src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_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_ElecEw_VdwCSTab_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_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     __m128i          ewitab;
  84     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  85     real             *ewtab;
  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     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     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 111     ewtab            = fr->ic->tabq_coul_FDV0;
 112     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 113     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 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             r00              = _mm_mul_ps(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             qq00             = _mm_mul_ps(iq0,jq0);
 207             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 208                                          vdwparam+vdwioffset0+vdwjidx0B,
 209                                          vdwparam+vdwioffset0+vdwjidx0C,
 210                                          vdwparam+vdwioffset0+vdwjidx0D,
 211                                          &c6_00,&c12_00);
 212
 213             /* Calculate table index by multiplying r with table scale and truncate to integer */
 214             rt               = _mm_mul_ps(r00,vftabscale);
 215             vfitab           = _mm_cvttps_epi32(rt);
 216             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 217             vfitab           = _mm_slli_epi32(vfitab,3);
 218
 219             /* EWALD ELECTROSTATICS */
 220
 221             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 222             ewrt             = _mm_mul_ps(r00,ewtabscale);
 223             ewitab           = _mm_cvttps_epi32(ewrt);
 224             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 225             ewitab           = _mm_slli_epi32(ewitab,2);
 226             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 227             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 228             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 229             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 230             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 231             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 232             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 233             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 234             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 235
 236             /* CUBIC SPLINE TABLE DISPERSION */
 237             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 238             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 239             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 240             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 241             _MM_TRANSPOSE4_PS(Y,F,G,H);
 242             Heps             = _mm_mul_ps(vfeps,H);
 243             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 244             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 245             vvdw6            = _mm_mul_ps(c6_00,VV);
 246             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 247             fvdw6            = _mm_mul_ps(c6_00,FF);
 248
 249             /* CUBIC SPLINE TABLE REPULSION */
 250             vfitab           = _mm_add_epi32(vfitab,ifour);
 251             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 252             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 253             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 254             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 255             _MM_TRANSPOSE4_PS(Y,F,G,H);
 256             Heps             = _mm_mul_ps(vfeps,H);
 257             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 258             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 259             vvdw12           = _mm_mul_ps(c12_00,VV);
 260             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 261             fvdw12           = _mm_mul_ps(c12_00,FF);
 262             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 263             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             velecsum         = _mm_add_ps(velecsum,velec);
 267             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 268
 269             fscal            = _mm_add_ps(felec,fvdw);
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm_mul_ps(fscal,dx00);
 273             ty               = _mm_mul_ps(fscal,dy00);
 274             tz               = _mm_mul_ps(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_add_ps(fix0,tx);
 278             fiy0             = _mm_add_ps(fiy0,ty);
 279             fiz0             = _mm_add_ps(fiz0,tz);
 280
 281             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 282                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 283                                                    tx,ty,tz);
 284
 285             /* Inner loop uses 75 flops */
 286         }
 287
 288         if(jidx<j_index_end)
 289         {
 290
 291             /* Get j neighbor index, and coordinate index */
 292             jnrA             = jjnr[jidx];
 293             jnrB             = jjnr[jidx+1];
 294             jnrC             = jjnr[jidx+2];
 295             jnrD             = jjnr[jidx+3];
 296
 297             /* Sign of each element will be negative for non-real atoms.
