src/mdlib/nbnxn_kernels/nbnxn_kernel_simd_4xn_inner.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
   5  * Copyright (c) 2001-2009, The GROMACS Development Team
   6  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   7  * David van der Spoel, Berk Hess, Erik Lindahl, and including many
   8  * others, as listed in the AUTHORS file in the top-level source
   9  * directory and at http://www.gromacs.org.
  10  *
  11  * GROMACS is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public License
  13  * as published by the Free Software Foundation; either version 2.1
  14  * of the License, or (at your option) any later version.
  15  *
  16  * GROMACS is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with GROMACS; if not, see
  23  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  24  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  25  *
  26  * If you want to redistribute modifications to GROMACS, please
  27  * consider that scientific software is very special. Version
  28  * control is crucial - bugs must be traceable. We will be happy to
  29  * consider code for inclusion in the official distribution, but
  30  * derived work must not be called official GROMACS. Details are found
  31  * in the README & COPYING files - if they are missing, get the
  32  * official version at http://www.gromacs.org.
  33  *
  34  * To help us fund GROMACS development, we humbly ask that you cite
  35  * the research papers on the package. Check out http://www.gromacs.org.
  36  */
  37
  38 /* This is the innermost loop contents for the 4 x N atom SIMD kernel.
  39  * This flavor of the kernel calculates interactions of 4 i-atoms
  40  * with N j-atoms stored in N wide SIMD registers.
  41  */
  42
  43
  44 /* When calculating RF or Ewald interactions we calculate the electrostatic
  45  * forces on excluded atom pairs here in the non-bonded loops.
  46  * But when energies and/or virial is required we calculate them
  47  * separately to as then it is easier to separate the energy and virial
  48  * contributions.
  49  */
  50 #if defined CHECK_EXCLS && defined CALC_COULOMB
  51 #define EXCL_FORCES
  52 #endif
  53
  54 /* Without exclusions and energies we only need to mask the cut-off,
  55  * this can be faster with blendv (only available with SSE4.1 and later).
  56  */
  57 #if !(defined CHECK_EXCLS || defined CALC_ENERGIES) && defined GMX_X86_SSE4_1 && !defined COUNT_PAIRS
  58 /* With RF and tabulated Coulomb we replace cmp+and with sub+blendv.
  59  * With gcc this is slower, except for RF on Sandy Bridge.
  60  * Tested with gcc 4.6.2, 4.6.3 and 4.7.1.
  61  */
  62 #if (defined CALC_COUL_RF || defined CALC_COUL_TAB) && (!defined __GNUC__ || (defined CALC_COUL_RF && defined GMX_X86_AVX_256))
  63 #define CUTOFF_BLENDV
  64 #endif
  65 /* With analytical Ewald we replace cmp+and+and with sub+blendv+blendv.
  66  * This is only faster with icc on Sandy Bridge (PS kernel slower than gcc 4.7).
  67  * Tested with icc 13.
  68  */
  69 #if defined CALC_COUL_EWALD && defined __INTEL_COMPILER && defined GMX_X86_AVX_256
  70 #define CUTOFF_BLENDV
  71 #endif
  72 #endif
  73
