src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_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_ElecCSTab_VdwCSTab_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_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     nvdwtype         = fr->ntype;
 101     vdwparam         = fr->nbfp;
 102     vdwtype          = mdatoms->typeA;
 103
 104     vftab            = kernel_data->table_elec_vdw->data;
 105     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 106
 107     /* Avoid stupid compiler warnings */
 108     jnrA = jnrB = jnrC = jnrD = 0;
 109     j_coord_offsetA = 0;
 110     j_coord_offsetB = 0;
 111     j_coord_offsetC = 0;
 112     j_coord_offsetD = 0;
 113
 114     outeriter        = 0;
 115     inneriter        = 0;
 116
 117     /* Start outer loop over neighborlists */
 118     for(iidx=0; iidx<nri; iidx++)
 119     {
 120         /* Load shift vector for this list */
 121         i_shift_offset   = DIM*shiftidx[iidx];
 122         shX              = shiftvec[i_shift_offset+XX];
 123         shY              = shiftvec[i_shift_offset+YY];
 124         shZ              = shiftvec[i_shift_offset+ZZ];
 125
 126         /* Load limits for loop over neighbors */
 127         j_index_start    = jindex[iidx];
 128         j_index_end      = jindex[iidx+1];
 129
 130         /* Get outer coordinate index */
 131         inr              = iinr[iidx];
 132         i_coord_offset   = DIM*inr;
 133
 134         /* Load i particle coords and add shift vector */
 135         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 136         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 137         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 138
 139         fix0             = _mm_setzero_ps();
 140         fiy0             = _mm_setzero_ps();
 141         fiz0             = _mm_setzero_ps();
 142
 143         /* Load parameters for i particles */
 144         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 145         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 146
 147         /* Reset potential sums */
 148         velecsum         = _mm_setzero_ps();
 149         vvdwsum          = _mm_setzero_ps();
 150
 151         /* Start inner kernel loop */
 152         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 153         {
 154
 155             /* Get j neighbor index, and coordinate index */
 156             jnrA             = jjnr[jidx];
 157             jnrB             = jjnr[jidx+1];
 158             jnrC             = jjnr[jidx+2];
 159             jnrD             = jjnr[jidx+3];
 160
 161             j_coord_offsetA  = DIM*jnrA;
 162             j_coord_offsetB  = DIM*jnrB;
 163             j_coord_offsetC  = DIM*jnrC;
 164             j_coord_offsetD  = DIM*jnrD;
 165
 166             /* load j atom coordinates */
 167             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 168                                               x+j_coord_offsetC,x+j_coord_offsetD,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _mm_sub_ps(ix0,jx0);
 173             dy00             = _mm_sub_ps(iy0,jy0);
 174             dz00             = _mm_sub_ps(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 178
 179             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 180
 181             /* Load parameters for j particles */
 182             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 183                                                               charge+jnrC+0,charge+jnrD+0);
 184             vdwjidx0A        = 2*vdwtype[jnrA+0];
 185             vdwjidx0B        = 2*vdwtype[jnrB+0];
 186             vdwjidx0C        = 2*vdwtype[jnrC+0];
 187             vdwjidx0D        = 2*vdwtype[jnrD+0];
 188
 189             /**************************
 190              * CALCULATE INTERACTIONS *
 191              **************************/
 192
 193             r00              = _mm_mul_ps(rsq00,rinv00);
 194
 195             /* Compute parameters for interactions between i and j atoms */
 196             qq00             = _mm_mul_ps(iq0,jq0);
 197             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 198                                          vdwparam+vdwioffset0+vdwjidx0B,
 199                                          vdwparam+vdwioffset0+vdwjidx0C,
 200                                          vdwparam+vdwioffset0+vdwjidx0D,
 201                                          &c6_00,&c12_00);
 202
 203             /* Calculate table index by multiplying r with table scale and truncate to integer */
 204             rt               = _mm_mul_ps(r00,vftabscale);
 205             vfitab           = _mm_cvttps_epi32(rt);
 206             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 207             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 208
 209             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 210             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 211             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 212             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 213             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 214             _MM_TRANSPOSE4_PS(Y,F,G,H);
 215             Heps             = _mm_mul_ps(vfeps,H);
 216             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 217             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 218             velec            = _mm_mul_ps(qq00,VV);
 219             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 220             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 221
 222             /* CUBIC SPLINE TABLE DISPERSION */
 223             vfitab           = _mm_add_epi32(vfitab,ifour);
