src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwLJ_GeomW3P1_sse4_1_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_double.h"
  34 #include "kernelutil_x86_sse4_1_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_sse4_1_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  84     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  85     __m128d          dummy_mask,cutoff_mask;
  86     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  87     __m128d          one     = _mm_set1_pd(1.0);
  88     __m128d          two     = _mm_set1_pd(2.0);
  89     x                = xx[0];
  90     f                = ff[0];
  91
  92     nri              = nlist->nri;
  93     iinr             = nlist->iinr;
  94     jindex           = nlist->jindex;
  95     jjnr             = nlist->jjnr;
  96     shiftidx         = nlist->shift;
  97     gid              = nlist->gid;
  98     shiftvec         = fr->shift_vec[0];
  99     fshift           = fr->fshift[0];
 100     facel            = _mm_set1_pd(fr->epsfac);
 101     charge           = mdatoms->chargeA;
 102     nvdwtype         = fr->ntype;
 103     vdwparam         = fr->nbfp;
 104     vdwtype          = mdatoms->typeA;
 105
 106     /* Setup water-specific parameters */
 107     inr              = nlist->iinr[0];
 108     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 109     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 110     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 111     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117
 118     outeriter        = 0;
 119     inneriter        = 0;
 120
 121     /* Start outer loop over neighborlists */
 122     for(iidx=0; iidx<nri; iidx++)
 123     {
 124         /* Load shift vector for this list */
 125         i_shift_offset   = DIM*shiftidx[iidx];
 126
 127         /* Load limits for loop over neighbors */
 128         j_index_start    = jindex[iidx];
 129         j_index_end      = jindex[iidx+1];
 130
 131         /* Get outer coordinate index */
 132         inr              = iinr[iidx];
 133         i_coord_offset   = DIM*inr;
 134
 135         /* Load i particle coords and add shift vector */
 136         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 137                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 138
 139         fix0             = _mm_setzero_pd();
 140         fiy0             = _mm_setzero_pd();
 141         fiz0             = _mm_setzero_pd();
 142         fix1             = _mm_setzero_pd();
 143         fiy1             = _mm_setzero_pd();
 144         fiz1             = _mm_setzero_pd();
 145         fix2             = _mm_setzero_pd();
 146         fiy2             = _mm_setzero_pd();
 147         fiz2             = _mm_setzero_pd();
 148
 149         /* Reset potential sums */
 150         velecsum         = _mm_setzero_pd();
 151         vvdwsum          = _mm_setzero_pd();
 152
 153         /* Start inner kernel loop */
 154         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 155         {
 156
 157             /* Get j neighbor index, and coordinate index */
 158             jnrA             = jjnr[jidx];
 159             jnrB             = jjnr[jidx+1];
 160             j_coord_offsetA  = DIM*jnrA;
 161             j_coord_offsetB  = DIM*jnrB;
 162
 163             /* load j atom coordinates */
 164             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 165                                               &jx0,&jy0,&jz0);
 166
 167             /* Calculate displacement vector */
 168             dx00             = _mm_sub_pd(ix0,jx0);
 169             dy00             = _mm_sub_pd(iy0,jy0);
 170             dz00             = _mm_sub_pd(iz0,jz0);
 171             dx10             = _mm_sub_pd(ix1,jx0);
 172             dy10             = _mm_sub_pd(iy1,jy0);
 173             dz10             = _mm_sub_pd(iz1,jz0);
 174             dx20             = _mm_sub_pd(ix2,jx0);
 175             dy20             = _mm_sub_pd(iy2,jy0);
 176             dz20             = _mm_sub_pd(iz2,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 180             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 181             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 182
 183             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 184             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 185             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 186
 187             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 188             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 189             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 190
 191             /* Load parameters for j particles */
 192             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 193             vdwjidx0A        = 2*vdwtype[jnrA+0];
 194             vdwjidx0B        = 2*vdwtype[jnrB+0];