  74         {
  75             int        cj,aj,ajx,ajy,ajz;
  76
  77 #ifdef ENERGY_GROUPS
  78             /* Energy group indices for two atoms packed into one int */
  79             int        egp_jj[UNROLLJ/2];
  80 #endif
  81
  82 #ifdef CHECK_EXCLS
  83             /* Interaction (non-exclusion) mask of all 1's or 0's */
  84             gmx_mm_pr  int_SSE0;
  85             gmx_mm_pr  int_SSE1;
  86             gmx_mm_pr  int_SSE2;
  87             gmx_mm_pr  int_SSE3;
  88 #endif
  89
  90             gmx_mm_pr  jxSSE,jySSE,jzSSE;
  91             gmx_mm_pr  dx_SSE0,dy_SSE0,dz_SSE0;
  92             gmx_mm_pr  dx_SSE1,dy_SSE1,dz_SSE1;
  93             gmx_mm_pr  dx_SSE2,dy_SSE2,dz_SSE2;
  94             gmx_mm_pr  dx_SSE3,dy_SSE3,dz_SSE3;
  95             gmx_mm_pr  tx_SSE0,ty_SSE0,tz_SSE0;
  96             gmx_mm_pr  tx_SSE1,ty_SSE1,tz_SSE1;
  97             gmx_mm_pr  tx_SSE2,ty_SSE2,tz_SSE2;
  98             gmx_mm_pr  tx_SSE3,ty_SSE3,tz_SSE3;
  99             gmx_mm_pr  rsq_SSE0,rinv_SSE0,rinvsq_SSE0;
 100             gmx_mm_pr  rsq_SSE1,rinv_SSE1,rinvsq_SSE1;
 101             gmx_mm_pr  rsq_SSE2,rinv_SSE2,rinvsq_SSE2;
 102             gmx_mm_pr  rsq_SSE3,rinv_SSE3,rinvsq_SSE3;
 103 #ifndef CUTOFF_BLENDV
 104             /* wco: within cut-off, mask of all 1's or 0's */
 105             gmx_mm_pr  wco_SSE0;
 106             gmx_mm_pr  wco_SSE1;
 107             gmx_mm_pr  wco_SSE2;
 108             gmx_mm_pr  wco_SSE3;
 109 #endif
 110 #ifdef VDW_CUTOFF_CHECK
 111             gmx_mm_pr  wco_vdw_SSE0;
 112             gmx_mm_pr  wco_vdw_SSE1;
 113 #ifndef HALF_LJ
 114             gmx_mm_pr  wco_vdw_SSE2;
 115             gmx_mm_pr  wco_vdw_SSE3;
 116 #endif
 117 #endif
 118 #ifdef CALC_COULOMB
 119 #ifdef CHECK_EXCLS
 120             /* 1/r masked with the interaction mask */
 121             gmx_mm_pr  rinv_ex_SSE0;
 122             gmx_mm_pr  rinv_ex_SSE1;
 123             gmx_mm_pr  rinv_ex_SSE2;
 124             gmx_mm_pr  rinv_ex_SSE3;
 125 #endif
 126             gmx_mm_pr  jq_SSE;
 127             gmx_mm_pr  qq_SSE0;
 128             gmx_mm_pr  qq_SSE1;
 129             gmx_mm_pr  qq_SSE2;
 130             gmx_mm_pr  qq_SSE3;
 131 #ifdef CALC_COUL_TAB
 132             /* The force (PME mesh force) we need to subtract from 1/r^2 */
 133             gmx_mm_pr  fsub_SSE0;
 134             gmx_mm_pr  fsub_SSE1;
 135             gmx_mm_pr  fsub_SSE2;
 136             gmx_mm_pr  fsub_SSE3;
 137 #endif
 138 #ifdef CALC_COUL_EWALD
 139             gmx_mm_pr  brsq_SSE0,brsq_SSE1,brsq_SSE2,brsq_SSE3;
 140             gmx_mm_pr  ewcorr_SSE0,ewcorr_SSE1,ewcorr_SSE2,ewcorr_SSE3;
 141 #endif
 142
 143             /* frcoul = (1/r - fsub)*r */
 144             gmx_mm_pr  frcoul_SSE0;
 145             gmx_mm_pr  frcoul_SSE1;
 146             gmx_mm_pr  frcoul_SSE2;
 147             gmx_mm_pr  frcoul_SSE3;
 148 #ifdef CALC_COUL_TAB
 149             /* For tables: r, rs=r/sp, rf=floor(rs), frac=rs-rf */
 150             gmx_mm_pr  r_SSE0,rs_SSE0,rf_SSE0,frac_SSE0;
 151             gmx_mm_pr  r_SSE1,rs_SSE1,rf_SSE1,frac_SSE1;
 152             gmx_mm_pr  r_SSE2,rs_SSE2,rf_SSE2,frac_SSE2;