 298              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 299              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 300              */
 301             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 302             jnrA       = (jnrA>=0) ? jnrA : 0;
 303             jnrB       = (jnrB>=0) ? jnrB : 0;
 304             jnrC       = (jnrC>=0) ? jnrC : 0;
 305             jnrD       = (jnrD>=0) ? jnrD : 0;
 306
 307             j_coord_offsetA  = DIM*jnrA;
 308             j_coord_offsetB  = DIM*jnrB;
 309             j_coord_offsetC  = DIM*jnrC;
 310             j_coord_offsetD  = DIM*jnrD;
 311
 312             /* load j atom coordinates */
 313             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 314                                               x+j_coord_offsetC,x+j_coord_offsetD,
 315                                               &jx0,&jy0,&jz0);
 316
 317             /* Calculate displacement vector */
 318             dx00             = _mm_sub_ps(ix0,jx0);
 319             dy00             = _mm_sub_ps(iy0,jy0);
 320             dz00             = _mm_sub_ps(iz0,jz0);
 321
 322             /* Calculate squared distance and things based on it */
 323             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 324
 325             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 326
 327             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 328
 329             /* Load parameters for j particles */
 330             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 331                                                               charge+jnrC+0,charge+jnrD+0);
 332             vdwjidx0A        = 2*vdwtype[jnrA+0];
 333             vdwjidx0B        = 2*vdwtype[jnrB+0];
 334             vdwjidx0C        = 2*vdwtype[jnrC+0];
 335             vdwjidx0D        = 2*vdwtype[jnrD+0];
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             r00              = _mm_mul_ps(rsq00,rinv00);
 342             r00              = _mm_andnot_ps(dummy_mask,r00);
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq00             = _mm_mul_ps(iq0,jq0);
 346             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 347                                          vdwparam+vdwioffset0+vdwjidx0B,
 348                                          vdwparam+vdwioffset0+vdwjidx0C,
 349                                          vdwparam+vdwioffset0+vdwjidx0D,
 350                                          &c6_00,&c12_00);
 351
 352             /* Calculate table index by multiplying r with table scale and truncate to integer */
 353             rt               = _mm_mul_ps(r00,vftabscale);
 354             vfitab           = _mm_cvttps_epi32(rt);
 355             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 356             vfitab           = _mm_slli_epi32(vfitab,3);
 357
 358             /* EWALD ELECTROSTATICS */
 359
 360             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 361             ewrt             = _mm_mul_ps(r00,ewtabscale);
 362             ewitab           = _mm_cvttps_epi32(ewrt);
 363             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 364             ewitab           = _mm_slli_epi32(ewitab,2);
 365             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 366             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 367             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 368             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 369             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 370             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 371             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 372             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 373             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 374
 375             /* CUBIC SPLINE TABLE DISPERSION */
 376             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 377             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 378             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 379             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 380             _MM_TRANSPOSE4_PS(Y,F,G,H);
 381             Heps             = _mm_mul_ps(vfeps,H);
 382             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 383             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 384             vvdw6            = _mm_mul_ps(c6_00,VV);
 385             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 386             fvdw6            = _mm_mul_ps(c6_00,FF);
 387
 388             /* CUBIC SPLINE TABLE REPULSION */
 389             vfitab           = _mm_add_epi32(vfitab,ifour);
 390             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 391             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 392             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 393             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 394             _MM_TRANSPOSE4_PS(Y,F,G,H);
 395             Heps             = _mm_mul_ps(vfeps,H);
 396             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 397             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 398             vvdw12           = _mm_mul_ps(c12_00,VV);
 399             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 400             fvdw12           = _mm_mul_ps(c12_00,FF);
 401             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 402             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _mm_andnot_ps(dummy_mask,velec);
 406             velecsum         = _mm_add_ps(velecsum,velec);
 407             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 408             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 409
 410             fscal            = _mm_add_ps(felec,fvdw);
 411
 412             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 413
 414             /* Calculate temporary vectorial force */
 415             tx               = _mm_mul_ps(fscal,dx00);
 416             ty               = _mm_mul_ps(fscal,dy00);
 417             tz               = _mm_mul_ps(fscal,dz00);
 418
 419             /* Update vectorial force */
 420             fix0             = _mm_add_ps(fix0,tx);
 421             fiy0             = _mm_add_ps(fiy0,ty);
 422             fiz0             = _mm_add_ps(fiz0,tz);
 423
 424             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 425                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 426                                                    tx,ty,tz);
 427
 428             /* Inner loop uses 76 flops */
 429         }
 430
 431         /* End of innermost loop */
 432
 433         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 434                                               f+i_coord_offset,fshift+i_shift_offset);
 435
 436         ggid                        = gid[iidx];
 437         /* Update potential energies */
 438         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 439         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 440
 441         /* Increment number of inner iterations */
 442         inneriter                  += j_index_end - j_index_start;
 443
 444         /* Outer loop uses 12 flops */
 445     }
 446
 447     /* Increment number of outer iterations */
 448     outeriter        += nri;
 449
 450     /* Update outer/inner flops */
 451
 452     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*12 + inneriter*76);
 453 }
 454 /*
 455  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse2_single
 456  * Electrostatics interaction: Ewald
 457  * VdW interaction:            CubicSplineTable
 458  * Geometry:                   Particle-Particle
 459  * Calculate force/pot:        Force
 460  */
 461 void
 462 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse2_single
 463                     (t_nblist * gmx_restrict                nlist,
 464                      rvec * gmx_restrict                    xx,
 465                      rvec * gmx_restrict                    ff,
 466                      t_forcerec * gmx_restrict              fr,
 467                      t_mdatoms * gmx_restrict               mdatoms,
 468                      nb_kernel_data_t * gmx_restrict        kernel_data,
 469                      t_nrnb * gmx_restrict                  nrnb)
 470 {
 471     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 472      * just 0 for non-waters.
