src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwNone_GeomW3P1_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_ElecRFCut_VdwNone_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_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              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  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     krf              = _mm_set1_ps(fr->ic->k_rf);
 100     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 101     crf              = _mm_set1_ps(fr->ic->c_rf);
 102
 103     /* Setup water-specific parameters */
 104     inr              = nlist->iinr[0];
 105     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 106     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 107     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 108
 109     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 110     rcutoff_scalar   = fr->rcoulomb;
 111     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 112     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 113
 114     /* Avoid stupid compiler warnings */
 115     jnrA = jnrB = jnrC = jnrD = 0;
 116     j_coord_offsetA = 0;
 117     j_coord_offsetB = 0;
 118     j_coord_offsetC = 0;
 119     j_coord_offsetD = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     for(iidx=0;iidx<4*DIM;iidx++)
 125     {
 126         scratch[iidx] = 0.0;
 127     }
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 145                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 146
 147         fix0             = _mm_setzero_ps();
 148         fiy0             = _mm_setzero_ps();
 149         fiz0             = _mm_setzero_ps();
 150         fix1             = _mm_setzero_ps();
 151         fiy1             = _mm_setzero_ps();
 152         fiz1             = _mm_setzero_ps();
 153         fix2             = _mm_setzero_ps();
 154         fiy2             = _mm_setzero_ps();
 155         fiz2             = _mm_setzero_ps();
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm_setzero_ps();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             jnrC             = jjnr[jidx+2];
 168             jnrD             = jjnr[jidx+3];
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171             j_coord_offsetC  = DIM*jnrC;
 172             j_coord_offsetD  = DIM*jnrD;
 173
 174             /* load j atom coordinates */
 175             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                               x+j_coord_offsetC,x+j_coord_offsetD,
 177                                               &jx0,&jy0,&jz0);
 178
 179             /* Calculate displacement vector */
 180             dx00             = _mm_sub_ps(ix0,jx0);
 181             dy00             = _mm_sub_ps(iy0,jy0);
 182             dz00             = _mm_sub_ps(iz0,jz0);
 183             dx10             = _mm_sub_ps(ix1,jx0);
 184             dy10             = _mm_sub_ps(iy1,jy0);
 185             dz10             = _mm_sub_ps(iz1,jz0);
 186             dx20             = _mm_sub_ps(ix2,jx0);
 187             dy20             = _mm_sub_ps(iy2,jy0);
 188             dz20             = _mm_sub_ps(iz2,jz0);
 189
 190             /* Calculate squared distance and things based on it */
 191             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 192             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 193             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 194
 195             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 196             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 197             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 198
 199             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 200             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 201             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 202
 203             /* Load parameters for j particles */
 204             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 205                                                               charge+jnrC+0,charge+jnrD+0);
 206
 207             fjx0             = _mm_setzero_ps();
 208             fjy0             = _mm_setzero_ps();
 209             fjz0             = _mm_setzero_ps();
 210
 211             /**************************
 212              * CALCULATE INTERACTIONS *
 213              **************************/
 214
 215             if (gmx_mm_any_lt(rsq00,rcutoff2))
 216             {
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq00             = _mm_mul_ps(iq0,jq0);
 220
 221             /* REACTION-FIELD ELECTROSTATICS */
 222             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 223             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 224
 225             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             velec            = _mm_and_ps(velec,cutoff_mask);
 229             velecsum         = _mm_add_ps(velecsum,velec);
 230
 231             fscal            = felec;
 232
 233             fscal            = _mm_and_ps(fscal,cutoff_mask);
 234
 235              /* Update vectorial force */
 236             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 237             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 238             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 239
 240             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 241             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 242             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 243
 244             }
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             if (gmx_mm_any_lt(rsq10,rcutoff2))
 251             {
 252
 253             /* Compute parameters for interactions between i and j atoms */
 254             qq10             = _mm_mul_ps(iq1,jq0);
 255
 256             /* REACTION-FIELD ELECTROSTATICS */
 257             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 258             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 259
 260             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             velec            = _mm_and_ps(velec,cutoff_mask);
