src/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRF_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_ElecRF_VdwNone_GeomP1P1_VF_sse2_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_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     krf              = _mm_set1_pd(fr->ic->k_rf);
  91     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
  92     crf              = _mm_set1_pd(fr->ic->c_rf);
  93
  94     /* Avoid stupid compiler warnings */
  95     jnrA = jnrB = 0;
  96     j_coord_offsetA = 0;
  97     j_coord_offsetB = 0;
  98
  99     outeriter        = 0;
 100     inneriter        = 0;
 101
 102     /* Start outer loop over neighborlists */
 103     for(iidx=0; iidx<nri; iidx++)
 104     {
 105         /* Load shift vector for this list */
 106         i_shift_offset   = DIM*shiftidx[iidx];
 107
 108         /* Load limits for loop over neighbors */
 109         j_index_start    = jindex[iidx];
 110         j_index_end      = jindex[iidx+1];
 111
 112         /* Get outer coordinate index */
 113         inr              = iinr[iidx];
 114         i_coord_offset   = DIM*inr;
 115
 116         /* Load i particle coords and add shift vector */
 117         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 118
 119         fix0             = _mm_setzero_pd();
 120         fiy0             = _mm_setzero_pd();
 121         fiz0             = _mm_setzero_pd();
 122
 123         /* Load parameters for i particles */
 124         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 125
 126         /* Reset potential sums */
 127         velecsum         = _mm_setzero_pd();
 128
 129         /* Start inner kernel loop */
 130         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 131         {
 132
 133             /* Get j neighbor index, and coordinate index */
 134             jnrA             = jjnr[jidx];
 135             jnrB             = jjnr[jidx+1];
 136             j_coord_offsetA  = DIM*jnrA;
 137             j_coord_offsetB  = DIM*jnrB;
 138
 139             /* load j atom coordinates */
 140             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 141                                               &jx0,&jy0,&jz0);
 142
 143             /* Calculate displacement vector */
 144             dx00             = _mm_sub_pd(ix0,jx0);
 145             dy00             = _mm_sub_pd(iy0,jy0);
 146             dz00             = _mm_sub_pd(iz0,jz0);
 147
 148             /* Calculate squared distance and things based on it */
 149             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 150
 151             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 152
 153             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 154
 155             /* Load parameters for j particles */
 156             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 157
 158             /**************************
 159              * CALCULATE INTERACTIONS *
 160              **************************/
 161
 162             /* Compute parameters for interactions between i and j atoms */
 163             qq00             = _mm_mul_pd(iq0,jq0);
 164
 165             /* REACTION-FIELD ELECTROSTATICS */
 166             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 167             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 168
 169             /* Update potential sum for this i atom from the interaction with this j atom. */
 170             velecsum         = _mm_add_pd(velecsum,velec);
 171
 172             fscal            = felec;
 173
 174             /* Calculate temporary vectorial force */
 175             tx               = _mm_mul_pd(fscal,dx00);
 176             ty               = _mm_mul_pd(fscal,dy00);
 177             tz               = _mm_mul_pd(fscal,dz00);
 178
 179             /* Update vectorial force */
 180             fix0             = _mm_add_pd(fix0,tx);
 181             fiy0             = _mm_add_pd(fiy0,ty);
 182             fiz0             = _mm_add_pd(fiz0,tz);
 183
 184             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 185
 186             /* Inner loop uses 32 flops */
 187         }
 188
 189         if(jidx<j_index_end)
 190         {
 191
 192             jnrA             = jjnr[jidx];
 193             j_coord_offsetA  = DIM*jnrA;
 194
 195             /* load j atom coordinates */
 196             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm_sub_pd(ix0,jx0);
 201             dy00             = _mm_sub_pd(iy0,jy0);
 202             dz00             = _mm_sub_pd(iz0,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 206
 207             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 208
 209             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 210
 211             /* Load parameters for j particles */
 212             jq0              = _mm_load_sd(charge+jnrA+0);
 213
 214             /**************************
 215              * CALCULATE INTERACTIONS *
 216              **************************/
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq00             = _mm_mul_pd(iq0,jq0);
 220
 221             /* REACTION-FIELD ELECTROSTATICS */
 222             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 223             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 224
