src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_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_ElecRF_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_ElecRF_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              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           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  72     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  73     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  74     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  75     real             *charge;
  76     int              nvdwtype;
  77     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  78     int              *vdwtype;
  79     real             *vdwparam;
  80     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  81     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  82     __m128i          vfitab;
  83     __m128i          ifour       = _mm_set1_epi32(4);
  84     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  85     real             *vftab;
  86     __m128           dummy_mask,cutoff_mask;
  87     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  88     __m128           one     = _mm_set1_ps(1.0);
  89     __m128           two     = _mm_set1_ps(2.0);
  90     x                = xx[0];
  91     f                = ff[0];
  92
  93     nri              = nlist->nri;
  94     iinr             = nlist->iinr;
  95     jindex           = nlist->jindex;
  96     jjnr             = nlist->jjnr;
  97     shiftidx         = nlist->shift;
  98     gid              = nlist->gid;
  99     shiftvec         = fr->shift_vec[0];
 100     fshift           = fr->fshift[0];
 101     facel            = _mm_set1_ps(fr->epsfac);
 102     charge           = mdatoms->chargeA;
 103     krf              = _mm_set1_ps(fr->ic->k_rf);
 104     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 105     crf              = _mm_set1_ps(fr->ic->c_rf);
 106     nvdwtype         = fr->ntype;
 107     vdwparam         = fr->nbfp;
 108     vdwtype          = mdatoms->typeA;
 109
 110     vftab            = kernel_data->table_vdw->data;
 111     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = jnrC = jnrD = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117     j_coord_offsetC = 0;
 118     j_coord_offsetD = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     for(iidx=0;iidx<4*DIM;iidx++)
 124     {
 125         scratch[iidx] = 0.0;
 126     }
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 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         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _mm_setzero_ps();
 146         fiy0             = _mm_setzero_ps();
 147         fiz0             = _mm_setzero_ps();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 151         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         velecsum         = _mm_setzero_ps();
 155         vvdwsum          = _mm_setzero_ps();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             jnrC             = jjnr[jidx+2];
 165             jnrD             = jjnr[jidx+3];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168             j_coord_offsetC  = DIM*jnrC;
 169             j_coord_offsetD  = DIM*jnrD;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               x+j_coord_offsetC,x+j_coord_offsetD,
 174                                               &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx00             = _mm_sub_ps(ix0,jx0);
 178             dy00             = _mm_sub_ps(iy0,jy0);
 179             dz00             = _mm_sub_ps(iz0,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 183
 184             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 185
 186             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 187
 188             /* Load parameters for j particles */
 189             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 190                                                               charge+jnrC+0,charge+jnrD+0);
 191             vdwjidx0A        = 2*vdwtype[jnrA+0];
 192             vdwjidx0B        = 2*vdwtype[jnrB+0];
 193             vdwjidx0C        = 2*vdwtype[jnrC+0];
 194             vdwjidx0D        = 2*vdwtype[jnrD+0];
 195
 196             /**************************
 197              * CALCULATE INTERACTIONS *
 198              **************************/
 199
 200             r00              = _mm_mul_ps(rsq00,rinv00);
 201
 202             /* Compute parameters for interactions between i and j atoms */
 203             qq00             = _mm_mul_ps(iq0,jq0);
 204             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 205                                          vdwparam+vdwioffset0+vdwjidx0B,
 206                                          vdwparam+vdwioffset0+vdwjidx0C,
 207                                          vdwparam+vdwioffset0+vdwjidx0D,
 208                                          &c6_00,&c12_00);
 209
 210             /* Calculate table index by multiplying r with table scale and truncate to integer */
 211             rt               = _mm_mul_ps(r00,vftabscale);
 212             vfitab           = _mm_cvttps_epi32(rt);
 213             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 214             vfitab           = _mm_slli_epi32(vfitab,3);
 215
 216             /* REACTION-FIELD ELECTROSTATICS */
 217             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 218             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 219
 220             /* CUBIC SPLINE TABLE DISPERSION */
