src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJEw_GeomP1P1_sse4_1_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 sse4_1_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sse4_1_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sse4_1_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128           c6grid_00;
  97     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  98     real             *vdwgridparam;
  99     __m128           one_half  = _mm_set1_ps(0.5);
 100     __m128           minus_one = _mm_set1_ps(-1.0);
 101     __m128i          ewitab;
 102     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 103     real             *ewtab;
 104     __m128           dummy_mask,cutoff_mask;
 105     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 106     __m128           one     = _mm_set1_ps(1.0);
 107     __m128           two     = _mm_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_ps(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124     vdwgridparam     = fr->ljpme_c6grid;
 125     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 126     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 127     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 128
 129     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 130     ewtab            = fr->ic->tabq_coul_FDV0;
 131     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 132     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = jnrC = jnrD = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138     j_coord_offsetC = 0;
 139     j_coord_offsetD = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 165
 166         fix0             = _mm_setzero_ps();
 167         fiy0             = _mm_setzero_ps();
 168         fiz0             = _mm_setzero_ps();
 169
 170         /* Load parameters for i particles */
 171         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 172         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_ps();
 176         vvdwsum          = _mm_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               x+j_coord_offsetC,x+j_coord_offsetD,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_ps(ix0,jx0);
 199             dy00             = _mm_sub_ps(iy0,jy0);
 200             dz00             = _mm_sub_ps(iz0,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 204
 205             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 206
 207             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 208
 209             /* Load parameters for j particles */
 210             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 211                                                               charge+jnrC+0,charge+jnrD+0);
 212             vdwjidx0A        = 2*vdwtype[jnrA+0];
 213             vdwjidx0B        = 2*vdwtype[jnrB+0];
 214             vdwjidx0C        = 2*vdwtype[jnrC+0];
 215             vdwjidx0D        = 2*vdwtype[jnrD+0];
 216
 217             /**************************
 218              * CALCULATE INTERACTIONS *
 219              **************************/
 220
 221             r00              = _mm_mul_ps(rsq00,rinv00);
 222
 223             /* Compute parameters for interactions between i and j atoms */
 224             qq00             = _mm_mul_ps(iq0,jq0);
 225             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 226                                          vdwparam+vdwioffset0+vdwjidx0B,
 227                                          vdwparam+vdwioffset0+vdwjidx0C,
 228                                          vdwparam+vdwioffset0+vdwjidx0D,
 229                                          &c6_00,&c12_00);
 230
 231             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 232                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 233                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 234                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 235
 236             /* EWALD ELECTROSTATICS */
 237
 238             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 239             ewrt             = _mm_mul_ps(r00,ewtabscale);
 240             ewitab           = _mm_cvttps_epi32(ewrt);
 241             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 242             ewitab           = _mm_slli_epi32(ewitab,2);
 243             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 244             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 245             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 246             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 247             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 248             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 249             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 250             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 251             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 252
 253             /* Analytical LJ-PME */
 254             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 255             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 256             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 257             exponent         = gmx_simd_exp_r(ewcljrsq);
 258             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 259             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 260             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 261             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 262             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 263             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 264             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 265             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);
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             velecsum         = _mm_add_ps(velecsum,velec);
 269             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 270
 271             fscal            = _mm_add_ps(felec,fvdw);
 272
 273             /* Calculate temporary vectorial force */
 274             tx               = _mm_mul_ps(fscal,dx00);
 275             ty               = _mm_mul_ps(fscal,dy00);
 276             tz               = _mm_mul_ps(fscal,dz00);
 277
 278             /* Update vectorial force */
 279             fix0             = _mm_add_ps(fix0,tx);
 280             fiy0             = _mm_add_ps(fiy0,ty);
 281             fiz0             = _mm_add_ps(fiz0,tz);
 282
 283             fjptrA             = f+j_coord_offsetA;
 284             fjptrB             = f+j_coord_offsetB;
 285             fjptrC             = f+j_coord_offsetC;
 286             fjptrD             = f+j_coord_offsetD;
 287             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 288
 289             /* Inner loop uses 69 flops */
 290         }
 291
 292         if(jidx<j_index_end)
 293         {
 294
 295             /* Get j neighbor index, and coordinate index */
 296             jnrlistA         = jjnr[jidx];
 297             jnrlistB         = jjnr[jidx+1];
 298             jnrlistC         = jjnr[jidx+2];
 299             jnrlistD         = jjnr[jidx+3];
 300             /* Sign of each element will be negative for non-real atoms.
