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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sse2_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_sse2_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             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 231                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 232                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 233                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 234
 235             /* EWALD ELECTROSTATICS */
 236
 237             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 238             ewrt             = _mm_mul_ps(r00,ewtabscale);
 239             ewitab           = _mm_cvttps_epi32(ewrt);
 240             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 241             ewitab           = _mm_slli_epi32(ewitab,2);
 242             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 243             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 244             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 245             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 246             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 247             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 248             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 249             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 250             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 251
 252             /* Analytical LJ-PME */
 253             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 254             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 255             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 256             exponent         = gmx_simd_exp_r(ewcljrsq);
 257             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 258             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 259             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 260             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 261             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 262             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 263             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 264             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);
 265
 266             /* Update potential sum for this i atom from the interaction with this j atom. */
 267             velecsum         = _mm_add_ps(velecsum,velec);
 268             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 269
 270             fscal            = _mm_add_ps(felec,fvdw);
 271
 272             /* Calculate temporary vectorial force */
 273             tx               = _mm_mul_ps(fscal,dx00);
 274             ty               = _mm_mul_ps(fscal,dy00);
 275             tz               = _mm_mul_ps(fscal,dz00);
 276
 277             /* Update vectorial force */
 278             fix0             = _mm_add_ps(fix0,tx);
 279             fiy0             = _mm_add_ps(fiy0,ty);
 280             fiz0             = _mm_add_ps(fiz0,tz);
 281
 282             fjptrA             = f+j_coord_offsetA;
 283             fjptrB             = f+j_coord_offsetB;
 284             fjptrC             = f+j_coord_offsetC;
 285             fjptrD             = f+j_coord_offsetD;
 286             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 287
 288             /* Inner loop uses 69 flops */
 289         }
 290
 291         if(jidx<j_index_end)
 292         {
 293
 294             /* Get j neighbor index, and coordinate index */
 295             jnrlistA         = jjnr[jidx];
 296             jnrlistB         = jjnr[jidx+1];
 297             jnrlistC         = jjnr[jidx+2];
 298             jnrlistD         = jjnr[jidx+3];
 299             /* Sign of each element will be negative for non-real atoms.
 300              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 301              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 302              */
 303             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 304             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 305             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 306             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 307             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 308             j_coord_offsetA  = DIM*jnrA;
 309             j_coord_offsetB  = DIM*jnrB;
 310             j_coord_offsetC  = DIM*jnrC;
 311             j_coord_offsetD  = DIM*jnrD;
 312
 313             /* load j atom coordinates */
 314             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 315                                               x+j_coord_offsetC,x+j_coord_offsetD,
 316                                               &jx0,&jy0,&jz0);
 317
 318             /* Calculate displacement vector */
 319             dx00             = _mm_sub_ps(ix0,jx0);
 320             dy00             = _mm_sub_ps(iy0,jy0);
 321             dz00             = _mm_sub_ps(iz0,jz0);
 322
 323             /* Calculate squared distance and things based on it */
 324             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 325
 326             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 327
 328             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 329
 330             /* Load parameters for j particles */
 331             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 332                                                               charge+jnrC+0,charge+jnrD+0);
 333             vdwjidx0A        = 2*vdwtype[jnrA+0];
 334             vdwjidx0B        = 2*vdwtype[jnrB+0];
 335             vdwjidx0C        = 2*vdwtype[jnrC+0];
 336             vdwjidx0D        = 2*vdwtype[jnrD+0];
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             r00              = _mm_mul_ps(rsq00,rinv00);
 343             r00              = _mm_andnot_ps(dummy_mask,r00);
 344
 345             /* Compute parameters for interactions between i and j atoms */
 346             qq00             = _mm_mul_ps(iq0,jq0);
 347             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 348                                          vdwparam+vdwioffset0+vdwjidx0B,
 349                                          vdwparam+vdwioffset0+vdwjidx0C,
 350                                          vdwparam+vdwioffset0+vdwjidx0D,
 351                                          &c6_00,&c12_00);
 352             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 353                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 354                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 355                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 356
 357             /* EWALD ELECTROSTATICS */
 358
 359             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 360             ewrt             = _mm_mul_ps(r00,ewtabscale);
 361             ewitab           = _mm_cvttps_epi32(ewrt);
 362             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 363             ewitab           = _mm_slli_epi32(ewitab,2);