 264             velecsum         = _mm_add_ps(velecsum,velec);
 265
 266             fscal            = felec;
 267
 268             fscal            = _mm_and_ps(fscal,cutoff_mask);
 269
 270              /* Update vectorial force */
 271             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 272             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 273             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 274
 275             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 276             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 277             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 278
 279             }
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             if (gmx_mm_any_lt(rsq20,rcutoff2))
 286             {
 287
 288             /* Compute parameters for interactions between i and j atoms */
 289             qq20             = _mm_mul_ps(iq2,jq0);
 290
 291             /* REACTION-FIELD ELECTROSTATICS */
 292             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 293             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 294
 295             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velec            = _mm_and_ps(velec,cutoff_mask);
 299             velecsum         = _mm_add_ps(velecsum,velec);
 300
 301             fscal            = felec;
 302
 303             fscal            = _mm_and_ps(fscal,cutoff_mask);
 304
 305              /* Update vectorial force */
 306             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 307             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 308             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 309
 310             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 311             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 312             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 313
 314             }
 315
 316             fjptrA             = f+j_coord_offsetA;
 317             fjptrB             = f+j_coord_offsetB;
 318             fjptrC             = f+j_coord_offsetC;
 319             fjptrD             = f+j_coord_offsetD;
 320
 321             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 322
 323             /* Inner loop uses 117 flops */
 324         }
 325
 326         if(jidx<j_index_end)
 327         {
 328
 329             /* Get j neighbor index, and coordinate index */
 330             jnrlistA         = jjnr[jidx];
 331             jnrlistB         = jjnr[jidx+1];
 332             jnrlistC         = jjnr[jidx+2];
 333             jnrlistD         = jjnr[jidx+3];
 334             /* Sign of each element will be negative for non-real atoms.
 335              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 336              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 337              */
 338             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 339             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 340             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 341             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 342             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 343             j_coord_offsetA  = DIM*jnrA;
 344             j_coord_offsetB  = DIM*jnrB;
 345             j_coord_offsetC  = DIM*jnrC;
 346             j_coord_offsetD  = DIM*jnrD;
 347
 348             /* load j atom coordinates */
 349             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 350                                               x+j_coord_offsetC,x+j_coord_offsetD,
 351                                               &jx0,&jy0,&jz0);
 352
 353             /* Calculate displacement vector */
 354             dx00             = _mm_sub_ps(ix0,jx0);
 355             dy00             = _mm_sub_ps(iy0,jy0);
 356             dz00             = _mm_sub_ps(iz0,jz0);
 357             dx10             = _mm_sub_ps(ix1,jx0);
 358             dy10             = _mm_sub_ps(iy1,jy0);
 359             dz10             = _mm_sub_ps(iz1,jz0);
 360             dx20             = _mm_sub_ps(ix2,jx0);
 361             dy20             = _mm_sub_ps(iy2,jy0);
 362             dz20             = _mm_sub_ps(iz2,jz0);
 363
 364             /* Calculate squared distance and things based on it */
 365             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 366             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 367             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 368
 369             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 370             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 371             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 372
 373             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 374             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 375             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 376
 377             /* Load parameters for j particles */
 378             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 379                                                               charge+jnrC+0,charge+jnrD+0);
 380
 381             fjx0             = _mm_setzero_ps();
 382             fjy0             = _mm_setzero_ps();
 383             fjz0             = _mm_setzero_ps();
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             if (gmx_mm_any_lt(rsq00,rcutoff2))
 390             {
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq00             = _mm_mul_ps(iq0,jq0);
 394
 395             /* REACTION-FIELD ELECTROSTATICS */
 396             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 397             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 398
 399             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 400
 401             /* Update potential sum for this i atom from the interaction with this j atom. */
 402             velec            = _mm_and_ps(velec,cutoff_mask);
 403             velec            = _mm_andnot_ps(dummy_mask,velec);
 404             velecsum         = _mm_add_ps(velecsum,velec);
