src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwNone_GeomW4P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single 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_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwNone_GeomW4P1_VF_sse2_single
  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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset1;
  67     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  68     int              vdwioffset2;
  69     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  70     int              vdwioffset3;
  71     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  75     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  76     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m128           dummy_mask,cutoff_mask;
  80     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  81     __m128           one     = _mm_set1_ps(1.0);
  82     __m128           two     = _mm_set1_ps(2.0);
  83     x                = xx[0];
  84     f                = ff[0];
  85
  86     nri              = nlist->nri;
  87     iinr             = nlist->iinr;
  88     jindex           = nlist->jindex;
  89     jjnr             = nlist->jjnr;
  90     shiftidx         = nlist->shift;
  91     gid              = nlist->gid;
  92     shiftvec         = fr->shift_vec[0];
  93     fshift           = fr->fshift[0];
  94     facel            = _mm_set1_ps(fr->epsfac);
  95     charge           = mdatoms->chargeA;
  96     krf              = _mm_set1_ps(fr->ic->k_rf);
  97     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
  98     crf              = _mm_set1_ps(fr->ic->c_rf);
  99
 100     /* Setup water-specific parameters */
 101     inr              = nlist->iinr[0];
 102     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 103     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 104     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 105
 106     /* Avoid stupid compiler warnings */
 107     jnrA = jnrB = jnrC = jnrD = 0;
 108     j_coord_offsetA = 0;
 109     j_coord_offsetB = 0;
 110     j_coord_offsetC = 0;
 111     j_coord_offsetD = 0;
 112
 113     outeriter        = 0;
 114     inneriter        = 0;
 115
 116     /* Start outer loop over neighborlists */
 117     for(iidx=0; iidx<nri; iidx++)
 118     {
 119         /* Load shift vector for this list */
 120         i_shift_offset   = DIM*shiftidx[iidx];
 121         shX              = shiftvec[i_shift_offset+XX];
 122         shY              = shiftvec[i_shift_offset+YY];
 123         shZ              = shiftvec[i_shift_offset+ZZ];
 124
 125         /* Load limits for loop over neighbors */
 126         j_index_start    = jindex[iidx];
 127         j_index_end      = jindex[iidx+1];
 128
 129         /* Get outer coordinate index */
 130         inr              = iinr[iidx];
 131         i_coord_offset   = DIM*inr;
 132
 133         /* Load i particle coords and add shift vector */
 134         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 135         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 136         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 137         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 138         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 139         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 140         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 141         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 142         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 143
 144         fix1             = _mm_setzero_ps();
 145         fiy1             = _mm_setzero_ps();
 146         fiz1             = _mm_setzero_ps();
 147         fix2             = _mm_setzero_ps();
 148         fiy2             = _mm_setzero_ps();
 149         fiz2             = _mm_setzero_ps();
 150         fix3             = _mm_setzero_ps();
 151         fiy3             = _mm_setzero_ps();
 152         fiz3             = _mm_setzero_ps();
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm_setzero_ps();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             jnrC             = jjnr[jidx+2];
 165             jnrD             = jjnr[jidx+3];
 166
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169             j_coord_offsetC  = DIM*jnrC;
 170             j_coord_offsetD  = DIM*jnrD;
 171
 172             /* load j atom coordinates */
 173             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 174                                               x+j_coord_offsetC,x+j_coord_offsetD,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx10             = _mm_sub_ps(ix1,jx0);
 179             dy10             = _mm_sub_ps(iy1,jy0);
 180             dz10             = _mm_sub_ps(iz1,jz0);
 181             dx20             = _mm_sub_ps(ix2,jx0);
 182             dy20             = _mm_sub_ps(iy2,jy0);
 183             dz20             = _mm_sub_ps(iz2,jz0);
 184             dx30             = _mm_sub_ps(ix3,jx0);
 185             dy30             = _mm_sub_ps(iy3,jy0);
 186             dz30             = _mm_sub_ps(iz3,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 190             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 191             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 192
 193             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 194             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 195             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 196
 197             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 198             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 199             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 200
 201             /* Load parameters for j particles */
 202             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 203                                                               charge+jnrC+0,charge+jnrD+0);
 204
 205             /**************************
 206              * CALCULATE INTERACTIONS *
 207              **************************/
 208
 209             /* Compute parameters for interactions between i and j atoms */
 210             qq10             = _mm_mul_ps(iq1,jq0);