 221             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 222             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 223             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 224             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 225             _MM_TRANSPOSE4_PS(Y,F,G,H);
 226             Heps             = _mm_mul_ps(vfeps,H);
 227             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 228             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 229             vvdw6            = _mm_mul_ps(c6_00,VV);
 230             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 231             fvdw6            = _mm_mul_ps(c6_00,FF);
 232
 233             /* CUBIC SPLINE TABLE REPULSION */
 234             vfitab           = _mm_add_epi32(vfitab,ifour);
 235             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 236             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 237             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 238             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 239             _MM_TRANSPOSE4_PS(Y,F,G,H);
 240             Heps             = _mm_mul_ps(vfeps,H);
 241             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 242             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 243             vvdw12           = _mm_mul_ps(c12_00,VV);
 244             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 245             fvdw12           = _mm_mul_ps(c12_00,FF);
 246             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 247             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 248
 249             /* Update potential sum for this i atom from the interaction with this j atom. */
 250             velecsum         = _mm_add_ps(velecsum,velec);
 251             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 252
 253             fscal            = _mm_add_ps(felec,fvdw);
 254
 255             /* Calculate temporary vectorial force */
 256             tx               = _mm_mul_ps(fscal,dx00);
 257             ty               = _mm_mul_ps(fscal,dy00);
 258             tz               = _mm_mul_ps(fscal,dz00);
 259
 260             /* Update vectorial force */
 261             fix0             = _mm_add_ps(fix0,tx);
 262             fiy0             = _mm_add_ps(fiy0,ty);
 263             fiz0             = _mm_add_ps(fiz0,tz);
 264
 265             fjptrA             = f+j_coord_offsetA;
 266             fjptrB             = f+j_coord_offsetB;
 267             fjptrC             = f+j_coord_offsetC;
 268             fjptrD             = f+j_coord_offsetD;
 269             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 270
 271             /* Inner loop uses 67 flops */
 272         }
 273
 274         if(jidx<j_index_end)
 275         {
 276
 277             /* Get j neighbor index, and coordinate index */
 278             jnrlistA         = jjnr[jidx];
 279             jnrlistB         = jjnr[jidx+1];
 280             jnrlistC         = jjnr[jidx+2];
 281             jnrlistD         = jjnr[jidx+3];
 282             /* Sign of each element will be negative for non-real atoms.
 283              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 284              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 285              */
 286             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 287             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 288             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 289             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 290             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 291             j_coord_offsetA  = DIM*jnrA;
 292             j_coord_offsetB  = DIM*jnrB;
 293             j_coord_offsetC  = DIM*jnrC;
 294             j_coord_offsetD  = DIM*jnrD;
 295
 296             /* load j atom coordinates */
 297             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 298                                               x+j_coord_offsetC,x+j_coord_offsetD,
 299                                               &jx0,&jy0,&jz0);
 300
 301             /* Calculate displacement vector */
 302             dx00             = _mm_sub_ps(ix0,jx0);
 303             dy00             = _mm_sub_ps(iy0,jy0);
 304             dz00             = _mm_sub_ps(iz0,jz0);
 305
 306             /* Calculate squared distance and things based on it */
 307             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 308
 309             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 310
 311             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 312
 313             /* Load parameters for j particles */
 314             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 315                                                               charge+jnrC+0,charge+jnrD+0);
 316             vdwjidx0A        = 2*vdwtype[jnrA+0];
 317             vdwjidx0B        = 2*vdwtype[jnrB+0];
 318             vdwjidx0C        = 2*vdwtype[jnrC+0];
 319             vdwjidx0D        = 2*vdwtype[jnrD+0];
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             r00              = _mm_mul_ps(rsq00,rinv00);
 326             r00              = _mm_andnot_ps(dummy_mask,r00);
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq00             = _mm_mul_ps(iq0,jq0);
 330             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 331                                          vdwparam+vdwioffset0+vdwjidx0B,
 332                                          vdwparam+vdwioffset0+vdwjidx0C,
