src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwNone_GeomP1P1_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_ElecRFCut_VdwNone_GeomP1P1_VF_sse4_1_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_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              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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
  95     rcutoff_scalar   = fr->rcoulomb;
  96     rcutoff          = _mm_set1_pd(rcutoff_scalar);
  97     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
  98
  99     /* Avoid stupid compiler warnings */
 100     jnrA = jnrB = 0;
 101     j_coord_offsetA = 0;
 102     j_coord_offsetB = 0;
 103
 104     outeriter        = 0;
 105     inneriter        = 0;
 106
 107     /* Start outer loop over neighborlists */
 108     for(iidx=0; iidx<nri; iidx++)
 109     {
 110         /* Load shift vector for this list */
 111         i_shift_offset   = DIM*shiftidx[iidx];
 112
 113         /* Load limits for loop over neighbors */
 114         j_index_start    = jindex[iidx];
 115         j_index_end      = jindex[iidx+1];
 116
 117         /* Get outer coordinate index */
 118         inr              = iinr[iidx];
 119         i_coord_offset   = DIM*inr;
 120
 121         /* Load i particle coords and add shift vector */
 122         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 123
 124         fix0             = _mm_setzero_pd();
 125         fiy0             = _mm_setzero_pd();
 126         fiz0             = _mm_setzero_pd();
 127
 128         /* Load parameters for i particles */
 129         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 130
 131         /* Reset potential sums */
 132         velecsum         = _mm_setzero_pd();
 133
 134         /* Start inner kernel loop */
 135         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 136         {
 137
 138             /* Get j neighbor index, and coordinate index */
 139             jnrA             = jjnr[jidx];
 140             jnrB             = jjnr[jidx+1];
 141             j_coord_offsetA  = DIM*jnrA;
 142             j_coord_offsetB  = DIM*jnrB;
 143
 144             /* load j atom coordinates */
 145             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 146                                               &jx0,&jy0,&jz0);
 147
 148             /* Calculate displacement vector */
 149             dx00             = _mm_sub_pd(ix0,jx0);
 150             dy00             = _mm_sub_pd(iy0,jy0);
 151             dz00             = _mm_sub_pd(iz0,jz0);
 152
 153             /* Calculate squared distance and things based on it */
 154             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 155
 156             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 157
 158             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 159
 160             /* Load parameters for j particles */
 161             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 162
 163             /**************************
 164              * CALCULATE INTERACTIONS *
 165              **************************/
 166
 167             if (gmx_mm_any_lt(rsq00,rcutoff2))
 168             {
 169
 170             /* Compute parameters for interactions between i and j atoms */
 171             qq00             = _mm_mul_pd(iq0,jq0);
 172
 173             /* REACTION-FIELD ELECTROSTATICS */
 174             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 175             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 176
 177             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 178
 179             /* Update potential sum for this i atom from the interaction with this j atom. */
 180             velec            = _mm_and_pd(velec,cutoff_mask);
 181             velecsum         = _mm_add_pd(velecsum,velec);
 182
 183             fscal            = felec;
 184
 185             fscal            = _mm_and_pd(fscal,cutoff_mask);
 186
 187             /* Calculate temporary vectorial force */
 188             tx               = _mm_mul_pd(fscal,dx00);
 189             ty               = _mm_mul_pd(fscal,dy00);
 190             tz               = _mm_mul_pd(fscal,dz00);
 191
 192             /* Update vectorial force */
 193             fix0             = _mm_add_pd(fix0,tx);
 194             fiy0             = _mm_add_pd(fiy0,ty);
 195             fiz0             = _mm_add_pd(fiz0,tz);
 196
 197             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 198
 199             }
 200
 201             /* Inner loop uses 36 flops */
 202         }
 203
 204         if(jidx<j_index_end)
 205         {
 206
 207             jnrA             = jjnr[jidx];
 208             j_coord_offsetA  = DIM*jnrA;
 209
 210             /* load j atom coordinates */
 211             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 212                                               &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx00             = _mm_sub_pd(ix0,jx0);
 216             dy00             = _mm_sub_pd(iy0,jy0);
 217             dz00             = _mm_sub_pd(iz0,jz0);
 218
 219             /* Calculate squared distance and things based on it */
 220             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 221
 222             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 223
 224             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 225
 226             /* Load parameters for j particles */
 227             jq0              = _mm_load_sd(charge+jnrA+0);
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             if (gmx_mm_any_lt(rsq00,rcutoff2))
 234             {
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             qq00             = _mm_mul_pd(iq0,jq0);
 238