 405
 406             fscal            = felec;
 407
 408             fscal            = _mm_and_ps(fscal,cutoff_mask);
 409
 410             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 411
 412              /* Update vectorial force */
 413             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 414             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 415             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 416
 417             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 418             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 419             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 420
 421             }
 422
 423             /**************************
 424              * CALCULATE INTERACTIONS *
 425              **************************/
 426
 427             if (gmx_mm_any_lt(rsq10,rcutoff2))
 428             {
 429
 430             /* Compute parameters for interactions between i and j atoms */
 431             qq10             = _mm_mul_ps(iq1,jq0);
 432
 433             /* REACTION-FIELD ELECTROSTATICS */
 434             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 435             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 436
 437             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 438
 439             /* Update potential sum for this i atom from the interaction with this j atom. */
 440             velec            = _mm_and_ps(velec,cutoff_mask);
 441             velec            = _mm_andnot_ps(dummy_mask,velec);
 442             velecsum         = _mm_add_ps(velecsum,velec);
 443
 444             fscal            = felec;
 445
 446             fscal            = _mm_and_ps(fscal,cutoff_mask);
 447
 448             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 449
 450              /* Update vectorial force */
 451             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 452             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 453             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 454
 455             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 456             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 457             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 458
 459             }
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             if (gmx_mm_any_lt(rsq20,rcutoff2))
 466             {
 467
 468             /* Compute parameters for interactions between i and j atoms */
 469             qq20             = _mm_mul_ps(iq2,jq0);
 470
 471             /* REACTION-FIELD ELECTROSTATICS */
 472             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 473             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 474
 475             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 476
 477             /* Update potential sum for this i atom from the interaction with this j atom. */
 478             velec            = _mm_and_ps(velec,cutoff_mask);
 479             velec            = _mm_andnot_ps(dummy_mask,velec);
 480             velecsum         = _mm_add_ps(velecsum,velec);
 481
 482             fscal            = felec;
 483
 484             fscal            = _mm_and_ps(fscal,cutoff_mask);
 485
 486             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 487
 488              /* Update vectorial force */
 489             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 490             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 491             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 492
 493             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 494             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 495             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 496
 497             }
 498
 499             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 500             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 501             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 502             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 503
 504             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 505
 506             /* Inner loop uses 117 flops */
 507         }
 508
 509         /* End of innermost loop */
 510
 511         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 512                                               f+i_coord_offset,fshift+i_shift_offset);
 513
 514         ggid                        = gid[iidx];
 515         /* Update potential energies */
 516         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 517
 518         /* Increment number of inner iterations */
 519         inneriter                  += j_index_end - j_index_start;
 520
 521         /* Outer loop uses 19 flops */
 522     }
 523
 524     /* Increment number of outer iterations */
 525     outeriter        += nri;
 526
 527     /* Update outer/inner flops */
 528
 529     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*117);
 530 }
 531 /*
 532  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_avx_128_fma_single
 533  * Electrostatics interaction: ReactionField
 534  * VdW interaction:            None
 535  * Geometry:                   Water3-Particle
 536  * Calculate force/pot:        Force
 537  */
 538 void
 539 nb_kernel_ElecRFCut_VdwNone_GeomW3P1_F_avx_128_fma_single
 540                     (t_nblist * gmx_restrict                nlist,
 541                      rvec * gmx_restrict                    xx,
 542                      rvec * gmx_restrict                    ff,
 543                      t_forcerec * gmx_restrict              fr,
 544                      t_mdatoms * gmx_restrict               mdatoms,
 545                      nb_kernel_data_t * gmx_restrict        kernel_data,
 546                      t_nrnb * gmx_restrict                  nrnb)
 547 {
 548     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 549      * just 0 for non-waters.
 550      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 551      * jnr indices corresponding to data put in the four positions in the SIMD register.
