src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwCSTab_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_ElecEw_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_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              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              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
  94     real             *vftab;
  95     __m128i          ewitab;
  96     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  98     real             *ewtab;
  99     __m128           dummy_mask,cutoff_mask;
 100     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 101     __m128           one     = _mm_set1_ps(1.0);
 102     __m128           two     = _mm_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_ps(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_vdw->data;
 121     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 122
 123     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 124     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 125     beta2            = _mm_mul_ps(beta,beta);
 126     beta3            = _mm_mul_ps(beta,beta2);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 136     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 170
 171         fix0             = _mm_setzero_ps();
 172         fiy0             = _mm_setzero_ps();
 173         fiz0             = _mm_setzero_ps();
 174         fix1             = _mm_setzero_ps();
 175         fiy1             = _mm_setzero_ps();
 176         fiz1             = _mm_setzero_ps();
 177         fix2             = _mm_setzero_ps();
 178         fiy2             = _mm_setzero_ps();
 179         fiz2             = _mm_setzero_ps();
 180         fix3             = _mm_setzero_ps();
 181         fiy3             = _mm_setzero_ps();
 182         fiz3             = _mm_setzero_ps();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_ps();
 186         vvdwsum          = _mm_setzero_ps();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                               x+j_coord_offsetC,x+j_coord_offsetD,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_ps(ix0,jx0);
 209             dy00             = _mm_sub_ps(iy0,jy0);
 210             dz00             = _mm_sub_ps(iz0,jz0);
 211             dx10             = _mm_sub_ps(ix1,jx0);
 212             dy10             = _mm_sub_ps(iy1,jy0);
 213             dz10             = _mm_sub_ps(iz1,jz0);
 214             dx20             = _mm_sub_ps(ix2,jx0);
 215             dy20             = _mm_sub_ps(iy2,jy0);
 216             dz20             = _mm_sub_ps(iz2,jz0);
 217             dx30             = _mm_sub_ps(ix3,jx0);
 218             dy30             = _mm_sub_ps(iy3,jy0);
 219             dz30             = _mm_sub_ps(iz3,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 223             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 224             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 225             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 226
 227             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 228             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 229             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 230             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 231
 232             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 233             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 234             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 235
 236             /* Load parameters for j particles */
 237             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 238                                                               charge+jnrC+0,charge+jnrD+0);
 239             vdwjidx0A        = 2*vdwtype[jnrA+0];
 240             vdwjidx0B        = 2*vdwtype[jnrB+0];
 241             vdwjidx0C        = 2*vdwtype[jnrC+0];
 242             vdwjidx0D        = 2*vdwtype[jnrD+0];
 243
 244             fjx0             = _mm_setzero_ps();
 245             fjy0             = _mm_setzero_ps();
 246             fjz0             = _mm_setzero_ps();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             r00              = _mm_mul_ps(rsq00,rinv00);
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 256                                          vdwparam+vdwioffset0+vdwjidx0B,
 257                                          vdwparam+vdwioffset0+vdwjidx0C,
 258                                          vdwparam+vdwioffset0+vdwjidx0D,
 259                                          &c6_00,&c12_00);
 260
 261             /* Calculate table index by multiplying r with table scale and truncate to integer */
 262             rt               = _mm_mul_ps(r00,vftabscale);
 263             vfitab           = _mm_cvttps_epi32(rt);
 264 #ifdef __XOP__
 265             vfeps            = _mm_frcz_ps(rt);
 266 #else
 267             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 268 #endif
 269             twovfeps         = _mm_add_ps(vfeps,vfeps);
 270             vfitab           = _mm_slli_epi32(vfitab,3);
 271
 272             /* CUBIC SPLINE TABLE DISPERSION */
 273             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 274             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 275             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 276             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 277             _MM_TRANSPOSE4_PS(Y,F,G,H);
 278             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 279             VV               = _mm_macc_ps(vfeps,Fp,Y);
 280             vvdw6            = _mm_mul_ps(c6_00,VV);
 281             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 282             fvdw6            = _mm_mul_ps(c6_00,FF);
 283
 284             /* CUBIC SPLINE TABLE REPULSION */
 285             vfitab           = _mm_add_epi32(vfitab,ifour);
 286             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 287             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 288             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 289             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 290             _MM_TRANSPOSE4_PS(Y,F,G,H);
