ff[0];
 716
 717     nri              = nlist->nri;
 718     iinr             = nlist->iinr;
 719     jindex           = nlist->jindex;
 720     jjnr             = nlist->jjnr;
 721     shiftidx         = nlist->shift;
 722     gid              = nlist->gid;
 723     shiftvec         = fr->shift_vec[0];
 724     fshift           = fr->fshift[0];
 725     facel            = _mm_set1_ps(fr->epsfac);
 726     charge           = mdatoms->chargeA;
 727     krf              = _mm_set1_ps(fr->ic->k_rf);
 728     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 729     crf              = _mm_set1_ps(fr->ic->c_rf);
 730     nvdwtype         = fr->ntype;
 731     vdwparam         = fr->nbfp;
 732     vdwtype          = mdatoms->typeA;
 733
 734     vftab            = kernel_data->table_vdw->data;
 735     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 736
 737     /* Setup water-specific parameters */
 738     inr              = nlist->iinr[0];
 739     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 740     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 741     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 742     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 743
 744     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 745     rcutoff_scalar   = fr->rcoulomb;
 746     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 747     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 748
 749     /* Avoid stupid compiler warnings */
 750     jnrA = jnrB = jnrC = jnrD = 0;
 751     j_coord_offsetA = 0;
 752     j_coord_offsetB = 0;
 753     j_coord_offsetC = 0;
 754     j_coord_offsetD = 0;
 755
 756     outeriter        = 0;
 757     inneriter        = 0;
 758
 759     for(iidx=0;iidx<4*DIM;iidx++)
 760     {
 761         scratch[iidx] = 0.0;
 762     }
 763
 764     /* Start outer loop over neighborlists */
 765     for(iidx=0; iidx<nri; iidx++)
 766     {
 767         /* Load shift vector for this list */
 768         i_shift_offset   = DIM*shiftidx[iidx];
 769
 770         /* Load limits for loop over neighbors */
 771         j_index_start    = jindex[iidx];
 772         j_index_end      = jindex[iidx+1];
 773
 774         /* Get outer coordinate index */
 775         inr              = iinr[iidx];
 776         i_coord_offset   = DIM*inr;
 777
 778         /* Load i particle coords and add shift vector */
 779         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 780                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 781
 782         fix0             = _mm_setzero_ps();
 783         fiy0             = _mm_setzero_ps();
 784         fiz0             = _mm_setzero_ps();
 785         fix1             = _mm_setzero_ps();
 786         fiy1             = _mm_setzero_ps();
 787         fiz1             = _mm_setzero_ps();
 788         fix2             = _mm_setzero_ps();
 789         fiy2             = _mm_setzero_ps();
 790         fiz2             = _mm_setzero_ps();
 791
 792         /* Start inner kernel loop */
 793         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 794         {
 795
 796             /* Get j neighbor index, and coordinate index */
 797             jnrA             = jjnr[jidx];
 798             jnrB             = jjnr[jidx+1];
 799             jnrC             = jjnr[jidx+2];
 800             jnrD             = jjnr[jidx+3];
 801             j_coord_offsetA  = DIM*jnrA;
 802             j_coord_offsetB  = DIM*jnrB;
 803             j_coord_offsetC  = DIM*jnrC;
 804             j_coord_offsetD  = DIM*jnrD;
 805
 806             /* load j atom coordinates */
 807             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 808                                               x+j_coord_offsetC,x+j_coord_offsetD,
 809                                               &jx0,&jy0,&jz0);
 810
 811             /* Calculate displacement vector */
 812             dx00             = _mm_sub_ps(ix0,jx0);
 813             dy00             = _mm_sub_ps(iy0,jy0);
 814             dz00             = _mm_sub_ps(iz0,jz0);
 815             dx10             = _mm_sub_ps(ix1,jx0);
 816             dy10             = _mm_sub_ps(iy1,jy0);
 817             dz10             = _mm_sub_ps(iz1,jz0);
 818             dx20             = _mm_sub_ps(ix2,jx0);
 819             dy20             = _mm_sub_ps(iy2,jy0);
 820             dz20             = _mm_sub_ps(iz2,jz0);
 821
 822             /* Calculate squared distance and things based on it */
 823             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 824             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 825             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 826
 827             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 828             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 829             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 830
 831             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 832             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 833             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 834
 835             /* Load parameters for j particles */
 836             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 837                                                               charge+jnrC+0,charge+jnrD+0);
 838             vdwjidx0A        = 2*vdwtype[jnrA+0];
 839             vdwjidx0B        = 2*vdwtype[jnrB+0];
 840             vdwjidx0C        = 2*vdwtype[jnrC+0];
 841             vdwjidx0D        = 2*vdwtype[jnrD+0];
 842
 843             fjx0             = _mm_setzero_ps();
 844             fjy0             = _mm_setzero_ps();
 845             fjz0             = _mm_setzero_ps();
 846
 847             /**************************
 848              * CALCULATE INTERACTIONS *
 849              **************************/
 850
 851             if (gmx_mm_any_lt(rsq00,rcutoff2))
 852             {
 853
 854             r00              = _mm_mul_ps(rsq00,rinv00);
 855
