src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwNone_GeomW4P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_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_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_avx_128_fma_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 AVX_128, 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset1;
  70     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  71     int              vdwioffset2;
  72     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  73     int              vdwioffset3;
  74     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     __m128           dummy_mask,cutoff_mask;
  83     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  84     __m128           one     = _mm_set1_ps(1.0);
  85     __m128           two     = _mm_set1_ps(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     facel            = _mm_set1_ps(fr->epsfac);
  98     charge           = mdatoms->chargeA;
  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     for(iidx=0;iidx<4*DIM;iidx++)
 117     {
 118         scratch[iidx] = 0.0;
 119     }
 120
 121     /* Start outer loop over neighborlists */
 122     for(iidx=0; iidx<nri; iidx++)
 123     {
 124         /* Load shift vector for this list */
 125         i_shift_offset   = DIM*shiftidx[iidx];
 126
 127         /* Load limits for loop over neighbors */
 128         j_index_start    = jindex[iidx];
 129         j_index_end      = jindex[iidx+1];
 130
 131         /* Get outer coordinate index */
 132         inr              = iinr[iidx];
 133         i_coord_offset   = DIM*inr;
 134
 135         /* Load i particle coords and add shift vector */
 136         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 137                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 138
 139         fix1             = _mm_setzero_ps();
 140         fiy1             = _mm_setzero_ps();
 141         fiz1             = _mm_setzero_ps();
 142         fix2             = _mm_setzero_ps();
 143         fiy2             = _mm_setzero_ps();
 144         fiz2             = _mm_setzero_ps();
 145         fix3             = _mm_setzero_ps();
 146         fiy3             = _mm_setzero_ps();
 147         fiz3             = _mm_setzero_ps();
 148
 149         /* Reset potential sums */
 150         velecsum         = _mm_setzero_ps();
 151
 152         /* Start inner kernel loop */
 153         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 154         {
 155
 156             /* Get j neighbor index, and coordinate index */
 157             jnrA             = jjnr[jidx];
 158             jnrB             = jjnr[jidx+1];
 159             jnrC             = jjnr[jidx+2];
 160             jnrD             = jjnr[jidx+3];
 161             j_coord_offsetA  = DIM*jnrA;
 162             j_coord_offsetB  = DIM*jnrB;
 163             j_coord_offsetC  = DIM*jnrC;
 164             j_coord_offsetD  = DIM*jnrD;
 165
 166             /* load j atom coordinates */
 167             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 168                                               x+j_coord_offsetC,x+j_coord_offsetD,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx10             = _mm_sub_ps(ix1,jx0);
 173             dy10             = _mm_sub_ps(iy1,jy0);
 174             dz10             = _mm_sub_ps(iz1,jz0);
 175             dx20             = _mm_sub_ps(ix2,jx0);
 176             dy20             = _mm_sub_ps(iy2,jy0);
 177             dz20             = _mm_sub_ps(iz2,jz0);
 178             dx30             = _mm_sub_ps(ix3,jx0);
 179             dy30             = _mm_sub_ps(iy3,jy0);
 180             dz30             = _mm_sub_ps(iz3,jz0);
 181
 182             /* Calculate squared distance and things based on it */
 183             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 184             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 185             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 186
 187             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 188             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 189             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 190
 191             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 192             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 193             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 194
 195             /* Load parameters for j particles */
 196             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 197                                                               charge+jnrC+0,charge+jnrD+0);
 198
 199             fjx0             = _mm_setzero_ps();
 200             fjy0             = _mm_setzero_ps();
 201             fjz0             = _mm_setzero_ps();
 202
 203             /**************************
 204              * CALCULATE INTERACTIONS *
 205              **************************/
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq10             = _mm_mul_ps(iq1,jq0);
 209
 210             /* COULOMB ELECTROSTATICS */
 211             velec            = _mm_mul_ps(qq10,rinv10);
 212             felec            = _mm_mul_ps(velec,rinvsq10);
 213
 214             /* Update potential sum for this i atom from the interaction with this j atom. */
 215             velecsum         = _mm_add_ps(velecsum,velec);
 216
 217             fscal            = felec;
 218
 219              /* Update vectorial force */
 220             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 221             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 222             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 223
 224             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 225             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 226             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             qq20             = _mm_mul_ps(iq2,jq0);
 234
 235             /* COULOMB ELECTROSTATICS */
 236             velec            = _mm_mul_ps(qq20,rinv20);
 237             felec            = _mm_mul_ps(velec,rinvsq20);
 238
 239             /* Update potential sum for this i atom from the interaction with this j atom. */
 240             velecsum         = _mm_add_ps(velecsum,velec);
 241
 242             fscal            = felec;
 243
 244              /* Update vectorial force */
 245             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 246             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 247             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 248