 291             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 292             VV               = _mm_macc_ps(vfeps,Fp,Y);
 293             vvdw12           = _mm_mul_ps(c12_00,VV);
 294             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 295             fvdw12           = _mm_mul_ps(c12_00,FF);
 296             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 297             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 301
 302             fscal            = fvdw;
 303
 304              /* Update vectorial force */
 305             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 306             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 307             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 308
 309             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 310             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 311             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             r10              = _mm_mul_ps(rsq10,rinv10);
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq10             = _mm_mul_ps(iq1,jq0);
 321
 322             /* EWALD ELECTROSTATICS */
 323
 324             /* Analytical PME correction */
 325             zeta2            = _mm_mul_ps(beta2,rsq10);
 326             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 327             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 328             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 329             felec            = _mm_mul_ps(qq10,felec);
 330             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 331             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 332             velec            = _mm_mul_ps(qq10,velec);
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velecsum         = _mm_add_ps(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339              /* Update vectorial force */
 340             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 341             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 342             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 343
 344             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 345             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 346             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             r20              = _mm_mul_ps(rsq20,rinv20);
 353
 354             /* Compute parameters for interactions between i and j atoms */
 355             qq20             = _mm_mul_ps(iq2,jq0);
 356
 357             /* EWALD ELECTROSTATICS */
 358
 359             /* Analytical PME correction */
 360             zeta2            = _mm_mul_ps(beta2,rsq20);
 361             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 362             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 363             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 364             felec            = _mm_mul_ps(qq20,felec);
 365             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 366             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 367             velec            = _mm_mul_ps(qq20,velec);
 368
 369             /* Update potential sum for this i atom from the interaction with this j atom. */
 370             velecsum         = _mm_add_ps(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374              /* Update vectorial force */
 375             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 376             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 377             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 378
 379             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 380             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 381             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 382
 383             /**************************
 384              * CALCULATE INTERACTIONS *
 385              **************************/
 386
 387             r30              = _mm_mul_ps(rsq30,rinv30);
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq30             = _mm_mul_ps(iq3,jq0);
 391
 392             /* EWALD ELECTROSTATICS */
 393
 394             /* Analytical PME correction */
 395             zeta2            = _mm_mul_ps(beta2,rsq30);
 396             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 397             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 398             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 399             felec            = _mm_mul_ps(qq30,felec);
 400             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 401             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 402             velec            = _mm_mul_ps(qq30,velec);
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velecsum         = _mm_add_ps(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409              /* Update vectorial force */
 410             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 411             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 412             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 413
 414             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 415             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 416             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 417
 418             fjptrA             = f+j_coord_offsetA;
 419             fjptrB             = f+j_coord_offsetB;
 420             fjptrC             = f+j_coord_offsetC;
 421             fjptrD             = f+j_coord_offsetD;
 422
 423             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 424
 425             /* Inner loop uses 146 flops */
 426         }
 427
 428         if(jidx<j_index_end)
 429         {
 430
 431             /* Get j neighbor index, and coordinate index */
 432             jnrlistA         = jjnr[jidx];
 433             jnrlistB         = jjnr[jidx+1];
 434             jnrlistC         = jjnr[jidx+2];
 435             jnrlistD         = jjnr[jidx+3];
 436             /* Sign of each element will be negative for non-real atoms.