 364             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 365             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 366             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 367             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 368             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 369             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 370             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 371             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 372             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 373
 374             /* Analytical LJ-PME */
 375             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 376             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 377             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 378             exponent         = gmx_simd_exp_r(ewcljrsq);
 379             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 380             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 381             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 382             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 383             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 384             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 385             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 386             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);
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velec            = _mm_andnot_ps(dummy_mask,velec);
 390             velecsum         = _mm_add_ps(velecsum,velec);
 391             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 392             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 393
 394             fscal            = _mm_add_ps(felec,fvdw);
 395
 396             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm_mul_ps(fscal,dx00);
 400             ty               = _mm_mul_ps(fscal,dy00);
 401             tz               = _mm_mul_ps(fscal,dz00);
 402
 403             /* Update vectorial force */
 404             fix0             = _mm_add_ps(fix0,tx);
 405             fiy0             = _mm_add_ps(fiy0,ty);
 406             fiz0             = _mm_add_ps(fiz0,tz);
 407
 408             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 409             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 410             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 411             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 412             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 413
 414             /* Inner loop uses 70 flops */
 415         }
 416
 417         /* End of innermost loop */
 418
 419         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 420                                               f+i_coord_offset,fshift+i_shift_offset);
 421
 422         ggid                        = gid[iidx];
 423         /* Update potential energies */
 424         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 425         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 426
 427         /* Increment number of inner iterations */
 428         inneriter                  += j_index_end - j_index_start;
 429
 430         /* Outer loop uses 9 flops */
 431     }
 432
 433     /* Increment number of outer iterations */
 434     outeriter        += nri;
 435
 436     /* Update outer/inner flops */
 437
 438     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*70);
 439 }
 440 /*
 441  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse2_single
 442  * Electrostatics interaction: Ewald
 443  * VdW interaction:            LJEwald
 444  * Geometry:                   Particle-Particle
 445  * Calculate force/pot:        Force
 446  */
 447 void
 448 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse2_single
 449                     (t_nblist                    * gmx_restrict       nlist,
 450                      rvec                        * gmx_restrict          xx,
 451                      rvec                        * gmx_restrict          ff,
 452                      t_forcerec                  * gmx_restrict          fr,
 453                      t_mdatoms                   * gmx_restrict     mdatoms,
 454                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 455                      t_nrnb                      * gmx_restrict        nrnb)
 456 {
 457     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 458      * just 0 for non-waters.
 459      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 460      * jnr indices corresponding to data put in the four positions in the SIMD register.
 461      */
 462     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 463     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 464     int              jnrA,jnrB,jnrC,jnrD;
 465     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 466     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 467     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 468     real             rcutoff_scalar;
 469     real             *shiftvec,*fshift,*x,*f;
 470     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 471     real             scratch[4*DIM];
 472     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 473     int              vdwioffset0;
 474     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 475     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 476     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 477     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 478     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 479     real             *charge;
 480     int              nvdwtype;
 481     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 482     int              *vdwtype;
 483     real             *vdwparam;
 484     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 485     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 486     __m128           c6grid_00;
 487     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 488     real             *vdwgridparam;
 489     __m128           one_half = _mm_set1_ps(0.5);
 490     __m128           minus_one = _mm_set1_ps(-1.0);
 491     __m128i          ewitab;
 492     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 493     real             *ewtab;
 494     __m128           dummy_mask,cutoff_mask;
 495     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 496     __m128           one     = _mm_set1_ps(1.0);
 497     __m128           two     = _mm_set1_ps(2.0);
 498     x                = xx[0];
 499     f                = ff[0];
 500
 501     nri              = nlist->nri;
 502     iinr             = nlist->iinr;
 503     jindex           = nlist->jindex;
 504     jjnr             = nlist->jjnr;
 505     shiftidx         = nlist->shift;
 506     gid              = nlist->gid;
 507     shiftvec         = fr->shift_vec[0];
 508     fshift           = fr->fshift[0];
 509     facel            = _mm_set1_ps(fr->epsfac);
 510     charge           = mdatoms->chargeA;
 511     nvdwtype         = fr->ntype;
 512     vdwparam         = fr->nbfp;
 513     vdwtype          = mdatoms->typeA;
 514     vdwgridparam     = fr->ljpme_c6grid;
