src/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCoul_VdwNone_GeomP1P1_sse2_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_double.h"
  34 #include "kernelutil_x86_sse2_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_sse2_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_sse2_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              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     __m128d          dummy_mask,cutoff_mask;
  74     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  75     __m128d          one     = _mm_set1_pd(1.0);
  76     __m128d          two     = _mm_set1_pd(2.0);
  77     x                = xx[0];
  78     f                = ff[0];
  79
  80     nri              = nlist->nri;
  81     iinr             = nlist->iinr;
  82     jindex           = nlist->jindex;
  83     jjnr             = nlist->jjnr;
  84     shiftidx         = nlist->shift;
  85     gid              = nlist->gid;
  86     shiftvec         = fr->shift_vec[0];
  87     fshift           = fr->fshift[0];
  88     facel            = _mm_set1_pd(fr->epsfac);
  89     charge           = mdatoms->chargeA;
  90
  91     /* Avoid stupid compiler warnings */
  92     jnrA = jnrB = 0;
  93     j_coord_offsetA = 0;
  94     j_coord_offsetB = 0;
  95
  96     outeriter        = 0;
  97     inneriter        = 0;
  98
  99     /* Start outer loop over neighborlists */
 100     for(iidx=0; iidx<nri; iidx++)
 101     {
 102         /* Load shift vector for this list */
 103         i_shift_offset   = DIM*shiftidx[iidx];
 104
 105         /* Load limits for loop over neighbors */
 106         j_index_start    = jindex[iidx];
 107         j_index_end      = jindex[iidx+1];
 108
 109         /* Get outer coordinate index */
 110         inr              = iinr[iidx];
 111         i_coord_offset   = DIM*inr;
 112
 113         /* Load i particle coords and add shift vector */
 114         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 115
 116         fix0             = _mm_setzero_pd();
 117         fiy0             = _mm_setzero_pd();
 118         fiz0             = _mm_setzero_pd();
 119
 120         /* Load parameters for i particles */
 121         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 122
 123         /* Reset potential sums */
 124         velecsum         = _mm_setzero_pd();
 125
 126         /* Start inner kernel loop */
 127         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 128         {
 129
 130             /* Get j neighbor index, and coordinate index */
 131             jnrA             = jjnr[jidx];
 132             jnrB             = jjnr[jidx+1];
 133             j_coord_offsetA  = DIM*jnrA;
 134             j_coord_offsetB  = DIM*jnrB;
 135
 136             /* load j atom coordinates */
 137             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 138                                               &jx0,&jy0,&jz0);
 139
 140             /* Calculate displacement vector */
 141             dx00             = _mm_sub_pd(ix0,jx0);
 142             dy00             = _mm_sub_pd(iy0,jy0);
 143             dz00             = _mm_sub_pd(iz0,jz0);
 144
 145             /* Calculate squared distance and things based on it */
 146             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 147
 148             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 149
 150             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 151
 152             /* Load parameters for j particles */
 153             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 154
 155             /**************************
 156              * CALCULATE INTERACTIONS *
 157              **************************/
 158
 159             /* Compute parameters for interactions between i and j atoms */
 160             qq00             = _mm_mul_pd(iq0,jq0);
 161
 162             /* COULOMB ELECTROSTATICS */
 163             velec            = _mm_mul_pd(qq00,rinv00);
 164             felec            = _mm_mul_pd(velec,rinvsq00);
 165
 166             /* Update potential sum for this i atom from the interaction with this j atom. */
 167             velecsum         = _mm_add_pd(velecsum,velec);
 168
 169             fscal            = felec;
 170
 171             /* Calculate temporary vectorial force */
 172             tx               = _mm_mul_pd(fscal,dx00);
 173             ty               = _mm_mul_pd(fscal,dy00);
 174             tz               = _mm_mul_pd(fscal,dz00);
 175
 176             /* Update vectorial force */
 177             fix0             = _mm_add_pd(fix0,tx);
 178             fiy0             = _mm_add_pd(fiy0,ty);
 179             fiz0             = _mm_add_pd(fiz0,tz);
 180
 181             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 182
 183             /* Inner loop uses 28 flops */
 184         }
 185
 186         if(jidx<j_index_end)
 187         {
 188
 189             jnrA             = jjnr[jidx];
 190             j_coord_offsetA  = DIM*jnrA;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_pd(ix0,jx0);
 198             dy00             = _mm_sub_pd(iy0,jy0);
 199             dz00             = _mm_sub_pd(iz0,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 203
 204             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 205
 206             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 207
 208             /* Load parameters for j particles */
 209             jq0              = _mm_load_sd(charge+jnrA+0);