 249             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 250             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 251             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             qq30             = _mm_mul_ps(iq3,jq0);
 259
 260             /* COULOMB ELECTROSTATICS */
 261             velec            = _mm_mul_ps(qq30,rinv30);
 262             felec            = _mm_mul_ps(velec,rinvsq30);
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             velecsum         = _mm_add_ps(velecsum,velec);
 266
 267             fscal            = felec;
 268
 269              /* Update vectorial force */
 270             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 271             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 272             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 273
 274             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 275             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 276             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 277
 278             fjptrA             = f+j_coord_offsetA;
 279             fjptrB             = f+j_coord_offsetB;
 280             fjptrC             = f+j_coord_offsetC;
 281             fjptrD             = f+j_coord_offsetD;
 282
 283             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 284
 285             /* Inner loop uses 93 flops */
 286         }
 287
 288         if(jidx<j_index_end)
 289         {
 290
 291             /* Get j neighbor index, and coordinate index */
 292             jnrlistA         = jjnr[jidx];
 293             jnrlistB         = jjnr[jidx+1];
 294             jnrlistC         = jjnr[jidx+2];
 295             jnrlistD         = jjnr[jidx+3];
 296             /* Sign of each element will be negative for non-real atoms.
 297              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 298              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 299              */
 300             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 301             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 302             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 303             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 304             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 305             j_coord_offsetA  = DIM*jnrA;
 306             j_coord_offsetB  = DIM*jnrB;
 307             j_coord_offsetC  = DIM*jnrC;
 308             j_coord_offsetD  = DIM*jnrD;
 309
 310             /* load j atom coordinates */
 311             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 312                                               x+j_coord_offsetC,x+j_coord_offsetD,
 313                                               &jx0,&jy0,&jz0);
 314
 315             /* Calculate displacement vector */
 316             dx10             = _mm_sub_ps(ix1,jx0);
 317             dy10             = _mm_sub_ps(iy1,jy0);
 318             dz10             = _mm_sub_ps(iz1,jz0);
 319             dx20             = _mm_sub_ps(ix2,jx0);
 320             dy20             = _mm_sub_ps(iy2,jy0);
 321             dz20             = _mm_sub_ps(iz2,jz0);
 322             dx30             = _mm_sub_ps(ix3,jx0);
 323             dy30             = _mm_sub_ps(iy3,jy0);
 324             dz30             = _mm_sub_ps(iz3,jz0);
 325
 326             /* Calculate squared distance and things based on it */
 327             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 328             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 329             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 330
 331             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 332             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 333             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 334
 335             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 336             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 337             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 338
 339             /* Load parameters for j particles */
 340             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 341                                                               charge+jnrC+0,charge+jnrD+0);
 342
 343             fjx0             = _mm_setzero_ps();
 344             fjy0             = _mm_setzero_ps();
 345             fjz0             = _mm_setzero_ps();
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq10             = _mm_mul_ps(iq1,jq0);
 353
 354             /* COULOMB ELECTROSTATICS */
 355             velec            = _mm_mul_ps(qq10,rinv10);
 356             felec            = _mm_mul_ps(velec,rinvsq10);
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velec            = _mm_andnot_ps(dummy_mask,velec);
 360             velecsum         = _mm_add_ps(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 365
 366              /* Update vectorial force */
 367             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 368             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 369             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 370
 371             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 372             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 373             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             /* Compute parameters for interactions between i and j atoms */
 380             qq20             = _mm_mul_ps(iq2,jq0);
 381
 382             /* COULOMB ELECTROSTATICS */
 383             velec            = _mm_mul_ps(qq20,rinv20);
 384             felec            = _mm_mul_ps(velec,rinvsq20);
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velec            = _mm_andnot_ps(dummy_mask,velec);
 388             velecsum         = _mm_add_ps(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 393
 394              /* Update vectorial force */
 395             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 396             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 397             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 398
 399             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 400             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 401             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 402
 403             /**************************
 404              * CALCULATE INTERACTIONS *
 405              **************************/
 406
 407             /* Compute parameters for interactions between i and j atoms */