 153             gmx_mm_pr  r_SSE3,rs_SSE3,rf_SSE3,frac_SSE3;
 154             /* Table index: rs truncated to an int */
 155 #if !(defined GMX_MM256_HERE && defined GMX_DOUBLE)
 156             gmx_epi32  ti_SSE0,ti_SSE1,ti_SSE2,ti_SSE3;
 157 #else
 158             __m128i    ti_SSE0,ti_SSE1,ti_SSE2,ti_SSE3;
 159 #endif
 160             /* Linear force table values */
 161             gmx_mm_pr  ctab0_SSE0,ctab1_SSE0;
 162             gmx_mm_pr  ctab0_SSE1,ctab1_SSE1;
 163             gmx_mm_pr  ctab0_SSE2,ctab1_SSE2;
 164             gmx_mm_pr  ctab0_SSE3,ctab1_SSE3;
 165 #ifdef CALC_ENERGIES
 166             /* Quadratic energy table value */
 167             gmx_mm_pr  ctabv_SSE0;
 168             gmx_mm_pr  ctabv_SSE1;
 169             gmx_mm_pr  ctabv_SSE2;
 170             gmx_mm_pr  ctabv_SSE3;
 171 #endif
 172 #endif
 173 #if defined CALC_ENERGIES && (defined CALC_COUL_EWALD || defined CALC_COUL_TAB)
 174             /* The potential (PME mesh) we need to subtract from 1/r */
 175             gmx_mm_pr  vc_sub_SSE0;
 176             gmx_mm_pr  vc_sub_SSE1;
 177             gmx_mm_pr  vc_sub_SSE2;
 178             gmx_mm_pr  vc_sub_SSE3;
 179 #endif
 180 #ifdef CALC_ENERGIES
 181             /* Electrostatic potential */
 182             gmx_mm_pr  vcoul_SSE0;
 183             gmx_mm_pr  vcoul_SSE1;
 184             gmx_mm_pr  vcoul_SSE2;
 185             gmx_mm_pr  vcoul_SSE3;
 186 #endif
 187 #endif
 188             /* The force times 1/r */
 189             gmx_mm_pr  fscal_SSE0;
 190             gmx_mm_pr  fscal_SSE1;
 191             gmx_mm_pr  fscal_SSE2;
 192             gmx_mm_pr  fscal_SSE3;
 193
 194 #ifdef CALC_LJ
 195 #ifdef LJ_COMB_LB
 196             /* LJ sigma_j/2 and sqrt(epsilon_j) */
 197             gmx_mm_pr  hsig_j_SSE,seps_j_SSE;
 198             /* LJ sigma_ij and epsilon_ij */
 199             gmx_mm_pr  sig_SSE0,eps_SSE0;
 200             gmx_mm_pr  sig_SSE1,eps_SSE1;
 201 #ifndef HALF_LJ
 202             gmx_mm_pr  sig_SSE2,eps_SSE2;
 203             gmx_mm_pr  sig_SSE3,eps_SSE3;
 204 #endif
 205 #ifdef CALC_ENERGIES
 206             gmx_mm_pr  sig2_SSE0,sig6_SSE0;
 207             gmx_mm_pr  sig2_SSE1,sig6_SSE1;
 208 #ifndef HALF_LJ
 209             gmx_mm_pr  sig2_SSE2,sig6_SSE2;
 210             gmx_mm_pr  sig2_SSE3,sig6_SSE3;
 211 #endif
 212 #endif /* LJ_COMB_LB */
 213 #endif /* CALC_LJ */
 214
 215 #ifdef LJ_COMB_GEOM
 216             gmx_mm_pr  c6s_j_SSE,c12s_j_SSE;
 217 #endif
 218
 219 #if defined LJ_COMB_GEOM || defined LJ_COMB_LB
 220             /* Index for loading LJ parameters, complicated when interleaving */
 221             int         aj2;
 222 #endif
 223
 224 #ifndef FIX_LJ_C
 225             /* LJ C6 and C12 parameters, used with geometric comb. rule */
 226             gmx_mm_pr  c6_SSE0,c12_SSE0;
 227             gmx_mm_pr  c6_SSE1,c12_SSE1;
 228 #ifndef HALF_LJ
 229             gmx_mm_pr  c6_SSE2,c12_SSE2;
 230             gmx_mm_pr  c6_SSE3,c12_SSE3;
 231 #endif
 232 #endif
 233
 234             /* Intermediate variables for LJ calculation */
 235 #ifndef LJ_COMB_LB
 236             gmx_mm_pr  rinvsix_SSE0;
 237             gmx_mm_pr  rinvsix_SSE1;