 224             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 225             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 226             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 227             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 228             _MM_TRANSPOSE4_PS(Y,F,G,H);
 229             Heps             = _mm_mul_ps(vfeps,H);
 230             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 231             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 232             vvdw6            = _mm_mul_ps(c6_00,VV);
 233             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 234             fvdw6            = _mm_mul_ps(c6_00,FF);
 235
 236             /* CUBIC SPLINE TABLE REPULSION */
 237             vfitab           = _mm_add_epi32(vfitab,ifour);
 238             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 239             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 240             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 241             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 242             _MM_TRANSPOSE4_PS(Y,F,G,H);
 243             Heps             = _mm_mul_ps(vfeps,H);
 244             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 245             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 246             vvdw12           = _mm_mul_ps(c12_00,VV);
 247             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 248             fvdw12           = _mm_mul_ps(c12_00,FF);
 249             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 250             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 251
 252             /* Update potential sum for this i atom from the interaction with this j atom. */
 253             velecsum         = _mm_add_ps(velecsum,velec);
 254             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 255
 256             fscal            = _mm_add_ps(felec,fvdw);
 257
 258             /* Calculate temporary vectorial force */
 259             tx               = _mm_mul_ps(fscal,dx00);
 260             ty               = _mm_mul_ps(fscal,dy00);
 261             tz               = _mm_mul_ps(fscal,dz00);
 262
 263             /* Update vectorial force */
 264             fix0             = _mm_add_ps(fix0,tx);
 265             fiy0             = _mm_add_ps(fiy0,ty);
 266             fiz0             = _mm_add_ps(fiz0,tz);
 267
 268             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 269                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 270                                                    tx,ty,tz);
 271
 272             /* Inner loop uses 73 flops */
 273         }
 274
 275         if(jidx<j_index_end)
 276         {
 277
 278             /* Get j neighbor index, and coordinate index */
 279             jnrA             = jjnr[jidx];
 280             jnrB             = jjnr[jidx+1];
 281             jnrC             = jjnr[jidx+2];
 282             jnrD             = jjnr[jidx+3];
 283
 284             /* Sign of each element will be negative for non-real atoms.
 285              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 286              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 287              */
 288             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 289             jnrA       = (jnrA>=0) ? jnrA : 0;
 290             jnrB       = (jnrB>=0) ? jnrB : 0;
 291             jnrC       = (jnrC>=0) ? jnrC : 0;
 292             jnrD       = (jnrD>=0) ? jnrD : 0;
 293
 294             j_coord_offsetA  = DIM*jnrA;
 295             j_coord_offsetB  = DIM*jnrB;
 296             j_coord_offsetC  = DIM*jnrC;
 297             j_coord_offsetD  = DIM*jnrD;
 298
 299             /* load j atom coordinates */
 300             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 301                                               x+j_coord_offsetC,x+j_coord_offsetD,
 302                                               &jx0,&jy0,&jz0);
 303
 304             /* Calculate displacement vector */
 305             dx00             = _mm_sub_ps(ix0,jx0);
 306             dy00             = _mm_sub_ps(iy0,jy0);
 307             dz00             = _mm_sub_ps(iz0,jz0);
 308
 309             /* Calculate squared distance and things based on it */
 310             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 311
 312             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 313
 314             /* Load parameters for j particles */
 315             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 316                                                               charge+jnrC+0,charge+jnrD+0);
 317             vdwjidx0A        = 2*vdwtype[jnrA+0];
 318             vdwjidx0B        = 2*vdwtype[jnrB+0];
 319             vdwjidx0C        = 2*vdwtype[jnrC+0];
 320             vdwjidx0D        = 2*vdwtype[jnrD+0];
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             r00              = _mm_mul_ps(rsq00,rinv00);
 327             r00              = _mm_andnot_ps(dummy_mask,r00);
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq00             = _mm_mul_ps(iq0,jq0);
 331             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 332                                          vdwparam+vdwioffset0+vdwjidx0B,
 333                                          vdwparam+vdwioffset0+vdwjidx0C,
 334                                          vdwparam+vdwioffset0+vdwjidx0D,
 335                                          &c6_00,&c12_00);
 336
 337             /* Calculate table index by multiplying r with table scale and truncate to integer */