 211
 212             /* REACTION-FIELD ELECTROSTATICS */
 213             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 214             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 215
 216             /* Update potential sum for this i atom from the interaction with this j atom. */
 217             velecsum         = _mm_add_ps(velecsum,velec);
 218
 219             fscal            = felec;
 220
 221             /* Calculate temporary vectorial force */
 222             tx               = _mm_mul_ps(fscal,dx10);
 223             ty               = _mm_mul_ps(fscal,dy10);
 224             tz               = _mm_mul_ps(fscal,dz10);
 225
 226             /* Update vectorial force */
 227             fix1             = _mm_add_ps(fix1,tx);
 228             fiy1             = _mm_add_ps(fiy1,ty);
 229             fiz1             = _mm_add_ps(fiz1,tz);
 230
 231             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 232                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 233                                                    tx,ty,tz);
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq20             = _mm_mul_ps(iq2,jq0);
 241
 242             /* REACTION-FIELD ELECTROSTATICS */
 243             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 244             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velecsum         = _mm_add_ps(velecsum,velec);
 248
 249             fscal            = felec;
 250
 251             /* Calculate temporary vectorial force */
 252             tx               = _mm_mul_ps(fscal,dx20);
 253             ty               = _mm_mul_ps(fscal,dy20);
 254             tz               = _mm_mul_ps(fscal,dz20);
 255
 256             /* Update vectorial force */
 257             fix2             = _mm_add_ps(fix2,tx);
 258             fiy2             = _mm_add_ps(fiy2,ty);
 259             fiz2             = _mm_add_ps(fiz2,tz);
 260
 261             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 262                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 263                                                    tx,ty,tz);
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq30             = _mm_mul_ps(iq3,jq0);
 271
 272             /* REACTION-FIELD ELECTROSTATICS */
 273             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 274             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velecsum         = _mm_add_ps(velecsum,velec);
 278
 279             fscal            = felec;
 280
 281             /* Calculate temporary vectorial force */
 282             tx               = _mm_mul_ps(fscal,dx30);
 283             ty               = _mm_mul_ps(fscal,dy30);
 284             tz               = _mm_mul_ps(fscal,dz30);
 285
 286             /* Update vectorial force */
 287             fix3             = _mm_add_ps(fix3,tx);
 288             fiy3             = _mm_add_ps(fiy3,ty);
 289             fiz3             = _mm_add_ps(fiz3,tz);
 290
 291             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 292                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 293                                                    tx,ty,tz);
 294
 295             /* Inner loop uses 96 flops */
 296         }
 297
 298         if(jidx<j_index_end)
 299         {
 300
 301             /* Get j neighbor index, and coordinate index */
 302             jnrA             = jjnr[jidx];
 303             jnrB             = jjnr[jidx+1];
 304             jnrC             = jjnr[jidx+2];
 305             jnrD             = jjnr[jidx+3];
 306
 307             /* Sign of each element will be negative for non-real atoms.
 308              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 309              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 310              */
 311             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 312             jnrA       = (jnrA>=0) ? jnrA : 0;
 313             jnrB       = (jnrB>=0) ? jnrB : 0;
 314             jnrC       = (jnrC>=0) ? jnrC : 0;
 315             jnrD       = (jnrD>=0) ? jnrD : 0;
 316
 317             j_coord_offsetA  = DIM*jnrA;
 318             j_coord_offsetB  = DIM*jnrB;
 319             j_coord_offsetC  = DIM*jnrC;
 320             j_coord_offsetD  = DIM*jnrD;
 321
 322             /* load j atom coordinates */
 323             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 324                                               x+j_coord_offsetC,x+j_coord_offsetD,
 325                                               &jx0,&jy0,&jz0);
 326
 327             /* Calculate displacement vector */
 328             dx10             = _mm_sub_ps(ix1,jx0);
 329             dy10             = _mm_sub_ps(iy1,jy0);
 330             dz10             = _mm_sub_ps(iz1,jz0);
 331             dx20             = _mm_sub_ps(ix2,jx0);
 332             dy20             = _mm_sub_ps(iy2,jy0);
 333             dz20             = _mm_sub_ps(iz2,jz0);
 334             dx30             = _mm_sub_ps(ix3,jx0);
 335             dy30             = _mm_sub_ps(iy3,jy0);
 336             dz30             = _mm_sub_ps(iz3,jz0);
 337
 338             /* Calculate squared distance and things based on it */
 339             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 340             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 341             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 342
 343             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 344             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 345             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 346
 347             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 348             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 349             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 350
 351             /* Load parameters for j particles */
 352             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 353                                                               charge+jnrC+0,charge+jnrD+0);
 354
 355             /**************************
 356              * CALCULATE INTERACTIONS *