 238 #ifndef HALF_LJ
 239             gmx_mm_pr  rinvsix_SSE2;
 240             gmx_mm_pr  rinvsix_SSE3;
 241 #endif
 242 #endif
 243 #ifdef LJ_COMB_LB
 244             gmx_mm_pr  sir_SSE0,sir2_SSE0,sir6_SSE0;
 245             gmx_mm_pr  sir_SSE1,sir2_SSE1,sir6_SSE1;
 246 #ifndef HALF_LJ
 247             gmx_mm_pr  sir_SSE2,sir2_SSE2,sir6_SSE2;
 248             gmx_mm_pr  sir_SSE3,sir2_SSE3,sir6_SSE3;
 249 #endif
 250 #endif
 251
 252             gmx_mm_pr  FrLJ6_SSE0,FrLJ12_SSE0;
 253             gmx_mm_pr  FrLJ6_SSE1,FrLJ12_SSE1;
 254 #ifndef HALF_LJ
 255             gmx_mm_pr  FrLJ6_SSE2,FrLJ12_SSE2;
 256             gmx_mm_pr  FrLJ6_SSE3,FrLJ12_SSE3;
 257 #endif
 258 #ifdef CALC_ENERGIES
 259             gmx_mm_pr  VLJ6_SSE0,VLJ12_SSE0,VLJ_SSE0;
 260             gmx_mm_pr  VLJ6_SSE1,VLJ12_SSE1,VLJ_SSE1;
 261 #ifndef HALF_LJ
 262             gmx_mm_pr  VLJ6_SSE2,VLJ12_SSE2,VLJ_SSE2;
 263             gmx_mm_pr  VLJ6_SSE3,VLJ12_SSE3,VLJ_SSE3;
 264 #endif
 265 #endif
 266 #endif /* CALC_LJ */
 267
 268             /* j-cluster index */
 269             cj            = l_cj[cjind].cj;
 270
 271             /* Atom indices (of the first atom in the cluster) */
 272             aj            = cj*UNROLLJ;
 273 #if defined CALC_LJ && (defined LJ_COMB_GEOM || defined LJ_COMB_LB)
 274 #if UNROLLJ == STRIDE
 275             aj2           = aj*2;
 276 #else
 277             aj2           = (cj>>1)*2*STRIDE + (cj & 1)*UNROLLJ;
 278 #endif
 279 #endif
 280 #if UNROLLJ == STRIDE
 281             ajx           = aj*DIM;
 282 #else
 283             ajx           = (cj>>1)*DIM*STRIDE + (cj & 1)*UNROLLJ;
 284 #endif
 285             ajy           = ajx + STRIDE;
 286             ajz           = ajy + STRIDE;
 287
 288 #ifdef CHECK_EXCLS
 289 #if defined GMX_X86_SSE2 && defined GMX_MM128_HERE
 290             {
 291                 /* Load integer interaction mask */
 292                 __m128i mask_int = _mm_set1_epi32(l_cj[cjind].excl);
 293
 294                 int_SSE0  = gmx_mm_castsi128_pr(_mm_cmpeq_epi32(_mm_andnot_si128(mask_int,mask0),zeroi_SSE));
 295                 int_SSE1  = gmx_mm_castsi128_pr(_mm_cmpeq_epi32(_mm_andnot_si128(mask_int,mask1),zeroi_SSE));
 296                 int_SSE2  = gmx_mm_castsi128_pr(_mm_cmpeq_epi32(_mm_andnot_si128(mask_int,mask2),zeroi_SSE));
 297                 int_SSE3  = gmx_mm_castsi128_pr(_mm_cmpeq_epi32(_mm_andnot_si128(mask_int,mask3),zeroi_SSE));
 298             }
 299 #endif
 300 #if defined GMX_X86_SSE2 && defined GMX_MM256_HERE
 301             {
 302 #ifndef GMX_DOUBLE
 303                 /* Load integer interaction mask */
 304                 /* With AVX there are no integer operations, so cast to real */
 305                 gmx_mm_pr mask_pr = gmx_mm_castsi256_pr(_mm256_set1_epi32(l_cj[cjind].excl));
 306                 /* We can't compare all 4*8=32 float bits: shift the mask */
 307                 gmx_mm_pr masksh_pr = gmx_mm_castsi256_pr(_mm256_set1_epi32(l_cj[cjind].excl>>(2*UNROLLJ)));
 308                 /* Intel Compiler version 12.1.3 20120130 is buggy: use cast.
 309                  * With gcc we don't need the cast, but it's faster.