 437              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 438              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 439              */
 440             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 441             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 442             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 443             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 444             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 445             j_coord_offsetA  = DIM*jnrA;
 446             j_coord_offsetB  = DIM*jnrB;
 447             j_coord_offsetC  = DIM*jnrC;
 448             j_coord_offsetD  = DIM*jnrD;
 449
 450             /* load j atom coordinates */
 451             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 452                                               x+j_coord_offsetC,x+j_coord_offsetD,
 453                                               &jx0,&jy0,&jz0);
 454
 455             /* Calculate displacement vector */
 456             dx00             = _mm_sub_ps(ix0,jx0);
 457             dy00             = _mm_sub_ps(iy0,jy0);
 458             dz00             = _mm_sub_ps(iz0,jz0);
 459             dx10             = _mm_sub_ps(ix1,jx0);
 460             dy10             = _mm_sub_ps(iy1,jy0);
 461             dz10             = _mm_sub_ps(iz1,jz0);
 462             dx20             = _mm_sub_ps(ix2,jx0);
 463             dy20             = _mm_sub_ps(iy2,jy0);
 464             dz20             = _mm_sub_ps(iz2,jz0);
 465             dx30             = _mm_sub_ps(ix3,jx0);
 466             dy30             = _mm_sub_ps(iy3,jy0);
 467             dz30             = _mm_sub_ps(iz3,jz0);
 468
 469             /* Calculate squared distance and things based on it */
 470             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 471             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 472             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 473             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 474
 475             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 476             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 477             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 478             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 479
 480             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 481             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 482             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 483
 484             /* Load parameters for j particles */
 485             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 486                                                               charge+jnrC+0,charge+jnrD+0);
 487             vdwjidx0A        = 2*vdwtype[jnrA+0];
 488             vdwjidx0B        = 2*vdwtype[jnrB+0];
 489             vdwjidx0C        = 2*vdwtype[jnrC+0];
 490             vdwjidx0D        = 2*vdwtype[jnrD+0];
 491
 492             fjx0             = _mm_setzero_ps();
 493             fjy0             = _mm_setzero_ps();
 494             fjz0             = _mm_setzero_ps();
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             r00              = _mm_mul_ps(rsq00,rinv00);
 501             r00              = _mm_andnot_ps(dummy_mask,r00);
 502
 503             /* Compute parameters for interactions between i and j atoms */
 504             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 505                                          vdwparam+vdwioffset0+vdwjidx0B,
 506                                          vdwparam+vdwioffset0+vdwjidx0C,
 507                                          vdwparam+vdwioffset0+vdwjidx0D,
 508                                          &c6_00,&c12_00);
 509
 510             /* Calculate table index by multiplying r with table scale and truncate to integer */
 511             rt               = _mm_mul_ps(r00,vftabscale);
 512             vfitab           = _mm_cvttps_epi32(rt);
 513 #ifdef __XOP__
 514             vfeps            = _mm_frcz_ps(rt);
 515 #else
 516             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 517 #endif
 518             twovfeps         = _mm_add_ps(vfeps,vfeps);
 519             vfitab           = _mm_slli_epi32(vfitab,3);
 520
 521             /* CUBIC SPLINE TABLE DISPERSION */
 522             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 523             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 524             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 525             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 526             _MM_TRANSPOSE4_PS(Y,F,G,H);
 527             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 528             VV               = _mm_macc_ps(vfeps,Fp,Y);
 529             vvdw6            = _mm_mul_ps(c6_00,VV);
 530             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 531             fvdw6            = _mm_mul_ps(c6_00,FF);
 532
 533             /* CUBIC SPLINE TABLE REPULSION */
 534             vfitab           = _mm_add_epi32(vfitab,ifour);
 535             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 536             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 537             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 538             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 539             _MM_TRANSPOSE4_PS(Y,F,G,H);
 540             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 541             VV               = _mm_macc_ps(vfeps,Fp,Y);
 542             vvdw12           = _mm_mul_ps(c12_00,VV);
 543             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 544             fvdw12           = _mm_mul_ps(c12_00,FF);
 545             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 546             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 547
 548             /* Update potential sum for this i atom from the interaction with this j atom. */
 549             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 550             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 551
 552             fscal            = fvdw;
 553
 554             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 555
 556              /* Update vectorial force */
 557             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 558             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 559             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 560
 561             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 562             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 563             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             r10              = _mm_mul_ps(rsq10,rinv10);
 570             r10              = _mm_andnot_ps(dummy_mask,r10);
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq10             = _mm_mul_ps(iq1,jq0);
 574
 575             /* EWALD ELECTROSTATICS */
 576
 577             /* Analytical PME correction */
