src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwNone_GeomP1P1_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_ElecCSTab_VdwNone_GeomP1P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  72     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  73     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  74     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  75     real             *charge;
  76     __m128i          vfitab;
  77     __m128i          ifour       = _mm_set1_epi32(4);
  78     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
  79     real             *vftab;
  80     __m128           dummy_mask,cutoff_mask;
  81     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  82     __m128           one     = _mm_set1_ps(1.0);
  83     __m128           two     = _mm_set1_ps(2.0);
  84     x                = xx[0];
  85     f                = ff[0];
  86
  87     nri              = nlist->nri;
  88     iinr             = nlist->iinr;
  89     jindex           = nlist->jindex;
  90     jjnr             = nlist->jjnr;
  91     shiftidx         = nlist->shift;
  92     gid              = nlist->gid;
  93     shiftvec         = fr->shift_vec[0];
  94     fshift           = fr->fshift[0];
  95     facel            = _mm_set1_ps(fr->epsfac);
  96     charge           = mdatoms->chargeA;
  97
  98     vftab            = kernel_data->table_elec->data;
  99     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 100
 101     /* Avoid stupid compiler warnings */
 102     jnrA = jnrB = jnrC = jnrD = 0;
 103     j_coord_offsetA = 0;
 104     j_coord_offsetB = 0;
 105     j_coord_offsetC = 0;
 106     j_coord_offsetD = 0;
 107
 108     outeriter        = 0;
 109     inneriter        = 0;
 110
 111     for(iidx=0;iidx<4*DIM;iidx++)
 112     {
 113         scratch[iidx] = 0.0;
 114     }
 115
 116     /* Start outer loop over neighborlists */
 117     for(iidx=0; iidx<nri; iidx++)
 118     {
 119         /* Load shift vector for this list */
 120         i_shift_offset   = DIM*shiftidx[iidx];
 121
 122         /* Load limits for loop over neighbors */
 123         j_index_start    = jindex[iidx];
 124         j_index_end      = jindex[iidx+1];
 125
 126         /* Get outer coordinate index */
 127         inr              = iinr[iidx];
 128         i_coord_offset   = DIM*inr;
 129
 130         /* Load i particle coords and add shift vector */
 131         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 132
 133         fix0             = _mm_setzero_ps();
 134         fiy0             = _mm_setzero_ps();
 135         fiz0             = _mm_setzero_ps();
 136
 137         /* Load parameters for i particles */
 138         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 139
 140         /* Reset potential sums */
 141         velecsum         = _mm_setzero_ps();
 142
 143         /* Start inner kernel loop */
 144         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 145         {
 146
 147             /* Get j neighbor index, and coordinate index */
 148             jnrA             = jjnr[jidx];
 149             jnrB             = jjnr[jidx+1];
 150             jnrC             = jjnr[jidx+2];
 151             jnrD             = jjnr[jidx+3];
 152             j_coord_offsetA  = DIM*jnrA;
 153             j_coord_offsetB  = DIM*jnrB;
 154             j_coord_offsetC  = DIM*jnrC;
 155             j_coord_offsetD  = DIM*jnrD;
 156
 157             /* load j atom coordinates */
 158             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 159                                               x+j_coord_offsetC,x+j_coord_offsetD,
 160                                               &jx0,&jy0,&jz0);
 161
 162             /* Calculate displacement vector */
 163             dx00             = _mm_sub_ps(ix0,jx0);
 164             dy00             = _mm_sub_ps(iy0,jy0);
 165             dz00             = _mm_sub_ps(iz0,jz0);
 166
 167             /* Calculate squared distance and things based on it */
 168             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 169
 170             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 171
 172             /* Load parameters for j particles */
 173             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 174                                                               charge+jnrC+0,charge+jnrD+0);
 175
 176             /**************************
 177              * CALCULATE INTERACTIONS *
 178              **************************/
 179
 180             r00              = _mm_mul_ps(rsq00,rinv00);
 181
 182             /* Compute parameters for interactions between i and j atoms */
 183             qq00             = _mm_mul_ps(iq0,jq0);
 184
 185             /* Calculate table index by multiplying r with table scale and truncate to integer */
 186             rt               = _mm_mul_ps(r00,vftabscale);
 187             vfitab           = _mm_cvttps_epi32(rt);
 188 #ifdef __XOP__
 189             vfeps            = _mm_frcz_ps(rt);
 190 #else
 191             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 192 #endif
 193             twovfeps         = _mm_add_ps(vfeps,vfeps);
 194             vfitab           = _mm_slli_epi32(vfitab,2);
 195
 196             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 197             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 198             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 199             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 200             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 201             _MM_TRANSPOSE4_PS(Y,F,G,H);
 202             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 203             VV               = _mm_macc_ps(vfeps,Fp,Y);
 204             velec            = _mm_mul_ps(qq00,VV);
 205             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 206             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 207
 208             /* Update potential sum for this i atom from the interaction with this j atom. */
 209             velecsum         = _mm_add_ps(velecsum,velec);
 210
 211             fscal            = felec;
 212
 213              /* Update vectorial force */
 214             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 215             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 216             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 217
 218             fjptrA             = f+j_coord_offsetA;
 219             fjptrB             = f+j_coord_offsetB;
 220             fjptrC             = f+j_coord_offsetC;
 221             fjptrD             = f+j_coord_offsetD;