 310                  */
 311 #define cast_cvt(x)  _mm256_cvtepi32_ps(_mm256_castps_si256(x))
 312                 int_SSE0  = gmx_cmpneq_pr(cast_cvt(gmx_and_pr(mask_pr,mask0)),zero_SSE);
 313                 int_SSE1  = gmx_cmpneq_pr(cast_cvt(gmx_and_pr(mask_pr,mask1)),zero_SSE);
 314                 int_SSE2  = gmx_cmpneq_pr(cast_cvt(gmx_and_pr(masksh_pr,mask0)),zero_SSE);
 315                 int_SSE3  = gmx_cmpneq_pr(cast_cvt(gmx_and_pr(masksh_pr,mask1)),zero_SSE);
 316 #undef cast_cvt
 317 #else
 318                 /* Load integer interaction mask */
 319                 /* With AVX there are no integer operations,
 320                  * and there is no int to double conversion, so cast to float
 321                  */
 322                 __m256 mask_ps = _mm256_castsi256_ps(_mm256_set1_epi32(l_cj[cjind].excl));
 323 #define cast_cvt(x)  _mm256_castps_pd(_mm256_cvtepi32_ps(_mm256_castps_si256(x)))
 324                 int_SSE0  = gmx_cmpneq_pr(cast_cvt(_mm256_and_ps(mask_ps,mask0)),zero_SSE);
 325                 int_SSE1  = gmx_cmpneq_pr(cast_cvt(_mm256_and_ps(mask_ps,mask1)),zero_SSE);
 326                 int_SSE2  = gmx_cmpneq_pr(cast_cvt(_mm256_and_ps(mask_ps,mask2)),zero_SSE);
 327                 int_SSE3  = gmx_cmpneq_pr(cast_cvt(_mm256_and_ps(mask_ps,mask3)),zero_SSE);
 328 #undef cast_cvt
 329 #endif
 330             }
 331 #endif
 332 #endif
 333             /* load j atom coordinates */
 334             jxSSE         = gmx_load_pr(x+ajx);
 335             jySSE         = gmx_load_pr(x+ajy);
 336             jzSSE         = gmx_load_pr(x+ajz);
 337
 338             /* Calculate distance */
 339             dx_SSE0       = gmx_sub_pr(ix_SSE0,jxSSE);
 340             dy_SSE0       = gmx_sub_pr(iy_SSE0,jySSE);
 341             dz_SSE0       = gmx_sub_pr(iz_SSE0,jzSSE);
 342             dx_SSE1       = gmx_sub_pr(ix_SSE1,jxSSE);
 343             dy_SSE1       = gmx_sub_pr(iy_SSE1,jySSE);
 344             dz_SSE1       = gmx_sub_pr(iz_SSE1,jzSSE);
 345             dx_SSE2       = gmx_sub_pr(ix_SSE2,jxSSE);
 346             dy_SSE2       = gmx_sub_pr(iy_SSE2,jySSE);
 347             dz_SSE2       = gmx_sub_pr(iz_SSE2,jzSSE);
 348             dx_SSE3       = gmx_sub_pr(ix_SSE3,jxSSE);
 349             dy_SSE3       = gmx_sub_pr(iy_SSE3,jySSE);
 350             dz_SSE3       = gmx_sub_pr(iz_SSE3,jzSSE);
 351
 352             /* rsq = dx*dx+dy*dy+dz*dz */
 353             rsq_SSE0      = gmx_calc_rsq_pr(dx_SSE0,dy_SSE0,dz_SSE0);
 354             rsq_SSE1      = gmx_calc_rsq_pr(dx_SSE1,dy_SSE1,dz_SSE1);
 355             rsq_SSE2      = gmx_calc_rsq_pr(dx_SSE2,dy_SSE2,dz_SSE2);
 356             rsq_SSE3      = gmx_calc_rsq_pr(dx_SSE3,dy_SSE3,dz_SSE3);
 357
 358 #ifndef CUTOFF_BLENDV
 359             wco_SSE0      = gmx_cmplt_pr(rsq_SSE0,rc2_SSE);
 360             wco_SSE1      = gmx_cmplt_pr(rsq_SSE1,rc2_SSE);
 361             wco_SSE2      = gmx_cmplt_pr(rsq_SSE2,rc2_SSE);
 362             wco_SSE3      = gmx_cmplt_pr(rsq_SSE3,rc2_SSE);
 363 #endif
 364
 365 #ifdef CHECK_EXCLS
 366 #ifdef EXCL_FORCES
 367             /* Only remove the (sub-)diagonal to avoid double counting */
 368 #if UNROLLJ == UNROLLI
 369             if (cj == ci_sh)
 370             {
 371                 wco_SSE0  = gmx_and_pr(wco_SSE0,diag_SSE0);
 372                 wco_SSE1  = gmx_and_pr(wco_SSE1,diag_SSE1);
 373                 wco_SSE2  = gmx_and_pr(wco_SSE2,diag_SSE2);
 374                 wco_SSE3  = gmx_and_pr(wco_SSE3,diag_SSE3);
 375             }
