src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_sse2_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 SSE, 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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  69     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  78     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  79     __m128i          vfitab;
  80     __m128i          ifour       = _mm_set1_epi32(4);
  81     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  82     real             *vftab;
  83     __m128           dummy_mask,cutoff_mask;
  84     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  85     __m128           one     = _mm_set1_ps(1.0);
  86     __m128           two     = _mm_set1_ps(2.0);
  87     x                = xx[0];
  88     f                = ff[0];
  89
  90     nri              = nlist->nri;
  91     iinr             = nlist->iinr;
  92     jindex           = nlist->jindex;
  93     jjnr             = nlist->jjnr;
  94     shiftidx         = nlist->shift;
  95     gid              = nlist->gid;
  96     shiftvec         = fr->shift_vec[0];
  97     fshift           = fr->fshift[0];
  98     facel            = _mm_set1_ps(fr->epsfac);
  99     charge           = mdatoms->chargeA;
 100     krf              = _mm_set1_ps(fr->ic->k_rf);
 101     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 102     crf              = _mm_set1_ps(fr->ic->c_rf);
 103     nvdwtype         = fr->ntype;
 104     vdwparam         = fr->nbfp;
 105     vdwtype          = mdatoms->typeA;
 106
 107     vftab            = kernel_data->table_vdw->data;
 108     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 109
 110     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 111     rcutoff_scalar   = fr->rcoulomb;
 112     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 113     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 114
 115     /* Avoid stupid compiler warnings */
 116     jnrA = jnrB = jnrC = jnrD = 0;
 117     j_coord_offsetA = 0;
 118     j_coord_offsetB = 0;
 119     j_coord_offsetC = 0;
 120     j_coord_offsetD = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     /* Start outer loop over neighborlists */
 126     for(iidx=0; iidx<nri; iidx++)
 127     {
 128         /* Load shift vector for this list */
 129         i_shift_offset   = DIM*shiftidx[iidx];
 130         shX              = shiftvec[i_shift_offset+XX];
 131         shY              = shiftvec[i_shift_offset+YY];
 132         shZ              = shiftvec[i_shift_offset+ZZ];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 144         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 145         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 146
 147         fix0             = _mm_setzero_ps();
 148         fiy0             = _mm_setzero_ps();
 149         fiz0             = _mm_setzero_ps();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 153         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155         /* Reset potential sums */
 156         velecsum         = _mm_setzero_ps();
 157         vvdwsum          = _mm_setzero_ps();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             jnrC             = jjnr[jidx+2];
 167             jnrD             = jjnr[jidx+3];
 168
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171             j_coord_offsetC  = DIM*jnrC;
 172             j_coord_offsetD  = DIM*jnrD;
 173
 174             /* load j atom coordinates */
 175             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                               x+j_coord_offsetC,x+j_coord_offsetD,
 177                                               &jx0,&jy0,&jz0);
 178
 179             /* Calculate displacement vector */
 180             dx00             = _mm_sub_ps(ix0,jx0);
 181             dy00             = _mm_sub_ps(iy0,jy0);
 182             dz00             = _mm_sub_ps(iz0,jz0);
 183
 184             /* Calculate squared distance and things based on it */
 185             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 186
 187             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 188
 189             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 190
 191             /* Load parameters for j particles */
 192             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 193                                                               charge+jnrC+0,charge+jnrD+0);
 194             vdwjidx0A        = 2*vdwtype[jnrA+0];
 195             vdwjidx0B        = 2*vdwtype[jnrB+0];
 196             vdwjidx0C        = 2*vdwtype[jnrC+0];
 197             vdwjidx0D        = 2*vdwtype[jnrD+0];
 198
 199             /**************************
 200              * CALCULATE INTERACTIONS *
 201              **************************/
 202
 203             if (gmx_mm_any_lt(rsq00,rcutoff2))
 204             {
 205
 206             r00              = _mm_mul_ps(rsq00,rinv00);
 207
 208             /* Compute parameters for interactions between i and j atoms */
 209             qq00             = _mm_mul_ps(iq0,jq0);
 210             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 211                                          vdwparam+vdwioffset0+vdwjidx0B,
 212                                          vdwparam+vdwioffset0+vdwjidx0C,
 213                                          vdwparam+vdwioffset0+vdwjidx0D,
 214                                          &c6_00,&c12_00);
 215
 216             /* Calculate table index by multiplying r with table scale and truncate to integer */
 217             rt               = _mm_mul_ps(r00,vftabscale);
 218             vfitab           = _mm_cvttps_epi32(rt);
 219             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 220             vfitab           = _mm_slli_epi32(vfitab,3);
 221
 222             /* REACTION-FIELD ELECTROSTATICS */