 578             zeta2            = _mm_mul_ps(beta2,rsq10);
 579             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 580             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 581             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 582             felec            = _mm_mul_ps(qq10,felec);
 583             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 584             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 585             velec            = _mm_mul_ps(qq10,velec);
 586
 587             /* Update potential sum for this i atom from the interaction with this j atom. */
 588             velec            = _mm_andnot_ps(dummy_mask,velec);
 589             velecsum         = _mm_add_ps(velecsum,velec);
 590
 591             fscal            = felec;
 592
 593             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 594
 595              /* Update vectorial force */
 596             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 597             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 598             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 599
 600             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 601             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 602             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             r20              = _mm_mul_ps(rsq20,rinv20);
 609             r20              = _mm_andnot_ps(dummy_mask,r20);
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq20             = _mm_mul_ps(iq2,jq0);
 613
 614             /* EWALD ELECTROSTATICS */
 615
 616             /* Analytical PME correction */
 617             zeta2            = _mm_mul_ps(beta2,rsq20);
 618             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 619             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 620             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 621             felec            = _mm_mul_ps(qq20,felec);
 622             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 623             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 624             velec            = _mm_mul_ps(qq20,velec);
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm_andnot_ps(dummy_mask,velec);
 628             velecsum         = _mm_add_ps(velecsum,velec);
 629
 630             fscal            = felec;
 631
 632             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 633
 634              /* Update vectorial force */
 635             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 636             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 637             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 638
 639             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 640             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 641             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             r30              = _mm_mul_ps(rsq30,rinv30);
 648             r30              = _mm_andnot_ps(dummy_mask,r30);
 649
 650             /* Compute parameters for interactions between i and j atoms */
 651             qq30             = _mm_mul_ps(iq3,jq0);
 652
 653             /* EWALD ELECTROSTATICS */
 654
 655             /* Analytical PME correction */
 656             zeta2            = _mm_mul_ps(beta2,rsq30);
 657             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 658             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 659             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 660             felec            = _mm_mul_ps(qq30,felec);
 661             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 662             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 663             velec            = _mm_mul_ps(qq30,velec);
 664
 665             /* Update potential sum for this i atom from the interaction with this j atom. */
 666             velec            = _mm_andnot_ps(dummy_mask,velec);
 667             velecsum         = _mm_add_ps(velecsum,velec);
 668
 669             fscal            = felec;
 670
 671             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 672
 673              /* Update vectorial force */
 674             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 675             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 676             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 677
 678             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 679             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 680             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 681
 682             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 683             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 684             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 685             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 686
 687             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 688
 689             /* Inner loop uses 150 flops */
 690         }
 691
 692         /* End of innermost loop */
 693
 694         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 695                                               f+i_coord_offset,fshift+i_shift_offset);
 696
 697         ggid                        = gid[iidx];
 698         /* Update potential energies */
 699         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 700         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 701
 702         /* Increment number of inner iterations */
 703         inneriter                  += j_index_end - j_index_start;
 704
 705         /* Outer loop uses 26 flops */
 706     }
 707
 708     /* Increment number of outer iterations */
 709     outeriter        += nri;
 710
 711     /* Update outer/inner flops */
 712
 713     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*150);
 714 }
 715 /*
 716  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 717  * Electrostatics interaction: Ewald
 718  * VdW interaction:            CubicSplineTable
 719  * Geometry:                   Water4-Particle
 720  * Calculate force/pot:        Force
 721  */
 722 void
 723 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 724                     (t_nblist * gmx_restrict                nlist,
 725                      rvec * gmx_restrict                    xx,
 726                      rvec * gmx_restrict                    ff,
 727                      t_forcerec * gmx_restrict              fr,
 728                      t_mdatoms * gmx_restrict               mdatoms,
 729                      nb_kernel_data_t * gmx_restrict        kernel_data,
 730                      t_nrnb * gmx_restrict                  nrnb)
 731 {
 732     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 733      * just 0 for non-waters.
 734      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 735      * jnr indices corresponding to data put in the four positions in the SIMD register.