 195
 196             fjx0             = _mm_setzero_pd();
 197             fjy0             = _mm_setzero_pd();
 198             fjz0             = _mm_setzero_pd();
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             qq00             = _mm_mul_pd(iq0,jq0);
 206             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 207                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 208
 209             /* COULOMB ELECTROSTATICS */
 210             velec            = _mm_mul_pd(qq00,rinv00);
 211             felec            = _mm_mul_pd(velec,rinvsq00);
 212
 213             /* LENNARD-JONES DISPERSION/REPULSION */
 214
 215             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 216             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 217             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 218             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 219             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 220
 221             /* Update potential sum for this i atom from the interaction with this j atom. */
 222             velecsum         = _mm_add_pd(velecsum,velec);
 223             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 224
 225             fscal            = _mm_add_pd(felec,fvdw);
 226
 227             /* Calculate temporary vectorial force */
 228             tx               = _mm_mul_pd(fscal,dx00);
 229             ty               = _mm_mul_pd(fscal,dy00);
 230             tz               = _mm_mul_pd(fscal,dz00);
 231
 232             /* Update vectorial force */
 233             fix0             = _mm_add_pd(fix0,tx);
 234             fiy0             = _mm_add_pd(fiy0,ty);
 235             fiz0             = _mm_add_pd(fiz0,tz);
 236
 237             fjx0             = _mm_add_pd(fjx0,tx);
 238             fjy0             = _mm_add_pd(fjy0,ty);
 239             fjz0             = _mm_add_pd(fjz0,tz);
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq10             = _mm_mul_pd(iq1,jq0);
 247
 248             /* COULOMB ELECTROSTATICS */
 249             velec            = _mm_mul_pd(qq10,rinv10);
 250             felec            = _mm_mul_pd(velec,rinvsq10);
 251
 252             /* Update potential sum for this i atom from the interaction with this j atom. */
 253             velecsum         = _mm_add_pd(velecsum,velec);
 254
 255             fscal            = felec;
 256
 257             /* Calculate temporary vectorial force */
 258             tx               = _mm_mul_pd(fscal,dx10);
 259             ty               = _mm_mul_pd(fscal,dy10);
 260             tz               = _mm_mul_pd(fscal,dz10);
 261
 262             /* Update vectorial force */
 263             fix1             = _mm_add_pd(fix1,tx);
 264             fiy1             = _mm_add_pd(fiy1,ty);
 265             fiz1             = _mm_add_pd(fiz1,tz);
 266
 267             fjx0             = _mm_add_pd(fjx0,tx);
 268             fjy0             = _mm_add_pd(fjy0,ty);
 269             fjz0             = _mm_add_pd(fjz0,tz);
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             qq20             = _mm_mul_pd(iq2,jq0);
 277
 278             /* COULOMB ELECTROSTATICS */
 279             velec            = _mm_mul_pd(qq20,rinv20);
 280             felec            = _mm_mul_pd(velec,rinvsq20);
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velecsum         = _mm_add_pd(velecsum,velec);
 284
 285             fscal            = felec;
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm_mul_pd(fscal,dx20);
 289             ty               = _mm_mul_pd(fscal,dy20);
 290             tz               = _mm_mul_pd(fscal,dz20);
 291
 292             /* Update vectorial force */
 293             fix2             = _mm_add_pd(fix2,tx);
 294             fiy2             = _mm_add_pd(fiy2,ty);
 295             fiz2             = _mm_add_pd(fiz2,tz);
 296
 297             fjx0             = _mm_add_pd(fjx0,tx);
 298             fjy0             = _mm_add_pd(fjy0,ty);
 299             fjz0             = _mm_add_pd(fjz0,tz);
 300
 301             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 302
 303             /* Inner loop uses 99 flops */
 304         }
 305
 306         if(jidx<j_index_end)
 307         {
 308
 309             jnrA             = jjnr[jidx];
 310             j_coord_offsetA  = DIM*jnrA;
 311
 312             /* load j atom coordinates */
 313             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 314                                               &jx0,&jy0,&jz0);
 315
 316             /* Calculate displacement vector */
 317             dx00             = _mm_sub_pd(ix0,jx0);
 318             dy00             = _mm_sub_pd(iy0,jy0);
 319             dz00             = _mm_sub_pd(iz0,jz0);
 320             dx10             = _mm_sub_pd(ix1,jx0);