 515     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 516     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 517     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 518
 519     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 520     ewtab            = fr->ic->tabq_coul_F;
 521     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 522     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 523
 524     /* Avoid stupid compiler warnings */
 525     jnrA = jnrB = jnrC = jnrD = 0;
 526     j_coord_offsetA = 0;
 527     j_coord_offsetB = 0;
 528     j_coord_offsetC = 0;
 529     j_coord_offsetD = 0;
 530
 531     outeriter        = 0;
 532     inneriter        = 0;
 533
 534     for(iidx=0;iidx<4*DIM;iidx++)
 535     {
 536         scratch[iidx] = 0.0;
 537     }
 538
 539     /* Start outer loop over neighborlists */
 540     for(iidx=0; iidx<nri; iidx++)
 541     {
 542         /* Load shift vector for this list */
 543         i_shift_offset   = DIM*shiftidx[iidx];
 544
 545         /* Load limits for loop over neighbors */
 546         j_index_start    = jindex[iidx];
 547         j_index_end      = jindex[iidx+1];
 548
 549         /* Get outer coordinate index */
 550         inr              = iinr[iidx];
 551         i_coord_offset   = DIM*inr;
 552
 553         /* Load i particle coords and add shift vector */
 554         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 555
 556         fix0             = _mm_setzero_ps();
 557         fiy0             = _mm_setzero_ps();
 558         fiz0             = _mm_setzero_ps();
 559
 560         /* Load parameters for i particles */
 561         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 562         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 563
 564         /* Start inner kernel loop */
 565         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 566         {
 567
 568             /* Get j neighbor index, and coordinate index */
 569             jnrA             = jjnr[jidx];
 570             jnrB             = jjnr[jidx+1];
 571             jnrC             = jjnr[jidx+2];
 572             jnrD             = jjnr[jidx+3];
 573             j_coord_offsetA  = DIM*jnrA;
 574             j_coord_offsetB  = DIM*jnrB;
 575             j_coord_offsetC  = DIM*jnrC;
 576             j_coord_offsetD  = DIM*jnrD;
 577
 578             /* load j atom coordinates */
 579             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 580                                               x+j_coord_offsetC,x+j_coord_offsetD,
 581                                               &jx0,&jy0,&jz0);
 582
 583             /* Calculate displacement vector */
 584             dx00             = _mm_sub_ps(ix0,jx0);
 585             dy00             = _mm_sub_ps(iy0,jy0);
 586             dz00             = _mm_sub_ps(iz0,jz0);
 587
 588             /* Calculate squared distance and things based on it */
 589             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 590
 591             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 592
 593             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 594
 595             /* Load parameters for j particles */
 596             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 597                                                               charge+jnrC+0,charge+jnrD+0);
 598             vdwjidx0A        = 2*vdwtype[jnrA+0];
 599             vdwjidx0B        = 2*vdwtype[jnrB+0];
 600             vdwjidx0C        = 2*vdwtype[jnrC+0];
 601             vdwjidx0D        = 2*vdwtype[jnrD+0];
 602
 603             /**************************
 604              * CALCULATE INTERACTIONS *
 605              **************************/
 606
 607             r00              = _mm_mul_ps(rsq00,rinv00);
 608
 609             /* Compute parameters for interactions between i and j atoms */
 610             qq00             = _mm_mul_ps(iq0,jq0);
 611             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 612                                          vdwparam+vdwioffset0+vdwjidx0B,
 613                                          vdwparam+vdwioffset0+vdwjidx0C,
 614                                          vdwparam+vdwioffset0+vdwjidx0D,
 615                                          &c6_00,&c12_00);
 616             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 617                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 618                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 619                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 620
 621             /* EWALD ELECTROSTATICS */
 622
 623             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 624             ewrt             = _mm_mul_ps(r00,ewtabscale);
 625             ewitab           = _mm_cvttps_epi32(ewrt);
 626             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 627             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 628                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 629                                          &ewtabF,&ewtabFn);
 630             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 631             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 632
 633             /* Analytical LJ-PME */
 634             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 635             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 636             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 637             exponent         = gmx_simd_exp_r(ewcljrsq);
 638             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 639             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 640             /* f6A = 6 * C6grid * (1 - poly) */
 641             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 642             /* f6B = C6grid * exponent * beta^6 */
 643             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 644             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 645             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);
 646
 647             fscal            = _mm_add_ps(felec,fvdw);
 648
 649             /* Calculate temporary vectorial force */
 650             tx               = _mm_mul_ps(fscal,dx00);
 651             ty               = _mm_mul_ps(fscal,dy00);
 652             tz               = _mm_mul_ps(fscal,dz00);
 653
 654             /* Update vectorial force */
 655             fix0             = _mm_add_ps(fix0,tx);
 656             fiy0             = _mm_add_ps(fiy0,ty);
 657             fiz0             = _mm_add_ps(fiz0,tz);
 658
 659             fjptrA             = f+j_coord_offsetA;
 660             fjptrB             = f+j_coord_offsetB;
 661             fjptrC             = f+j_coord_offsetC;
 662             fjptrD             = f+j_coord_offsetD;
 663             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 664
 665             /* Inner loop uses 59 flops */
 666         }
 667
 668         if(jidx<j_index_end)
 669         {
 670
 671             /* Get j neighbor index, and coordinate index */
 672             jnrlistA         = jjnr[jidx];
 673             jnrlistB         = jjnr[jidx+1];
 674             jnrlistC         = jjnr[jidx+2];
 675             jnrlistD         = jjnr[jidx+3];
 676             /* Sign of each element will be negative for non-real atoms.