 736      */
 737     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 738     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 739     int              jnrA,jnrB,jnrC,jnrD;
 740     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 741     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 742     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 743     real             rcutoff_scalar;
 744     real             *shiftvec,*fshift,*x,*f;
 745     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 746     real             scratch[4*DIM];
 747     __m128           fscal,rcutoff,rcutoff2,jidxall;
 748     int              vdwioffset0;
 749     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 750     int              vdwioffset1;
 751     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 752     int              vdwioffset2;
 753     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 754     int              vdwioffset3;
 755     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 756     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 757     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 758     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 759     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 760     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 761     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 762     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 763     real             *charge;
 764     int              nvdwtype;
 765     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 766     int              *vdwtype;
 767     real             *vdwparam;
 768     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 769     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 770     __m128i          vfitab;
 771     __m128i          ifour       = _mm_set1_epi32(4);
 772     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 773     real             *vftab;
 774     __m128i          ewitab;
 775     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 776     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 777     real             *ewtab;
 778     __m128           dummy_mask,cutoff_mask;
 779     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 780     __m128           one     = _mm_set1_ps(1.0);
 781     __m128           two     = _mm_set1_ps(2.0);
 782     x                = xx[0];
 783     f                = ff[0];
 784
 785     nri              = nlist->nri;
 786     iinr             = nlist->iinr;
 787     jindex           = nlist->jindex;
 788     jjnr             = nlist->jjnr;
 789     shiftidx         = nlist->shift;
 790     gid              = nlist->gid;
 791     shiftvec         = fr->shift_vec[0];
 792     fshift           = fr->fshift[0];
 793     facel            = _mm_set1_ps(fr->epsfac);
 794     charge           = mdatoms->chargeA;
 795     nvdwtype         = fr->ntype;
 796     vdwparam         = fr->nbfp;
 797     vdwtype          = mdatoms->typeA;
 798
 799     vftab            = kernel_data->table_vdw->data;
 800     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 801
 802     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 803     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 804     beta2            = _mm_mul_ps(beta,beta);
 805     beta3            = _mm_mul_ps(beta,beta2);
 806     ewtab            = fr->ic->tabq_coul_F;
 807     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 808     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 809
 810     /* Setup water-specific parameters */
 811     inr              = nlist->iinr[0];
 812     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 813     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 814     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 815     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 816
 817     /* Avoid stupid compiler warnings */
 818     jnrA = jnrB = jnrC = jnrD = 0;
 819     j_coord_offsetA = 0;
 820     j_coord_offsetB = 0;
 821     j_coord_offsetC = 0;
 822     j_coord_offsetD = 0;
 823
 824     outeriter        = 0;
 825     inneriter        = 0;
 826
 827     for(iidx=0;iidx<4*DIM;iidx++)
 828     {
 829         scratch[iidx] = 0.0;
 830     }
 831
 832     /* Start outer loop over neighborlists */
 833     for(iidx=0; iidx<nri; iidx++)
 834     {
 835         /* Load shift vector for this list */
 836         i_shift_offset   = DIM*shiftidx[iidx];
 837
 838         /* Load limits for loop over neighbors */
 839         j_index_start    = jindex[iidx];
 840         j_index_end      = jindex[iidx+1];
 841
 842         /* Get outer coordinate index */
 843         inr              = iinr[iidx];
 844         i_coord_offset   = DIM*inr;
 845
 846         /* Load i particle coords and add shift vector */
 847         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 848                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 849
 850         fix0             = _mm_setzero_ps();
 851         fiy0             = _mm_setzero_ps();
 852         fiz0             = _mm_setzero_ps();
 853         fix1             = _mm_setzero_ps();
 854         fiy1             = _mm_setzero_ps();
 855         fiz1             = _mm_setzero_ps();
 856         fix2             = _mm_setzero_ps();
 857         fiy2             = _mm_setzero_ps();
 858         fiz2             = _mm_setzero_ps();
 859         fix3             = _mm_setzero_ps();
 860         fiy3             = _mm_setzero_ps();
 861         fiz3             = _mm_setzero_ps();
 862
 863         /* Start inner kernel loop */
 864         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 865         {
 866
 867             /* Get j neighbor index, and coordinate index */
 868             jnrA             = jjnr[jidx];
 869             jnrB             = jjnr[jidx+1];
 870             jnrC             = jjnr[jidx+2];
 871             jnrD             = jjnr[jidx+3];
 872             j_coord_offsetA  = DIM*jnrA;
 873             j_coord_offsetB  = DIM*jnrB;
 874             j_coord_offsetC  = DIM*jnrC;
 875             j_coord_offsetD  = DIM*jnrD;
 876
 877             /* load j atom coordinates */
 878             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 879                                               x+j_coord_offsetC,x+j_coord_offsetD,
 880                                               &jx0,&jy0,&jz0);
 881
 882             /* Calculate displacement vector */
 883             dx00             = _mm_sub_ps(ix0,jx0);
 884             dy00             = _mm_sub_ps(iy0,jy0);
 885             dz00             = _mm_sub_ps(iz0,jz0);
 886             dx10             = _mm_sub_ps(ix1,jx0);
 887             dy10             = _mm_sub_ps(iy1,jy0);
 888             dz10             = _mm_sub_ps(iz1,jz0);
 889             dx20             = _mm_sub_ps(ix2,jx0);
 890             dy20             = _mm_sub_ps(iy2,jy0);