 239             /* REACTION-FIELD ELECTROSTATICS */
 240             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 241             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 242
 243             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 244
 245             /* Update potential sum for this i atom from the interaction with this j atom. */
 246             velec            = _mm_and_pd(velec,cutoff_mask);
 247             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 248             velecsum         = _mm_add_pd(velecsum,velec);
 249
 250             fscal            = felec;
 251
 252             fscal            = _mm_and_pd(fscal,cutoff_mask);
 253
 254             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 255
 256             /* Calculate temporary vectorial force */
 257             tx               = _mm_mul_pd(fscal,dx00);
 258             ty               = _mm_mul_pd(fscal,dy00);
 259             tz               = _mm_mul_pd(fscal,dz00);
 260
 261             /* Update vectorial force */
 262             fix0             = _mm_add_pd(fix0,tx);
 263             fiy0             = _mm_add_pd(fiy0,ty);
 264             fiz0             = _mm_add_pd(fiz0,tz);
 265
 266             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 267
 268             }
 269
 270             /* Inner loop uses 36 flops */
 271         }
 272
 273         /* End of innermost loop */
 274
 275         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 276                                               f+i_coord_offset,fshift+i_shift_offset);
 277
 278         ggid                        = gid[iidx];
 279         /* Update potential energies */
 280         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 281
 282         /* Increment number of inner iterations */
 283         inneriter                  += j_index_end - j_index_start;
 284
 285         /* Outer loop uses 8 flops */
 286     }
 287
 288     /* Increment number of outer iterations */
 289     outeriter        += nri;
 290
 291     /* Update outer/inner flops */
 292
 293     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
 294 }
 295 /*
 296  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse4_1_double
 297  * Electrostatics interaction: ReactionField
 298  * VdW interaction:            None
 299  * Geometry:                   Particle-Particle
 300  * Calculate force/pot:        Force
 301  */
 302 void
 303 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse4_1_double
 304                     (t_nblist * gmx_restrict                nlist,
 305                      rvec * gmx_restrict                    xx,
 306                      rvec * gmx_restrict                    ff,
 307                      t_forcerec * gmx_restrict              fr,
 308                      t_mdatoms * gmx_restrict               mdatoms,
 309                      nb_kernel_data_t * gmx_restrict        kernel_data,
 310                      t_nrnb * gmx_restrict                  nrnb)
 311 {
 312     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 313      * just 0 for non-waters.
 314      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 315      * jnr indices corresponding to data put in the four positions in the SIMD register.
 316      */
 317     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 318     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 319     int              jnrA,jnrB;
 320     int              j_coord_offsetA,j_coord_offsetB;
 321     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 322     real             rcutoff_scalar;
 323     real             *shiftvec,*fshift,*x,*f;
 324     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 325     int              vdwioffset0;
 326     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 327     int              vdwjidx0A,vdwjidx0B;
 328     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 329     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 330     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 331     real             *charge;
 332     __m128d          dummy_mask,cutoff_mask;
 333     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 334     __m128d          one     = _mm_set1_pd(1.0);
 335     __m128d          two     = _mm_set1_pd(2.0);
 336     x                = xx[0];
 337     f                = ff[0];
 338
 339     nri              = nlist->nri;
 340     iinr             = nlist->iinr;
 341     jindex           = nlist->jindex;
 342     jjnr             = nlist->jjnr;
 343     shiftidx         = nlist->shift;
 344     gid              = nlist->gid;
 345     shiftvec         = fr->shift_vec[0];
 346     fshift           = fr->fshift[0];
 347     facel            = _mm_set1_pd(fr->epsfac);
 348     charge           = mdatoms->chargeA;
 349     krf              = _mm_set1_pd(fr->ic->k_rf);
 350     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 351     crf              = _mm_set1_pd(fr->ic->c_rf);
 352
 353     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 354     rcutoff_scalar   = fr->rcoulomb;
 355     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 356     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 357
 358     /* Avoid stupid compiler warnings */
 359     jnrA = jnrB = 0;
 360     j_coord_offsetA = 0;
 361     j_coord_offsetB = 0;
 362
 363     outeriter        = 0;
 364     inneriter        = 0;
 365
 366     /* Start outer loop over neighborlists */
 367     for(iidx=0; iidx<nri; iidx++)
 368     {
 369         /* Load shift vector for this list */
 370         i_shift_offset   = DIM*shiftidx[iidx];
 371
 372         /* Load limits for loop over neighbors */
 373         j_index_start    = jindex[iidx];
 374         j_index_end      = jindex[iidx+1];