 223             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 224             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 225
 226             /* CUBIC SPLINE TABLE DISPERSION */
 227             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 228             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 229             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 230             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 231             _MM_TRANSPOSE4_PS(Y,F,G,H);
 232             Heps             = _mm_mul_ps(vfeps,H);
 233             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 234             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 235             vvdw6            = _mm_mul_ps(c6_00,VV);
 236             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 237             fvdw6            = _mm_mul_ps(c6_00,FF);
 238
 239             /* CUBIC SPLINE TABLE REPULSION */
 240             vfitab           = _mm_add_epi32(vfitab,ifour);
 241             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 242             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 243             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 244             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 245             _MM_TRANSPOSE4_PS(Y,F,G,H);
 246             Heps             = _mm_mul_ps(vfeps,H);
 247             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 248             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 249             vvdw12           = _mm_mul_ps(c12_00,VV);
 250             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 251             fvdw12           = _mm_mul_ps(c12_00,FF);
 252             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 253             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 254
 255             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 256
 257             /* Update potential sum for this i atom from the interaction with this j atom. */
 258             velec            = _mm_and_ps(velec,cutoff_mask);
 259             velecsum         = _mm_add_ps(velecsum,velec);
 260             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 261             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 262
 263             fscal            = _mm_add_ps(felec,fvdw);
 264
 265             fscal            = _mm_and_ps(fscal,cutoff_mask);
 266
 267             /* Calculate temporary vectorial force */
 268             tx               = _mm_mul_ps(fscal,dx00);
 269             ty               = _mm_mul_ps(fscal,dy00);
 270             tz               = _mm_mul_ps(fscal,dz00);
 271
 272             /* Update vectorial force */
 273             fix0             = _mm_add_ps(fix0,tx);
 274             fiy0             = _mm_add_ps(fiy0,ty);
 275             fiz0             = _mm_add_ps(fiz0,tz);
 276
 277             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 278                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 279                                                    tx,ty,tz);
 280
 281             }
 282
 283             /* Inner loop uses 72 flops */
 284         }
 285
 286         if(jidx<j_index_end)
 287         {
 288
 289             /* Get j neighbor index, and coordinate index */
 290             jnrA             = jjnr[jidx];
 291             jnrB             = jjnr[jidx+1];
 292             jnrC             = jjnr[jidx+2];
 293             jnrD             = jjnr[jidx+3];
 294
 295             /* Sign of each element will be negative for non-real atoms.
 296              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 297              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 298              */
 299             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 300             jnrA       = (jnrA>=0) ? jnrA : 0;
 301             jnrB       = (jnrB>=0) ? jnrB : 0;
 302             jnrC       = (jnrC>=0) ? jnrC : 0;
 303             jnrD       = (jnrD>=0) ? jnrD : 0;
 304
 305             j_coord_offsetA  = DIM*jnrA;
 306             j_coord_offsetB  = DIM*jnrB;
 307             j_coord_offsetC  = DIM*jnrC;
 308             j_coord_offsetD  = DIM*jnrD;
 309
 310             /* load j atom coordinates */
 311             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 312                                               x+j_coord_offsetC,x+j_coord_offsetD,
 313                                               &jx0,&jy0,&jz0);
 314
 315             /* Calculate displacement vector */
 316             dx00             = _mm_sub_ps(ix0,jx0);
 317             dy00             = _mm_sub_ps(iy0,jy0);
 318             dz00             = _mm_sub_ps(iz0,jz0);
 319
 320             /* Calculate squared distance and things based on it */
 321             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 322
 323             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 324
 325             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 326
 327             /* Load parameters for j particles */
 328             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 329                                                               charge+jnrC+0,charge+jnrD+0);
 330             vdwjidx0A        = 2*vdwtype[jnrA+0];
 331             vdwjidx0B        = 2*vdwtype[jnrB+0];
 332             vdwjidx0C        = 2*vdwtype[jnrC+0];
 333             vdwjidx0D        = 2*vdwtype[jnrD+0];
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             if (gmx_mm_any_lt(rsq00,rcutoff2))
 340             {
 341
 342             r00              = _mm_mul_ps(rsq00,rinv00);
 343             r00              = _mm_andnot_ps(dummy_mask,r00);
 344
 345             /* Compute parameters for interactions between i and j atoms */
 346             qq00             = _mm_mul_ps(iq0,jq0);