 225             /* Update potential sum for this i atom from the interaction with this j atom. */
 226             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 227             velecsum         = _mm_add_pd(velecsum,velec);
 228
 229             fscal            = felec;
 230
 231             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 232
 233             /* Calculate temporary vectorial force */
 234             tx               = _mm_mul_pd(fscal,dx00);
 235             ty               = _mm_mul_pd(fscal,dy00);
 236             tz               = _mm_mul_pd(fscal,dz00);
 237
 238             /* Update vectorial force */
 239             fix0             = _mm_add_pd(fix0,tx);
 240             fiy0             = _mm_add_pd(fiy0,ty);
 241             fiz0             = _mm_add_pd(fiz0,tz);
 242
 243             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 244
 245             /* Inner loop uses 32 flops */
 246         }
 247
 248         /* End of innermost loop */
 249
 250         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 251                                               f+i_coord_offset,fshift+i_shift_offset);
 252
 253         ggid                        = gid[iidx];
 254         /* Update potential energies */
 255         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 256
 257         /* Increment number of inner iterations */
 258         inneriter                  += j_index_end - j_index_start;
 259
 260         /* Outer loop uses 8 flops */
 261     }
 262
 263     /* Increment number of outer iterations */
 264     outeriter        += nri;
 265
 266     /* Update outer/inner flops */
 267
 268     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
 269 }
 270 /*
 271  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse2_double
 272  * Electrostatics interaction: ReactionField
 273  * VdW interaction:            None
 274  * Geometry:                   Particle-Particle
 275  * Calculate force/pot:        Force
 276  */
 277 void
 278 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse2_double
 279                     (t_nblist * gmx_restrict                nlist,
 280                      rvec * gmx_restrict                    xx,
 281                      rvec * gmx_restrict                    ff,
 282                      t_forcerec * gmx_restrict              fr,
 283                      t_mdatoms * gmx_restrict               mdatoms,
 284                      nb_kernel_data_t * gmx_restrict        kernel_data,
 285                      t_nrnb * gmx_restrict                  nrnb)
 286 {
 287     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 288      * just 0 for non-waters.
 289      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 290      * jnr indices corresponding to data put in the four positions in the SIMD register.
 291      */
 292     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 293     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 294     int              jnrA,jnrB;
 295     int              j_coord_offsetA,j_coord_offsetB;
 296     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 297     real             rcutoff_scalar;
 298     real             *shiftvec,*fshift,*x,*f;
 299     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 300     int              vdwioffset0;
 301     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 302     int              vdwjidx0A,vdwjidx0B;
 303     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 304     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 305     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 306     real             *charge;
 307     __m128d          dummy_mask,cutoff_mask;
 308     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 309     __m128d          one     = _mm_set1_pd(1.0);
 310     __m128d          two     = _mm_set1_pd(2.0);
 311     x                = xx[0];
 312     f                = ff[0];
 313
 314     nri              = nlist->nri;
 315     iinr             = nlist->iinr;
 316     jindex           = nlist->jindex;
 317     jjnr             = nlist->jjnr;
 318     shiftidx         = nlist->shift;
 319     gid              = nlist->gid;
 320     shiftvec         = fr->shift_vec[0];
 321     fshift           = fr->fshift[0];
 322     facel            = _mm_set1_pd(fr->epsfac);
 323     charge           = mdatoms->chargeA;
 324     krf              = _mm_set1_pd(fr->ic->k_rf);
 325     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 326     crf              = _mm_set1_pd(fr->ic->c_rf);
 327
 328     /* Avoid stupid compiler warnings */
 329     jnrA = jnrB = 0;
 330     j_coord_offsetA = 0;
 331     j_coord_offsetB = 0;
 332
 333     outeriter        = 0;
 334     inneriter        = 0;
 335
 336     /* Start outer loop over neighborlists */
 337     for(iidx=0; iidx<nri; iidx++)
 338     {
 339         /* Load shift vector for this list */
 340         i_shift_offset   = DIM*shiftidx[iidx];
 341
 342         /* Load limits for loop over neighbors */
 343         j_index_start    = jindex[iidx];
 344         j_index_end      = jindex[iidx+1];
 345
 346         /* Get outer coordinate index */
 347         inr              = iinr[iidx];
 348         i_coord_offset   = DIM*inr;
 349
 350         /* Load i particle coords and add shift vector */