 222             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 223                                                    _mm_mul_ps(dx00,fscal),
 224                                                    _mm_mul_ps(dy00,fscal),
 225                                                    _mm_mul_ps(dz00,fscal));
 226
 227             /* Inner loop uses 46 flops */
 228         }
 229
 230         if(jidx<j_index_end)
 231         {
 232
 233             /* Get j neighbor index, and coordinate index */
 234             jnrlistA         = jjnr[jidx];
 235             jnrlistB         = jjnr[jidx+1];
 236             jnrlistC         = jjnr[jidx+2];
 237             jnrlistD         = jjnr[jidx+3];
 238             /* Sign of each element will be negative for non-real atoms.
 239              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 240              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 241              */
 242             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 243             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 244             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 245             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 246             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 247             j_coord_offsetA  = DIM*jnrA;
 248             j_coord_offsetB  = DIM*jnrB;
 249             j_coord_offsetC  = DIM*jnrC;
 250             j_coord_offsetD  = DIM*jnrD;
 251
 252             /* load j atom coordinates */
 253             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 254                                               x+j_coord_offsetC,x+j_coord_offsetD,
 255                                               &jx0,&jy0,&jz0);
 256
 257             /* Calculate displacement vector */
 258             dx00             = _mm_sub_ps(ix0,jx0);
 259             dy00             = _mm_sub_ps(iy0,jy0);
 260             dz00             = _mm_sub_ps(iz0,jz0);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 264
 265             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 266
 267             /* Load parameters for j particles */
 268             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 269                                                               charge+jnrC+0,charge+jnrD+0);
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             r00              = _mm_mul_ps(rsq00,rinv00);
 276             r00              = _mm_andnot_ps(dummy_mask,r00);
 277
 278             /* Compute parameters for interactions between i and j atoms */
 279             qq00             = _mm_mul_ps(iq0,jq0);
 280
 281             /* Calculate table index by multiplying r with table scale and truncate to integer */
 282             rt               = _mm_mul_ps(r00,vftabscale);
 283             vfitab           = _mm_cvttps_epi32(rt);
 284 #ifdef __XOP__
 285             vfeps            = _mm_frcz_ps(rt);
 286 #else
 287             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 288 #endif
 289             twovfeps         = _mm_add_ps(vfeps,vfeps);
 290             vfitab           = _mm_slli_epi32(vfitab,2);
 291
 292             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 293             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 294             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 295             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 296             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 297             _MM_TRANSPOSE4_PS(Y,F,G,H);
 298             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 299             VV               = _mm_macc_ps(vfeps,Fp,Y);
 300             velec            = _mm_mul_ps(qq00,VV);
 301             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 302             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velec            = _mm_andnot_ps(dummy_mask,velec);
 306             velecsum         = _mm_add_ps(velecsum,velec);
 307
 308             fscal            = felec;
 309
 310             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 311
 312              /* Update vectorial force */
 313             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 314             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 315             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 316
 317             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 318             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 319             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 320             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 321             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 322                                                    _mm_mul_ps(dx00,fscal),
 323                                                    _mm_mul_ps(dy00,fscal),
 324                                                    _mm_mul_ps(dz00,fscal));
 325
 326             /* Inner loop uses 47 flops */
 327         }
 328
 329         /* End of innermost loop */
 330
 331         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 332                                               f+i_coord_offset,fshift+i_shift_offset);
 333
 334         ggid                        = gid[iidx];
 335         /* Update potential energies */
 336         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 337
 338         /* Increment number of inner iterations */
 339         inneriter                  += j_index_end - j_index_start;
 340
 341         /* Outer loop uses 8 flops */
 342     }
 343
 344     /* Increment number of outer iterations */
 345     outeriter        += nri;
 346
 347     /* Update outer/inner flops */
 348
 349     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*47);
 350 }
 351 /*
 352  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_avx_128_fma_single
 353  * Electrostatics interaction: CubicSplineTable
 354  * VdW interaction:            None
 355  * Geometry:                   Particle-Particle
 356  * Calculate force/pot:        Force
 357  */
 358 void
 359 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_avx_128_fma_single
 360                     (t_nblist * gmx_restrict                nlist,
 361                      rvec * gmx_restrict                    xx,
 362                      rvec * gmx_restrict                    ff,
 363                      t_forcerec * gmx_restrict              fr,
 364                      t_mdatoms * gmx_restrict               mdatoms,
 365                      nb_kernel_data_t * gmx_restrict        kernel_data,
 366                      t_nrnb * gmx_restrict                  nrnb)
 367 {
 368     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 369      * just 0 for non-waters.