 333                                          vdwparam+vdwioffset0+vdwjidx0D,
 334                                          &c6_00,&c12_00);
 335
 336             /* Calculate table index by multiplying r with table scale and truncate to integer */
 337             rt               = _mm_mul_ps(r00,vftabscale);
 338             vfitab           = _mm_cvttps_epi32(rt);
 339             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 340             vfitab           = _mm_slli_epi32(vfitab,3);
 341
 342             /* REACTION-FIELD ELECTROSTATICS */
 343             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 344             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 345
 346             /* CUBIC SPLINE TABLE DISPERSION */
 347             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 348             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 349             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 350             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 351             _MM_TRANSPOSE4_PS(Y,F,G,H);
 352             Heps             = _mm_mul_ps(vfeps,H);
 353             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 354             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 355             vvdw6            = _mm_mul_ps(c6_00,VV);
 356             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 357             fvdw6            = _mm_mul_ps(c6_00,FF);
 358
 359             /* CUBIC SPLINE TABLE REPULSION */
 360             vfitab           = _mm_add_epi32(vfitab,ifour);
 361             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 362             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 363             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 364             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 365             _MM_TRANSPOSE4_PS(Y,F,G,H);
 366             Heps             = _mm_mul_ps(vfeps,H);
 367             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 368             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 369             vvdw12           = _mm_mul_ps(c12_00,VV);
 370             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 371             fvdw12           = _mm_mul_ps(c12_00,FF);
 372             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 373             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velec            = _mm_andnot_ps(dummy_mask,velec);
 377             velecsum         = _mm_add_ps(velecsum,velec);
 378             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 379             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 380
 381             fscal            = _mm_add_ps(felec,fvdw);
 382
 383             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm_mul_ps(fscal,dx00);
 387             ty               = _mm_mul_ps(fscal,dy00);
 388             tz               = _mm_mul_ps(fscal,dz00);
 389
 390             /* Update vectorial force */
 391             fix0             = _mm_add_ps(fix0,tx);
 392             fiy0             = _mm_add_ps(fiy0,ty);
 393             fiz0             = _mm_add_ps(fiz0,tz);
 394
 395             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 396             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 397             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 398             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 399             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 400
 401             /* Inner loop uses 68 flops */
 402         }
 403
 404         /* End of innermost loop */
 405
 406         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 407                                               f+i_coord_offset,fshift+i_shift_offset);
 408
 409         ggid                        = gid[iidx];
 410         /* Update potential energies */
 411         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 412         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 413
 414         /* Increment number of inner iterations */
 415         inneriter                  += j_index_end - j_index_start;
 416
 417         /* Outer loop uses 9 flops */
 418     }
 419
 420     /* Increment number of outer iterations */
 421     outeriter        += nri;
 422
 423     /* Update outer/inner flops */
 424
 425     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
 426 }
 427 /*
 428  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse2_single
 429  * Electrostatics interaction: ReactionField
 430  * VdW interaction:            CubicSplineTable
 431  * Geometry:                   Particle-Particle
 432  * Calculate force/pot:        Force
 433  */
 434 void
 435 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse2_single
 436                     (t_nblist * gmx_restrict                nlist,
 437                      rvec * gmx_restrict                    xx,
 438                      rvec * gmx_restrict                    ff,
 439                      t_forcerec * gmx_restrict              fr,
 440                      t_mdatoms * gmx_restrict               mdatoms,
 441                      nb_kernel_data_t * gmx_restrict        kernel_data,
 442                      t_nrnb * gmx_restrict                  nrnb)
 443 {
 444     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 445      * just 0 for non-waters.
 446      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 447      * jnr indices corresponding to data put in the four positions in the SIMD register.
