src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwLJEw_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  97     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  98     __m128           c6grid_00;
  99     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 100     real             *vdwgridparam;
 101     __m128           one_half = _mm_set1_ps(0.5);
 102     __m128           minus_one = _mm_set1_ps(-1.0);
 103     __m128i          ewitab;
 104     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 105     real             *ewtab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126     vdwgridparam     = fr->ljpme_c6grid;
 127     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 128     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 129     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 130
 131     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 132     ewtab            = fr->ic->tabq_coul_FDV0;
 133     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 134     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     for(iidx=0;iidx<4*DIM;iidx++)
 147     {
 148         scratch[iidx] = 0.0;
 149     }
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 167
 168         fix0             = _mm_setzero_ps();
 169         fiy0             = _mm_setzero_ps();
 170         fiz0             = _mm_setzero_ps();
 171
 172         /* Load parameters for i particles */
 173         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 174         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_ps();
 178         vvdwsum          = _mm_setzero_ps();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             jnrC             = jjnr[jidx+2];
 188             jnrD             = jjnr[jidx+3];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193
 194             /* load j atom coordinates */
 195             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                               x+j_coord_offsetC,x+j_coord_offsetD,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm_sub_ps(ix0,jx0);
 201             dy00             = _mm_sub_ps(iy0,jy0);
 202             dz00             = _mm_sub_ps(iz0,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 206
 207             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 208
 209             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 210
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 213                                                               charge+jnrC+0,charge+jnrD+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216             vdwjidx0C        = 2*vdwtype[jnrC+0];
 217             vdwjidx0D        = 2*vdwtype[jnrD+0];
 218
 219             /**************************
 220              * CALCULATE INTERACTIONS *
 221              **************************/
 222
 223             r00              = _mm_mul_ps(rsq00,rinv00);
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq00             = _mm_mul_ps(iq0,jq0);
 227             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 228                                          vdwparam+vdwioffset0+vdwjidx0B,
 229                                          vdwparam+vdwioffset0+vdwjidx0C,
 230                                          vdwparam+vdwioffset0+vdwjidx0D,
 231                                          &c6_00,&c12_00);
 232             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 233                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 234                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 235                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 236
 237             /* EWALD ELECTROSTATICS */
 238
 239             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 240             ewrt             = _mm_mul_ps(r00,ewtabscale);
 241             ewitab           = _mm_cvttps_epi32(ewrt);
 242             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 243             ewitab           = _mm_slli_epi32(ewitab,2);
 244             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 245             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 246             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 247             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 248             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 249             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 250             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 251             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 252             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 253
 254             /* Analytical LJ-PME */
 255             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 256             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 257             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 258             exponent         = gmx_simd_exp_r(ewcljrsq);
 259             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 260             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 261             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 262             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 263             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 264             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 265             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 266             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             velecsum         = _mm_add_ps(velecsum,velec);
 270             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 271
 272             fscal            = _mm_add_ps(felec,fvdw);
 273
 274             /* Calculate temporary vectorial force */
 275             tx               = _mm_mul_ps(fscal,dx00);
 276             ty               = _mm_mul_ps(fscal,dy00);
 277             tz               = _mm_mul_ps(fscal,dz00);
 278
 279             /* Update vectorial force */
 280             fix0             = _mm_add_ps(fix0,tx);
 281             fiy0             = _mm_add_ps(fiy0,ty);
 282             fiz0             = _mm_add_ps(fiz0,tz);
 283
 284             fjptrA             = f+j_coord_offsetA;
 285             fjptrB             = f+j_coord_offsetB;
 286             fjptrC             = f+j_coord_offsetC;
 287             fjptrD             = f+j_coord_offsetD;
 288             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 289
 290             /* Inner loop uses 69 flops */
 291         }
 292
 293         if(jidx<j_index_end)
 294         {
 295
 296             /* Get j neighbor index, and coordinate index */
 297             jnrlistA         = jjnr[jidx];
 298             jnrlistB         = jjnr[jidx+1];
 299             jnrlistC         = jjnr[jidx+2];
 300             jnrlistD         = jjnr[jidx+3];
 301             /* Sign of each element will be negative for non-real atoms.