 408             qq30             = _mm_mul_ps(iq3,jq0);
 409
 410             /* COULOMB ELECTROSTATICS */
 411             velec            = _mm_mul_ps(qq30,rinv30);
 412             felec            = _mm_mul_ps(velec,rinvsq30);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velec            = _mm_andnot_ps(dummy_mask,velec);
 416             velecsum         = _mm_add_ps(velecsum,velec);
 417
 418             fscal            = felec;
 419
 420             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 421
 422              /* Update vectorial force */
 423             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 424             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 425             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 426
 427             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 428             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 429             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 430
 431             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 432             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 433             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 434             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 435
 436             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 437
 438             /* Inner loop uses 93 flops */
 439         }
 440
 441         /* End of innermost loop */
 442
 443         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 444                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 445
 446         ggid                        = gid[iidx];
 447         /* Update potential energies */
 448         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 449
 450         /* Increment number of inner iterations */
 451         inneriter                  += j_index_end - j_index_start;
 452
 453         /* Outer loop uses 19 flops */
 454     }
 455
 456     /* Increment number of outer iterations */
 457     outeriter        += nri;
 458
 459     /* Update outer/inner flops */
 460
 461     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*93);
 462 }
 463 /*
 464  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_128_fma_single
 465  * Electrostatics interaction: Coulomb
 466  * VdW interaction:            None
 467  * Geometry:                   Water4-Particle
 468  * Calculate force/pot:        Force
 469  */
 470 void
 471 nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_avx_128_fma_single
 472                     (t_nblist * gmx_restrict                nlist,
 473                      rvec * gmx_restrict                    xx,
 474                      rvec * gmx_restrict                    ff,
 475                      t_forcerec * gmx_restrict              fr,
 476                      t_mdatoms * gmx_restrict               mdatoms,
 477                      nb_kernel_data_t * gmx_restrict        kernel_data,
 478                      t_nrnb * gmx_restrict                  nrnb)
 479 {
 480     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 481      * just 0 for non-waters.
 482      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 483      * jnr indices corresponding to data put in the four positions in the SIMD register.
 484      */
 485     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 486     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 487     int              jnrA,jnrB,jnrC,jnrD;
 488     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 489     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 490     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 491     real             rcutoff_scalar;
 492     real             *shiftvec,*fshift,*x,*f;
 493     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 494     real             scratch[4*DIM];
 495     __m128           fscal,rcutoff,rcutoff2,jidxall;
 496     int              vdwioffset1;
 497     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 498     int              vdwioffset2;
 499     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 500     int              vdwioffset3;
 501     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 502     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 503     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 504     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 505     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 506     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 507     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 508     real             *charge;
 509     __m128           dummy_mask,cutoff_mask;
 510     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 511     __m128           one     = _mm_set1_ps(1.0);
 512     __m128           two     = _mm_set1_ps(2.0);
 513     x                = xx[0];
 514     f                = ff[0];
 515
 516     nri              = nlist->nri;
 517     iinr             = nlist->iinr;
 518     jindex           = nlist->jindex;
 519     jjnr             = nlist->jjnr;
 520     shiftidx         = nlist->shift;
 521     gid              = nlist->gid;
 522     shiftvec         = fr->shift_vec[0];
 523     fshift           = fr->fshift[0];
 524     facel            = _mm_set1_ps(fr->epsfac);
 525     charge           = mdatoms->chargeA;
 526
 527     /* Setup water-specific parameters */
 528     inr              = nlist->iinr[0];
 529     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 530     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 531     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 532
 533     /* Avoid stupid compiler warnings */
 534     jnrA = jnrB = jnrC = jnrD = 0;
 535     j_coord_offsetA = 0;
 536     j_coord_offsetB = 0;
 537     j_coord_offsetC = 0;
 538     j_coord_offsetD = 0;
 539
 540     outeriter        = 0;
 541     inneriter        = 0;
 542
 543     for(iidx=0;iidx<4*DIM;iidx++)
 544     {
 545         scratch[iidx] = 0.0;
 546     }
 547
 548     /* Start outer loop over neighborlists */
 549     for(iidx=0; iidx<nri; iidx++)
 550     {
 551         /* Load shift vector for this list */
 552         i_shift_offset   = DIM*shiftidx[iidx];
 553
 554         /* Load limits for loop over neighbors */
 555         j_index_start    = jindex[iidx];
 556         j_index_end      = jindex[iidx+1];
 557
 558         /* Get outer coordinate index */
 559         inr              = iinr[iidx];
 560         i_coord_offset   = DIM*inr;
 561
 562         /* Load i particle coords and add shift vector */
 563         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 564                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 565