 357              **************************/
 358
 359             /* Compute parameters for interactions between i and j atoms */
 360             qq10             = _mm_mul_ps(iq1,jq0);
 361
 362             /* REACTION-FIELD ELECTROSTATICS */
 363             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 364             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 365
 366             /* Update potential sum for this i atom from the interaction with this j atom. */
 367             velec            = _mm_andnot_ps(dummy_mask,velec);
 368             velecsum         = _mm_add_ps(velecsum,velec);
 369
 370             fscal            = felec;
 371
 372             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 373
 374             /* Calculate temporary vectorial force */
 375             tx               = _mm_mul_ps(fscal,dx10);
 376             ty               = _mm_mul_ps(fscal,dy10);
 377             tz               = _mm_mul_ps(fscal,dz10);
 378
 379             /* Update vectorial force */
 380             fix1             = _mm_add_ps(fix1,tx);
 381             fiy1             = _mm_add_ps(fiy1,ty);
 382             fiz1             = _mm_add_ps(fiz1,tz);
 383
 384             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 385                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 386                                                    tx,ty,tz);
 387
 388             /**************************
 389              * CALCULATE INTERACTIONS *
 390              **************************/
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq20             = _mm_mul_ps(iq2,jq0);
 394
 395             /* REACTION-FIELD ELECTROSTATICS */
 396             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 397             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 398
 399             /* Update potential sum for this i atom from the interaction with this j atom. */
 400             velec            = _mm_andnot_ps(dummy_mask,velec);
 401             velecsum         = _mm_add_ps(velecsum,velec);
 402
 403             fscal            = felec;
 404
 405             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 406
 407             /* Calculate temporary vectorial force */
 408             tx               = _mm_mul_ps(fscal,dx20);
 409             ty               = _mm_mul_ps(fscal,dy20);
 410             tz               = _mm_mul_ps(fscal,dz20);
 411
 412             /* Update vectorial force */
 413             fix2             = _mm_add_ps(fix2,tx);
 414             fiy2             = _mm_add_ps(fiy2,ty);
 415             fiz2             = _mm_add_ps(fiz2,tz);
 416
 417             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 418                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 419                                                    tx,ty,tz);
 420
 421             /**************************
 422              * CALCULATE INTERACTIONS *
 423              **************************/
 424
 425             /* Compute parameters for interactions between i and j atoms */
 426             qq30             = _mm_mul_ps(iq3,jq0);
 427
 428             /* REACTION-FIELD ELECTROSTATICS */
 429             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 430             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 431
 432             /* Update potential sum for this i atom from the interaction with this j atom. */
 433             velec            = _mm_andnot_ps(dummy_mask,velec);
 434             velecsum         = _mm_add_ps(velecsum,velec);
 435
 436             fscal            = felec;
 437
 438             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 439
 440             /* Calculate temporary vectorial force */
 441             tx               = _mm_mul_ps(fscal,dx30);
 442             ty               = _mm_mul_ps(fscal,dy30);
 443             tz               = _mm_mul_ps(fscal,dz30);
 444
 445             /* Update vectorial force */
 446             fix3             = _mm_add_ps(fix3,tx);
 447             fiy3             = _mm_add_ps(fiy3,ty);
 448             fiz3             = _mm_add_ps(fiz3,tz);
 449
 450             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 451                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 452                                                    tx,ty,tz);
 453
 454             /* Inner loop uses 96 flops */
 455         }
 456
 457         /* End of innermost loop */
 458
 459         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 460                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 461
 462         ggid                        = gid[iidx];
 463         /* Update potential energies */
 464         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 465
 466         /* Increment number of inner iterations */
 467         inneriter                  += j_index_end - j_index_start;
 468
 469         /* Outer loop uses 28 flops */
 470     }
 471
 472     /* Increment number of outer iterations */
 473     outeriter        += nri;
 474
 475     /* Update outer/inner flops */
 476
 477     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*28 + inneriter*96);
 478 }
 479 /*
 480  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_F_sse2_single
 481  * Electrostatics interaction: ReactionField
 482  * VdW interaction:            None
 483  * Geometry:                   Water4-Particle
 484  * Calculate force/pot:        Force
 485  */
 486 void
 487 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_sse2_single
 488                     (t_nblist * gmx_restrict                nlist,
 489                      rvec * gmx_restrict                    xx,
 490                      rvec * gmx_restrict                    ff,
 491                      t_forcerec * gmx_restrict              fr,
 492                      t_mdatoms * gmx_restrict               mdatoms,
 493                      nb_kernel_data_t * gmx_restrict        kernel_data,
 494                      t_nrnb * gmx_restrict                  nrnb)
 495 {
 496     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 497      * just 0 for non-waters.