 891             dz20             = _mm_sub_ps(iz2,jz0);
 892             dx30             = _mm_sub_ps(ix3,jx0);
 893             dy30             = _mm_sub_ps(iy3,jy0);
 894             dz30             = _mm_sub_ps(iz3,jz0);
 895
 896             /* Calculate squared distance and things based on it */
 897             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 898             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 899             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 900             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 901
 902             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 903             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 904             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 905             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 906
 907             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 908             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 909             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 910
 911             /* Load parameters for j particles */
 912             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 913                                                               charge+jnrC+0,charge+jnrD+0);
 914             vdwjidx0A        = 2*vdwtype[jnrA+0];
 915             vdwjidx0B        = 2*vdwtype[jnrB+0];
 916             vdwjidx0C        = 2*vdwtype[jnrC+0];
 917             vdwjidx0D        = 2*vdwtype[jnrD+0];
 918
 919             fjx0             = _mm_setzero_ps();
 920             fjy0             = _mm_setzero_ps();
 921             fjz0             = _mm_setzero_ps();
 922
 923             /**************************
 924              * CALCULATE INTERACTIONS *
 925              **************************/
 926
 927             r00              = _mm_mul_ps(rsq00,rinv00);
 928
 929             /* Compute parameters for interactions between i and j atoms */
 930             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 931                                          vdwparam+vdwioffset0+vdwjidx0B,
 932                                          vdwparam+vdwioffset0+vdwjidx0C,
 933                                          vdwparam+vdwioffset0+vdwjidx0D,
 934                                          &c6_00,&c12_00);
 935
 936             /* Calculate table index by multiplying r with table scale and truncate to integer */
 937             rt               = _mm_mul_ps(r00,vftabscale);
 938             vfitab           = _mm_cvttps_epi32(rt);
 939 #ifdef __XOP__
 940             vfeps            = _mm_frcz_ps(rt);
 941 #else
 942             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 943 #endif
 944             twovfeps         = _mm_add_ps(vfeps,vfeps);
 945             vfitab           = _mm_slli_epi32(vfitab,3);
 946
 947             /* CUBIC SPLINE TABLE DISPERSION */
 948             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 949             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 950             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 951             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 952             _MM_TRANSPOSE4_PS(Y,F,G,H);
 953             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 954             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 955             fvdw6            = _mm_mul_ps(c6_00,FF);
 956
 957             /* CUBIC SPLINE TABLE REPULSION */
 958             vfitab           = _mm_add_epi32(vfitab,ifour);
 959             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 960             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 961             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 962             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 963             _MM_TRANSPOSE4_PS(Y,F,G,H);
 964             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 965             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 966             fvdw12           = _mm_mul_ps(c12_00,FF);
 967             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 968
 969             fscal            = fvdw;
 970
 971              /* Update vectorial force */
 972             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 973             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 974             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 975
 976             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 977             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 978             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 979
 980             /**************************
 981              * CALCULATE INTERACTIONS *
 982              **************************/
 983
 984             r10              = _mm_mul_ps(rsq10,rinv10);
 985
 986             /* Compute parameters for interactions between i and j atoms */
 987             qq10             = _mm_mul_ps(iq1,jq0);
 988
 989             /* EWALD ELECTROSTATICS */
 990
 991             /* Analytical PME correction */
 992             zeta2            = _mm_mul_ps(beta2,rsq10);
 993             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 994             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 995             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 996             felec            = _mm_mul_ps(qq10,felec);
 997
 998             fscal            = felec;
 999
1000              /* Update vectorial force */
1001             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1002             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1003             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1004
1005             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1006             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1007             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1008
1009             /**************************
1010              * CALCULATE INTERACTIONS *
1011              **************************/
1012
1013             r20              = _mm_mul_ps(rsq20,rinv20);
1014
1015             /* Compute parameters for interactions between i and j atoms */
1016             qq20             = _mm_mul_ps(iq2,jq0);
1017
1018             /* EWALD ELECTROSTATICS */
1019
1020             /* Analytical PME correction */
1021             zeta2            = _mm_mul_ps(beta2,rsq20);
1022             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1023             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1024             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1025             felec            = _mm_mul_ps(qq20,felec);
1026
1027             fscal            = felec;
1028
1029              /* Update vectorial force */