 677              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 678              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 679              */
 680             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 681             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 682             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 683             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 684             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 685             j_coord_offsetA  = DIM*jnrA;
 686             j_coord_offsetB  = DIM*jnrB;
 687             j_coord_offsetC  = DIM*jnrC;
 688             j_coord_offsetD  = DIM*jnrD;
 689
 690             /* load j atom coordinates */
 691             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 692                                               x+j_coord_offsetC,x+j_coord_offsetD,
 693                                               &jx0,&jy0,&jz0);
 694
 695             /* Calculate displacement vector */
 696             dx00             = _mm_sub_ps(ix0,jx0);
 697             dy00             = _mm_sub_ps(iy0,jy0);
 698             dz00             = _mm_sub_ps(iz0,jz0);
 699
 700             /* Calculate squared distance and things based on it */
 701             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 702
 703             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 704
 705             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 706
 707             /* Load parameters for j particles */
 708             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 709                                                               charge+jnrC+0,charge+jnrD+0);
 710             vdwjidx0A        = 2*vdwtype[jnrA+0];
 711             vdwjidx0B        = 2*vdwtype[jnrB+0];
 712             vdwjidx0C        = 2*vdwtype[jnrC+0];
 713             vdwjidx0D        = 2*vdwtype[jnrD+0];
 714
 715             /**************************
 716              * CALCULATE INTERACTIONS *
 717              **************************/
 718
 719             r00              = _mm_mul_ps(rsq00,rinv00);
 720             r00              = _mm_andnot_ps(dummy_mask,r00);
 721
 722             /* Compute parameters for interactions between i and j atoms */
 723             qq00             = _mm_mul_ps(iq0,jq0);
 724             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 725                                          vdwparam+vdwioffset0+vdwjidx0B,
 726                                          vdwparam+vdwioffset0+vdwjidx0C,
 727                                          vdwparam+vdwioffset0+vdwjidx0D,
 728                                          &c6_00,&c12_00);
 729             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 730                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 731                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 732                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 733
 734             /* EWALD ELECTROSTATICS */
 735
 736             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 737             ewrt             = _mm_mul_ps(r00,ewtabscale);
 738             ewitab           = _mm_cvttps_epi32(ewrt);
 739             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 740             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 741                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 742                                          &ewtabF,&ewtabFn);
 743             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 744             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 745
 746             /* Analytical LJ-PME */
 747             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 748             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 749             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 750             exponent         = gmx_simd_exp_r(ewcljrsq);
 751             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 752             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 753             /* f6A = 6 * C6grid * (1 - poly) */
 754             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 755             /* f6B = C6grid * exponent * beta^6 */
 756             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 757             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 758             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);
 759
 760             fscal            = _mm_add_ps(felec,fvdw);
 761
 762             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 763
 764             /* Calculate temporary vectorial force */
 765             tx               = _mm_mul_ps(fscal,dx00);
 766             ty               = _mm_mul_ps(fscal,dy00);
 767             tz               = _mm_mul_ps(fscal,dz00);
 768
 769             /* Update vectorial force */
 770             fix0             = _mm_add_ps(fix0,tx);
 771             fiy0             = _mm_add_ps(fiy0,ty);
 772             fiz0             = _mm_add_ps(fiz0,tz);
 773
 774             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 775             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 776             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 777             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 778             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 779
 780             /* Inner loop uses 60 flops */
 781         }
 782
 783         /* End of innermost loop */
 784
 785         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 786                                               f+i_coord_offset,fshift+i_shift_offset);
 787
 788         /* Increment number of inner iterations */
 789         inneriter                  += j_index_end - j_index_start;
 790
 791         /* Outer loop uses 7 flops */
 792     }
 793
 794     /* Increment number of outer iterations */
 795     outeriter        += nri;
 796
 797     /* Update outer/inner flops */
 798
 799     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 800 }