 473      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 474      * jnr indices corresponding to data put in the four positions in the SIMD register.
 475      */
 476     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 477     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 478     int              jnrA,jnrB,jnrC,jnrD;
 479     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 480     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 481     real             shX,shY,shZ,rcutoff_scalar;
 482     real             *shiftvec,*fshift,*x,*f;
 483     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 484     int              vdwioffset0;
 485     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 486     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 487     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 488     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 489     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 490     real             *charge;
 491     int              nvdwtype;
 492     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 493     int              *vdwtype;
 494     real             *vdwparam;
 495     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 496     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 497     __m128i          vfitab;
 498     __m128i          ifour       = _mm_set1_epi32(4);
 499     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 500     real             *vftab;
 501     __m128i          ewitab;
 502     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 503     real             *ewtab;
 504     __m128           dummy_mask,cutoff_mask;
 505     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 506     __m128           one     = _mm_set1_ps(1.0);
 507     __m128           two     = _mm_set1_ps(2.0);
 508     x                = xx[0];
 509     f                = ff[0];
 510
 511     nri              = nlist->nri;
 512     iinr             = nlist->iinr;
 513     jindex           = nlist->jindex;
 514     jjnr             = nlist->jjnr;
 515     shiftidx         = nlist->shift;
 516     gid              = nlist->gid;
 517     shiftvec         = fr->shift_vec[0];
 518     fshift           = fr->fshift[0];
 519     facel            = _mm_set1_ps(fr->epsfac);
 520     charge           = mdatoms->chargeA;
 521     nvdwtype         = fr->ntype;
 522     vdwparam         = fr->nbfp;
 523     vdwtype          = mdatoms->typeA;
 524
 525     vftab            = kernel_data->table_vdw->data;
 526     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 527
 528     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 529     ewtab            = fr->ic->tabq_coul_F;
 530     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 531     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 532
 533     /* Avoid stupid compiler warnings */
 534     jnrA = jnrB = jnrC = jnrD = 0;
 535     j_coord_offsetA = 0;
 536     j_coord_offsetB = 0;
 537     j_coord_offsetC = 0;
 538     j_coord_offsetD = 0;
 539
 540     outeriter        = 0;
 541     inneriter        = 0;
 542
 543     /* Start outer loop over neighborlists */
 544     for(iidx=0; iidx<nri; iidx++)
 545     {
 546         /* Load shift vector for this list */
 547         i_shift_offset   = DIM*shiftidx[iidx];
 548         shX              = shiftvec[i_shift_offset+XX];
 549         shY              = shiftvec[i_shift_offset+YY];
 550         shZ              = shiftvec[i_shift_offset+ZZ];
 551
 552         /* Load limits for loop over neighbors */
 553         j_index_start    = jindex[iidx];
 554         j_index_end      = jindex[iidx+1];
 555
 556         /* Get outer coordinate index */
 557         inr              = iinr[iidx];
 558         i_coord_offset   = DIM*inr;
 559
 560         /* Load i particle coords and add shift vector */
 561         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 562         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 563         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 564
 565         fix0             = _mm_setzero_ps();
 566         fiy0             = _mm_setzero_ps();
 567         fiz0             = _mm_setzero_ps();
 568
 569         /* Load parameters for i particles */
 570         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 571         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 572
 573         /* Start inner kernel loop */
 574         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 575         {
 576
 577             /* Get j neighbor index, and coordinate index */
 578             jnrA             = jjnr[jidx];
 579             jnrB             = jjnr[jidx+1];
 580             jnrC             = jjnr[jidx+2];
 581             jnrD             = jjnr[jidx+3];
 582
 583             j_coord_offsetA  = DIM*jnrA;
 584             j_coord_offsetB  = DIM*jnrB;
 585             j_coord_offsetC  = DIM*jnrC;
 586             j_coord_offsetD  = DIM*jnrD;
 587
 588             /* load j atom coordinates */
 589             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 590                                               x+j_coord_offsetC,x+j_coord_offsetD,
 591                                               &jx0,&jy0,&jz0);
 592
 593             /* Calculate displacement vector */