1030             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1031             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1032             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1033
1034             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1035             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1036             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             r30              = _mm_mul_ps(rsq30,rinv30);
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             qq30             = _mm_mul_ps(iq3,jq0);
1046
1047             /* EWALD ELECTROSTATICS */
1048
1049             /* Analytical PME correction */
1050             zeta2            = _mm_mul_ps(beta2,rsq30);
1051             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1052             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1053             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1054             felec            = _mm_mul_ps(qq30,felec);
1055
1056             fscal            = felec;
1057
1058              /* Update vectorial force */
1059             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1060             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1061             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1062
1063             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1064             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1065             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1066
1067             fjptrA             = f+j_coord_offsetA;
1068             fjptrB             = f+j_coord_offsetB;
1069             fjptrC             = f+j_coord_offsetC;
1070             fjptrD             = f+j_coord_offsetD;
1071
1072             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1073
1074             /* Inner loop uses 135 flops */
1075         }
1076
1077         if(jidx<j_index_end)
1078         {
1079
1080             /* Get j neighbor index, and coordinate index */
1081             jnrlistA         = jjnr[jidx];
1082             jnrlistB         = jjnr[jidx+1];
1083             jnrlistC         = jjnr[jidx+2];
1084             jnrlistD         = jjnr[jidx+3];
1085             /* Sign of each element will be negative for non-real atoms.
1086              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1087              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1088              */
1089             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1090             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1091             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1092             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1093             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1094             j_coord_offsetA  = DIM*jnrA;
1095             j_coord_offsetB  = DIM*jnrB;
1096             j_coord_offsetC  = DIM*jnrC;
1097             j_coord_offsetD  = DIM*jnrD;
1098
1099             /* load j atom coordinates */
1100             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1101                                               x+j_coord_offsetC,x+j_coord_offsetD,
1102                                               &jx0,&jy0,&jz0);
1103
1104             /* Calculate displacement vector */
1105             dx00             = _mm_sub_ps(ix0,jx0);
1106             dy00             = _mm_sub_ps(iy0,jy0);
1107             dz00             = _mm_sub_ps(iz0,jz0);
1108             dx10             = _mm_sub_ps(ix1,jx0);
1109             dy10             = _mm_sub_ps(iy1,jy0);
1110             dz10             = _mm_sub_ps(iz1,jz0);
1111             dx20             = _mm_sub_ps(ix2,jx0);
1112             dy20             = _mm_sub_ps(iy2,jy0);
1113             dz20             = _mm_sub_ps(iz2,jz0);
1114             dx30             = _mm_sub_ps(ix3,jx0);
1115             dy30             = _mm_sub_ps(iy3,jy0);
1116             dz30             = _mm_sub_ps(iz3,jz0);
1117
1118             /* Calculate squared distance and things based on it */
1119             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1120             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1121             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1122             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1123
1124             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1125             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1126             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1127             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1128
1129             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1130             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1131             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1132
1133             /* Load parameters for j particles */
1134             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1135                                                               charge+jnrC+0,charge+jnrD+0);
1136             vdwjidx0A        = 2*vdwtype[jnrA+0];
1137             vdwjidx0B        = 2*vdwtype[jnrB+0];
1138             vdwjidx0C        = 2*vdwtype[jnrC+0];
1139             vdwjidx0D        = 2*vdwtype[jnrD+0];
1140
1141             fjx0             = _mm_setzero_ps();
1142             fjy0             = _mm_setzero_ps();
1143             fjz0             = _mm_setzero_ps();
1144
1145             /**************************
1146              * CALCULATE INTERACTIONS *
1147              **************************/
1148
1149             r00              = _mm_mul_ps(rsq00,rinv00);
1150             r00              = _mm_andnot_ps(dummy_mask,r00);
1151
1152             /* Compute parameters for interactions between i and j atoms */
1153             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1154                                          vdwparam+vdwioffset0+vdwjidx0B,
1155                                          vdwparam+vdwioffset0+vdwjidx0C,
1156                                          vdwparam+vdwioffset0+vdwjidx0D,
1157                                          &c6_00,&c12_00);
1158
1159             /* Calculate table index by multiplying r with table scale and truncate to integer */
1160             rt               = _mm_mul_ps(r00,vftabscale);
1161             vfitab           = _mm_cvttps_epi32(rt);
1162 #ifdef __XOP__
1163             vfeps            = _mm_frcz_ps(rt);
1164 #else
1165             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1166 #endif
1167             twovfeps         = _mm_add_ps(vfeps,vfeps);
1168             vfitab           = _mm_slli_epi32(vfitab,3);
1169
1170             /* CUBIC SPLINE TABLE DISPERSION */
1171             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1172             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1173             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1174             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1175             _MM_TRANSPOSE4_PS(Y,F,G,H);