 338             rt               = _mm_mul_ps(r00,vftabscale);
 339             vfitab           = _mm_cvttps_epi32(rt);
 340             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 341             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 342
 343             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 344             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 345             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 346             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 347             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 348             _MM_TRANSPOSE4_PS(Y,F,G,H);
 349             Heps             = _mm_mul_ps(vfeps,H);
 350             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 351             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 352             velec            = _mm_mul_ps(qq00,VV);
 353             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 354             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 355
 356             /* CUBIC SPLINE TABLE DISPERSION */
 357             vfitab           = _mm_add_epi32(vfitab,ifour);
 358             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 359             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 360             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 361             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 362             _MM_TRANSPOSE4_PS(Y,F,G,H);
 363             Heps             = _mm_mul_ps(vfeps,H);
 364             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 365             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 366             vvdw6            = _mm_mul_ps(c6_00,VV);
 367             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 368             fvdw6            = _mm_mul_ps(c6_00,FF);
 369
 370             /* CUBIC SPLINE TABLE REPULSION */
 371             vfitab           = _mm_add_epi32(vfitab,ifour);
 372             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 373             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 374             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 375             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 376             _MM_TRANSPOSE4_PS(Y,F,G,H);
 377             Heps             = _mm_mul_ps(vfeps,H);
 378             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 379             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 380             vvdw12           = _mm_mul_ps(c12_00,VV);
 381             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 382             fvdw12           = _mm_mul_ps(c12_00,FF);
 383             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 384             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velec            = _mm_andnot_ps(dummy_mask,velec);
 388             velecsum         = _mm_add_ps(velecsum,velec);
 389             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 390             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 391
 392             fscal            = _mm_add_ps(felec,fvdw);
 393
 394             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm_mul_ps(fscal,dx00);
 398             ty               = _mm_mul_ps(fscal,dy00);
 399             tz               = _mm_mul_ps(fscal,dz00);
 400
 401             /* Update vectorial force */
 402             fix0             = _mm_add_ps(fix0,tx);
 403             fiy0             = _mm_add_ps(fiy0,ty);
 404             fiz0             = _mm_add_ps(fiz0,tz);
 405
 406             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 407                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 408                                                    tx,ty,tz);
 409
 410             /* Inner loop uses 74 flops */
 411         }
 412
 413         /* End of innermost loop */
 414
 415         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 416                                               f+i_coord_offset,fshift+i_shift_offset);
 417
 418         ggid                        = gid[iidx];
 419         /* Update potential energies */
 420         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 421         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 422
 423         /* Increment number of inner iterations */
 424         inneriter                  += j_index_end - j_index_start;
 425
 426         /* Outer loop uses 12 flops */
 427     }
 428
 429     /* Increment number of outer iterations */
 430     outeriter        += nri;
 431
 432     /* Update outer/inner flops */
 433
 434     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*12 + inneriter*74);
 435 }
 436 /*
 437  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sse2_single
 438  * Electrostatics interaction: CubicSplineTable
 439  * VdW interaction:            CubicSplineTable
 440  * Geometry:                   Particle-Particle
 441  * Calculate force/pot:        Force
 442  */
 443 void
 444 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sse2_single
 445                     (t_nblist * gmx_restrict                nlist,
 446                      rvec * gmx_restrict                    xx,
 447                      rvec * gmx_restrict                    ff,
 448                      t_forcerec * gmx_restrict              fr,
 449                      t_mdatoms * gmx_restrict               mdatoms,
 450                      nb_kernel_data_t * gmx_restrict        kernel_data,
 451                      t_nrnb * gmx_restrict                  nrnb)
 452 {
 453     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 454      * just 0 for non-waters.