 856             /* Compute parameters for interactions between i and j atoms */
 857             qq00             = _mm_mul_ps(iq0,jq0);
 858             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 859                                          vdwparam+vdwioffset0+vdwjidx0B,
 860                                          vdwparam+vdwioffset0+vdwjidx0C,
 861                                          vdwparam+vdwioffset0+vdwjidx0D,
 862                                          &c6_00,&c12_00);
 863
 864             /* Calculate table index by multiplying r with table scale and truncate to integer */
 865             rt               = _mm_mul_ps(r00,vftabscale);
 866             vfitab           = _mm_cvttps_epi32(rt);
 867 #ifdef __XOP__
 868             vfeps            = _mm_frcz_ps(rt);
 869 #else
 870             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 871 #endif
 872             twovfeps         = _mm_add_ps(vfeps,vfeps);
 873             vfitab           = _mm_slli_epi32(vfitab,3);
 874
 875             /* REACTION-FIELD ELECTROSTATICS */
 876             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 877
 878             /* CUBIC SPLINE TABLE DISPERSION */
 879             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 880             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 881             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 882             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 883             _MM_TRANSPOSE4_PS(Y,F,G,H);
 884             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 885             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 886             fvdw6            = _mm_mul_ps(c6_00,FF);
 887
 888             /* CUBIC SPLINE TABLE REPULSION */
 889             vfitab           = _mm_add_epi32(vfitab,ifour);
 890             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 891             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 892             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 893             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 894             _MM_TRANSPOSE4_PS(Y,F,G,H);
 895             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 896             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 897             fvdw12           = _mm_mul_ps(c12_00,FF);
 898             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 899
 900             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 901
 902             fscal            = _mm_add_ps(felec,fvdw);
 903
 904             fscal            = _mm_and_ps(fscal,cutoff_mask);
 905
 906              /* Update vectorial force */
 907             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 908             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 909             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 910
 911             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 912             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 913             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 914
 915             }
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             if (gmx_mm_any_lt(rsq10,rcutoff2))
 922             {
 923
 924             /* Compute parameters for interactions between i and j atoms */
 925             qq10             = _mm_mul_ps(iq1,jq0);
 926
 927             /* REACTION-FIELD ELECTROSTATICS */
 928             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 929
 930             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 931
 932             fscal            = felec;
 933
 934             fscal            = _mm_and_ps(fscal,cutoff_mask);
 935
 936              /* Update vectorial force */
 937             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 938             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 939             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 940
 941             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 942             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 943             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 944
 945             }
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             if (gmx_mm_any_lt(rsq20,rcutoff2))
 952             {
 953
 954             /* Compute parameters for interactions between i and j atoms */
 955             qq20             = _mm_mul_ps(iq2,jq0);
 956
 957             /* REACTION-FIELD ELECTROSTATICS */
 958             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 959
 960             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 961
 962             fscal            = felec;
 963
 964             fscal            = _mm_and_ps(fscal,cutoff_mask);
 965
 966              /* Update vectorial force */
 967             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 968             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 969             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 970
 971             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 972             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 973             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 974
 975             }
 976
 977             fjptrA             = f+j_coord_offsetA;
 978             fjptrB             = f+j_coord_offsetB;
 979             fjptrC             = f+j_coord_offsetC;
 980             fjptrD             = f+j_coord_offsetD;
 981
 982             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 983
 984             /* Inner loop uses 126 flops */
 985         }
 986
 987         if(jidx<j_index_end)
 988         {
 989
 990             /* Get j neighbor index, and coordinate index */
 991             jnrlistA         = jjnr[jidx];
 992             jnrlistB         = jjnr[jidx+1];
 993             jnrlistC         = jjnr[jidx+2];
 994             jnrlistD         = jjnr[jidx+3];
 995             /* Sign of each element will be negative for non-real atoms.