 552      */
 553     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 554     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 555     int              jnrA,jnrB,jnrC,jnrD;
 556     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 557     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 558     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 559     real             rcutoff_scalar;
 560     real             *shiftvec,*fshift,*x,*f;
 561     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 562     real             scratch[4*DIM];
 563     __m128           fscal,rcutoff,rcutoff2,jidxall;
 564     int              vdwioffset0;
 565     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 566     int              vdwioffset1;
 567     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 568     int              vdwioffset2;
 569     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 570     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 571     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 572     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 573     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 574     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 575     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 576     real             *charge;
 577     __m128           dummy_mask,cutoff_mask;
 578     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 579     __m128           one     = _mm_set1_ps(1.0);
 580     __m128           two     = _mm_set1_ps(2.0);
 581     x                = xx[0];
 582     f                = ff[0];
 583
 584     nri              = nlist->nri;
 585     iinr             = nlist->iinr;
 586     jindex           = nlist->jindex;
 587     jjnr             = nlist->jjnr;
 588     shiftidx         = nlist->shift;
 589     gid              = nlist->gid;
 590     shiftvec         = fr->shift_vec[0];
 591     fshift           = fr->fshift[0];
 592     facel            = _mm_set1_ps(fr->epsfac);
 593     charge           = mdatoms->chargeA;
 594     krf              = _mm_set1_ps(fr->ic->k_rf);
 595     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 596     crf              = _mm_set1_ps(fr->ic->c_rf);
 597
 598     /* Setup water-specific parameters */
 599     inr              = nlist->iinr[0];
 600     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 601     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 602     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 603
 604     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 605     rcutoff_scalar   = fr->rcoulomb;
 606     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 607     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 608
 609     /* Avoid stupid compiler warnings */
 610     jnrA = jnrB = jnrC = jnrD = 0;
 611     j_coord_offsetA = 0;
 612     j_coord_offsetB = 0;
 613     j_coord_offsetC = 0;
 614     j_coord_offsetD = 0;
 615
 616     outeriter        = 0;
 617     inneriter        = 0;
 618
 619     for(iidx=0;iidx<4*DIM;iidx++)
 620     {
 621         scratch[iidx] = 0.0;
 622     }
 623
 624     /* Start outer loop over neighborlists */
 625     for(iidx=0; iidx<nri; iidx++)
 626     {
 627         /* Load shift vector for this list */
 628         i_shift_offset   = DIM*shiftidx[iidx];
 629
 630         /* Load limits for loop over neighbors */
 631         j_index_start    = jindex[iidx];
 632         j_index_end      = jindex[iidx+1];
 633
 634         /* Get outer coordinate index */
 635         inr              = iinr[iidx];
 636         i_coord_offset   = DIM*inr;
 637
 638         /* Load i particle coords and add shift vector */
 639         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 640                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 641
 642         fix0             = _mm_setzero_ps();
 643         fiy0             = _mm_setzero_ps();
 644         fiz0             = _mm_setzero_ps();
 645         fix1             = _mm_setzero_ps();
 646         fiy1             = _mm_setzero_ps();
 647         fiz1             = _mm_setzero_ps();
 648         fix2             = _mm_setzero_ps();
 649         fiy2             = _mm_setzero_ps();
 650         fiz2             = _mm_setzero_ps();
 651
 652         /* Start inner kernel loop */
 653         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 654         {
 655
 656             /* Get j neighbor index, and coordinate index */
 657             jnrA             = jjnr[jidx];
 658             jnrB             = jjnr[jidx+1];
 659             jnrC             = jjnr[jidx+2];
 660             jnrD             = jjnr[jidx+3];
 661             j_coord_offsetA  = DIM*jnrA;
 662             j_coord_offsetB  = DIM*jnrB;
 663             j_coord_offsetC  = DIM*jnrC;
 664             j_coord_offsetD  = DIM*jnrD;
 665
 666             /* load j atom coordinates */
 667             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 668                                               x+j_coord_offsetC,x+j_coord_offsetD,
 669                                               &jx0,&jy0,&jz0);
 670
 671             /* Calculate displacement vector */
 672             dx00             = _mm_sub_ps(ix0,jx0);
 673             dy00             = _mm_sub_ps(iy0,jy0);
 674             dz00             = _mm_sub_ps(iz0,jz0);
 675             dx10             = _mm_sub_ps(ix1,jx0);
 676             dy10             = _mm_sub_ps(iy1,jy0);
 677             dz10             = _mm_sub_ps(iz1,jz0);
 678             dx20             = _mm_sub_ps(ix2,jx0);