 302              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 303              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 304              */
 305             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 306             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 307             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 308             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 309             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 310             j_coord_offsetA  = DIM*jnrA;
 311             j_coord_offsetB  = DIM*jnrB;
 312             j_coord_offsetC  = DIM*jnrC;
 313             j_coord_offsetD  = DIM*jnrD;
 314
 315             /* load j atom coordinates */
 316             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 317                                               x+j_coord_offsetC,x+j_coord_offsetD,
 318                                               &jx0,&jy0,&jz0);
 319
 320             /* Calculate displacement vector */
 321             dx00             = _mm_sub_ps(ix0,jx0);
 322             dy00             = _mm_sub_ps(iy0,jy0);
 323             dz00             = _mm_sub_ps(iz0,jz0);
 324
 325             /* Calculate squared distance and things based on it */
 326             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 327
 328             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 329
 330             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 331
 332             /* Load parameters for j particles */
 333             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 334                                                               charge+jnrC+0,charge+jnrD+0);
 335             vdwjidx0A        = 2*vdwtype[jnrA+0];
 336             vdwjidx0B        = 2*vdwtype[jnrB+0];
 337             vdwjidx0C        = 2*vdwtype[jnrC+0];
 338             vdwjidx0D        = 2*vdwtype[jnrD+0];
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             r00              = _mm_mul_ps(rsq00,rinv00);
 345             r00              = _mm_andnot_ps(dummy_mask,r00);
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq00             = _mm_mul_ps(iq0,jq0);
 349             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 350                                          vdwparam+vdwioffset0+vdwjidx0B,
 351                                          vdwparam+vdwioffset0+vdwjidx0C,
 352                                          vdwparam+vdwioffset0+vdwjidx0D,
 353                                          &c6_00,&c12_00);
 354             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 355                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 356                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 357                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 358
 359             /* EWALD ELECTROSTATICS */
 360
 361             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 362             ewrt             = _mm_mul_ps(r00,ewtabscale);
 363             ewitab           = _mm_cvttps_epi32(ewrt);
 364             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 365             ewitab           = _mm_slli_epi32(ewitab,2);
 366             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 367             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 368             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 369             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 370             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 371             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 372             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 373             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 374             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 375
 376             /* Analytical LJ-PME */
 377             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 378             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 379             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 380             exponent         = gmx_simd_exp_r(ewcljrsq);
 381             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 382             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 383             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 384             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 385             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 386             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 387             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 388             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velec            = _mm_andnot_ps(dummy_mask,velec);
 392             velecsum         = _mm_add_ps(velecsum,velec);
 393             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 394             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 395
 396             fscal            = _mm_add_ps(felec,fvdw);
 397
 398             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 399
 400             /* Calculate temporary vectorial force */
 401             tx               = _mm_mul_ps(fscal,dx00);
 402             ty               = _mm_mul_ps(fscal,dy00);
 403             tz               = _mm_mul_ps(fscal,dz00);
 404
 405             /* Update vectorial force */
 406             fix0             = _mm_add_ps(fix0,tx);
 407             fiy0             = _mm_add_ps(fiy0,ty);
 408             fiz0             = _mm_add_ps(fiz0,tz);
 409
 410             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 411             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 412             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 413             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 414             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 415
 416             /* Inner loop uses 70 flops */
 417         }
 418
 419         /* End of innermost loop */
 420
 421         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 422                                               f+i_coord_offset,fshift+i_shift_offset);
 423
 424         ggid                        = gid[iidx];
 425         /* Update potential energies */
 426         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 427         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 428
 429         /* Increment number of inner iterations */
 430         inneriter                  += j_index_end - j_index_start;
 431
 432         /* Outer loop uses 9 flops */
 433     }
 434
 435     /* Increment number of outer iterations */
 436     outeriter        += nri;
 437
 438     /* Update outer/inner flops */
 439
 440     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*70);
 441 }
 442 /*
 443  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse2_single
 444  * Electrostatics interaction: Ewald
 445  * VdW interaction:            LJEwald
 446  * Geometry:                   Particle-Particle
 447  * Calculate force/pot:        Force
 448  */
 449 void
 450 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse2_single
 451                     (t_nblist                    * gmx_restrict       nlist,
 452                      rvec                        * gmx_restrict          xx,
 453                      rvec                        * gmx_restrict          ff,
 454                      t_forcerec                  * gmx_restrict          fr,
 455                      t_mdatoms                   * gmx_restrict     mdatoms,
 456                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 457                      t_nrnb                      * gmx_restrict        nrnb)
 458 {
 459     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 460      * just 0 for non-waters.
