src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCoul_VdwNone_GeomP1P1_sse2_double.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_double 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_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_sse2_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_sse2_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     __m128d          dummy_mask,cutoff_mask;
  88     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  89     __m128d          one     = _mm_set1_pd(1.0);
  90     __m128d          two     = _mm_set1_pd(2.0);
  91     x                = xx[0];
  92     f                = ff[0];
  93
  94     nri              = nlist->nri;
  95     iinr             = nlist->iinr;
  96     jindex           = nlist->jindex;
  97     jjnr             = nlist->jjnr;
  98     shiftidx         = nlist->shift;
  99     gid              = nlist->gid;
 100     shiftvec         = fr->shift_vec[0];
 101     fshift           = fr->fshift[0];
 102     facel            = _mm_set1_pd(fr->epsfac);
 103     charge           = mdatoms->chargeA;
 104
 105     /* Avoid stupid compiler warnings */
 106     jnrA = jnrB = 0;
 107     j_coord_offsetA = 0;
 108     j_coord_offsetB = 0;
 109
 110     outeriter        = 0;
 111     inneriter        = 0;
 112
 113     /* Start outer loop over neighborlists */
 114     for(iidx=0; iidx<nri; iidx++)
 115     {
 116         /* Load shift vector for this list */
 117         i_shift_offset   = DIM*shiftidx[iidx];
 118
 119         /* Load limits for loop over neighbors */
 120         j_index_start    = jindex[iidx];
 121         j_index_end      = jindex[iidx+1];
 122
 123         /* Get outer coordinate index */
 124         inr              = iinr[iidx];
 125         i_coord_offset   = DIM*inr;
 126
 127         /* Load i particle coords and add shift vector */
 128         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 129
 130         fix0             = _mm_setzero_pd();
 131         fiy0             = _mm_setzero_pd();
 132         fiz0             = _mm_setzero_pd();
 133
 134         /* Load parameters for i particles */
 135         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 136
 137         /* Reset potential sums */
 138         velecsum         = _mm_setzero_pd();
 139
 140         /* Start inner kernel loop */
 141         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 142         {
 143
 144             /* Get j neighbor index, and coordinate index */
 145             jnrA             = jjnr[jidx];
 146             jnrB             = jjnr[jidx+1];
 147             j_coord_offsetA  = DIM*jnrA;
 148             j_coord_offsetB  = DIM*jnrB;
 149
 150             /* load j atom coordinates */
 151             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 152                                               &jx0,&jy0,&jz0);
 153
 154             /* Calculate displacement vector */
 155             dx00             = _mm_sub_pd(ix0,jx0);
 156             dy00             = _mm_sub_pd(iy0,jy0);
 157             dz00             = _mm_sub_pd(iz0,jz0);
 158
 159             /* Calculate squared distance and things based on it */
 160             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 161
 162             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 163
 164             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 165
 166             /* Load parameters for j particles */
 167             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 168
 169             /**************************
 170              * CALCULATE INTERACTIONS *
 171              **************************/
 172
 173             /* Compute parameters for interactions between i and j atoms */
 174             qq00             = _mm_mul_pd(iq0,jq0);
 175
 176             /* COULOMB ELECTROSTATICS */
 177             velec            = _mm_mul_pd(qq00,rinv00);
 178             felec            = _mm_mul_pd(velec,rinvsq00);
 179
 180             /* Update potential sum for this i atom from the interaction with this j atom. */
 181             velecsum         = _mm_add_pd(velecsum,velec);
 182
 183             fscal            = felec;
 184
 185             /* Calculate temporary vectorial force */
 186             tx               = _mm_mul_pd(fscal,dx00);
 187             ty               = _mm_mul_pd(fscal,dy00);
 188             tz               = _mm_mul_pd(fscal,dz00);
 189
 190             /* Update vectorial force */
 191             fix0             = _mm_add_pd(fix0,tx);
 192             fiy0             = _mm_add_pd(fiy0,ty);
 193             fiz0             = _mm_add_pd(fiz0,tz);
 194
 195             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 196
 197             /* Inner loop uses 28 flops */
 198         }
 199
 200         if(jidx<j_index_end)
 201         {
 202
 203             jnrA             = jjnr[jidx];
 204             j_coord_offsetA  = DIM*jnrA;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 208                                               &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm_sub_pd(ix0,jx0);
 212             dy00             = _mm_sub_pd(iy0,jy0);
 213             dz00             = _mm_sub_pd(iz0,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 217
 218             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 219
 220             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 221
 222             /* Load parameters for j particles */