 375
 376         /* Get outer coordinate index */
 377         inr              = iinr[iidx];
 378         i_coord_offset   = DIM*inr;
 379
 380         /* Load i particle coords and add shift vector */
 381         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 382
 383         fix0             = _mm_setzero_pd();
 384         fiy0             = _mm_setzero_pd();
 385         fiz0             = _mm_setzero_pd();
 386
 387         /* Load parameters for i particles */
 388         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 389
 390         /* Start inner kernel loop */
 391         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 392         {
 393
 394             /* Get j neighbor index, and coordinate index */
 395             jnrA             = jjnr[jidx];
 396             jnrB             = jjnr[jidx+1];
 397             j_coord_offsetA  = DIM*jnrA;
 398             j_coord_offsetB  = DIM*jnrB;
 399
 400             /* load j atom coordinates */
 401             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 402                                               &jx0,&jy0,&jz0);
 403
 404             /* Calculate displacement vector */
 405             dx00             = _mm_sub_pd(ix0,jx0);
 406             dy00             = _mm_sub_pd(iy0,jy0);
 407             dz00             = _mm_sub_pd(iz0,jz0);
 408
 409             /* Calculate squared distance and things based on it */
 410             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 411
 412             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 413
 414             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 415
 416             /* Load parameters for j particles */
 417             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             if (gmx_mm_any_lt(rsq00,rcutoff2))
 424             {
 425
 426             /* Compute parameters for interactions between i and j atoms */
 427             qq00             = _mm_mul_pd(iq0,jq0);
 428
 429             /* REACTION-FIELD ELECTROSTATICS */
 430             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 431
 432             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 433
 434             fscal            = felec;
 435
 436             fscal            = _mm_and_pd(fscal,cutoff_mask);
 437
 438             /* Calculate temporary vectorial force */
 439             tx               = _mm_mul_pd(fscal,dx00);
 440             ty               = _mm_mul_pd(fscal,dy00);
 441             tz               = _mm_mul_pd(fscal,dz00);
 442
 443             /* Update vectorial force */
 444             fix0             = _mm_add_pd(fix0,tx);
 445             fiy0             = _mm_add_pd(fiy0,ty);
 446             fiz0             = _mm_add_pd(fiz0,tz);
 447
 448             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 449
 450             }
 451
 452             /* Inner loop uses 30 flops */
 453         }
 454
 455         if(jidx<j_index_end)
 456         {
 457
 458             jnrA             = jjnr[jidx];
 459             j_coord_offsetA  = DIM*jnrA;
 460
 461             /* load j atom coordinates */
 462             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 463                                               &jx0,&jy0,&jz0);
 464
 465             /* Calculate displacement vector */
 466             dx00             = _mm_sub_pd(ix0,jx0);
 467             dy00             = _mm_sub_pd(iy0,jy0);
 468             dz00             = _mm_sub_pd(iz0,jz0);
 469
 470             /* Calculate squared distance and things based on it */
 471             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 472
 473             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 474
 475             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 476
 477             /* Load parameters for j particles */
 478             jq0              = _mm_load_sd(charge+jnrA+0);
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             if (gmx_mm_any_lt(rsq00,rcutoff2))
 485             {
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             qq00             = _mm_mul_pd(iq0,jq0);
 489
 490             /* REACTION-FIELD ELECTROSTATICS */
 491             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 492
 493             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 494
 495             fscal            = felec;
 496
 497             fscal            = _mm_and_pd(fscal,cutoff_mask);
 498
 499             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 500
 501             /* Calculate temporary vectorial force */
 502             tx               = _mm_mul_pd(fscal,dx00);
 503             ty               = _mm_mul_pd(fscal,dy00);
 504             tz               = _mm_mul_pd(fscal,dz00);
 505
 506             /* Update vectorial force */
 507             fix0             = _mm_add_pd(fix0,tx);
 508             fiy0             = _mm_add_pd(fiy0,ty);
 509             fiz0             = _mm_add_pd(fiz0,tz);
 510
 511             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 512
 513             }
 514
 515             /* Inner loop uses 30 flops */
 516         }
 517
 518         /* End of innermost loop */
 519
 520         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 521                                               f+i_coord_offset,fshift+i_shift_offset);
 522
 523         /* Increment number of inner iterations */
 524         inneriter                  += j_index_end - j_index_start;
 525
 526         /* Outer loop uses 7 flops */
 527     }
 528
 529     /* Increment number of outer iterations */
 530     outeriter        += nri;
 531
 532     /* Update outer/inner flops */
 533
 534     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 535 }