 370      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 371      * jnr indices corresponding to data put in the four positions in the SIMD register.
 372      */
 373     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 374     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 375     int              jnrA,jnrB,jnrC,jnrD;
 376     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 377     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 378     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 379     real             rcutoff_scalar;
 380     real             *shiftvec,*fshift,*x,*f;
 381     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 382     real             scratch[4*DIM];
 383     __m128           fscal,rcutoff,rcutoff2,jidxall;
 384     int              vdwioffset0;
 385     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 386     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 387     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 388     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 389     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 390     real             *charge;
 391     __m128i          vfitab;
 392     __m128i          ifour       = _mm_set1_epi32(4);
 393     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 394     real             *vftab;
 395     __m128           dummy_mask,cutoff_mask;
 396     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 397     __m128           one     = _mm_set1_ps(1.0);
 398     __m128           two     = _mm_set1_ps(2.0);
 399     x                = xx[0];
 400     f                = ff[0];
 401
 402     nri              = nlist->nri;
 403     iinr             = nlist->iinr;
 404     jindex           = nlist->jindex;
 405     jjnr             = nlist->jjnr;
 406     shiftidx         = nlist->shift;
 407     gid              = nlist->gid;
 408     shiftvec         = fr->shift_vec[0];
 409     fshift           = fr->fshift[0];
 410     facel            = _mm_set1_ps(fr->epsfac);
 411     charge           = mdatoms->chargeA;
 412
 413     vftab            = kernel_data->table_elec->data;
 414     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 415
 416     /* Avoid stupid compiler warnings */
 417     jnrA = jnrB = jnrC = jnrD = 0;
 418     j_coord_offsetA = 0;
 419     j_coord_offsetB = 0;
 420     j_coord_offsetC = 0;
 421     j_coord_offsetD = 0;
 422
 423     outeriter        = 0;
 424     inneriter        = 0;
 425
 426     for(iidx=0;iidx<4*DIM;iidx++)
 427     {
 428         scratch[iidx] = 0.0;
 429     }
 430
 431     /* Start outer loop over neighborlists */
 432     for(iidx=0; iidx<nri; iidx++)
 433     {
 434         /* Load shift vector for this list */
 435         i_shift_offset   = DIM*shiftidx[iidx];
 436
 437         /* Load limits for loop over neighbors */
 438         j_index_start    = jindex[iidx];
 439         j_index_end      = jindex[iidx+1];
 440
 441         /* Get outer coordinate index */
 442         inr              = iinr[iidx];
 443         i_coord_offset   = DIM*inr;
 444
 445         /* Load i particle coords and add shift vector */
 446         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 447
 448         fix0             = _mm_setzero_ps();
 449         fiy0             = _mm_setzero_ps();
 450         fiz0             = _mm_setzero_ps();
 451
 452         /* Load parameters for i particles */
 453         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 454
 455         /* Start inner kernel loop */
 456         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 457         {
 458
 459             /* Get j neighbor index, and coordinate index */
 460             jnrA             = jjnr[jidx];
 461             jnrB             = jjnr[jidx+1];
 462             jnrC             = jjnr[jidx+2];
 463             jnrD             = jjnr[jidx+3];
 464             j_coord_offsetA  = DIM*jnrA;
 465             j_coord_offsetB  = DIM*jnrB;
 466             j_coord_offsetC  = DIM*jnrC;
 467             j_coord_offsetD  = DIM*jnrD;
 468
 469             /* load j atom coordinates */
 470             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 471                                               x+j_coord_offsetC,x+j_coord_offsetD,
 472                                               &jx0,&jy0,&jz0);
 473
 474             /* Calculate displacement vector */
 475             dx00             = _mm_sub_ps(ix0,jx0);
 476             dy00             = _mm_sub_ps(iy0,jy0);
 477             dz00             = _mm_sub_ps(iz0,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 481
 482             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 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
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             r00              = _mm_mul_ps(rsq00,rinv00);
 493
 494             /* Compute parameters for interactions between i and j atoms */
 495             qq00             = _mm_mul_ps(iq0,jq0);
 496
 497             /* Calculate table index by multiplying r with table scale and truncate to integer */
 498             rt               = _mm_mul_ps(r00,vftabscale);
 499             vfitab           = _mm_cvttps_epi32(rt);
 500 #ifdef __XOP__
 501             vfeps            = _mm_frcz_ps(rt);
 502 #else
 503             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 504 #endif
 505             twovfeps         = _mm_add_ps(vfeps,vfeps);
 506             vfitab           = _mm_slli_epi32(vfitab,2);
 507
 508             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 509             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 510             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 511             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 512             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 513             _MM_TRANSPOSE4_PS(Y,F,G,H);
 514             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 515             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 516             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 517