 210
 211             /**************************
 212              * CALCULATE INTERACTIONS *
 213              **************************/
 214
 215             /* Compute parameters for interactions between i and j atoms */
 216             qq00             = _mm_mul_pd(iq0,jq0);
 217
 218             /* COULOMB ELECTROSTATICS */
 219             velec            = _mm_mul_pd(qq00,rinv00);
 220             felec            = _mm_mul_pd(velec,rinvsq00);
 221
 222             /* Update potential sum for this i atom from the interaction with this j atom. */
 223             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 224             velecsum         = _mm_add_pd(velecsum,velec);
 225
 226             fscal            = felec;
 227
 228             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 229
 230             /* Calculate temporary vectorial force */
 231             tx               = _mm_mul_pd(fscal,dx00);
 232             ty               = _mm_mul_pd(fscal,dy00);
 233             tz               = _mm_mul_pd(fscal,dz00);
 234
 235             /* Update vectorial force */
 236             fix0             = _mm_add_pd(fix0,tx);
 237             fiy0             = _mm_add_pd(fiy0,ty);
 238             fiz0             = _mm_add_pd(fiz0,tz);
 239
 240             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 241
 242             /* Inner loop uses 28 flops */
 243         }
 244
 245         /* End of innermost loop */
 246
 247         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 248                                               f+i_coord_offset,fshift+i_shift_offset);
 249
 250         ggid                        = gid[iidx];
 251         /* Update potential energies */
 252         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 253
 254         /* Increment number of inner iterations */
 255         inneriter                  += j_index_end - j_index_start;
 256
 257         /* Outer loop uses 8 flops */
 258     }
 259
 260     /* Increment number of outer iterations */
 261     outeriter        += nri;
 262
 263     /* Update outer/inner flops */
 264
 265     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*28);
 266 }
 267 /*
 268  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_sse2_double
 269  * Electrostatics interaction: Coulomb
 270  * VdW interaction:            None
 271  * Geometry:                   Particle-Particle
 272  * Calculate force/pot:        Force
 273  */
 274 void
 275 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_sse2_double
 276                     (t_nblist * gmx_restrict                nlist,
 277                      rvec * gmx_restrict                    xx,
 278                      rvec * gmx_restrict                    ff,
 279                      t_forcerec * gmx_restrict              fr,
 280                      t_mdatoms * gmx_restrict               mdatoms,
 281                      nb_kernel_data_t * gmx_restrict        kernel_data,
 282                      t_nrnb * gmx_restrict                  nrnb)
 283 {
 284     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 285      * just 0 for non-waters.
 286      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 287      * jnr indices corresponding to data put in the four positions in the SIMD register.
 288      */
 289     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 290     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 291     int              jnrA,jnrB;
 292     int              j_coord_offsetA,j_coord_offsetB;
 293     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 294     real             rcutoff_scalar;
 295     real             *shiftvec,*fshift,*x,*f;
 296     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 297     int              vdwioffset0;
 298     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 299     int              vdwjidx0A,vdwjidx0B;
 300     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 301     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 302     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 303     real             *charge;
 304     __m128d          dummy_mask,cutoff_mask;
 305     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 306     __m128d          one     = _mm_set1_pd(1.0);
 307     __m128d          two     = _mm_set1_pd(2.0);
 308     x                = xx[0];
 309     f                = ff[0];
 310
 311     nri              = nlist->nri;
 312     iinr             = nlist->iinr;
 313     jindex           = nlist->jindex;
 314     jjnr             = nlist->jjnr;
 315     shiftidx         = nlist->shift;
 316     gid              = nlist->gid;
 317     shiftvec         = fr->shift_vec[0];
 318     fshift           = fr->fshift[0];
 319     facel            = _mm_set1_pd(fr->epsfac);
 320     charge           = mdatoms->chargeA;
 321
 322     /* Avoid stupid compiler warnings */
 323     jnrA = jnrB = 0;
 324     j_coord_offsetA = 0;
 325     j_coord_offsetB = 0;
 326
 327     outeriter        = 0;
 328     inneriter        = 0;
 329
 330     /* Start outer loop over neighborlists */
 331     for(iidx=0; iidx<nri; iidx++)
 332     {
 333         /* Load shift vector for this list */
 334         i_shift_offset   = DIM*shiftidx[iidx];
 335
 336         /* Load limits for loop over neighbors */
 337         j_index_start    = jindex[iidx];
 338         j_index_end      = jindex[iidx+1];
 339
 340         /* Get outer coordinate index */