 455      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 456      * jnr indices corresponding to data put in the four positions in the SIMD register.
 457      */
 458     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 459     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 460     int              jnrA,jnrB,jnrC,jnrD;
 461     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 462     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 463     real             shX,shY,shZ,rcutoff_scalar;
 464     real             *shiftvec,*fshift,*x,*f;
 465     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 466     int              vdwioffset0;
 467     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 468     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 469     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 470     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 471     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 472     real             *charge;
 473     int              nvdwtype;
 474     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 475     int              *vdwtype;
 476     real             *vdwparam;
 477     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 478     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 479     __m128i          vfitab;
 480     __m128i          ifour       = _mm_set1_epi32(4);
 481     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 482     real             *vftab;
 483     __m128           dummy_mask,cutoff_mask;
 484     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 485     __m128           one     = _mm_set1_ps(1.0);
 486     __m128           two     = _mm_set1_ps(2.0);
 487     x                = xx[0];
 488     f                = ff[0];
 489
 490     nri              = nlist->nri;
 491     iinr             = nlist->iinr;
 492     jindex           = nlist->jindex;
 493     jjnr             = nlist->jjnr;
 494     shiftidx         = nlist->shift;
 495     gid              = nlist->gid;
 496     shiftvec         = fr->shift_vec[0];
 497     fshift           = fr->fshift[0];
 498     facel            = _mm_set1_ps(fr->epsfac);
 499     charge           = mdatoms->chargeA;
 500     nvdwtype         = fr->ntype;
 501     vdwparam         = fr->nbfp;
 502     vdwtype          = mdatoms->typeA;
 503
 504     vftab            = kernel_data->table_elec_vdw->data;
 505     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 506
 507     /* Avoid stupid compiler warnings */
 508     jnrA = jnrB = jnrC = jnrD = 0;
 509     j_coord_offsetA = 0;
 510     j_coord_offsetB = 0;
 511     j_coord_offsetC = 0;
 512     j_coord_offsetD = 0;
 513
 514     outeriter        = 0;
 515     inneriter        = 0;
 516
 517     /* Start outer loop over neighborlists */
 518     for(iidx=0; iidx<nri; iidx++)
 519     {
 520         /* Load shift vector for this list */
 521         i_shift_offset   = DIM*shiftidx[iidx];
 522         shX              = shiftvec[i_shift_offset+XX];
 523         shY              = shiftvec[i_shift_offset+YY];
 524         shZ              = shiftvec[i_shift_offset+ZZ];
 525
 526         /* Load limits for loop over neighbors */
 527         j_index_start    = jindex[iidx];
 528         j_index_end      = jindex[iidx+1];
 529
 530         /* Get outer coordinate index */
 531         inr              = iinr[iidx];
 532         i_coord_offset   = DIM*inr;
 533
 534         /* Load i particle coords and add shift vector */
 535         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 536         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 537         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 538
 539         fix0             = _mm_setzero_ps();
 540         fiy0             = _mm_setzero_ps();
 541         fiz0             = _mm_setzero_ps();
 542
 543         /* Load parameters for i particles */
 544         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 545         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 546
 547         /* Start inner kernel loop */
 548         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 549         {
 550
 551             /* Get j neighbor index, and coordinate index */
 552             jnrA             = jjnr[jidx];
 553             jnrB             = jjnr[jidx+1];
 554             jnrC             = jjnr[jidx+2];
 555             jnrD             = jjnr[jidx+3];
 556
 557             j_coord_offsetA  = DIM*jnrA;
 558             j_coord_offsetB  = DIM*jnrB;
 559             j_coord_offsetC  = DIM*jnrC;
 560             j_coord_offsetD  = DIM*jnrD;
 561
 562             /* load j atom coordinates */
 563             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 564                                               x+j_coord_offsetC,x+j_coord_offsetD,
 565                                               &jx0,&jy0,&jz0);
 566
 567             /* Calculate displacement vector */
 568             dx00             = _mm_sub_ps(ix0,jx0);
 569             dy00             = _mm_sub_ps(iy0,jy0);