 301              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 302              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 303              */
 304             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 305             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 306             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 307             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 308             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 309             j_coord_offsetA  = DIM*jnrA;
 310             j_coord_offsetB  = DIM*jnrB;
 311             j_coord_offsetC  = DIM*jnrC;
 312             j_coord_offsetD  = DIM*jnrD;
 313
 314             /* load j atom coordinates */
 315             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 316                                               x+j_coord_offsetC,x+j_coord_offsetD,
 317                                               &jx0,&jy0,&jz0);
 318
 319             /* Calculate displacement vector */
 320             dx00             = _mm_sub_ps(ix0,jx0);
 321             dy00             = _mm_sub_ps(iy0,jy0);
 322             dz00             = _mm_sub_ps(iz0,jz0);
 323
 324             /* Calculate squared distance and things based on it */
 325             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 326
 327             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 328
 329             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 330
 331             /* Load parameters for j particles */
 332             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 333                                                               charge+jnrC+0,charge+jnrD+0);
 334             vdwjidx0A        = 2*vdwtype[jnrA+0];
 335             vdwjidx0B        = 2*vdwtype[jnrB+0];
 336             vdwjidx0C        = 2*vdwtype[jnrC+0];
 337             vdwjidx0D        = 2*vdwtype[jnrD+0];
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             r00              = _mm_mul_ps(rsq00,rinv00);
 344             r00              = _mm_andnot_ps(dummy_mask,r00);
 345
 346             /* Compute parameters for interactions between i and j atoms */
 347             qq00             = _mm_mul_ps(iq0,jq0);
 348             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 349                                          vdwparam+vdwioffset0+vdwjidx0B,
 350                                          vdwparam+vdwioffset0+vdwjidx0C,
 351                                          vdwparam+vdwioffset0+vdwjidx0D,
 352                                          &c6_00,&c12_00);
 353
 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_round_ps(ewrt, _MM_FROUND_FLOOR));
 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_sse4_1_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_sse4_1_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
 619             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 620                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 621                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 622                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 623
 624             /* EWALD ELECTROSTATICS */
 625
 626             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 627             ewrt             = _mm_mul_ps(r00,ewtabscale);
 628             ewitab           = _mm_cvttps_epi32(ewrt);
 629             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 630             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 631                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 632                                          &ewtabF,&ewtabFn);
 633             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 634             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 635
 636             /* Analytical LJ-PME */
 637             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 638             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 639             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 640             exponent         = gmx_simd_exp_r(ewcljrsq);
 641             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 642             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 643             /* f6A = 6 * C6grid * (1 - poly) */
 644             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 645             /* f6B = C6grid * exponent * beta^6 */
 646             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 647             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 648             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);
 649
 650             fscal            = _mm_add_ps(felec,fvdw);
 651
 652             /* Calculate temporary vectorial force */
 653             tx               = _mm_mul_ps(fscal,dx00);
 654             ty               = _mm_mul_ps(fscal,dy00);
 655             tz               = _mm_mul_ps(fscal,dz00);
 656
 657             /* Update vectorial force */
 658             fix0             = _mm_add_ps(fix0,tx);
 659             fiy0             = _mm_add_ps(fiy0,ty);
 660             fiz0             = _mm_add_ps(fiz0,tz);
 661
 662             fjptrA             = f+j_coord_offsetA;
 663             fjptrB             = f+j_coord_offsetB;
 664             fjptrC             = f+j_coord_offsetC;
 665             fjptrD             = f+j_coord_offsetD;
 666             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 667
 668             /* Inner loop uses 59 flops */
 669         }
 670
 671         if(jidx<j_index_end)
 672         {
 673
 674             /* Get j neighbor index, and coordinate index */
 675             jnrlistA         = jjnr[jidx];
 676             jnrlistB         = jjnr[jidx+1];
 677             jnrlistC         = jjnr[jidx+2];
 678             jnrlistD         = jjnr[jidx+3];
 679             /* Sign of each element will be negative for non-real atoms.