 223             jq0              = _mm_load_sd(charge+jnrA+0);
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq00             = _mm_mul_pd(iq0,jq0);
 231
 232             /* COULOMB ELECTROSTATICS */
 233             velec            = _mm_mul_pd(qq00,rinv00);
 234             felec            = _mm_mul_pd(velec,rinvsq00);
 235
 236             /* Update potential sum for this i atom from the interaction with this j atom. */
 237             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 238             velecsum         = _mm_add_pd(velecsum,velec);
 239
 240             fscal            = felec;
 241
 242             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 243
 244             /* Calculate temporary vectorial force */
 245             tx               = _mm_mul_pd(fscal,dx00);
 246             ty               = _mm_mul_pd(fscal,dy00);
 247             tz               = _mm_mul_pd(fscal,dz00);
 248
 249             /* Update vectorial force */
 250             fix0             = _mm_add_pd(fix0,tx);
 251             fiy0             = _mm_add_pd(fiy0,ty);
 252             fiz0             = _mm_add_pd(fiz0,tz);
 253
 254             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 255
 256             /* Inner loop uses 28 flops */
 257         }
 258
 259         /* End of innermost loop */
 260
 261         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 262                                               f+i_coord_offset,fshift+i_shift_offset);
 263
 264         ggid                        = gid[iidx];
 265         /* Update potential energies */
 266         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 267
 268         /* Increment number of inner iterations */
 269         inneriter                  += j_index_end - j_index_start;
 270
 271         /* Outer loop uses 8 flops */
 272     }
 273
 274     /* Increment number of outer iterations */
 275     outeriter        += nri;
 276
 277     /* Update outer/inner flops */
 278
 279     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*28);
 280 }
 281 /*
 282  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_sse2_double
 283  * Electrostatics interaction: Coulomb
 284  * VdW interaction:            None
 285  * Geometry:                   Particle-Particle
 286  * Calculate force/pot:        Force
 287  */
 288 void
 289 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_sse2_double
 290                     (t_nblist                    * gmx_restrict       nlist,
 291                      rvec                        * gmx_restrict          xx,
 292                      rvec                        * gmx_restrict          ff,
 293                      t_forcerec                  * gmx_restrict          fr,
 294                      t_mdatoms                   * gmx_restrict     mdatoms,
 295                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 296                      t_nrnb                      * gmx_restrict        nrnb)
 297 {
 298     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 299      * just 0 for non-waters.
 300      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 301      * jnr indices corresponding to data put in the four positions in the SIMD register.
 302      */
 303     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 304     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 305     int              jnrA,jnrB;
 306     int              j_coord_offsetA,j_coord_offsetB;
 307     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 308     real             rcutoff_scalar;
 309     real             *shiftvec,*fshift,*x,*f;
 310     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 311     int              vdwioffset0;
 312     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 313     int              vdwjidx0A,vdwjidx0B;
 314     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 315     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 316     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 317     real             *charge;
 318     __m128d          dummy_mask,cutoff_mask;
 319     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 320     __m128d          one     = _mm_set1_pd(1.0);
 321     __m128d          two     = _mm_set1_pd(2.0);
 322     x                = xx[0];
 323     f                = ff[0];
 324
 325     nri              = nlist->nri;
 326     iinr             = nlist->iinr;
 327     jindex           = nlist->jindex;
 328     jjnr             = nlist->jjnr;
 329     shiftidx         = nlist->shift;
 330     gid              = nlist->gid;
 331     shiftvec         = fr->shift_vec[0];
 332     fshift           = fr->fshift[0];
 333     facel            = _mm_set1_pd(fr->epsfac);
 334     charge           = mdatoms->chargeA;
 335
 336     /* Avoid stupid compiler warnings */
 337     jnrA = jnrB = 0;
 338     j_coord_offsetA = 0;
 339     j_coord_offsetB = 0;
 340
 341     outeriter        = 0;
 342     inneriter        = 0;
 343
 344     /* Start outer loop over neighborlists */
 345     for(iidx=0; iidx<nri; iidx++)
 346     {
 347         /* Load shift vector for this list */
 348         i_shift_offset   = DIM*shiftidx[iidx];
 349
 350         /* Load limits for loop over neighbors */
 351         j_index_start    = jindex[iidx];
 352         j_index_end      = jindex[iidx+1];
 353
 354         /* Get outer coordinate index */
 355         inr              = iinr[iidx];
 356         i_coord_offset   = DIM*inr;
 357