 498      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 499      * jnr indices corresponding to data put in the four positions in the SIMD register.
 500      */
 501     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 502     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 503     int              jnrA,jnrB,jnrC,jnrD;
 504     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 505     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 506     real             shX,shY,shZ,rcutoff_scalar;
 507     real             *shiftvec,*fshift,*x,*f;
 508     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 509     int              vdwioffset1;
 510     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 511     int              vdwioffset2;
 512     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 513     int              vdwioffset3;
 514     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 515     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 516     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 517     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 518     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 519     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 520     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 521     real             *charge;
 522     __m128           dummy_mask,cutoff_mask;
 523     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 524     __m128           one     = _mm_set1_ps(1.0);
 525     __m128           two     = _mm_set1_ps(2.0);
 526     x                = xx[0];
 527     f                = ff[0];
 528
 529     nri              = nlist->nri;
 530     iinr             = nlist->iinr;
 531     jindex           = nlist->jindex;
 532     jjnr             = nlist->jjnr;
 533     shiftidx         = nlist->shift;
 534     gid              = nlist->gid;
 535     shiftvec         = fr->shift_vec[0];
 536     fshift           = fr->fshift[0];
 537     facel            = _mm_set1_ps(fr->epsfac);
 538     charge           = mdatoms->chargeA;
 539     krf              = _mm_set1_ps(fr->ic->k_rf);
 540     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 541     crf              = _mm_set1_ps(fr->ic->c_rf);
 542
 543     /* Setup water-specific parameters */
 544     inr              = nlist->iinr[0];
 545     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 546     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 547     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 548
 549     /* Avoid stupid compiler warnings */
 550     jnrA = jnrB = jnrC = jnrD = 0;
 551     j_coord_offsetA = 0;
 552     j_coord_offsetB = 0;
 553     j_coord_offsetC = 0;
 554     j_coord_offsetD = 0;
 555
 556     outeriter        = 0;
 557     inneriter        = 0;
 558
 559     /* Start outer loop over neighborlists */
 560     for(iidx=0; iidx<nri; iidx++)
 561     {
 562         /* Load shift vector for this list */
 563         i_shift_offset   = DIM*shiftidx[iidx];
 564         shX              = shiftvec[i_shift_offset+XX];
 565         shY              = shiftvec[i_shift_offset+YY];
 566         shZ              = shiftvec[i_shift_offset+ZZ];
 567
 568         /* Load limits for loop over neighbors */
 569         j_index_start    = jindex[iidx];
 570         j_index_end      = jindex[iidx+1];
 571
 572         /* Get outer coordinate index */
 573         inr              = iinr[iidx];
 574         i_coord_offset   = DIM*inr;
 575
 576         /* Load i particle coords and add shift vector */
 577         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 578         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 579         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 580         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 581         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 582         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 583         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 584         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 585         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 586
 587         fix1             = _mm_setzero_ps();
 588         fiy1             = _mm_setzero_ps();
 589         fiz1             = _mm_setzero_ps();
 590         fix2             = _mm_setzero_ps();
 591         fiy2             = _mm_setzero_ps();
 592         fiz2             = _mm_setzero_ps();
 593         fix3             = _mm_setzero_ps();
 594         fiy3             = _mm_setzero_ps();
 595         fiz3             = _mm_setzero_ps();
 596
 597         /* Start inner kernel loop */
 598         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 599         {
 600
 601             /* Get j neighbor index, and coordinate index */
 602             jnrA             = jjnr[jidx];
 603             jnrB             = jjnr[jidx+1];
 604             jnrC             = jjnr[jidx+2];
 605             jnrD             = jjnr[jidx+3];
 606
 607             j_coord_offsetA  = DIM*jnrA;
 608             j_coord_offsetB  = DIM*jnrB;
 609             j_coord_offsetC  = DIM*jnrC;
 610             j_coord_offsetD  = DIM*jnrD;
 611
 612             /* load j atom coordinates */
 613             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 614                                               x+j_coord_offsetC,x+j_coord_offsetD,
 615                                               &jx0,&jy0,&jz0);