 566         fix1             = _mm_setzero_ps();
 567         fiy1             = _mm_setzero_ps();
 568         fiz1             = _mm_setzero_ps();
 569         fix2             = _mm_setzero_ps();
 570         fiy2             = _mm_setzero_ps();
 571         fiz2             = _mm_setzero_ps();
 572         fix3             = _mm_setzero_ps();
 573         fiy3             = _mm_setzero_ps();
 574         fiz3             = _mm_setzero_ps();
 575
 576         /* Start inner kernel loop */
 577         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 578         {
 579
 580             /* Get j neighbor index, and coordinate index */
 581             jnrA             = jjnr[jidx];
 582             jnrB             = jjnr[jidx+1];
 583             jnrC             = jjnr[jidx+2];
 584             jnrD             = jjnr[jidx+3];
 585             j_coord_offsetA  = DIM*jnrA;
 586             j_coord_offsetB  = DIM*jnrB;
 587             j_coord_offsetC  = DIM*jnrC;
 588             j_coord_offsetD  = DIM*jnrD;
 589
 590             /* load j atom coordinates */
 591             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 592                                               x+j_coord_offsetC,x+j_coord_offsetD,
 593                                               &jx0,&jy0,&jz0);
 594
 595             /* Calculate displacement vector */
 596             dx10             = _mm_sub_ps(ix1,jx0);
 597             dy10             = _mm_sub_ps(iy1,jy0);
 598             dz10             = _mm_sub_ps(iz1,jz0);
 599             dx20             = _mm_sub_ps(ix2,jx0);
 600             dy20             = _mm_sub_ps(iy2,jy0);
 601             dz20             = _mm_sub_ps(iz2,jz0);
 602             dx30             = _mm_sub_ps(ix3,jx0);
 603             dy30             = _mm_sub_ps(iy3,jy0);
 604             dz30             = _mm_sub_ps(iz3,jz0);
 605
 606             /* Calculate squared distance and things based on it */
 607             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 608             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 609             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 610
 611             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 612             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 613             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 614
 615             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 616             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 617             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 618
 619             /* Load parameters for j particles */
 620             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 621                                                               charge+jnrC+0,charge+jnrD+0);
 622
 623             fjx0             = _mm_setzero_ps();
 624             fjy0             = _mm_setzero_ps();
 625             fjz0             = _mm_setzero_ps();
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             /* Compute parameters for interactions between i and j atoms */
 632             qq10             = _mm_mul_ps(iq1,jq0);
 633
 634             /* COULOMB ELECTROSTATICS */
 635             velec            = _mm_mul_ps(qq10,rinv10);
 636             felec            = _mm_mul_ps(velec,rinvsq10);
 637
 638             fscal            = felec;
 639
 640              /* Update vectorial force */
 641             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 642             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 643             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 644
 645             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 646             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 647             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 648
 649             /**************************
 650              * CALCULATE INTERACTIONS *
 651              **************************/
 652
 653             /* Compute parameters for interactions between i and j atoms */
 654             qq20             = _mm_mul_ps(iq2,jq0);
 655
 656             /* COULOMB ELECTROSTATICS */
 657             velec            = _mm_mul_ps(qq20,rinv20);
 658             felec            = _mm_mul_ps(velec,rinvsq20);
 659
 660             fscal            = felec;
 661
 662              /* Update vectorial force */
 663             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 664             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 665             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 666
 667             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 668             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 669             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 670
 671             /**************************
 672              * CALCULATE INTERACTIONS *
 673              **************************/
 674
 675             /* Compute parameters for interactions between i and j atoms */
 676             qq30             = _mm_mul_ps(iq3,jq0);
 677
 678             /* COULOMB ELECTROSTATICS */
 679             velec            = _mm_mul_ps(qq30,rinv30);
 680             felec            = _mm_mul_ps(velec,rinvsq30);
 681
 682             fscal            = felec;
 683
 684              /* Update vectorial force */
 685             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 686             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 687             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 688
 689             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 690             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 691             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 692
 693             fjptrA             = f+j_coord_offsetA;
 694             fjptrB             = f+j_coord_offsetB;
 695             fjptrC             = f+j_coord_offsetC;
 696             fjptrD             = f+j_coord_offsetD;
 697
 698             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 699
 700             /* Inner loop uses 90 flops */
 701         }
 702
 703         if(jidx<j_index_end)
 704         {
 705
 706             /* Get j neighbor index, and coordinate index */
 707             jnrlistA         = jjnr[jidx];
 708             jnrlistB         = jjnr[jidx+1];
 709             jnrlistC         = jjnr[jidx+2];
 710             jnrlistD         = jjnr[jidx+3];
 711             /* Sign of each element will be negative for non-real atoms.