 351         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 352
 353         fix0             = _mm_setzero_pd();
 354         fiy0             = _mm_setzero_pd();
 355         fiz0             = _mm_setzero_pd();
 356
 357         /* Load parameters for i particles */
 358         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 359
 360         /* Start inner kernel loop */
 361         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 362         {
 363
 364             /* Get j neighbor index, and coordinate index */
 365             jnrA             = jjnr[jidx];
 366             jnrB             = jjnr[jidx+1];
 367             j_coord_offsetA  = DIM*jnrA;
 368             j_coord_offsetB  = DIM*jnrB;
 369
 370             /* load j atom coordinates */
 371             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 372                                               &jx0,&jy0,&jz0);
 373
 374             /* Calculate displacement vector */
 375             dx00             = _mm_sub_pd(ix0,jx0);
 376             dy00             = _mm_sub_pd(iy0,jy0);
 377             dz00             = _mm_sub_pd(iz0,jz0);
 378
 379             /* Calculate squared distance and things based on it */
 380             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 381
 382             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 383
 384             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 385
 386             /* Load parameters for j particles */
 387             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 388
 389             /**************************
 390              * CALCULATE INTERACTIONS *
 391              **************************/
 392
 393             /* Compute parameters for interactions between i and j atoms */
 394             qq00             = _mm_mul_pd(iq0,jq0);
 395
 396             /* REACTION-FIELD ELECTROSTATICS */
 397             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 398
 399             fscal            = felec;
 400
 401             /* Calculate temporary vectorial force */
 402             tx               = _mm_mul_pd(fscal,dx00);
 403             ty               = _mm_mul_pd(fscal,dy00);
 404             tz               = _mm_mul_pd(fscal,dz00);
 405
 406             /* Update vectorial force */
 407             fix0             = _mm_add_pd(fix0,tx);
 408             fiy0             = _mm_add_pd(fiy0,ty);
 409             fiz0             = _mm_add_pd(fiz0,tz);
 410
 411             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 412
 413             /* Inner loop uses 27 flops */
 414         }
 415
 416         if(jidx<j_index_end)
 417         {
 418
 419             jnrA             = jjnr[jidx];
 420             j_coord_offsetA  = DIM*jnrA;
 421
 422             /* load j atom coordinates */
 423             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 424                                               &jx0,&jy0,&jz0);
 425
 426             /* Calculate displacement vector */
 427             dx00             = _mm_sub_pd(ix0,jx0);
 428             dy00             = _mm_sub_pd(iy0,jy0);
 429             dz00             = _mm_sub_pd(iz0,jz0);
 430
 431             /* Calculate squared distance and things based on it */
 432             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 433
 434             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 435
 436             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 437
 438             /* Load parameters for j particles */
 439             jq0              = _mm_load_sd(charge+jnrA+0);
 440
 441             /**************************
 442              * CALCULATE INTERACTIONS *
 443              **************************/
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq00             = _mm_mul_pd(iq0,jq0);
 447
 448             /* REACTION-FIELD ELECTROSTATICS */
 449             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 450
 451             fscal            = felec;
 452
 453             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 454
 455             /* Calculate temporary vectorial force */
 456             tx               = _mm_mul_pd(fscal,dx00);
 457             ty               = _mm_mul_pd(fscal,dy00);
 458             tz               = _mm_mul_pd(fscal,dz00);
 459
 460             /* Update vectorial force */
 461             fix0             = _mm_add_pd(fix0,tx);
 462             fiy0             = _mm_add_pd(fiy0,ty);
 463             fiz0             = _mm_add_pd(fiz0,tz);
 464
 465             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 466
 467             /* Inner loop uses 27 flops */
 468         }
 469
 470         /* End of innermost loop */
 471
 472         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 473                                               f+i_coord_offset,fshift+i_shift_offset);
 474
 475         /* Increment number of inner iterations */
 476         inneriter                  += j_index_end - j_index_start;
 477
 478         /* Outer loop uses 7 flops */
 479     }
 480
 481     /* Increment number of outer iterations */
 482     outeriter        += nri;
 483
 484     /* Update outer/inner flops */
 485
 486     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 487 }