 321             dy10             = _mm_sub_pd(iy1,jy0);
 322             dz10             = _mm_sub_pd(iz1,jz0);
 323             dx20             = _mm_sub_pd(ix2,jx0);
 324             dy20             = _mm_sub_pd(iy2,jy0);
 325             dz20             = _mm_sub_pd(iz2,jz0);
 326
 327             /* Calculate squared distance and things based on it */
 328             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 329             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 330             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 331
 332             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 333             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 334             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 335
 336             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 337             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 338             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 339
 340             /* Load parameters for j particles */
 341             jq0              = _mm_load_sd(charge+jnrA+0);
 342             vdwjidx0A        = 2*vdwtype[jnrA+0];
 343
 344             fjx0             = _mm_setzero_pd();
 345             fjy0             = _mm_setzero_pd();
 346             fjz0             = _mm_setzero_pd();
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq00             = _mm_mul_pd(iq0,jq0);
 354             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 355
 356             /* COULOMB ELECTROSTATICS */
 357             velec            = _mm_mul_pd(qq00,rinv00);
 358             felec            = _mm_mul_pd(velec,rinvsq00);
 359
 360             /* LENNARD-JONES DISPERSION/REPULSION */
 361
 362             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 363             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 364             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 365             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 366             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 370             velecsum         = _mm_add_pd(velecsum,velec);
 371             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 372             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 373
 374             fscal            = _mm_add_pd(felec,fvdw);
 375
 376             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 377
 378             /* Calculate temporary vectorial force */
 379             tx               = _mm_mul_pd(fscal,dx00);
 380             ty               = _mm_mul_pd(fscal,dy00);
 381             tz               = _mm_mul_pd(fscal,dz00);
 382
 383             /* Update vectorial force */
 384             fix0             = _mm_add_pd(fix0,tx);
 385             fiy0             = _mm_add_pd(fiy0,ty);
 386             fiz0             = _mm_add_pd(fiz0,tz);
 387
 388             fjx0             = _mm_add_pd(fjx0,tx);
 389             fjy0             = _mm_add_pd(fjy0,ty);
 390             fjz0             = _mm_add_pd(fjz0,tz);
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             /* Compute parameters for interactions between i and j atoms */
 397             qq10             = _mm_mul_pd(iq1,jq0);
 398
 399             /* COULOMB ELECTROSTATICS */
 400             velec            = _mm_mul_pd(qq10,rinv10);
 401             felec            = _mm_mul_pd(velec,rinvsq10);
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 405             velecsum         = _mm_add_pd(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 410
 411             /* Calculate temporary vectorial force */
 412             tx               = _mm_mul_pd(fscal,dx10);
 413             ty               = _mm_mul_pd(fscal,dy10);
 414             tz               = _mm_mul_pd(fscal,dz10);
 415
 416             /* Update vectorial force */
 417             fix1             = _mm_add_pd(fix1,tx);
 418             fiy1             = _mm_add_pd(fiy1,ty);
 419             fiz1             = _mm_add_pd(fiz1,tz);
 420
 421             fjx0             = _mm_add_pd(fjx0,tx);
 422             fjy0             = _mm_add_pd(fjy0,ty);
 423             fjz0             = _mm_add_pd(fjz0,tz);
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             /* Compute parameters for interactions between i and j atoms */
 430             qq20             = _mm_mul_pd(iq2,jq0);
 431
 432             /* COULOMB ELECTROSTATICS */
 433             velec            = _mm_mul_pd(qq20,rinv20);
 434             felec            = _mm_mul_pd(velec,rinvsq20);
 435
 436             /* Update potential sum for this i atom from the interaction with this j atom. */
 437             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 438             velecsum         = _mm_add_pd(velecsum,velec);
 439
 440             fscal            = felec;
 441