 347             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 348                                          vdwparam+vdwioffset0+vdwjidx0B,
 349                                          vdwparam+vdwioffset0+vdwjidx0C,
 350                                          vdwparam+vdwioffset0+vdwjidx0D,
 351                                          &c6_00,&c12_00);
 352
 353             /* Calculate table index by multiplying r with table scale and truncate to integer */
 354             rt               = _mm_mul_ps(r00,vftabscale);
 355             vfitab           = _mm_cvttps_epi32(rt);
 356             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 357             vfitab           = _mm_slli_epi32(vfitab,3);
 358
 359             /* REACTION-FIELD ELECTROSTATICS */
 360             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 361             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 362
 363             /* CUBIC SPLINE TABLE DISPERSION */
 364             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 365             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 366             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 367             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 368             _MM_TRANSPOSE4_PS(Y,F,G,H);
 369             Heps             = _mm_mul_ps(vfeps,H);
 370             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 371             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 372             vvdw6            = _mm_mul_ps(c6_00,VV);
 373             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 374             fvdw6            = _mm_mul_ps(c6_00,FF);
 375
 376             /* CUBIC SPLINE TABLE REPULSION */
 377             vfitab           = _mm_add_epi32(vfitab,ifour);
 378             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 379             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 380             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 381             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 382             _MM_TRANSPOSE4_PS(Y,F,G,H);
 383             Heps             = _mm_mul_ps(vfeps,H);
 384             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 385             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 386             vvdw12           = _mm_mul_ps(c12_00,VV);
 387             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 388             fvdw12           = _mm_mul_ps(c12_00,FF);
 389             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 390             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 391
 392             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velec            = _mm_and_ps(velec,cutoff_mask);
 396             velec            = _mm_andnot_ps(dummy_mask,velec);
 397             velecsum         = _mm_add_ps(velecsum,velec);
 398             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 399             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 400             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 401
 402             fscal            = _mm_add_ps(felec,fvdw);
 403
 404             fscal            = _mm_and_ps(fscal,cutoff_mask);
 405
 406             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 407
 408             /* Calculate temporary vectorial force */
 409             tx               = _mm_mul_ps(fscal,dx00);
 410             ty               = _mm_mul_ps(fscal,dy00);
 411             tz               = _mm_mul_ps(fscal,dz00);
 412
 413             /* Update vectorial force */
 414             fix0             = _mm_add_ps(fix0,tx);
 415             fiy0             = _mm_add_ps(fiy0,ty);
 416             fiz0             = _mm_add_ps(fiz0,tz);
 417
 418             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 419                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 420                                                    tx,ty,tz);
 421
 422             }
 423
 424             /* Inner loop uses 73 flops */
 425         }
 426
 427         /* End of innermost loop */
 428
 429         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 430                                               f+i_coord_offset,fshift+i_shift_offset);
 431
 432         ggid                        = gid[iidx];
 433         /* Update potential energies */
 434         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 435         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 436
 437         /* Increment number of inner iterations */
 438         inneriter                  += j_index_end - j_index_start;
 439
 440         /* Outer loop uses 12 flops */
 441     }
 442
 443     /* Increment number of outer iterations */
 444     outeriter        += nri;
 445
 446     /* Update outer/inner flops */
 447
 448     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*12 + inneriter*73);
 449 }
 450 /*
 451  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse2_single
 452  * Electrostatics interaction: ReactionField
 453  * VdW interaction:            CubicSplineTable
 454  * Geometry:                   Particle-Particle
 455  * Calculate force/pot:        Force
 456  */
 457 void
 458 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse2_single
 459                     (t_nblist * gmx_restrict                nlist,
 460                      rvec * gmx_restrict                    xx,
 461                      rvec * gmx_restrict                    ff,
 462                      t_forcerec * gmx_restrict              fr,
 463                      t_mdatoms * gmx_restrict               mdatoms,
 464                      nb_kernel_data_t * gmx_restrict        kernel_data,
 465                      t_nrnb * gmx_restrict                  nrnb)
 466 {
 467     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 468      * just 0 for non-waters.