1176             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1177             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1178             fvdw6            = _mm_mul_ps(c6_00,FF);
1179
1180             /* CUBIC SPLINE TABLE REPULSION */
1181             vfitab           = _mm_add_epi32(vfitab,ifour);
1182             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1183             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1184             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1185             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1186             _MM_TRANSPOSE4_PS(Y,F,G,H);
1187             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1188             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1189             fvdw12           = _mm_mul_ps(c12_00,FF);
1190             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1191
1192             fscal            = fvdw;
1193
1194             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1195
1196              /* Update vectorial force */
1197             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1198             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1199             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1200
1201             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1202             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1203             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1204
1205             /**************************
1206              * CALCULATE INTERACTIONS *
1207              **************************/
1208
1209             r10              = _mm_mul_ps(rsq10,rinv10);
1210             r10              = _mm_andnot_ps(dummy_mask,r10);
1211
1212             /* Compute parameters for interactions between i and j atoms */
1213             qq10             = _mm_mul_ps(iq1,jq0);
1214
1215             /* EWALD ELECTROSTATICS */
1216
1217             /* Analytical PME correction */
1218             zeta2            = _mm_mul_ps(beta2,rsq10);
1219             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1220             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1221             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1222             felec            = _mm_mul_ps(qq10,felec);
1223
1224             fscal            = felec;
1225
1226             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1227
1228              /* Update vectorial force */
1229             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1230             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1231             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1232
1233             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1234             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1235             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1236
1237             /**************************
1238              * CALCULATE INTERACTIONS *
1239              **************************/
1240
1241             r20              = _mm_mul_ps(rsq20,rinv20);
1242             r20              = _mm_andnot_ps(dummy_mask,r20);
1243
1244             /* Compute parameters for interactions between i and j atoms */
1245             qq20             = _mm_mul_ps(iq2,jq0);
1246
1247             /* EWALD ELECTROSTATICS */
1248
1249             /* Analytical PME correction */
1250             zeta2            = _mm_mul_ps(beta2,rsq20);
1251             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1252             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1253             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1254             felec            = _mm_mul_ps(qq20,felec);
1255
1256             fscal            = felec;
1257
1258             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1259
1260              /* Update vectorial force */
1261             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1262             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1263             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1264
1265             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1266             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1267             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1268
1269             /**************************
1270              * CALCULATE INTERACTIONS *
1271              **************************/
1272
1273             r30              = _mm_mul_ps(rsq30,rinv30);
1274             r30              = _mm_andnot_ps(dummy_mask,r30);
1275
1276             /* Compute parameters for interactions between i and j atoms */
1277             qq30             = _mm_mul_ps(iq3,jq0);
1278
1279             /* EWALD ELECTROSTATICS */
1280
1281             /* Analytical PME correction */
1282             zeta2            = _mm_mul_ps(beta2,rsq30);
1283             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1284             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1285             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1286             felec            = _mm_mul_ps(qq30,felec);
1287
1288             fscal            = felec;
1289
1290             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1291
1292              /* Update vectorial force */
1293             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1294             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1295             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1296
1297             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1298             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1299             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1300
1301             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1302             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1303             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1304             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1305
1306             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1307
1308             /* Inner loop uses 139 flops */
1309         }
1310
1311         /* End of innermost loop */
1312
1313         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1314                                               f+i_coord_offset,fshift+i_shift_offset);
1315
1316         /* Increment number of inner iterations */
1317         inneriter                  += j_index_end - j_index_start;
1318
1319         /* Outer loop uses 24 flops */
1320     }
1321
1322     /* Increment number of outer iterations */
1323     outeriter        += nri;
1324
1325     /* Update outer/inner flops */
1326
1327     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*139);
1328 }