 376 #else
 377 #if UNROLLJ < UNROLLI
 378             if (cj == ci_sh*2)
 379             {
 380                 wco_SSE0  = gmx_and_pr(wco_SSE0,diag0_SSE0);
 381                 wco_SSE1  = gmx_and_pr(wco_SSE1,diag0_SSE1);
 382                 wco_SSE2  = gmx_and_pr(wco_SSE2,diag0_SSE2);
 383                 wco_SSE3  = gmx_and_pr(wco_SSE3,diag0_SSE3);
 384             }
 385             if (cj == ci_sh*2 + 1)
 386             {
 387                 wco_SSE0  = gmx_and_pr(wco_SSE0,diag1_SSE0);
 388                 wco_SSE1  = gmx_and_pr(wco_SSE1,diag1_SSE1);
 389                 wco_SSE2  = gmx_and_pr(wco_SSE2,diag1_SSE2);
 390                 wco_SSE3  = gmx_and_pr(wco_SSE3,diag1_SSE3);
 391             }
 392 #else
 393             if (cj*2 == ci_sh)
 394             {
 395                 wco_SSE0  = gmx_and_pr(wco_SSE0,diag0_SSE0);
 396                 wco_SSE1  = gmx_and_pr(wco_SSE1,diag0_SSE1);
 397                 wco_SSE2  = gmx_and_pr(wco_SSE2,diag0_SSE2);
 398                 wco_SSE3  = gmx_and_pr(wco_SSE3,diag0_SSE3);
 399             }
 400             else if (cj*2 + 1 == ci_sh)
 401             {
 402                 wco_SSE0  = gmx_and_pr(wco_SSE0,diag1_SSE0);
 403                 wco_SSE1  = gmx_and_pr(wco_SSE1,diag1_SSE1);
 404                 wco_SSE2  = gmx_and_pr(wco_SSE2,diag1_SSE2);
 405                 wco_SSE3  = gmx_and_pr(wco_SSE3,diag1_SSE3);
 406             }
 407 #endif
 408 #endif
 409 #else /* EXCL_FORCES */
 410             /* Remove all excluded atom pairs from the list */
 411             wco_SSE0      = gmx_and_pr(wco_SSE0,int_SSE0);
 412             wco_SSE1      = gmx_and_pr(wco_SSE1,int_SSE1);
 413             wco_SSE2      = gmx_and_pr(wco_SSE2,int_SSE2);
 414             wco_SSE3      = gmx_and_pr(wco_SSE3,int_SSE3);
 415 #endif
 416 #endif
 417
 418 #ifdef COUNT_PAIRS
 419             {
 420                 int i,j;
 421                 real tmp[UNROLLJ];
 422                 for(i=0; i<UNROLLI; i++)
 423                 {
 424                     gmx_storeu_pr(tmp,i==0 ? wco_SSE0 : (i==1 ? wco_SSE1 : (i==2 ? wco_SSE2 : wco_SSE3)));
 425                     for(j=0; j<UNROLLJ; j++)
 426                     {
 427                         if (!(tmp[j] == 0))
 428                         {
 429                             npair++;
 430                         }
 431                     }
 432                 }
 433             }
 434 #endif
 435
 436 #ifdef CHECK_EXCLS
 437             /* For excluded pairs add a small number to avoid r^-6 = NaN */
 438             rsq_SSE0      = gmx_add_pr(rsq_SSE0,gmx_andnot_pr(int_SSE0,avoid_sing_SSE));
 439             rsq_SSE1      = gmx_add_pr(rsq_SSE1,gmx_andnot_pr(int_SSE1,avoid_sing_SSE));
 440             rsq_SSE2      = gmx_add_pr(rsq_SSE2,gmx_andnot_pr(int_SSE2,avoid_sing_SSE));
 441             rsq_SSE3      = gmx_add_pr(rsq_SSE3,gmx_andnot_pr(int_SSE3,avoid_sing_SSE));
 442 #endif
 443
 444             /* Calculate 1/r */
 445 #ifndef GMX_DOUBLE
 446             rinv_SSE0     = gmx_invsqrt_pr(rsq_SSE0);
 447             rinv_SSE1     = gmx_invsqrt_pr(rsq_SSE1);
 448             rinv_SSE2     = gmx_invsqrt_pr(rsq_SSE2);
 449             rinv_SSE3     = gmx_invsqrt_pr(rsq_SSE3);
 450 #else
 451             GMX_MM_INVSQRT2_PD(rsq_SSE0,rsq_SSE1,rinv_SSE0,rinv_SSE1);
 452             GMX_MM_INVSQRT2_PD(rsq_SSE2,rsq_SSE3,rinv_SSE2,rinv_SSE3);
 453 #endif
 454
 455 #ifdef CALC_COULOMB
 456             /* Load parameters for j atom */