 442             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 443
 444             /* Calculate temporary vectorial force */
 445             tx               = _mm_mul_pd(fscal,dx20);
 446             ty               = _mm_mul_pd(fscal,dy20);
 447             tz               = _mm_mul_pd(fscal,dz20);
 448
 449             /* Update vectorial force */
 450             fix2             = _mm_add_pd(fix2,tx);
 451             fiy2             = _mm_add_pd(fiy2,ty);
 452             fiz2             = _mm_add_pd(fiz2,tz);
 453
 454             fjx0             = _mm_add_pd(fjx0,tx);
 455             fjy0             = _mm_add_pd(fjy0,ty);
 456             fjz0             = _mm_add_pd(fjz0,tz);
 457
 458             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 459
 460             /* Inner loop uses 99 flops */
 461         }
 462
 463         /* End of innermost loop */
 464
 465         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 466                                               f+i_coord_offset,fshift+i_shift_offset);
 467
 468         ggid                        = gid[iidx];
 469         /* Update potential energies */
 470         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 471         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 472
 473         /* Increment number of inner iterations */
 474         inneriter                  += j_index_end - j_index_start;
 475
 476         /* Outer loop uses 20 flops */
 477     }
 478
 479     /* Increment number of outer iterations */
 480     outeriter        += nri;
 481
 482     /* Update outer/inner flops */
 483
 484     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*99);
 485 }
 486 /*
 487  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_double
 488  * Electrostatics interaction: Coulomb
 489  * VdW interaction:            LennardJones
 490  * Geometry:                   Water3-Particle
 491  * Calculate force/pot:        Force
 492  */
 493 void
 494 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_sse4_1_double
 495                     (t_nblist * gmx_restrict                nlist,
 496                      rvec * gmx_restrict                    xx,
 497                      rvec * gmx_restrict                    ff,
 498                      t_forcerec * gmx_restrict              fr,
 499                      t_mdatoms * gmx_restrict               mdatoms,
 500                      nb_kernel_data_t * gmx_restrict        kernel_data,
 501                      t_nrnb * gmx_restrict                  nrnb)
 502 {
 503     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 504      * just 0 for non-waters.
 505      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 506      * jnr indices corresponding to data put in the four positions in the SIMD register.
 507      */
 508     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 509     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 510     int              jnrA,jnrB;
 511     int              j_coord_offsetA,j_coord_offsetB;
 512     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 513     real             rcutoff_scalar;
 514     real             *shiftvec,*fshift,*x,*f;
 515     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 516     int              vdwioffset0;
 517     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 518     int              vdwioffset1;
 519     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 520     int              vdwioffset2;
 521     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 522     int              vdwjidx0A,vdwjidx0B;
 523     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 524     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 525     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 526     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 527     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 528     real             *charge;
 529     int              nvdwtype;
 530     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 531     int              *vdwtype;
 532     real             *vdwparam;
 533     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 534     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 535     __m128d          dummy_mask,cutoff_mask;
 536     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 537     __m128d          one     = _mm_set1_pd(1.0);
 538     __m128d          two     = _mm_set1_pd(2.0);
 539     x                = xx[0];
 540     f                = ff[0];
 541
 542     nri              = nlist->nri;
 543     iinr             = nlist->iinr;
 544     jindex           = nlist->jindex;
 545     jjnr             = nlist->jjnr;
 546     shiftidx         = nlist->shift;
 547     gid              = nlist->gid;
 548     shiftvec         = fr->shift_vec[0];
 549     fshift           = fr->fshift[0];
 550     facel            = _mm_set1_pd(fr->epsfac);