 570             dz00             = _mm_sub_ps(iz0,jz0);
 571
 572             /* Calculate squared distance and things based on it */
 573             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 574
 575             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 576
 577             /* Load parameters for j particles */
 578             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 579                                                               charge+jnrC+0,charge+jnrD+0);
 580             vdwjidx0A        = 2*vdwtype[jnrA+0];
 581             vdwjidx0B        = 2*vdwtype[jnrB+0];
 582             vdwjidx0C        = 2*vdwtype[jnrC+0];
 583             vdwjidx0D        = 2*vdwtype[jnrD+0];
 584
 585             /**************************
 586              * CALCULATE INTERACTIONS *
 587              **************************/
 588
 589             r00              = _mm_mul_ps(rsq00,rinv00);
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq00             = _mm_mul_ps(iq0,jq0);
 593             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 594                                          vdwparam+vdwioffset0+vdwjidx0B,
 595                                          vdwparam+vdwioffset0+vdwjidx0C,
 596                                          vdwparam+vdwioffset0+vdwjidx0D,
 597                                          &c6_00,&c12_00);
 598
 599             /* Calculate table index by multiplying r with table scale and truncate to integer */
 600             rt               = _mm_mul_ps(r00,vftabscale);
 601             vfitab           = _mm_cvttps_epi32(rt);
 602             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 603             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 604
 605             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 606             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 607             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 608             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 609             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 610             _MM_TRANSPOSE4_PS(Y,F,G,H);
 611             Heps             = _mm_mul_ps(vfeps,H);
 612             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 613             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 614             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 615
 616             /* CUBIC SPLINE TABLE DISPERSION */
 617             vfitab           = _mm_add_epi32(vfitab,ifour);
 618             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 619             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 620             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 621             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 622             _MM_TRANSPOSE4_PS(Y,F,G,H);
 623             Heps             = _mm_mul_ps(vfeps,H);
 624             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 625             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 626             fvdw6            = _mm_mul_ps(c6_00,FF);
 627
 628             /* CUBIC SPLINE TABLE REPULSION */
 629             vfitab           = _mm_add_epi32(vfitab,ifour);
 630             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 631             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 632             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 633             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 634             _MM_TRANSPOSE4_PS(Y,F,G,H);
 635             Heps             = _mm_mul_ps(vfeps,H);
 636             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 637             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 638             fvdw12           = _mm_mul_ps(c12_00,FF);
 639             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 640
 641             fscal            = _mm_add_ps(felec,fvdw);
 642
 643             /* Calculate temporary vectorial force */
 644             tx               = _mm_mul_ps(fscal,dx00);
 645             ty               = _mm_mul_ps(fscal,dy00);
 646             tz               = _mm_mul_ps(fscal,dz00);
 647
 648             /* Update vectorial force */
 649             fix0             = _mm_add_ps(fix0,tx);
 650             fiy0             = _mm_add_ps(fiy0,ty);
 651             fiz0             = _mm_add_ps(fiz0,tz);
 652
 653             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 654                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 655                                                    tx,ty,tz);
 656
 657             /* Inner loop uses 61 flops */
 658         }
 659
 660         if(jidx<j_index_end)
 661         {
 662
 663             /* Get j neighbor index, and coordinate index */
 664             jnrA             = jjnr[jidx];
 665             jnrB             = jjnr[jidx+1];
 666             jnrC             = jjnr[jidx+2];
 667             jnrD             = jjnr[jidx+3];
 668
 669             /* Sign of each element will be negative for non-real atoms.