 594             dx00             = _mm_sub_ps(ix0,jx0);
 595             dy00             = _mm_sub_ps(iy0,jy0);
 596             dz00             = _mm_sub_ps(iz0,jz0);
 597
 598             /* Calculate squared distance and things based on it */
 599             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 600
 601             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 602
 603             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 604
 605             /* Load parameters for j particles */
 606             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 607                                                               charge+jnrC+0,charge+jnrD+0);
 608             vdwjidx0A        = 2*vdwtype[jnrA+0];
 609             vdwjidx0B        = 2*vdwtype[jnrB+0];
 610             vdwjidx0C        = 2*vdwtype[jnrC+0];
 611             vdwjidx0D        = 2*vdwtype[jnrD+0];
 612
 613             /**************************
 614              * CALCULATE INTERACTIONS *
 615              **************************/
 616
 617             r00              = _mm_mul_ps(rsq00,rinv00);
 618
 619             /* Compute parameters for interactions between i and j atoms */
 620             qq00             = _mm_mul_ps(iq0,jq0);
 621             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 622                                          vdwparam+vdwioffset0+vdwjidx0B,
 623                                          vdwparam+vdwioffset0+vdwjidx0C,
 624                                          vdwparam+vdwioffset0+vdwjidx0D,
 625                                          &c6_00,&c12_00);
 626
 627             /* Calculate table index by multiplying r with table scale and truncate to integer */
 628             rt               = _mm_mul_ps(r00,vftabscale);
 629             vfitab           = _mm_cvttps_epi32(rt);
 630             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 631             vfitab           = _mm_slli_epi32(vfitab,3);
 632
 633             /* EWALD ELECTROSTATICS */
 634
 635             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 636             ewrt             = _mm_mul_ps(r00,ewtabscale);
 637             ewitab           = _mm_cvttps_epi32(ewrt);
 638             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 639             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 640                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 641                                          &ewtabF,&ewtabFn);
 642             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 643             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 644
 645             /* CUBIC SPLINE TABLE DISPERSION */
 646             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 647             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 648             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 649             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 650             _MM_TRANSPOSE4_PS(Y,F,G,H);
 651             Heps             = _mm_mul_ps(vfeps,H);
 652             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 653             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 654             fvdw6            = _mm_mul_ps(c6_00,FF);
 655
 656             /* CUBIC SPLINE TABLE REPULSION */
 657             vfitab           = _mm_add_epi32(vfitab,ifour);
 658             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 659             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 660             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 661             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 662             _MM_TRANSPOSE4_PS(Y,F,G,H);
 663             Heps             = _mm_mul_ps(vfeps,H);
 664             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 665             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 666             fvdw12           = _mm_mul_ps(c12_00,FF);
 667             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 668
 669             fscal            = _mm_add_ps(felec,fvdw);
 670
 671             /* Calculate temporary vectorial force */
 672             tx               = _mm_mul_ps(fscal,dx00);
 673             ty               = _mm_mul_ps(fscal,dy00);
 674             tz               = _mm_mul_ps(fscal,dz00);
 675
 676             /* Update vectorial force */
 677             fix0             = _mm_add_ps(fix0,tx);
 678             fiy0             = _mm_add_ps(fiy0,ty);
 679             fiz0             = _mm_add_ps(fiz0,tz);
 680
 681             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 682                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 683                                                    tx,ty,tz);
 684
 685             /* Inner loop uses 62 flops */
 686         }
 687
 688         if(jidx<j_index_end)
 689         {
 690
 691             /* Get j neighbor index, and coordinate index */
 692             jnrA             = jjnr[jidx];
 693             jnrB             = jjnr[jidx+1];
 694             jnrC             = jjnr[jidx+2];
 695             jnrD             = jjnr[jidx+3];
 696
 697             /* Sign of each element will be negative for non-real atoms.