 341         inr              = iinr[iidx];
 342         i_coord_offset   = DIM*inr;
 343
 344         /* Load i particle coords and add shift vector */
 345         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 346
 347         fix0             = _mm_setzero_pd();
 348         fiy0             = _mm_setzero_pd();
 349         fiz0             = _mm_setzero_pd();
 350
 351         /* Load parameters for i particles */
 352         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 353
 354         /* Start inner kernel loop */
 355         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 356         {
 357
 358             /* Get j neighbor index, and coordinate index */
 359             jnrA             = jjnr[jidx];
 360             jnrB             = jjnr[jidx+1];
 361             j_coord_offsetA  = DIM*jnrA;
 362             j_coord_offsetB  = DIM*jnrB;
 363
 364             /* load j atom coordinates */
 365             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 366                                               &jx0,&jy0,&jz0);
 367
 368             /* Calculate displacement vector */
 369             dx00             = _mm_sub_pd(ix0,jx0);
 370             dy00             = _mm_sub_pd(iy0,jy0);
 371             dz00             = _mm_sub_pd(iz0,jz0);
 372
 373             /* Calculate squared distance and things based on it */
 374             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 375
 376             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 377
 378             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 379
 380             /* Load parameters for j particles */
 381             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 382
 383             /**************************
 384              * CALCULATE INTERACTIONS *
 385              **************************/
 386
 387             /* Compute parameters for interactions between i and j atoms */
 388             qq00             = _mm_mul_pd(iq0,jq0);
 389
 390             /* COULOMB ELECTROSTATICS */
 391             velec            = _mm_mul_pd(qq00,rinv00);
 392             felec            = _mm_mul_pd(velec,rinvsq00);
 393
 394             fscal            = felec;
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm_mul_pd(fscal,dx00);
 398             ty               = _mm_mul_pd(fscal,dy00);
 399             tz               = _mm_mul_pd(fscal,dz00);
 400
 401             /* Update vectorial force */
 402             fix0             = _mm_add_pd(fix0,tx);
 403             fiy0             = _mm_add_pd(fiy0,ty);
 404             fiz0             = _mm_add_pd(fiz0,tz);
 405
 406             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 407
 408             /* Inner loop uses 27 flops */
 409         }
 410
 411         if(jidx<j_index_end)
 412         {
 413
 414             jnrA             = jjnr[jidx];
 415             j_coord_offsetA  = DIM*jnrA;
 416
 417             /* load j atom coordinates */
 418             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 419                                               &jx0,&jy0,&jz0);
 420
 421             /* Calculate displacement vector */
 422             dx00             = _mm_sub_pd(ix0,jx0);
 423             dy00             = _mm_sub_pd(iy0,jy0);
 424             dz00             = _mm_sub_pd(iz0,jz0);
 425
 426             /* Calculate squared distance and things based on it */
 427             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 428
 429             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 430
 431             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 432
 433             /* Load parameters for j particles */
 434             jq0              = _mm_load_sd(charge+jnrA+0);
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             /* Compute parameters for interactions between i and j atoms */
 441             qq00             = _mm_mul_pd(iq0,jq0);
 442
 443             /* COULOMB ELECTROSTATICS */
 444             velec            = _mm_mul_pd(qq00,rinv00);
 445             felec            = _mm_mul_pd(velec,rinvsq00);
 446
 447             fscal            = felec;
 448
 449             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 450
 451             /* Calculate temporary vectorial force */
 452             tx               = _mm_mul_pd(fscal,dx00);
 453             ty               = _mm_mul_pd(fscal,dy00);
 454             tz               = _mm_mul_pd(fscal,dz00);
 455
 456             /* Update vectorial force */
 457             fix0             = _mm_add_pd(fix0,tx);
 458             fiy0             = _mm_add_pd(fiy0,ty);
 459             fiz0             = _mm_add_pd(fiz0,tz);
 460
 461             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 462
 463             /* Inner loop uses 27 flops */
 464         }
 465
 466         /* End of innermost loop */
 467
 468         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 469                                               f+i_coord_offset,fshift+i_shift_offset);
 470
 471         /* Increment number of inner iterations */
 472         inneriter                  += j_index_end - j_index_start;
 473
 474         /* Outer loop uses 7 flops */
 475     }
 476
 477     /* Increment number of outer iterations */
 478     outeriter        += nri;
 479
 480     /* Update outer/inner flops */
 481
 482     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 483 }