 469      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 470      * jnr indices corresponding to data put in the four positions in the SIMD register.
 471      */
 472     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 473     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 474     int              jnrA,jnrB,jnrC,jnrD;
 475     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 476     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 477     real             shX,shY,shZ,rcutoff_scalar;
 478     real             *shiftvec,*fshift,*x,*f;
 479     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 480     int              vdwioffset0;
 481     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 482     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 483     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 484     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 485     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 486     real             *charge;
 487     int              nvdwtype;
 488     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 489     int              *vdwtype;
 490     real             *vdwparam;
 491     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 492     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 493     __m128i          vfitab;
 494     __m128i          ifour       = _mm_set1_epi32(4);
 495     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 496     real             *vftab;
 497     __m128           dummy_mask,cutoff_mask;
 498     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 499     __m128           one     = _mm_set1_ps(1.0);
 500     __m128           two     = _mm_set1_ps(2.0);
 501     x                = xx[0];
 502     f                = ff[0];
 503
 504     nri              = nlist->nri;
 505     iinr             = nlist->iinr;
 506     jindex           = nlist->jindex;
 507     jjnr             = nlist->jjnr;
 508     shiftidx         = nlist->shift;
 509     gid              = nlist->gid;
 510     shiftvec         = fr->shift_vec[0];
 511     fshift           = fr->fshift[0];
 512     facel            = _mm_set1_ps(fr->epsfac);
 513     charge           = mdatoms->chargeA;
 514     krf              = _mm_set1_ps(fr->ic->k_rf);
 515     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 516     crf              = _mm_set1_ps(fr->ic->c_rf);
 517     nvdwtype         = fr->ntype;
 518     vdwparam         = fr->nbfp;
 519     vdwtype          = mdatoms->typeA;
 520
 521     vftab            = kernel_data->table_vdw->data;
 522     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 523
 524     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 525     rcutoff_scalar   = fr->rcoulomb;
 526     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 527     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 528
 529     /* Avoid stupid compiler warnings */
 530     jnrA = jnrB = jnrC = jnrD = 0;
 531     j_coord_offsetA = 0;
 532     j_coord_offsetB = 0;
 533     j_coord_offsetC = 0;
 534     j_coord_offsetD = 0;
 535
 536     outeriter        = 0;
 537     inneriter        = 0;
 538
 539     /* Start outer loop over neighborlists */
 540     for(iidx=0; iidx<nri; iidx++)
 541     {
 542         /* Load shift vector for this list */
 543         i_shift_offset   = DIM*shiftidx[iidx];
 544         shX              = shiftvec[i_shift_offset+XX];
 545         shY              = shiftvec[i_shift_offset+YY];
 546         shZ              = shiftvec[i_shift_offset+ZZ];
 547
 548         /* Load limits for loop over neighbors */
 549         j_index_start    = jindex[iidx];
 550         j_index_end      = jindex[iidx+1];
 551
 552         /* Get outer coordinate index */
 553         inr              = iinr[iidx];
 554         i_coord_offset   = DIM*inr;
 555
 556         /* Load i particle coords and add shift vector */
 557         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 558         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 559         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 560
 561         fix0             = _mm_setzero_ps();
 562         fiy0             = _mm_setzero_ps();
 563         fiz0             = _mm_setzero_ps();
 564
 565         /* Load parameters for i particles */
 566         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 567         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 568
 569         /* Start inner kernel loop */
 570         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 571         {
 572
 573             /* Get j neighbor index, and coordinate index */
 574             jnrA             = jjnr[jidx];
 575             jnrB             = jjnr[jidx+1];
 576             jnrC             = jjnr[jidx+2];
 577             jnrD             = jjnr[jidx+3];
 578
 579             j_coord_offsetA  = DIM*jnrA;
 580             j_coord_offsetB  = DIM*jnrB;
 581             j_coord_offsetC  = DIM*jnrC;
 582             j_coord_offsetD  = DIM*jnrD;
 583
 584             /* load j atom coordinates */