 551     charge           = mdatoms->chargeA;
 552     nvdwtype         = fr->ntype;
 553     vdwparam         = fr->nbfp;
 554     vdwtype          = mdatoms->typeA;
 555
 556     /* Setup water-specific parameters */
 557     inr              = nlist->iinr[0];
 558     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 559     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 560     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 561     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 562
 563     /* Avoid stupid compiler warnings */
 564     jnrA = jnrB = 0;
 565     j_coord_offsetA = 0;
 566     j_coord_offsetB = 0;
 567
 568     outeriter        = 0;
 569     inneriter        = 0;
 570
 571     /* Start outer loop over neighborlists */
 572     for(iidx=0; iidx<nri; iidx++)
 573     {
 574         /* Load shift vector for this list */
 575         i_shift_offset   = DIM*shiftidx[iidx];
 576
 577         /* Load limits for loop over neighbors */
 578         j_index_start    = jindex[iidx];
 579         j_index_end      = jindex[iidx+1];
 580
 581         /* Get outer coordinate index */
 582         inr              = iinr[iidx];
 583         i_coord_offset   = DIM*inr;
 584
 585         /* Load i particle coords and add shift vector */
 586         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 587                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 588
 589         fix0             = _mm_setzero_pd();
 590         fiy0             = _mm_setzero_pd();
 591         fiz0             = _mm_setzero_pd();
 592         fix1             = _mm_setzero_pd();
 593         fiy1             = _mm_setzero_pd();
 594         fiz1             = _mm_setzero_pd();
 595         fix2             = _mm_setzero_pd();
 596         fiy2             = _mm_setzero_pd();
 597         fiz2             = _mm_setzero_pd();
 598
 599         /* Start inner kernel loop */
 600         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 601         {
 602
 603             /* Get j neighbor index, and coordinate index */
 604             jnrA             = jjnr[jidx];
 605             jnrB             = jjnr[jidx+1];
 606             j_coord_offsetA  = DIM*jnrA;
 607             j_coord_offsetB  = DIM*jnrB;
 608
 609             /* load j atom coordinates */
 610             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 611                                               &jx0,&jy0,&jz0);
 612
 613             /* Calculate displacement vector */
 614             dx00             = _mm_sub_pd(ix0,jx0);
 615             dy00             = _mm_sub_pd(iy0,jy0);
 616             dz00             = _mm_sub_pd(iz0,jz0);
 617             dx10             = _mm_sub_pd(ix1,jx0);
 618             dy10             = _mm_sub_pd(iy1,jy0);
 619             dz10             = _mm_sub_pd(iz1,jz0);
 620             dx20             = _mm_sub_pd(ix2,jx0);
 621             dy20             = _mm_sub_pd(iy2,jy0);
 622             dz20             = _mm_sub_pd(iz2,jz0);
 623
 624             /* Calculate squared distance and things based on it */
 625             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 626             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 627             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 628
 629             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 630             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 631             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 632
 633             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 634             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 635             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 636
 637             /* Load parameters for j particles */
 638             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 639             vdwjidx0A        = 2*vdwtype[jnrA+0];
 640             vdwjidx0B        = 2*vdwtype[jnrB+0];
 641
 642             fjx0             = _mm_setzero_pd();
 643             fjy0             = _mm_setzero_pd();
 644             fjz0             = _mm_setzero_pd();
 645
 646             /**************************
 647              * CALCULATE INTERACTIONS *
 648              **************************/
 649
 650             /* Compute parameters for interactions between i and j atoms */
 651             qq00             = _mm_mul_pd(iq0,jq0);
 652             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 653                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 654
 655             /* COULOMB ELECTROSTATICS */
 656             velec            = _mm_mul_pd(qq00,rinv00);
 657             felec            = _mm_mul_pd(velec,rinvsq00);
 658
 659             /* LENNARD-JONES DISPERSION/REPULSION */
 660
 661             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 662             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 663