 670              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 671              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 672              */
 673             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 674             jnrA       = (jnrA>=0) ? jnrA : 0;
 675             jnrB       = (jnrB>=0) ? jnrB : 0;
 676             jnrC       = (jnrC>=0) ? jnrC : 0;
 677             jnrD       = (jnrD>=0) ? jnrD : 0;
 678
 679             j_coord_offsetA  = DIM*jnrA;
 680             j_coord_offsetB  = DIM*jnrB;
 681             j_coord_offsetC  = DIM*jnrC;
 682             j_coord_offsetD  = DIM*jnrD;
 683
 684             /* load j atom coordinates */
 685             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 686                                               x+j_coord_offsetC,x+j_coord_offsetD,
 687                                               &jx0,&jy0,&jz0);
 688
 689             /* Calculate displacement vector */
 690             dx00             = _mm_sub_ps(ix0,jx0);
 691             dy00             = _mm_sub_ps(iy0,jy0);
 692             dz00             = _mm_sub_ps(iz0,jz0);
 693
 694             /* Calculate squared distance and things based on it */
 695             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 696
 697             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 698
 699             /* Load parameters for j particles */
 700             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 701                                                               charge+jnrC+0,charge+jnrD+0);
 702             vdwjidx0A        = 2*vdwtype[jnrA+0];
 703             vdwjidx0B        = 2*vdwtype[jnrB+0];
 704             vdwjidx0C        = 2*vdwtype[jnrC+0];
 705             vdwjidx0D        = 2*vdwtype[jnrD+0];
 706
 707             /**************************
 708              * CALCULATE INTERACTIONS *
 709              **************************/
 710
 711             r00              = _mm_mul_ps(rsq00,rinv00);
 712             r00              = _mm_andnot_ps(dummy_mask,r00);
 713
 714             /* Compute parameters for interactions between i and j atoms */
 715             qq00             = _mm_mul_ps(iq0,jq0);
 716             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 717                                          vdwparam+vdwioffset0+vdwjidx0B,
 718                                          vdwparam+vdwioffset0+vdwjidx0C,
 719                                          vdwparam+vdwioffset0+vdwjidx0D,
 720                                          &c6_00,&c12_00);
 721
 722             /* Calculate table index by multiplying r with table scale and truncate to integer */
 723             rt               = _mm_mul_ps(r00,vftabscale);
 724             vfitab           = _mm_cvttps_epi32(rt);
 725             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 726             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 727
 728             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 729             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 730             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 731             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 732             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 733             _MM_TRANSPOSE4_PS(Y,F,G,H);
 734             Heps             = _mm_mul_ps(vfeps,H);
 735             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 736             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 737             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 738
 739             /* CUBIC SPLINE TABLE DISPERSION */
 740             vfitab           = _mm_add_epi32(vfitab,ifour);
 741             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 742             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 743             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 744             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 745             _MM_TRANSPOSE4_PS(Y,F,G,H);
 746             Heps             = _mm_mul_ps(vfeps,H);
 747             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 748             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 749             fvdw6            = _mm_mul_ps(c6_00,FF);
 750
 751             /* CUBIC SPLINE TABLE REPULSION */
 752             vfitab           = _mm_add_epi32(vfitab,ifour);
 753             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 754             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 755             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 756             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 757             _MM_TRANSPOSE4_PS(Y,F,G,H);
 758             Heps             = _mm_mul_ps(vfeps,H);
 759             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 760             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 761             fvdw12           = _mm_mul_ps(c12_00,FF);
 762             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 763
 764             fscal            = _mm_add_ps(felec,fvdw);
 765
 766             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 767
 768             /* Calculate temporary vectorial force */
 769             tx               = _mm_mul_ps(fscal,dx00);
 770             ty               = _mm_mul_ps(fscal,dy00);
 771             tz               = _mm_mul_ps(fscal,dz00);
 772
 773             /* Update vectorial force */
 774             fix0             = _mm_add_ps(fix0,tx);
 775             fiy0             = _mm_add_ps(fiy0,ty);
 776             fiz0             = _mm_add_ps(fiz0,tz);
 777
 778             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 779                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 780                                                    tx,ty,tz);
 781
 782             /* Inner loop uses 62 flops */
 783         }
 784
 785         /* End of innermost loop */
 786
 787         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 788                                               f+i_coord_offset,fshift+i_shift_offset);
 789
 790         /* Increment number of inner iterations */
 791         inneriter                  += j_index_end - j_index_start;
 792
 793         /* Outer loop uses 10 flops */
 794     }
 795
 796     /* Increment number of outer iterations */
 797     outeriter        += nri;
 798
 799     /* Update outer/inner flops */
 800
 801     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*10 + inneriter*62);
 802 }