 585             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 586                                               x+j_coord_offsetC,x+j_coord_offsetD,
 587                                               &jx0,&jy0,&jz0);
 588
 589             /* Calculate displacement vector */
 590             dx00             = _mm_sub_ps(ix0,jx0);
 591             dy00             = _mm_sub_ps(iy0,jy0);
 592             dz00             = _mm_sub_ps(iz0,jz0);
 593
 594             /* Calculate squared distance and things based on it */
 595             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 596
 597             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 598
 599             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 600
 601             /* Load parameters for j particles */
 602             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 603                                                               charge+jnrC+0,charge+jnrD+0);
 604             vdwjidx0A        = 2*vdwtype[jnrA+0];
 605             vdwjidx0B        = 2*vdwtype[jnrB+0];
 606             vdwjidx0C        = 2*vdwtype[jnrC+0];
 607             vdwjidx0D        = 2*vdwtype[jnrD+0];
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             if (gmx_mm_any_lt(rsq00,rcutoff2))
 614             {
 615
 616             r00              = _mm_mul_ps(rsq00,rinv00);
 617
 618             /* Compute parameters for interactions between i and j atoms */
 619             qq00             = _mm_mul_ps(iq0,jq0);
 620             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 621                                          vdwparam+vdwioffset0+vdwjidx0B,
 622                                          vdwparam+vdwioffset0+vdwjidx0C,
 623                                          vdwparam+vdwioffset0+vdwjidx0D,
 624                                          &c6_00,&c12_00);
 625
 626             /* Calculate table index by multiplying r with table scale and truncate to integer */
 627             rt               = _mm_mul_ps(r00,vftabscale);
 628             vfitab           = _mm_cvttps_epi32(rt);
 629             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 630             vfitab           = _mm_slli_epi32(vfitab,3);
 631
 632             /* REACTION-FIELD ELECTROSTATICS */
 633             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 634
 635             /* CUBIC SPLINE TABLE DISPERSION */
 636             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 637             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 638             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 639             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 640             _MM_TRANSPOSE4_PS(Y,F,G,H);
 641             Heps             = _mm_mul_ps(vfeps,H);
 642             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 643             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 644             fvdw6            = _mm_mul_ps(c6_00,FF);
 645
 646             /* CUBIC SPLINE TABLE REPULSION */
 647             vfitab           = _mm_add_epi32(vfitab,ifour);
 648             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 649             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 650             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 651             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 652             _MM_TRANSPOSE4_PS(Y,F,G,H);
 653             Heps             = _mm_mul_ps(vfeps,H);
 654             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 655             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 656             fvdw12           = _mm_mul_ps(c12_00,FF);
 657             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 658
 659             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 660
 661             fscal            = _mm_add_ps(felec,fvdw);
 662
 663             fscal            = _mm_and_ps(fscal,cutoff_mask);
 664
 665             /* Calculate temporary vectorial force */
 666             tx               = _mm_mul_ps(fscal,dx00);
 667             ty               = _mm_mul_ps(fscal,dy00);
 668             tz               = _mm_mul_ps(fscal,dz00);
 669
 670             /* Update vectorial force */
 671             fix0             = _mm_add_ps(fix0,tx);
 672             fiy0             = _mm_add_ps(fiy0,ty);
 673             fiz0             = _mm_add_ps(fiz0,tz);
 674
 675             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 676                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 677                                                    tx,ty,tz);
 678
 679             }
 680
 681             /* Inner loop uses 57 flops */
 682         }
 683
 684         if(jidx<j_index_end)
 685         {
 686
 687             /* Get j neighbor index, and coordinate index */
 688             jnrA             = jjnr[jidx];
 689             jnrB             = jjnr[jidx+1];
 690             jnrC             = jjnr[jidx+2];
 691             jnrD             = jjnr[jidx+3];
 692
 693             /* Sign of each element will be negative for non-real atoms.