 616
 617             /* Calculate displacement vector */
 618             dx10             = _mm_sub_ps(ix1,jx0);
 619             dy10             = _mm_sub_ps(iy1,jy0);
 620             dz10             = _mm_sub_ps(iz1,jz0);
 621             dx20             = _mm_sub_ps(ix2,jx0);
 622             dy20             = _mm_sub_ps(iy2,jy0);
 623             dz20             = _mm_sub_ps(iz2,jz0);
 624             dx30             = _mm_sub_ps(ix3,jx0);
 625             dy30             = _mm_sub_ps(iy3,jy0);
 626             dz30             = _mm_sub_ps(iz3,jz0);
 627
 628             /* Calculate squared distance and things based on it */
 629             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 630             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 631             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 632
 633             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 634             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 635             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 636
 637             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 638             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 639             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 640
 641             /* Load parameters for j particles */
 642             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 643                                                               charge+jnrC+0,charge+jnrD+0);
 644
 645             /**************************
 646              * CALCULATE INTERACTIONS *
 647              **************************/
 648
 649             /* Compute parameters for interactions between i and j atoms */
 650             qq10             = _mm_mul_ps(iq1,jq0);
 651
 652             /* REACTION-FIELD ELECTROSTATICS */
 653             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 654
 655             fscal            = felec;
 656
 657             /* Calculate temporary vectorial force */
 658             tx               = _mm_mul_ps(fscal,dx10);
 659             ty               = _mm_mul_ps(fscal,dy10);
 660             tz               = _mm_mul_ps(fscal,dz10);
 661
 662             /* Update vectorial force */
 663             fix1             = _mm_add_ps(fix1,tx);
 664             fiy1             = _mm_add_ps(fiy1,ty);
 665             fiz1             = _mm_add_ps(fiz1,tz);
 666
 667             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 668                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 669                                                    tx,ty,tz);
 670
 671             /**************************
 672              * CALCULATE INTERACTIONS *
 673              **************************/
 674
 675             /* Compute parameters for interactions between i and j atoms */
 676             qq20             = _mm_mul_ps(iq2,jq0);
 677
 678             /* REACTION-FIELD ELECTROSTATICS */
 679             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 680
 681             fscal            = felec;
 682
 683             /* Calculate temporary vectorial force */
 684             tx               = _mm_mul_ps(fscal,dx20);
 685             ty               = _mm_mul_ps(fscal,dy20);
 686             tz               = _mm_mul_ps(fscal,dz20);
 687
 688             /* Update vectorial force */
 689             fix2             = _mm_add_ps(fix2,tx);
 690             fiy2             = _mm_add_ps(fiy2,ty);
 691             fiz2             = _mm_add_ps(fiz2,tz);
 692
 693             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 694                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 695                                                    tx,ty,tz);
 696
 697             /**************************
 698              * CALCULATE INTERACTIONS *
 699              **************************/
 700
 701             /* Compute parameters for interactions between i and j atoms */
 702             qq30             = _mm_mul_ps(iq3,jq0);
 703
 704             /* REACTION-FIELD ELECTROSTATICS */
 705             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 706
 707             fscal            = felec;
 708
 709             /* Calculate temporary vectorial force */
 710             tx               = _mm_mul_ps(fscal,dx30);
 711             ty               = _mm_mul_ps(fscal,dy30);
 712             tz               = _mm_mul_ps(fscal,dz30);
 713
 714             /* Update vectorial force */
 715             fix3             = _mm_add_ps(fix3,tx);
 716             fiy3             = _mm_add_ps(fiy3,ty);
 717             fiz3             = _mm_add_ps(fiz3,tz);
 718
 719             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 720                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 721                                                    tx,ty,tz);
 722
 723             /* Inner loop uses 81 flops */
 724         }
 725
 726         if(jidx<j_index_end)
 727         {
 728
 729             /* Get j neighbor index, and coordinate index */
 730             jnrA             = jjnr[jidx];
 731             jnrB             = jjnr[jidx+1];
 732             jnrC             = jjnr[jidx+2];
 733             jnrD             = jjnr[jidx+3];
 734
 735             /* Sign of each element will be negative for non-real atoms.