 448      */
 449     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 450     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 451     int              jnrA,jnrB,jnrC,jnrD;
 452     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 453     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 454     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 455     real             rcutoff_scalar;
 456     real             *shiftvec,*fshift,*x,*f;
 457     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 458     real             scratch[4*DIM];
 459     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 460     int              vdwioffset0;
 461     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 462     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 463     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 464     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 465     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 466     real             *charge;
 467     int              nvdwtype;
 468     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 469     int              *vdwtype;
 470     real             *vdwparam;
 471     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 472     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 473     __m128i          vfitab;
 474     __m128i          ifour       = _mm_set1_epi32(4);
 475     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 476     real             *vftab;
 477     __m128           dummy_mask,cutoff_mask;
 478     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 479     __m128           one     = _mm_set1_ps(1.0);
 480     __m128           two     = _mm_set1_ps(2.0);
 481     x                = xx[0];
 482     f                = ff[0];
 483
 484     nri              = nlist->nri;
 485     iinr             = nlist->iinr;
 486     jindex           = nlist->jindex;
 487     jjnr             = nlist->jjnr;
 488     shiftidx         = nlist->shift;
 489     gid              = nlist->gid;
 490     shiftvec         = fr->shift_vec[0];
 491     fshift           = fr->fshift[0];
 492     facel            = _mm_set1_ps(fr->epsfac);
 493     charge           = mdatoms->chargeA;
 494     krf              = _mm_set1_ps(fr->ic->k_rf);
 495     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 496     crf              = _mm_set1_ps(fr->ic->c_rf);
 497     nvdwtype         = fr->ntype;
 498     vdwparam         = fr->nbfp;
 499     vdwtype          = mdatoms->typeA;
 500
 501     vftab            = kernel_data->table_vdw->data;
 502     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 503
 504     /* Avoid stupid compiler warnings */
 505     jnrA = jnrB = jnrC = jnrD = 0;
 506     j_coord_offsetA = 0;
 507     j_coord_offsetB = 0;
 508     j_coord_offsetC = 0;
 509     j_coord_offsetD = 0;
 510
 511     outeriter        = 0;
 512     inneriter        = 0;
 513
 514     for(iidx=0;iidx<4*DIM;iidx++)
 515     {
 516         scratch[iidx] = 0.0;
 517     }
 518
 519     /* Start outer loop over neighborlists */
 520     for(iidx=0; iidx<nri; iidx++)
 521     {
 522         /* Load shift vector for this list */
 523         i_shift_offset   = DIM*shiftidx[iidx];
 524
 525         /* Load limits for loop over neighbors */
 526         j_index_start    = jindex[iidx];
 527         j_index_end      = jindex[iidx+1];
 528
 529         /* Get outer coordinate index */
 530         inr              = iinr[iidx];
 531         i_coord_offset   = DIM*inr;
 532
 533         /* Load i particle coords and add shift vector */
 534         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 535
 536         fix0             = _mm_setzero_ps();
 537         fiy0             = _mm_setzero_ps();
 538         fiz0             = _mm_setzero_ps();
 539
 540         /* Load parameters for i particles */
 541         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 542         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 543
 544         /* Start inner kernel loop */
 545         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 546         {
 547
 548             /* Get j neighbor index, and coordinate index */
 549             jnrA             = jjnr[jidx];
 550             jnrB             = jjnr[jidx+1];
 551             jnrC             = jjnr[jidx+2];
 552             jnrD             = jjnr[jidx+3];
 553             j_coord_offsetA  = DIM*jnrA;
 554             j_coord_offsetB  = DIM*jnrB;
 555             j_coord_offsetC  = DIM*jnrC;
 556             j_coord_offsetD  = DIM*jnrD;
 557
 558             /* load j atom coordinates */
 559             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 560                                               x+j_coord_offsetC,x+j_coord_offsetD,
 561                                               &jx0,&jy0,&jz0);