 694              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 695              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 696              */
 697             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 698             jnrA       = (jnrA>=0) ? jnrA : 0;
 699             jnrB       = (jnrB>=0) ? jnrB : 0;
 700             jnrC       = (jnrC>=0) ? jnrC : 0;
 701             jnrD       = (jnrD>=0) ? jnrD : 0;
 702
 703             j_coord_offsetA  = DIM*jnrA;
 704             j_coord_offsetB  = DIM*jnrB;
 705             j_coord_offsetC  = DIM*jnrC;
 706             j_coord_offsetD  = DIM*jnrD;
 707
 708             /* load j atom coordinates */
 709             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 710                                               x+j_coord_offsetC,x+j_coord_offsetD,
 711                                               &jx0,&jy0,&jz0);
 712
 713             /* Calculate displacement vector */
 714             dx00             = _mm_sub_ps(ix0,jx0);
 715             dy00             = _mm_sub_ps(iy0,jy0);
 716             dz00             = _mm_sub_ps(iz0,jz0);
 717
 718             /* Calculate squared distance and things based on it */
 719             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 720
 721             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 722
 723             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 724
 725             /* Load parameters for j particles */
 726             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 727                                                               charge+jnrC+0,charge+jnrD+0);
 728             vdwjidx0A        = 2*vdwtype[jnrA+0];
 729             vdwjidx0B        = 2*vdwtype[jnrB+0];
 730             vdwjidx0C        = 2*vdwtype[jnrC+0];
 731             vdwjidx0D        = 2*vdwtype[jnrD+0];
 732
 733             /**************************
 734              * CALCULATE INTERACTIONS *
 735              **************************/
 736
 737             if (gmx_mm_any_lt(rsq00,rcutoff2))
 738             {
 739
 740             r00              = _mm_mul_ps(rsq00,rinv00);
 741             r00              = _mm_andnot_ps(dummy_mask,r00);
 742
 743             /* Compute parameters for interactions between i and j atoms */
 744             qq00             = _mm_mul_ps(iq0,jq0);
 745             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 746                                          vdwparam+vdwioffset0+vdwjidx0B,
 747                                          vdwparam+vdwioffset0+vdwjidx0C,
 748                                          vdwparam+vdwioffset0+vdwjidx0D,
 749                                          &c6_00,&c12_00);
 750
 751             /* Calculate table index by multiplying r with table scale and truncate to integer */
 752             rt               = _mm_mul_ps(r00,vftabscale);
 753             vfitab           = _mm_cvttps_epi32(rt);
 754             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 755             vfitab           = _mm_slli_epi32(vfitab,3);
 756
 757             /* REACTION-FIELD ELECTROSTATICS */
 758             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 759
 760             /* CUBIC SPLINE TABLE DISPERSION */
 761             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 762             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 763             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 764             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 765             _MM_TRANSPOSE4_PS(Y,F,G,H);
 766             Heps             = _mm_mul_ps(vfeps,H);
 767             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 768             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 769             fvdw6            = _mm_mul_ps(c6_00,FF);
 770
 771             /* CUBIC SPLINE TABLE REPULSION */
 772             vfitab           = _mm_add_epi32(vfitab,ifour);
 773             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 774             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 775             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 776             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 777             _MM_TRANSPOSE4_PS(Y,F,G,H);
 778             Heps             = _mm_mul_ps(vfeps,H);
 779             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 780             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 781             fvdw12           = _mm_mul_ps(c12_00,FF);
 782             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 783
 784             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 785
 786             fscal            = _mm_add_ps(felec,fvdw);
 787
 788             fscal            = _mm_and_ps(fscal,cutoff_mask);
 789
 790             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 791
 792             /* Calculate temporary vectorial force */
 793             tx               = _mm_mul_ps(fscal,dx00);
 794             ty               = _mm_mul_ps(fscal,dy00);
 795             tz               = _mm_mul_ps(fscal,dz00);
 796
 797             /* Update vectorial force */
 798             fix0             = _mm_add_ps(fix0,tx);
 799             fiy0             = _mm_add_ps(fiy0,ty);
 800             fiz0             = _mm_add_ps(fiz0,tz);
 801
 802             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 803                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 804                                                    tx,ty,tz);
 805
 806             }
 807
 808             /* Inner loop uses 58 flops */
 809         }
 810
 811         /* End of innermost loop */
 812
 813         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 814                                               f+i_coord_offset,fshift+i_shift_offset);
 815
 816         /* Increment number of inner iterations */
 817         inneriter                  += j_index_end - j_index_start;
 818
 819         /* Outer loop uses 10 flops */
 820     }
 821
 822     /* Increment number of outer iterations */
 823     outeriter        += nri;
 824
 825     /* Update outer/inner flops */
 826
 827     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*10 + inneriter*58);
 828 }