 679             dy20             = _mm_sub_ps(iy2,jy0);
 680             dz20             = _mm_sub_ps(iz2,jz0);
 681
 682             /* Calculate squared distance and things based on it */
 683             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 684             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 685             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 686
 687             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 688             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 689             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 690
 691             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 692             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 693             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 694
 695             /* Load parameters for j particles */
 696             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 697                                                               charge+jnrC+0,charge+jnrD+0);
 698
 699             fjx0             = _mm_setzero_ps();
 700             fjy0             = _mm_setzero_ps();
 701             fjz0             = _mm_setzero_ps();
 702
 703             /**************************
 704              * CALCULATE INTERACTIONS *
 705              **************************/
 706
 707             if (gmx_mm_any_lt(rsq00,rcutoff2))
 708             {
 709
 710             /* Compute parameters for interactions between i and j atoms */
 711             qq00             = _mm_mul_ps(iq0,jq0);
 712
 713             /* REACTION-FIELD ELECTROSTATICS */
 714             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 715
 716             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 717
 718             fscal            = felec;
 719
 720             fscal            = _mm_and_ps(fscal,cutoff_mask);
 721
 722              /* Update vectorial force */
 723             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 724             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 725             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 726
 727             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 728             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 729             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 730
 731             }
 732
 733             /**************************
 734              * CALCULATE INTERACTIONS *
 735              **************************/
 736
 737             if (gmx_mm_any_lt(rsq10,rcutoff2))
 738             {
 739
 740             /* Compute parameters for interactions between i and j atoms */
 741             qq10             = _mm_mul_ps(iq1,jq0);
 742
 743             /* REACTION-FIELD ELECTROSTATICS */
 744             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 745
 746             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 747
 748             fscal            = felec;
 749
 750             fscal            = _mm_and_ps(fscal,cutoff_mask);
 751
 752              /* Update vectorial force */
 753             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 754             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 755             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 756
 757             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 758             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 759             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 760
 761             }
 762
 763             /**************************
 764              * CALCULATE INTERACTIONS *
 765              **************************/
 766
 767             if (gmx_mm_any_lt(rsq20,rcutoff2))
 768             {
 769
 770             /* Compute parameters for interactions between i and j atoms */
 771             qq20             = _mm_mul_ps(iq2,jq0);
 772
 773             /* REACTION-FIELD ELECTROSTATICS */
 774             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 775
 776             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 777
 778             fscal            = felec;
 779
 780             fscal            = _mm_and_ps(fscal,cutoff_mask);
 781
 782              /* Update vectorial force */
 783             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 784             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 785             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 786
 787             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 788             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 789             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 790
 791             }
 792
 793             fjptrA             = f+j_coord_offsetA;
 794             fjptrB             = f+j_coord_offsetB;
 795             fjptrC             = f+j_coord_offsetC;
 796             fjptrD             = f+j_coord_offsetD;
 797
 798             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 799
 800             /* Inner loop uses 99 flops */
 801         }
 802
 803         if(jidx<j_index_end)
 804         {
 805
 806             /* Get j neighbor index, and coordinate index */
 807             jnrlistA         = jjnr[jidx];
 808             jnrlistB         = jjnr[jidx+1];
 809             jnrlistC         = jjnr[jidx+2];
 810             jnrlistD         = jjnr[jidx+3];
 811             /* Sign of each element will be negative for non-real atoms.