 680              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 681              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 682              */
 683             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 684             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 685             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 686             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 687             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 688             j_coord_offsetA  = DIM*jnrA;
 689             j_coord_offsetB  = DIM*jnrB;
 690             j_coord_offsetC  = DIM*jnrC;
 691             j_coord_offsetD  = DIM*jnrD;
 692
 693             /* load j atom coordinates */
 694             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 695                                               x+j_coord_offsetC,x+j_coord_offsetD,
 696                                               &jx0,&jy0,&jz0);
 697
 698             /* Calculate displacement vector */
 699             dx00             = _mm_sub_ps(ix0,jx0);
 700             dy00             = _mm_sub_ps(iy0,jy0);
 701             dz00             = _mm_sub_ps(iz0,jz0);
 702
 703             /* Calculate squared distance and things based on it */
 704             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 705
 706             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 707
 708             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 709
 710             /* Load parameters for j particles */
 711             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 712                                                               charge+jnrC+0,charge+jnrD+0);
 713             vdwjidx0A        = 2*vdwtype[jnrA+0];
 714             vdwjidx0B        = 2*vdwtype[jnrB+0];
 715             vdwjidx0C        = 2*vdwtype[jnrC+0];
 716             vdwjidx0D        = 2*vdwtype[jnrD+0];
 717
 718             /**************************
 719              * CALCULATE INTERACTIONS *
 720              **************************/
 721
 722             r00              = _mm_mul_ps(rsq00,rinv00);
 723             r00              = _mm_andnot_ps(dummy_mask,r00);
 724
 725             /* Compute parameters for interactions between i and j atoms */
 726             qq00             = _mm_mul_ps(iq0,jq0);
 727             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 728                                          vdwparam+vdwioffset0+vdwjidx0B,
 729                                          vdwparam+vdwioffset0+vdwjidx0C,
 730                                          vdwparam+vdwioffset0+vdwjidx0D,
 731                                          &c6_00,&c12_00);
 732
 733             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 734                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 735                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 736                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 737
 738             /* EWALD ELECTROSTATICS */
 739
 740             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 741             ewrt             = _mm_mul_ps(r00,ewtabscale);
 742             ewitab           = _mm_cvttps_epi32(ewrt);
 743             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 744             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 745                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 746                                          &ewtabF,&ewtabFn);
 747             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 748             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 749
 750             /* Analytical LJ-PME */
 751             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 752             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 753             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 754             exponent         = gmx_simd_exp_r(ewcljrsq);
 755             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 756             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 757             /* f6A = 6 * C6grid * (1 - poly) */
 758             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 759             /* f6B = C6grid * exponent * beta^6 */
 760             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 761             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 762             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);
 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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 779             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 780             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 781             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 782             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 783
 784             /* Inner loop uses 60 flops */
 785         }
 786
 787         /* End of innermost loop */
 788
 789         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 790                                               f+i_coord_offset,fshift+i_shift_offset);
 791
 792         /* Increment number of inner iterations */
 793         inneriter                  += j_index_end - j_index_start;
 794
 795         /* Outer loop uses 7 flops */
 796     }
 797
 798     /* Increment number of outer iterations */
 799     outeriter        += nri;
 800
 801     /* Update outer/inner flops */
 802
 803     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 804 }