 457             jq_SSE        = gmx_load_pr(q+aj);
 458             qq_SSE0       = gmx_mul_pr(iq_SSE0,jq_SSE);
 459             qq_SSE1       = gmx_mul_pr(iq_SSE1,jq_SSE);
 460             qq_SSE2       = gmx_mul_pr(iq_SSE2,jq_SSE);
 461             qq_SSE3       = gmx_mul_pr(iq_SSE3,jq_SSE);
 462 #endif
 463
 464 #ifdef CALC_LJ
 465
 466 #if !defined LJ_COMB_GEOM && !defined LJ_COMB_LB && !defined FIX_LJ_C
 467             load_lj_pair_params(nbfp0,type,aj,c6_SSE0,c12_SSE0);
 468             load_lj_pair_params(nbfp1,type,aj,c6_SSE1,c12_SSE1);
 469 #ifndef HALF_LJ
 470             load_lj_pair_params(nbfp2,type,aj,c6_SSE2,c12_SSE2);
 471             load_lj_pair_params(nbfp3,type,aj,c6_SSE3,c12_SSE3);
 472 #endif
 473 #endif /* not defined any LJ rule */
 474
 475 #ifdef LJ_COMB_GEOM
 476             c6s_j_SSE     = gmx_load_pr(ljc+aj2+0);
 477             c12s_j_SSE    = gmx_load_pr(ljc+aj2+STRIDE);
 478             c6_SSE0       = gmx_mul_pr(c6s_SSE0 ,c6s_j_SSE );
 479             c6_SSE1       = gmx_mul_pr(c6s_SSE1 ,c6s_j_SSE );
 480 #ifndef HALF_LJ
 481             c6_SSE2       = gmx_mul_pr(c6s_SSE2 ,c6s_j_SSE );
 482             c6_SSE3       = gmx_mul_pr(c6s_SSE3 ,c6s_j_SSE );
 483 #endif
 484             c12_SSE0      = gmx_mul_pr(c12s_SSE0,c12s_j_SSE);
 485             c12_SSE1      = gmx_mul_pr(c12s_SSE1,c12s_j_SSE);
 486 #ifndef HALF_LJ
 487             c12_SSE2      = gmx_mul_pr(c12s_SSE2,c12s_j_SSE);
 488             c12_SSE3      = gmx_mul_pr(c12s_SSE3,c12s_j_SSE);
 489 #endif
 490 #endif /* LJ_COMB_GEOM */
 491
 492 #ifdef LJ_COMB_LB
 493             hsig_j_SSE    = gmx_load_pr(ljc+aj2+0);
 494             seps_j_SSE    = gmx_load_pr(ljc+aj2+STRIDE);
 495
 496             sig_SSE0      = gmx_add_pr(hsig_i_SSE0,hsig_j_SSE);
 497             sig_SSE1      = gmx_add_pr(hsig_i_SSE1,hsig_j_SSE);
 498             eps_SSE0      = gmx_mul_pr(seps_i_SSE0,seps_j_SSE);
 499             eps_SSE1      = gmx_mul_pr(seps_i_SSE1,seps_j_SSE);
 500 #ifndef HALF_LJ
 501             sig_SSE2      = gmx_add_pr(hsig_i_SSE2,hsig_j_SSE);
 502             sig_SSE3      = gmx_add_pr(hsig_i_SSE3,hsig_j_SSE);
 503             eps_SSE2      = gmx_mul_pr(seps_i_SSE2,seps_j_SSE);
 504             eps_SSE3      = gmx_mul_pr(seps_i_SSE3,seps_j_SSE);
 505 #endif
 506 #endif /* LJ_COMB_LB */
 507
 508 #endif /* CALC_LJ */
 509
 510 #ifndef CUTOFF_BLENDV
 511             rinv_SSE0     = gmx_and_pr(rinv_SSE0,wco_SSE0);
 512             rinv_SSE1     = gmx_and_pr(rinv_SSE1,wco_SSE1);
 513             rinv_SSE2     = gmx_and_pr(rinv_SSE2,wco_SSE2);
 514             rinv_SSE3     = gmx_and_pr(rinv_SSE3,wco_SSE3);
 515 #else
 516             /* We only need to mask for the cut-off: blendv is faster */
 517             rinv_SSE0     = gmx_blendv_pr(rinv_SSE0,zero_SSE,gmx_sub_pr(rc2_SSE,rsq_SSE0));
 518             rinv_SSE1     = gmx_blendv_pr(rinv_SSE1,zero_SSE,gmx_sub_pr(rc2_SSE,rsq_SSE1));
 519             rinv_SSE2     = gmx_blendv_pr(rinv_SSE2,zero_SSE,gmx_sub_pr(rc2_SSE,rsq_SSE2));
 520             rinv_SSE3     = gmx_blendv_pr(rinv_SSE3,zero_SSE,gmx_sub_pr(rc2_SSE,rsq_SSE3));
 521 #endif
 522
 523             rinvsq_SSE0   = gmx_mul_pr(rinv_SSE0,rinv_SSE0);
 524             rinvsq_SSE1   = gmx_mul_pr(rinv_SSE1,rinv_SSE1);
 525             rinvsq_SSE2   = gmx_mul_pr(rinv_SSE2,rinv_SSE2);
 526             rinvsq_SSE3   = gmx_mul_pr(rinv_SSE3,rinv_SSE3);
 527
 528 #ifdef CALC_COULOMB
 529             /* Note that here we calculate force*r, not the usual force/r.