 461      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 462      * jnr indices corresponding to data put in the four positions in the SIMD register.
 463      */
 464     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 465     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 466     int              jnrA,jnrB,jnrC,jnrD;
 467     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 468     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 469     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 470     real             rcutoff_scalar;
 471     real             *shiftvec,*fshift,*x,*f;
 472     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 473     real             scratch[4*DIM];
 474     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 475     int              vdwioffset0;
 476     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 477     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 478     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 479     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 480     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 481     real             *charge;
 482     int              nvdwtype;
 483     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 484     int              *vdwtype;
 485     real             *vdwparam;
 486     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 487     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 488     __m128           c6grid_00;
 489     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 490     real             *vdwgridparam;
 491     __m128           one_half = _mm_set1_ps(0.5);
 492     __m128           minus_one = _mm_set1_ps(-1.0);
 493     __m128i          ewitab;
 494     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 495     real             *ewtab;
 496     __m128           dummy_mask,cutoff_mask;
 497     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 498     __m128           one     = _mm_set1_ps(1.0);
 499     __m128           two     = _mm_set1_ps(2.0);
 500     x                = xx[0];
 501     f                = ff[0];
 502
 503     nri              = nlist->nri;
 504     iinr             = nlist->iinr;
 505     jindex           = nlist->jindex;
 506     jjnr             = nlist->jjnr;
 507     shiftidx         = nlist->shift;
 508     gid              = nlist->gid;
 509     shiftvec         = fr->shift_vec[0];
 510     fshift           = fr->fshift[0];
 511     facel            = _mm_set1_ps(fr->epsfac);
 512     charge           = mdatoms->chargeA;
 513     nvdwtype         = fr->ntype;
 514     vdwparam         = fr->nbfp;
 515     vdwtype          = mdatoms->typeA;
 516     vdwgridparam     = fr->ljpme_c6grid;
 517     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 518     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 519     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 520
 521     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 522     ewtab            = fr->ic->tabq_coul_F;
 523     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 524     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 525
 526     /* Avoid stupid compiler warnings */
 527     jnrA = jnrB = jnrC = jnrD = 0;
 528     j_coord_offsetA = 0;
 529     j_coord_offsetB = 0;
 530     j_coord_offsetC = 0;
 531     j_coord_offsetD = 0;
 532
 533     outeriter        = 0;
 534     inneriter        = 0;
 535
 536     for(iidx=0;iidx<4*DIM;iidx++)
 537     {
 538         scratch[iidx] = 0.0;
 539     }
 540
 541     /* Start outer loop over neighborlists */
 542     for(iidx=0; iidx<nri; iidx++)
 543     {
 544         /* Load shift vector for this list */
 545         i_shift_offset   = DIM*shiftidx[iidx];
 546
 547         /* Load limits for loop over neighbors */
 548         j_index_start    = jindex[iidx];
 549         j_index_end      = jindex[iidx+1];
 550
 551         /* Get outer coordinate index */
 552         inr              = iinr[iidx];
 553         i_coord_offset   = DIM*inr;
 554
 555         /* Load i particle coords and add shift vector */
 556         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 557
 558         fix0             = _mm_setzero_ps();
 559         fiy0             = _mm_setzero_ps();
 560         fiz0             = _mm_setzero_ps();
 561
 562         /* Load parameters for i particles */
 563         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 564         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 565
 566         /* Start inner kernel loop */
 567         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 568         {
 569
 570             /* Get j neighbor index, and coordinate index */
 571             jnrA             = jjnr[jidx];
 572             jnrB             = jjnr[jidx+1];
 573             jnrC             = jjnr[jidx+2];
 574             jnrD             = jjnr[jidx+3];
 575             j_coord_offsetA  = DIM*jnrA;
 576             j_coord_offsetB  = DIM*jnrB;