 812              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 813              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 814              */
 815             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 816             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 817             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 818             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 819             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 820             j_coord_offsetA  = DIM*jnrA;
 821             j_coord_offsetB  = DIM*jnrB;
 822             j_coord_offsetC  = DIM*jnrC;
 823             j_coord_offsetD  = DIM*jnrD;
 824
 825             /* load j atom coordinates */
 826             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 827                                               x+j_coord_offsetC,x+j_coord_offsetD,
 828                                               &jx0,&jy0,&jz0);
 829
 830             /* Calculate displacement vector */
 831             dx00             = _mm_sub_ps(ix0,jx0);
 832             dy00             = _mm_sub_ps(iy0,jy0);
 833             dz00             = _mm_sub_ps(iz0,jz0);
 834             dx10             = _mm_sub_ps(ix1,jx0);
 835             dy10             = _mm_sub_ps(iy1,jy0);
 836             dz10             = _mm_sub_ps(iz1,jz0);
 837             dx20             = _mm_sub_ps(ix2,jx0);
 838             dy20             = _mm_sub_ps(iy2,jy0);
 839             dz20             = _mm_sub_ps(iz2,jz0);
 840
 841             /* Calculate squared distance and things based on it */
 842             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 843             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 844             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 845
 846             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 847             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 848             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 849
 850             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 851             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 852             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 853
 854             /* Load parameters for j particles */
 855             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 856                                                               charge+jnrC+0,charge+jnrD+0);
 857
 858             fjx0             = _mm_setzero_ps();
 859             fjy0             = _mm_setzero_ps();
 860             fjz0             = _mm_setzero_ps();
 861
 862             /**************************
 863              * CALCULATE INTERACTIONS *
 864              **************************/
 865
 866             if (gmx_mm_any_lt(rsq00,rcutoff2))
 867             {
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq00             = _mm_mul_ps(iq0,jq0);
 871
 872             /* REACTION-FIELD ELECTROSTATICS */
 873             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 874
 875             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 876
 877             fscal            = felec;
 878
 879             fscal            = _mm_and_ps(fscal,cutoff_mask);
 880
 881             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 882
 883              /* Update vectorial force */
 884             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 885             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 886             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 887
 888             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 889             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 890             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 891
 892             }
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             if (gmx_mm_any_lt(rsq10,rcutoff2))
 899             {
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             qq10             = _mm_mul_ps(iq1,jq0);
 903
 904             /* REACTION-FIELD ELECTROSTATICS */
 905             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 906
 907             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 908
 909             fscal            = felec;
 910
 911             fscal            = _mm_and_ps(fscal,cutoff_mask);
 912
 913             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 914
 915              /* Update vectorial force */
 916             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 917             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 918             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 919
 920             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 921             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 922             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 923
 924             }
 925
 926             /**************************
 927              * CALCULATE INTERACTIONS *
 928              **************************/
 929
 930             if (gmx_mm_any_lt(rsq20,rcutoff2))
 931             {
 932
 933             /* Compute parameters for interactions between i and j atoms */
 934             qq20             = _mm_mul_ps(iq2,jq0);
 935
 936             /* REACTION-FIELD ELECTROSTATICS */
 937             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 938
 939             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 940
 941             fscal            = felec;
 942
 943             fscal            = _mm_and_ps(fscal,cutoff_mask);
 944
 945             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 946
 947              /* Update vectorial force */
 948             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 949             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 950             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 951
 952             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 953             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 954             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 955
 956             }
 957
 958             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 959             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 960             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 961             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 962
 963             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 964
 965             /* Inner loop uses 99 flops */
 966         }
 967
 968         /* End of innermost loop */
 969
 970         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 971                                               f+i_coord_offset,fshift+i_shift_offset);
 972
 973         /* Increment number of inner iterations */
 974         inneriter                  += j_index_end - j_index_start;
 975
 976         /* Outer loop uses 18 flops */
 977     }
 978
 979     /* Increment number of outer iterations */
 980     outeriter        += nri;
 981
 982     /* Update outer/inner flops */
 983
 984     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*99);
 985 }