 664             fscal            = _mm_add_pd(felec,fvdw);
 665
 666             /* Calculate temporary vectorial force */
 667             tx               = _mm_mul_pd(fscal,dx00);
 668             ty               = _mm_mul_pd(fscal,dy00);
 669             tz               = _mm_mul_pd(fscal,dz00);
 670
 671             /* Update vectorial force */
 672             fix0             = _mm_add_pd(fix0,tx);
 673             fiy0             = _mm_add_pd(fiy0,ty);
 674             fiz0             = _mm_add_pd(fiz0,tz);
 675
 676             fjx0             = _mm_add_pd(fjx0,tx);
 677             fjy0             = _mm_add_pd(fjy0,ty);
 678             fjz0             = _mm_add_pd(fjz0,tz);
 679
 680             /**************************
 681              * CALCULATE INTERACTIONS *
 682              **************************/
 683
 684             /* Compute parameters for interactions between i and j atoms */
 685             qq10             = _mm_mul_pd(iq1,jq0);
 686
 687             /* COULOMB ELECTROSTATICS */
 688             velec            = _mm_mul_pd(qq10,rinv10);
 689             felec            = _mm_mul_pd(velec,rinvsq10);
 690
 691             fscal            = felec;
 692
 693             /* Calculate temporary vectorial force */
 694             tx               = _mm_mul_pd(fscal,dx10);
 695             ty               = _mm_mul_pd(fscal,dy10);
 696             tz               = _mm_mul_pd(fscal,dz10);
 697
 698             /* Update vectorial force */
 699             fix1             = _mm_add_pd(fix1,tx);
 700             fiy1             = _mm_add_pd(fiy1,ty);
 701             fiz1             = _mm_add_pd(fiz1,tz);
 702
 703             fjx0             = _mm_add_pd(fjx0,tx);
 704             fjy0             = _mm_add_pd(fjy0,ty);
 705             fjz0             = _mm_add_pd(fjz0,tz);
 706
 707             /**************************
 708              * CALCULATE INTERACTIONS *
 709              **************************/
 710
 711             /* Compute parameters for interactions between i and j atoms */
 712             qq20             = _mm_mul_pd(iq2,jq0);
 713
 714             /* COULOMB ELECTROSTATICS */
 715             velec            = _mm_mul_pd(qq20,rinv20);
 716             felec            = _mm_mul_pd(velec,rinvsq20);
 717
 718             fscal            = felec;
 719
 720             /* Calculate temporary vectorial force */
 721             tx               = _mm_mul_pd(fscal,dx20);
 722             ty               = _mm_mul_pd(fscal,dy20);
 723             tz               = _mm_mul_pd(fscal,dz20);
 724
 725             /* Update vectorial force */
 726             fix2             = _mm_add_pd(fix2,tx);
 727             fiy2             = _mm_add_pd(fiy2,ty);
 728             fiz2             = _mm_add_pd(fiz2,tz);
 729
 730             fjx0             = _mm_add_pd(fjx0,tx);
 731             fjy0             = _mm_add_pd(fjy0,ty);
 732             fjz0             = _mm_add_pd(fjz0,tz);
 733
 734             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 735
 736             /* Inner loop uses 91 flops */
 737         }
 738
 739         if(jidx<j_index_end)
 740         {
 741
 742             jnrA             = jjnr[jidx];
 743             j_coord_offsetA  = DIM*jnrA;
 744
 745             /* load j atom coordinates */
 746             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 747                                               &jx0,&jy0,&jz0);
 748
 749             /* Calculate displacement vector */
 750             dx00             = _mm_sub_pd(ix0,jx0);
 751             dy00             = _mm_sub_pd(iy0,jy0);
 752             dz00             = _mm_sub_pd(iz0,jz0);
 753             dx10             = _mm_sub_pd(ix1,jx0);
 754             dy10             = _mm_sub_pd(iy1,jy0);
 755             dz10             = _mm_sub_pd(iz1,jz0);
 756             dx20             = _mm_sub_pd(ix2,jx0);
 757             dy20             = _mm_sub_pd(iy2,jy0);
 758             dz20             = _mm_sub_pd(iz2,jz0);
 759
 760             /* Calculate squared distance and things based on it */
 761             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 762             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 763             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 764
 765             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 766             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 767             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 768
 769             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 770             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 771             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 772
 773             /* Load parameters for j particles */