 736              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 737              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 738              */
 739             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 740             jnrA       = (jnrA>=0) ? jnrA : 0;
 741             jnrB       = (jnrB>=0) ? jnrB : 0;
 742             jnrC       = (jnrC>=0) ? jnrC : 0;
 743             jnrD       = (jnrD>=0) ? jnrD : 0;
 744
 745             j_coord_offsetA  = DIM*jnrA;
 746             j_coord_offsetB  = DIM*jnrB;
 747             j_coord_offsetC  = DIM*jnrC;
 748             j_coord_offsetD  = DIM*jnrD;
 749
 750             /* load j atom coordinates */
 751             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 752                                               x+j_coord_offsetC,x+j_coord_offsetD,
 753                                               &jx0,&jy0,&jz0);
 754
 755             /* Calculate displacement vector */
 756             dx10             = _mm_sub_ps(ix1,jx0);
 757             dy10             = _mm_sub_ps(iy1,jy0);
 758             dz10             = _mm_sub_ps(iz1,jz0);
 759             dx20             = _mm_sub_ps(ix2,jx0);
 760             dy20             = _mm_sub_ps(iy2,jy0);
 761             dz20             = _mm_sub_ps(iz2,jz0);
 762             dx30             = _mm_sub_ps(ix3,jx0);
 763             dy30             = _mm_sub_ps(iy3,jy0);
 764             dz30             = _mm_sub_ps(iz3,jz0);
 765
 766             /* Calculate squared distance and things based on it */
 767             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 768             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 769             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 770
 771             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 772             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 773             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 774
 775             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 776             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 777             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 778
 779             /* Load parameters for j particles */
 780             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 781                                                               charge+jnrC+0,charge+jnrD+0);
 782
 783             /**************************
 784              * CALCULATE INTERACTIONS *
 785              **************************/
 786
 787             /* Compute parameters for interactions between i and j atoms */
 788             qq10             = _mm_mul_ps(iq1,jq0);
 789
 790             /* REACTION-FIELD ELECTROSTATICS */
 791             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 792
 793             fscal            = felec;
 794
 795             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 796
 797             /* Calculate temporary vectorial force */
 798             tx               = _mm_mul_ps(fscal,dx10);
 799             ty               = _mm_mul_ps(fscal,dy10);
 800             tz               = _mm_mul_ps(fscal,dz10);
 801
 802             /* Update vectorial force */
 803             fix1             = _mm_add_ps(fix1,tx);
 804             fiy1             = _mm_add_ps(fiy1,ty);
 805             fiz1             = _mm_add_ps(fiz1,tz);
 806
 807             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 808                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 809                                                    tx,ty,tz);
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             /* Compute parameters for interactions between i and j atoms */
 816             qq20             = _mm_mul_ps(iq2,jq0);
 817
 818             /* REACTION-FIELD ELECTROSTATICS */
 819             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 820
 821             fscal            = felec;
 822
 823             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 824
 825             /* Calculate temporary vectorial force */
 826             tx               = _mm_mul_ps(fscal,dx20);
 827             ty               = _mm_mul_ps(fscal,dy20);
 828             tz               = _mm_mul_ps(fscal,dz20);
 829
 830             /* Update vectorial force */
 831             fix2             = _mm_add_ps(fix2,tx);
 832             fiy2             = _mm_add_ps(fiy2,ty);
 833             fiz2             = _mm_add_ps(fiz2,tz);
 834
 835             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 836                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 837                                                    tx,ty,tz);
 838
 839             /**************************
 840              * CALCULATE INTERACTIONS *
 841              **************************/
 842
 843             /* Compute parameters for interactions between i and j atoms */
 844             qq30             = _mm_mul_ps(iq3,jq0);
 845
 846             /* REACTION-FIELD ELECTROSTATICS */
 847             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 848
 849             fscal            = felec;
 850
 851             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 852
 853             /* Calculate temporary vectorial force */
 854             tx               = _mm_mul_ps(fscal,dx30);
 855             ty               = _mm_mul_ps(fscal,dy30);
 856             tz               = _mm_mul_ps(fscal,dz30);
 857
 858             /* Update vectorial force */
 859             fix3             = _mm_add_ps(fix3,tx);
 860             fiy3             = _mm_add_ps(fiy3,ty);
 861             fiz3             = _mm_add_ps(fiz3,tz);
 862
 863             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 864                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 865                                                    tx,ty,tz);
 866
 867             /* Inner loop uses 81 flops */
 868         }
 869
 870         /* End of innermost loop */
 871
 872         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 873                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 874
 875         /* Increment number of inner iterations */
 876         inneriter                  += j_index_end - j_index_start;
 877
 878         /* Outer loop uses 27 flops */
 879     }
 880
 881     /* Increment number of outer iterations */
 882     outeriter        += nri;
 883
 884     /* Update outer/inner flops */
 885
 886     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*27 + inneriter*81);
 887 }