 518             fscal            = felec;
 519
 520              /* Update vectorial force */
 521             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 522             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 523             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 524
 525             fjptrA             = f+j_coord_offsetA;
 526             fjptrB             = f+j_coord_offsetB;
 527             fjptrC             = f+j_coord_offsetC;
 528             fjptrD             = f+j_coord_offsetD;
 529             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 530                                                    _mm_mul_ps(dx00,fscal),
 531                                                    _mm_mul_ps(dy00,fscal),
 532                                                    _mm_mul_ps(dz00,fscal));
 533
 534             /* Inner loop uses 42 flops */
 535         }
 536
 537         if(jidx<j_index_end)
 538         {
 539
 540             /* Get j neighbor index, and coordinate index */
 541             jnrlistA         = jjnr[jidx];
 542             jnrlistB         = jjnr[jidx+1];
 543             jnrlistC         = jjnr[jidx+2];
 544             jnrlistD         = jjnr[jidx+3];
 545             /* Sign of each element will be negative for non-real atoms.
 546              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 547              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 548              */
 549             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 550             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 551             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 552             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 553             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 554             j_coord_offsetA  = DIM*jnrA;
 555             j_coord_offsetB  = DIM*jnrB;
 556             j_coord_offsetC  = DIM*jnrC;
 557             j_coord_offsetD  = DIM*jnrD;
 558
 559             /* load j atom coordinates */
 560             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 561                                               x+j_coord_offsetC,x+j_coord_offsetD,
 562                                               &jx0,&jy0,&jz0);
 563
 564             /* Calculate displacement vector */
 565             dx00             = _mm_sub_ps(ix0,jx0);
 566             dy00             = _mm_sub_ps(iy0,jy0);
 567             dz00             = _mm_sub_ps(iz0,jz0);
 568
 569             /* Calculate squared distance and things based on it */
 570             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 571
 572             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 573
 574             /* Load parameters for j particles */
 575             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 576                                                               charge+jnrC+0,charge+jnrD+0);
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             r00              = _mm_mul_ps(rsq00,rinv00);
 583             r00              = _mm_andnot_ps(dummy_mask,r00);
 584
 585             /* Compute parameters for interactions between i and j atoms */
 586             qq00             = _mm_mul_ps(iq0,jq0);
 587
 588             /* Calculate table index by multiplying r with table scale and truncate to integer */
 589             rt               = _mm_mul_ps(r00,vftabscale);
 590             vfitab           = _mm_cvttps_epi32(rt);
 591 #ifdef __XOP__
 592             vfeps            = _mm_frcz_ps(rt);
 593 #else
 594             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 595 #endif
 596             twovfeps         = _mm_add_ps(vfeps,vfeps);
 597             vfitab           = _mm_slli_epi32(vfitab,2);
 598
 599             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 600             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 601             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 602             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 603             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 604             _MM_TRANSPOSE4_PS(Y,F,G,H);
 605             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 606             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 607             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 612
 613              /* Update vectorial force */
 614             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 615             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 616             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 617
 618             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 619             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 620             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 621             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 622             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 623                                                    _mm_mul_ps(dx00,fscal),
 624                                                    _mm_mul_ps(dy00,fscal),
 625                                                    _mm_mul_ps(dz00,fscal));
 626
 627             /* Inner loop uses 43 flops */
 628         }
 629
 630         /* End of innermost loop */
 631
 632         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 633                                               f+i_coord_offset,fshift+i_shift_offset);
 634
 635         /* Increment number of inner iterations */
 636         inneriter                  += j_index_end - j_index_start;
 637
 638         /* Outer loop uses 7 flops */
 639     }
 640
 641     /* Increment number of outer iterations */
 642     outeriter        += nri;
 643
 644     /* Update outer/inner flops */
 645
 646     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*43);
 647 }