 358         /* Load i particle coords and add shift vector */
 359         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 360
 361         fix0             = _mm_setzero_pd();
 362         fiy0             = _mm_setzero_pd();
 363         fiz0             = _mm_setzero_pd();
 364
 365         /* Load parameters for i particles */
 366         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 367
 368         /* Start inner kernel loop */
 369         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 370         {
 371
 372             /* Get j neighbor index, and coordinate index */
 373             jnrA             = jjnr[jidx];
 374             jnrB             = jjnr[jidx+1];
 375             j_coord_offsetA  = DIM*jnrA;
 376             j_coord_offsetB  = DIM*jnrB;
 377
 378             /* load j atom coordinates */
 379             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 380                                               &jx0,&jy0,&jz0);
 381
 382             /* Calculate displacement vector */
 383             dx00             = _mm_sub_pd(ix0,jx0);
 384             dy00             = _mm_sub_pd(iy0,jy0);
 385             dz00             = _mm_sub_pd(iz0,jz0);
 386
 387             /* Calculate squared distance and things based on it */
 388             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 389
 390             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 391
 392             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 393
 394             /* Load parameters for j particles */
 395             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 396
 397             /**************************
 398              * CALCULATE INTERACTIONS *
 399              **************************/
 400
 401             /* Compute parameters for interactions between i and j atoms */
 402             qq00             = _mm_mul_pd(iq0,jq0);
 403
 404             /* COULOMB ELECTROSTATICS */
 405             velec            = _mm_mul_pd(qq00,rinv00);
 406             felec            = _mm_mul_pd(velec,rinvsq00);
 407
 408             fscal            = felec;
 409
 410             /* Calculate temporary vectorial force */
 411             tx               = _mm_mul_pd(fscal,dx00);
 412             ty               = _mm_mul_pd(fscal,dy00);
 413             tz               = _mm_mul_pd(fscal,dz00);
 414
 415             /* Update vectorial force */
 416             fix0             = _mm_add_pd(fix0,tx);
 417             fiy0             = _mm_add_pd(fiy0,ty);
 418             fiz0             = _mm_add_pd(fiz0,tz);
 419
 420             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 421
 422             /* Inner loop uses 27 flops */
 423         }
 424
 425         if(jidx<j_index_end)
 426         {
 427
 428             jnrA             = jjnr[jidx];
 429             j_coord_offsetA  = DIM*jnrA;
 430
 431             /* load j atom coordinates */
 432             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 433                                               &jx0,&jy0,&jz0);
 434
 435             /* Calculate displacement vector */
 436             dx00             = _mm_sub_pd(ix0,jx0);
 437             dy00             = _mm_sub_pd(iy0,jy0);
 438             dz00             = _mm_sub_pd(iz0,jz0);
 439
 440             /* Calculate squared distance and things based on it */
 441             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 442
 443             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 444
 445             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 446
 447             /* Load parameters for j particles */
 448             jq0              = _mm_load_sd(charge+jnrA+0);
 449
 450             /**************************
 451              * CALCULATE INTERACTIONS *
 452              **************************/
 453
 454             /* Compute parameters for interactions between i and j atoms */
 455             qq00             = _mm_mul_pd(iq0,jq0);
 456
 457             /* COULOMB ELECTROSTATICS */
 458             velec            = _mm_mul_pd(qq00,rinv00);
 459             felec            = _mm_mul_pd(velec,rinvsq00);
 460
 461             fscal            = felec;
 462
 463             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 464
 465             /* Calculate temporary vectorial force */
 466             tx               = _mm_mul_pd(fscal,dx00);
 467             ty               = _mm_mul_pd(fscal,dy00);
 468             tz               = _mm_mul_pd(fscal,dz00);
 469
 470             /* Update vectorial force */
 471             fix0             = _mm_add_pd(fix0,tx);
 472             fiy0             = _mm_add_pd(fiy0,ty);
 473             fiz0             = _mm_add_pd(fiz0,tz);
 474
 475             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 476
 477             /* Inner loop uses 27 flops */
 478         }
 479
 480         /* End of innermost loop */
 481
 482         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 483                                               f+i_coord_offset,fshift+i_shift_offset);
 484
 485         /* Increment number of inner iterations */
 486         inneriter                  += j_index_end - j_index_start;
 487
 488         /* Outer loop uses 7 flops */
 489     }
 490
 491     /* Increment number of outer iterations */
 492     outeriter        += nri;
 493
 494     /* Update outer/inner flops */
 495
 496     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 497 }