 562
 563             /* Calculate displacement vector */
 564             dx00             = _mm_sub_ps(ix0,jx0);
 565             dy00             = _mm_sub_ps(iy0,jy0);
 566             dz00             = _mm_sub_ps(iz0,jz0);
 567
 568             /* Calculate squared distance and things based on it */
 569             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 570
 571             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 572
 573             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 574
 575             /* Load parameters for j particles */
 576             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 577                                                               charge+jnrC+0,charge+jnrD+0);
 578             vdwjidx0A        = 2*vdwtype[jnrA+0];
 579             vdwjidx0B        = 2*vdwtype[jnrB+0];
 580             vdwjidx0C        = 2*vdwtype[jnrC+0];
 581             vdwjidx0D        = 2*vdwtype[jnrD+0];
 582
 583             /**************************
 584              * CALCULATE INTERACTIONS *
 585              **************************/
 586
 587             r00              = _mm_mul_ps(rsq00,rinv00);
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq00             = _mm_mul_ps(iq0,jq0);
 591             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 592                                          vdwparam+vdwioffset0+vdwjidx0B,
 593                                          vdwparam+vdwioffset0+vdwjidx0C,
 594                                          vdwparam+vdwioffset0+vdwjidx0D,
 595                                          &c6_00,&c12_00);
 596
 597             /* Calculate table index by multiplying r with table scale and truncate to integer */
 598             rt               = _mm_mul_ps(r00,vftabscale);
 599             vfitab           = _mm_cvttps_epi32(rt);
 600             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 601             vfitab           = _mm_slli_epi32(vfitab,3);
 602
 603             /* REACTION-FIELD ELECTROSTATICS */
 604             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 605
 606             /* CUBIC SPLINE TABLE DISPERSION */
 607             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 608             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 609             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 610             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 611             _MM_TRANSPOSE4_PS(Y,F,G,H);
 612             Heps             = _mm_mul_ps(vfeps,H);
 613             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 614             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 615             fvdw6            = _mm_mul_ps(c6_00,FF);
 616
 617             /* CUBIC SPLINE TABLE REPULSION */
 618             vfitab           = _mm_add_epi32(vfitab,ifour);
 619             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 620             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 621             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 622             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 623             _MM_TRANSPOSE4_PS(Y,F,G,H);
 624             Heps             = _mm_mul_ps(vfeps,H);
 625             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 626             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 627             fvdw12           = _mm_mul_ps(c12_00,FF);
 628             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 629
 630             fscal            = _mm_add_ps(felec,fvdw);
 631
 632             /* Calculate temporary vectorial force */
 633             tx               = _mm_mul_ps(fscal,dx00);
 634             ty               = _mm_mul_ps(fscal,dy00);
 635             tz               = _mm_mul_ps(fscal,dz00);
 636
 637             /* Update vectorial force */
 638             fix0             = _mm_add_ps(fix0,tx);
 639             fiy0             = _mm_add_ps(fiy0,ty);
 640             fiz0             = _mm_add_ps(fiz0,tz);
 641
 642             fjptrA             = f+j_coord_offsetA;
 643             fjptrB             = f+j_coord_offsetB;
 644             fjptrC             = f+j_coord_offsetC;
 645             fjptrD             = f+j_coord_offsetD;
 646             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 647
 648             /* Inner loop uses 54 flops */
 649         }
 650
 651         if(jidx<j_index_end)
 652         {
 653
 654             /* Get j neighbor index, and coordinate index */
 655             jnrlistA         = jjnr[jidx];
 656             jnrlistB         = jjnr[jidx+1];
 657             jnrlistC         = jjnr[jidx+2];
 658             jnrlistD         = jjnr[jidx+3];
 659             /* Sign of each element will be negative for non-real atoms.