 996              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 997              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 998              */
 999             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1000             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1001             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1002             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1003             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1004             j_coord_offsetA  = DIM*jnrA;
1005             j_coord_offsetB  = DIM*jnrB;
1006             j_coord_offsetC  = DIM*jnrC;
1007             j_coord_offsetD  = DIM*jnrD;
1008
1009             /* load j atom coordinates */
1010             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1011                                               x+j_coord_offsetC,x+j_coord_offsetD,
1012                                               &jx0,&jy0,&jz0);
1013
1014             /* Calculate displacement vector */
1015             dx00             = _mm_sub_ps(ix0,jx0);
1016             dy00             = _mm_sub_ps(iy0,jy0);
1017             dz00             = _mm_sub_ps(iz0,jz0);
1018             dx10             = _mm_sub_ps(ix1,jx0);
1019             dy10             = _mm_sub_ps(iy1,jy0);
1020             dz10             = _mm_sub_ps(iz1,jz0);
1021             dx20             = _mm_sub_ps(ix2,jx0);
1022             dy20             = _mm_sub_ps(iy2,jy0);
1023             dz20             = _mm_sub_ps(iz2,jz0);
1024
1025             /* Calculate squared distance and things based on it */
1026             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1027             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1028             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1029
1030             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1031             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1032             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1033
1034             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1035             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1036             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1037
1038             /* Load parameters for j particles */
1039             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1040                                                               charge+jnrC+0,charge+jnrD+0);
1041             vdwjidx0A        = 2*vdwtype[jnrA+0];
1042             vdwjidx0B        = 2*vdwtype[jnrB+0];
1043             vdwjidx0C        = 2*vdwtype[jnrC+0];
1044             vdwjidx0D        = 2*vdwtype[jnrD+0];
1045
1046             fjx0             = _mm_setzero_ps();
1047             fjy0             = _mm_setzero_ps();
1048             fjz0             = _mm_setzero_ps();
1049
1050             /**************************
1051              * CALCULATE INTERACTIONS *
1052              **************************/
1053
1054             if (gmx_mm_any_lt(rsq00,rcutoff2))
1055             {
1056
1057             r00              = _mm_mul_ps(rsq00,rinv00);
1058             r00              = _mm_andnot_ps(dummy_mask,r00);
1059
1060             /* Compute parameters for interactions between i and j atoms */
1061             qq00             = _mm_mul_ps(iq0,jq0);
1062             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1063                                          vdwparam+vdwioffset0+vdwjidx0B,
1064                                          vdwparam+vdwioffset0+vdwjidx0C,
1065                                          vdwparam+vdwioffset0+vdwjidx0D,
1066                                          &c6_00,&c12_00);
1067
1068             /* Calculate table index by multiplying r with table scale and truncate to integer */
1069             rt               = _mm_mul_ps(r00,vftabscale);
1070             vfitab           = _mm_cvttps_epi32(rt);
1071 #ifdef __XOP__
1072             vfeps            = _mm_frcz_ps(rt);
1073 #else
1074             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1075 #endif
1076             twovfeps         = _mm_add_ps(vfeps,vfeps);
1077             vfitab           = _mm_slli_epi32(vfitab,3);
1078
1079             /* REACTION-FIELD ELECTROSTATICS */
1080             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1081
1082             /* CUBIC SPLINE TABLE DISPERSION */
1083             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1084             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1085             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1086             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1087             _MM_TRANSPOSE4_PS(Y,F,G,H);
1088             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1089             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1090             fvdw6            = _mm_mul_ps(c6_00,FF);
1091
1092             /* CUBIC SPLINE TABLE REPULSION */
1093             vfitab           = _mm_add_epi32(vfitab,ifour);
1094             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1095             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1096             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1097             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1098             _MM_TRANSPOSE4_PS(Y,F,G,H);
1099             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1100             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1101             fvdw12           = _mm_mul_ps(c12_00,FF);
1102             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1103
1104             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1105
1106             fscal            = _mm_add_ps(felec,fvdw);
1107
1108             fscal            = _mm_and_ps(fscal,cutoff_mask);
1109
1110             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1111
1112              /* Update vectorial force */
1113             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1114             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1115             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1116
1117             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1118             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1119             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1120
1121             }
1122
1123             /**************************
1124              * CALCULATE INTERACTIONS *
1125              **************************/
1126
1127             if (gmx_mm_any_lt(rsq10,rcutoff2))
1128             {
1129
1130             /* Compute parameters for interactions between i and j atoms */
1131             qq10             = _mm_mul_ps(iq1,jq0);
1132
1133             /* REACTION-FIELD ELECTROSTATICS */
1134             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1135
1136             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1137
1138             fscal            = felec;
1139
1140             fscal            = _mm_and_ps(fscal,cutoff_mask);
1141
1142             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1143
1144              /* Update vectorial force */
1145             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1146             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1147             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1148
1149             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1150             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1151             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1152
1153             }
1154
1155             /**************************
1156              * CALCULATE INTERACTIONS *
1157              **************************/
1158
1159             if (gmx_mm_any_lt(rsq20,rcutoff2))
1160             {
1161
1162             /* Compute parameters for interactions between i and j atoms */
1163             qq20             = _mm_mul_ps(iq2,jq0);
1164
1165             /* REACTION-FIELD ELECTROSTATICS */
1166             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1167
1168             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1169
1170             fscal            = felec;
1171
1172             fscal            = _mm_and_ps(fscal,cutoff_mask);
1173
1174             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1175
1176              /* Update vectorial force */
1177             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1178             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1179             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1180
1181             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1182             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1183             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1184
1185             }
1186
1187             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1188             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1189             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1190             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1191
1192             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1193
1194             /* Inner loop uses 127 flops */
1195         }
1196
1197         /* End of innermost loop */
1198
1199         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1200                                               f+i_coord_offset,fshift+i_shift_offset);
1201
1202         /* Increment number of inner iterations */
1203         inneriter                  += j_index_end - j_index_start;
1204
1205         /* Outer loop uses 18 flops */
1206     }
1207
1208     /* Increment number of outer iterations */
1209     outeriter        += nri;
1210
1211     /* Update outer/inner flops */
1212
1213     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);
1214 }