 577             j_coord_offsetC  = DIM*jnrC;
 578             j_coord_offsetD  = DIM*jnrD;
 579
 580             /* load j atom coordinates */
 581             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 582                                               x+j_coord_offsetC,x+j_coord_offsetD,
 583                                               &jx0,&jy0,&jz0);
 584
 585             /* Calculate displacement vector */
 586             dx00             = _mm_sub_ps(ix0,jx0);
 587             dy00             = _mm_sub_ps(iy0,jy0);
 588             dz00             = _mm_sub_ps(iz0,jz0);
 589
 590             /* Calculate squared distance and things based on it */
 591             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 592
 593             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 594
 595             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 596
 597             /* Load parameters for j particles */
 598             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 599                                                               charge+jnrC+0,charge+jnrD+0);
 600             vdwjidx0A        = 2*vdwtype[jnrA+0];
 601             vdwjidx0B        = 2*vdwtype[jnrB+0];
 602             vdwjidx0C        = 2*vdwtype[jnrC+0];
 603             vdwjidx0D        = 2*vdwtype[jnrD+0];
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             r00              = _mm_mul_ps(rsq00,rinv00);
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq00             = _mm_mul_ps(iq0,jq0);
 613             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 614                                          vdwparam+vdwioffset0+vdwjidx0B,
 615                                          vdwparam+vdwioffset0+vdwjidx0C,
 616                                          vdwparam+vdwioffset0+vdwjidx0D,
 617                                          &c6_00,&c12_00);
 618             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 619                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 620                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 621                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 622
 623             /* EWALD ELECTROSTATICS */
 624
 625             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 626             ewrt             = _mm_mul_ps(r00,ewtabscale);
 627             ewitab           = _mm_cvttps_epi32(ewrt);
 628             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 629             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 630                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 631                                          &ewtabF,&ewtabFn);
 632             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 633             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 634
 635             /* Analytical LJ-PME */
 636             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 637             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 638             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 639             exponent         = gmx_simd_exp_r(ewcljrsq);
 640             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 641             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 642             /* f6A = 6 * C6grid * (1 - poly) */
 643             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 644             /* f6B = C6grid * exponent * beta^6 */
 645             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 646             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 647             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 648
 649             fscal            = _mm_add_ps(felec,fvdw);
 650
 651             /* Calculate temporary vectorial force */
 652             tx               = _mm_mul_ps(fscal,dx00);
 653             ty               = _mm_mul_ps(fscal,dy00);
 654             tz               = _mm_mul_ps(fscal,dz00);
 655
 656             /* Update vectorial force */
 657             fix0             = _mm_add_ps(fix0,tx);
 658             fiy0             = _mm_add_ps(fiy0,ty);
 659             fiz0             = _mm_add_ps(fiz0,tz);
 660
 661             fjptrA             = f+j_coord_offsetA;
 662             fjptrB             = f+j_coord_offsetB;
 663             fjptrC             = f+j_coord_offsetC;
 664             fjptrD             = f+j_coord_offsetD;
 665             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 666
 667             /* Inner loop uses 59 flops */
 668         }
 669
 670         if(jidx<j_index_end)
 671         {
 672
 673             /* Get j neighbor index, and coordinate index */
 674             jnrlistA         = jjnr[jidx];
 675             jnrlistB         = jjnr[jidx+1];
 676             jnrlistC         = jjnr[jidx+2];
 677             jnrlistD         = jjnr[jidx+3];
 678             /* Sign of each element will be negative for non-real atoms.