 660              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 661              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 662              */
 663             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 664             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 665             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 666             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 667             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 668             j_coord_offsetA  = DIM*jnrA;
 669             j_coord_offsetB  = DIM*jnrB;
 670             j_coord_offsetC  = DIM*jnrC;
 671             j_coord_offsetD  = DIM*jnrD;
 672
 673             /* load j atom coordinates */
 674             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 675                                               x+j_coord_offsetC,x+j_coord_offsetD,
 676                                               &jx0,&jy0,&jz0);
 677
 678             /* Calculate displacement vector */
 679             dx00             = _mm_sub_ps(ix0,jx0);
 680             dy00             = _mm_sub_ps(iy0,jy0);
 681             dz00             = _mm_sub_ps(iz0,jz0);
 682
 683             /* Calculate squared distance and things based on it */
 684             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 685
 686             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 687
 688             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 689
 690             /* Load parameters for j particles */
 691             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 692                                                               charge+jnrC+0,charge+jnrD+0);
 693             vdwjidx0A        = 2*vdwtype[jnrA+0];
 694             vdwjidx0B        = 2*vdwtype[jnrB+0];
 695             vdwjidx0C        = 2*vdwtype[jnrC+0];
 696             vdwjidx0D        = 2*vdwtype[jnrD+0];
 697
 698             /**************************
 699              * CALCULATE INTERACTIONS *
 700              **************************/
 701
 702             r00              = _mm_mul_ps(rsq00,rinv00);
 703             r00              = _mm_andnot_ps(dummy_mask,r00);
 704
 705             /* Compute parameters for interactions between i and j atoms */
 706             qq00             = _mm_mul_ps(iq0,jq0);
 707             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 708                                          vdwparam+vdwioffset0+vdwjidx0B,
 709                                          vdwparam+vdwioffset0+vdwjidx0C,
 710                                          vdwparam+vdwioffset0+vdwjidx0D,
 711                                          &c6_00,&c12_00);
 712
 713             /* Calculate table index by multiplying r with table scale and truncate to integer */
 714             rt               = _mm_mul_ps(r00,vftabscale);
 715             vfitab           = _mm_cvttps_epi32(rt);
 716             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 717             vfitab           = _mm_slli_epi32(vfitab,3);
 718
 719             /* REACTION-FIELD ELECTROSTATICS */
 720             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 721
 722             /* CUBIC SPLINE TABLE DISPERSION */
 723             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 724             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 725             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 726             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 727             _MM_TRANSPOSE4_PS(Y,F,G,H);
 728             Heps             = _mm_mul_ps(vfeps,H);
 729             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 730             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 731             fvdw6            = _mm_mul_ps(c6_00,FF);
 732
 733             /* CUBIC SPLINE TABLE REPULSION */
 734             vfitab           = _mm_add_epi32(vfitab,ifour);
 735             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 736             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 737             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 738             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 739             _MM_TRANSPOSE4_PS(Y,F,G,H);
 740             Heps             = _mm_mul_ps(vfeps,H);
 741             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 742             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 743             fvdw12           = _mm_mul_ps(c12_00,FF);
 744             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 745
 746             fscal            = _mm_add_ps(felec,fvdw);
 747
 748             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 749
 750             /* Calculate temporary vectorial force */
 751             tx               = _mm_mul_ps(fscal,dx00);
 752             ty               = _mm_mul_ps(fscal,dy00);
 753             tz               = _mm_mul_ps(fscal,dz00);
 754
 755             /* Update vectorial force */
 756             fix0             = _mm_add_ps(fix0,tx);
 757             fiy0             = _mm_add_ps(fiy0,ty);
 758             fiz0             = _mm_add_ps(fiz0,tz);
 759
 760             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 761             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 762             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 763             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 764             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 765
 766             /* Inner loop uses 55 flops */
 767         }
 768
 769         /* End of innermost loop */
 770
 771         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 772                                               f+i_coord_offset,fshift+i_shift_offset);
 773
 774         /* Increment number of inner iterations */
 775         inneriter                  += j_index_end - j_index_start;
 776
 777         /* Outer loop uses 7 flops */
 778     }
 779
 780     /* Increment number of outer iterations */
 781     outeriter        += nri;
 782
 783     /* Update outer/inner flops */
 784
 785     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*55);
 786 }