 774             jq0              = _mm_load_sd(charge+jnrA+0);
 775             vdwjidx0A        = 2*vdwtype[jnrA+0];
 776
 777             fjx0             = _mm_setzero_pd();
 778             fjy0             = _mm_setzero_pd();
 779             fjz0             = _mm_setzero_pd();
 780
 781             /**************************
 782              * CALCULATE INTERACTIONS *
 783              **************************/
 784
 785             /* Compute parameters for interactions between i and j atoms */
 786             qq00             = _mm_mul_pd(iq0,jq0);
 787             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 788
 789             /* COULOMB ELECTROSTATICS */
 790             velec            = _mm_mul_pd(qq00,rinv00);
 791             felec            = _mm_mul_pd(velec,rinvsq00);
 792
 793             /* LENNARD-JONES DISPERSION/REPULSION */
 794
 795             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 796             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 797
 798             fscal            = _mm_add_pd(felec,fvdw);
 799
 800             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 801
 802             /* Calculate temporary vectorial force */
 803             tx               = _mm_mul_pd(fscal,dx00);
 804             ty               = _mm_mul_pd(fscal,dy00);
 805             tz               = _mm_mul_pd(fscal,dz00);
 806
 807             /* Update vectorial force */
 808             fix0             = _mm_add_pd(fix0,tx);
 809             fiy0             = _mm_add_pd(fiy0,ty);
 810             fiz0             = _mm_add_pd(fiz0,tz);
 811
 812             fjx0             = _mm_add_pd(fjx0,tx);
 813             fjy0             = _mm_add_pd(fjy0,ty);
 814             fjz0             = _mm_add_pd(fjz0,tz);
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             /* Compute parameters for interactions between i and j atoms */
 821             qq10             = _mm_mul_pd(iq1,jq0);
 822
 823             /* COULOMB ELECTROSTATICS */
 824             velec            = _mm_mul_pd(qq10,rinv10);
 825             felec            = _mm_mul_pd(velec,rinvsq10);
 826
 827             fscal            = felec;
 828
 829             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 830
 831             /* Calculate temporary vectorial force */
 832             tx               = _mm_mul_pd(fscal,dx10);
 833             ty               = _mm_mul_pd(fscal,dy10);
 834             tz               = _mm_mul_pd(fscal,dz10);
 835
 836             /* Update vectorial force */
 837             fix1             = _mm_add_pd(fix1,tx);
 838             fiy1             = _mm_add_pd(fiy1,ty);
 839             fiz1             = _mm_add_pd(fiz1,tz);
 840
 841             fjx0             = _mm_add_pd(fjx0,tx);
 842             fjy0             = _mm_add_pd(fjy0,ty);
 843             fjz0             = _mm_add_pd(fjz0,tz);
 844
 845             /**************************
 846              * CALCULATE INTERACTIONS *
 847              **************************/
 848
 849             /* Compute parameters for interactions between i and j atoms */
 850             qq20             = _mm_mul_pd(iq2,jq0);
 851
 852             /* COULOMB ELECTROSTATICS */
 853             velec            = _mm_mul_pd(qq20,rinv20);
 854             felec            = _mm_mul_pd(velec,rinvsq20);
 855
 856             fscal            = felec;
 857
 858             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 859
 860             /* Calculate temporary vectorial force */
 861             tx               = _mm_mul_pd(fscal,dx20);
 862             ty               = _mm_mul_pd(fscal,dy20);
 863             tz               = _mm_mul_pd(fscal,dz20);
 864
 865             /* Update vectorial force */
 866             fix2             = _mm_add_pd(fix2,tx);
 867             fiy2             = _mm_add_pd(fiy2,ty);
 868             fiz2             = _mm_add_pd(fiz2,tz);
 869
 870             fjx0             = _mm_add_pd(fjx0,tx);
 871             fjy0             = _mm_add_pd(fjy0,ty);
 872             fjz0             = _mm_add_pd(fjz0,tz);
 873
 874             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 875
 876             /* Inner loop uses 91 flops */
 877         }
 878
 879         /* End of innermost loop */
 880
 881         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 882                                               f+i_coord_offset,fshift+i_shift_offset);
 883
 884         /* Increment number of inner iterations */
 885         inneriter                  += j_index_end - j_index_start;
 886
 887         /* Outer loop uses 18 flops */
 888     }
 889
 890     /* Increment number of outer iterations */
 891     outeriter        += nri;
 892
 893     /* Update outer/inner flops */
 894
 895     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*91);
 896 }