 679              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 680              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 681              */
 682             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 683             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 684             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 685             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 686             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 687             j_coord_offsetA  = DIM*jnrA;
 688             j_coord_offsetB  = DIM*jnrB;
 689             j_coord_offsetC  = DIM*jnrC;
 690             j_coord_offsetD  = DIM*jnrD;
 691
 692             /* load j atom coordinates */
 693             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 694                                               x+j_coord_offsetC,x+j_coord_offsetD,
 695                                               &jx0,&jy0,&jz0);
 696
 697             /* Calculate displacement vector */
 698             dx00             = _mm_sub_ps(ix0,jx0);
 699             dy00             = _mm_sub_ps(iy0,jy0);
 700             dz00             = _mm_sub_ps(iz0,jz0);
 701
 702             /* Calculate squared distance and things based on it */
 703             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 704
 705             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 706
 707             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 708
 709             /* Load parameters for j particles */
 710             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 711                                                               charge+jnrC+0,charge+jnrD+0);
 712             vdwjidx0A        = 2*vdwtype[jnrA+0];
 713             vdwjidx0B        = 2*vdwtype[jnrB+0];
 714             vdwjidx0C        = 2*vdwtype[jnrC+0];
 715             vdwjidx0D        = 2*vdwtype[jnrD+0];
 716
 717             /**************************
 718              * CALCULATE INTERACTIONS *
 719              **************************/
 720
 721             r00              = _mm_mul_ps(rsq00,rinv00);
 722             r00              = _mm_andnot_ps(dummy_mask,r00);
 723
 724             /* Compute parameters for interactions between i and j atoms */
 725             qq00             = _mm_mul_ps(iq0,jq0);
 726             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 727                                          vdwparam+vdwioffset0+vdwjidx0B,
 728                                          vdwparam+vdwioffset0+vdwjidx0C,
 729                                          vdwparam+vdwioffset0+vdwjidx0D,
 730                                          &c6_00,&c12_00);
 731             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 732                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 733                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 734                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 735
 736             /* EWALD ELECTROSTATICS */
 737
 738             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 739             ewrt             = _mm_mul_ps(r00,ewtabscale);
 740             ewitab           = _mm_cvttps_epi32(ewrt);
 741             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 742             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 743                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 744                                          &ewtabF,&ewtabFn);
 745             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 746             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 747
 748             /* Analytical LJ-PME */
 749             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 750             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 751             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 752             exponent         = gmx_simd_exp_r(ewcljrsq);
 753             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 754             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 755             /* f6A = 6 * C6grid * (1 - poly) */
 756             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 757             /* f6B = C6grid * exponent * beta^6 */
 758             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 759             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 760             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 761
 762             fscal            = _mm_add_ps(felec,fvdw);
 763
 764             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 765
 766             /* Calculate temporary vectorial force */
 767             tx               = _mm_mul_ps(fscal,dx00);
 768             ty               = _mm_mul_ps(fscal,dy00);
 769             tz               = _mm_mul_ps(fscal,dz00);
 770
 771             /* Update vectorial force */
 772             fix0             = _mm_add_ps(fix0,tx);
 773             fiy0             = _mm_add_ps(fiy0,ty);
 774             fiz0             = _mm_add_ps(fiz0,tz);
 775
 776             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 777             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 778             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 779             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 780             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 781
 782             /* Inner loop uses 60 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 7 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*7 + inneriter*60);
 802 }