 698              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 699              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 700              */
 701             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 702             jnrA       = (jnrA>=0) ? jnrA : 0;
 703             jnrB       = (jnrB>=0) ? jnrB : 0;
 704             jnrC       = (jnrC>=0) ? jnrC : 0;
 705             jnrD       = (jnrD>=0) ? jnrD : 0;
 706
 707             j_coord_offsetA  = DIM*jnrA;
 708             j_coord_offsetB  = DIM*jnrB;
 709             j_coord_offsetC  = DIM*jnrC;
 710             j_coord_offsetD  = DIM*jnrD;
 711
 712             /* load j atom coordinates */
 713             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 714                                               x+j_coord_offsetC,x+j_coord_offsetD,
 715                                               &jx0,&jy0,&jz0);
 716
 717             /* Calculate displacement vector */
 718             dx00             = _mm_sub_ps(ix0,jx0);
 719             dy00             = _mm_sub_ps(iy0,jy0);
 720             dz00             = _mm_sub_ps(iz0,jz0);
 721
 722             /* Calculate squared distance and things based on it */
 723             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 724
 725             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 726
 727             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 728
 729             /* Load parameters for j particles */
 730             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 731                                                               charge+jnrC+0,charge+jnrD+0);
 732             vdwjidx0A        = 2*vdwtype[jnrA+0];
 733             vdwjidx0B        = 2*vdwtype[jnrB+0];
 734             vdwjidx0C        = 2*vdwtype[jnrC+0];
 735             vdwjidx0D        = 2*vdwtype[jnrD+0];
 736
 737             /**************************
 738              * CALCULATE INTERACTIONS *
 739              **************************/
 740
 741             r00              = _mm_mul_ps(rsq00,rinv00);
 742             r00              = _mm_andnot_ps(dummy_mask,r00);
 743
 744             /* Compute parameters for interactions between i and j atoms */
 745             qq00             = _mm_mul_ps(iq0,jq0);
 746             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 747                                          vdwparam+vdwioffset0+vdwjidx0B,
 748                                          vdwparam+vdwioffset0+vdwjidx0C,
 749                                          vdwparam+vdwioffset0+vdwjidx0D,
 750                                          &c6_00,&c12_00);
 751
 752             /* Calculate table index by multiplying r with table scale and truncate to integer */
 753             rt               = _mm_mul_ps(r00,vftabscale);
 754             vfitab           = _mm_cvttps_epi32(rt);
 755             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 756             vfitab           = _mm_slli_epi32(vfitab,3);
 757
 758             /* EWALD ELECTROSTATICS */
 759
 760             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 761             ewrt             = _mm_mul_ps(r00,ewtabscale);
 762             ewitab           = _mm_cvttps_epi32(ewrt);
 763             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 764             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 765                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 766                                          &ewtabF,&ewtabFn);
 767             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 768             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 769
 770             /* CUBIC SPLINE TABLE DISPERSION */
 771             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 772             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 773             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 774             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 775             _MM_TRANSPOSE4_PS(Y,F,G,H);
 776             Heps             = _mm_mul_ps(vfeps,H);
 777             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 778             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 779             fvdw6            = _mm_mul_ps(c6_00,FF);
 780
 781             /* CUBIC SPLINE TABLE REPULSION */
 782             vfitab           = _mm_add_epi32(vfitab,ifour);
 783             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 784             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 785             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 786             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 787             _MM_TRANSPOSE4_PS(Y,F,G,H);
 788             Heps             = _mm_mul_ps(vfeps,H);
 789             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 790             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 791             fvdw12           = _mm_mul_ps(c12_00,FF);
 792             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 793
 794             fscal            = _mm_add_ps(felec,fvdw);
 795
 796             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 797
 798             /* Calculate temporary vectorial force */
 799             tx               = _mm_mul_ps(fscal,dx00);
 800             ty               = _mm_mul_ps(fscal,dy00);
 801             tz               = _mm_mul_ps(fscal,dz00);
 802
 803             /* Update vectorial force */
 804             fix0             = _mm_add_ps(fix0,tx);
 805             fiy0             = _mm_add_ps(fiy0,ty);
 806             fiz0             = _mm_add_ps(fiz0,tz);
 807
 808             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 809                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 810                                                    tx,ty,tz);
 811
 812             /* Inner loop uses 63 flops */
 813         }
 814
 815         /* End of innermost loop */
 816
 817         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 818                                               f+i_coord_offset,fshift+i_shift_offset);
 819
 820         /* Increment number of inner iterations */
 821         inneriter                  += j_index_end - j_index_start;
 822
 823         /* Outer loop uses 10 flops */
 824     }
 825
 826     /* Increment number of outer iterations */
 827     outeriter        += nri;
 828
 829     /* Update outer/inner flops */
 830
 831     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*10 + inneriter*63);
 832 }