 712              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 713              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 714              */
 715             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 716             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 717             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 718             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 719             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 720             j_coord_offsetA  = DIM*jnrA;
 721             j_coord_offsetB  = DIM*jnrB;
 722             j_coord_offsetC  = DIM*jnrC;
 723             j_coord_offsetD  = DIM*jnrD;
 724
 725             /* load j atom coordinates */
 726             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 727                                               x+j_coord_offsetC,x+j_coord_offsetD,
 728                                               &jx0,&jy0,&jz0);
 729
 730             /* Calculate displacement vector */
 731             dx10             = _mm_sub_ps(ix1,jx0);
 732             dy10             = _mm_sub_ps(iy1,jy0);
 733             dz10             = _mm_sub_ps(iz1,jz0);
 734             dx20             = _mm_sub_ps(ix2,jx0);
 735             dy20             = _mm_sub_ps(iy2,jy0);
 736             dz20             = _mm_sub_ps(iz2,jz0);
 737             dx30             = _mm_sub_ps(ix3,jx0);
 738             dy30             = _mm_sub_ps(iy3,jy0);
 739             dz30             = _mm_sub_ps(iz3,jz0);
 740
 741             /* Calculate squared distance and things based on it */
 742             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 743             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 744             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 745
 746             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 747             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 748             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 749
 750             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 751             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 752             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 753
 754             /* Load parameters for j particles */
 755             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 756                                                               charge+jnrC+0,charge+jnrD+0);
 757
 758             fjx0             = _mm_setzero_ps();
 759             fjy0             = _mm_setzero_ps();
 760             fjz0             = _mm_setzero_ps();
 761
 762             /**************************
 763              * CALCULATE INTERACTIONS *
 764              **************************/
 765
 766             /* Compute parameters for interactions between i and j atoms */
 767             qq10             = _mm_mul_ps(iq1,jq0);
 768
 769             /* COULOMB ELECTROSTATICS */
 770             velec            = _mm_mul_ps(qq10,rinv10);
 771             felec            = _mm_mul_ps(velec,rinvsq10);
 772
 773             fscal            = felec;
 774
 775             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 776
 777              /* Update vectorial force */
 778             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 779             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 780             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 781
 782             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 783             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 784             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 785
 786             /**************************
 787              * CALCULATE INTERACTIONS *
 788              **************************/
 789
 790             /* Compute parameters for interactions between i and j atoms */
 791             qq20             = _mm_mul_ps(iq2,jq0);
 792
 793             /* COULOMB ELECTROSTATICS */
 794             velec            = _mm_mul_ps(qq20,rinv20);
 795             felec            = _mm_mul_ps(velec,rinvsq20);
 796
 797             fscal            = felec;
 798
 799             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 800
 801              /* Update vectorial force */
 802             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 803             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 804             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 805
 806             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 807             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 808             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 809
 810             /**************************
 811              * CALCULATE INTERACTIONS *
 812              **************************/
 813
 814             /* Compute parameters for interactions between i and j atoms */
 815             qq30             = _mm_mul_ps(iq3,jq0);
 816
 817             /* COULOMB ELECTROSTATICS */
 818             velec            = _mm_mul_ps(qq30,rinv30);
 819             felec            = _mm_mul_ps(velec,rinvsq30);
 820
 821             fscal            = felec;
 822
 823             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 824
 825              /* Update vectorial force */
 826             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 827             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 828             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 829
 830             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 831             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 832             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 833
 834             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 835             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 836             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 837             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 838
 839             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 840
 841             /* Inner loop uses 90 flops */
 842         }
 843
 844         /* End of innermost loop */
 845
 846         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 847                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 848
 849         /* Increment number of inner iterations */
 850         inneriter                  += j_index_end - j_index_start;
 851
 852         /* Outer loop uses 18 flops */
 853     }
 854
 855     /* Increment number of outer iterations */
 856     outeriter        += nri;
 857
 858     /* Update outer/inner flops */
 859
 860     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*90);
 861 }