 530              * This allows avoiding masking the reaction-field contribution,
 531              * as frcoul is later multiplied by rinvsq which has been
 532              * masked with the cut-off check.
 533              */
 534
 535 #ifdef EXCL_FORCES
 536             /* Only add 1/r for non-excluded atom pairs */
 537             rinv_ex_SSE0  = gmx_and_pr(rinv_SSE0,int_SSE0);
 538             rinv_ex_SSE1  = gmx_and_pr(rinv_SSE1,int_SSE1);
 539             rinv_ex_SSE2  = gmx_and_pr(rinv_SSE2,int_SSE2);
 540             rinv_ex_SSE3  = gmx_and_pr(rinv_SSE3,int_SSE3);
 541 #else
 542             /* No exclusion forces, we always need 1/r */
 543 #define     rinv_ex_SSE0    rinv_SSE0
 544 #define     rinv_ex_SSE1    rinv_SSE1
 545 #define     rinv_ex_SSE2    rinv_SSE2
 546 #define     rinv_ex_SSE3    rinv_SSE3
 547 #endif
 548
 549 #ifdef CALC_COUL_RF
 550             /* Electrostatic interactions */
 551             frcoul_SSE0   = gmx_mul_pr(qq_SSE0,gmx_add_pr(rinv_ex_SSE0,gmx_mul_pr(rsq_SSE0,mrc_3_SSE)));
 552             frcoul_SSE1   = gmx_mul_pr(qq_SSE1,gmx_add_pr(rinv_ex_SSE1,gmx_mul_pr(rsq_SSE1,mrc_3_SSE)));
 553             frcoul_SSE2   = gmx_mul_pr(qq_SSE2,gmx_add_pr(rinv_ex_SSE2,gmx_mul_pr(rsq_SSE2,mrc_3_SSE)));
 554             frcoul_SSE3   = gmx_mul_pr(qq_SSE3,gmx_add_pr(rinv_ex_SSE3,gmx_mul_pr(rsq_SSE3,mrc_3_SSE)));
 555
 556 #ifdef CALC_ENERGIES
 557             vcoul_SSE0    = gmx_mul_pr(qq_SSE0,gmx_add_pr(rinv_ex_SSE0,gmx_add_pr(gmx_mul_pr(rsq_SSE0,hrc_3_SSE),moh_rc_SSE)));
 558             vcoul_SSE1    = gmx_mul_pr(qq_SSE1,gmx_add_pr(rinv_ex_SSE1,gmx_add_pr(gmx_mul_pr(rsq_SSE1,hrc_3_SSE),moh_rc_SSE)));
 559             vcoul_SSE2    = gmx_mul_pr(qq_SSE2,gmx_add_pr(rinv_ex_SSE2,gmx_add_pr(gmx_mul_pr(rsq_SSE2,hrc_3_SSE),moh_rc_SSE)));
 560             vcoul_SSE3    = gmx_mul_pr(qq_SSE3,gmx_add_pr(rinv_ex_SSE3,gmx_add_pr(gmx_mul_pr(rsq_SSE3,hrc_3_SSE),moh_rc_SSE)));
 561 #endif
 562 #endif
 563
 564 #ifdef CALC_COUL_EWALD
 565             /* We need to mask (or limit) rsq for the cut-off,
 566              * as large distances can cause an overflow in gmx_pmecorrF/V.
 567              */
 568 #ifndef CUTOFF_BLENDV
 569             brsq_SSE0     = gmx_mul_pr(beta2_SSE,gmx_and_pr(rsq_SSE0,wco_SSE0));
 570             brsq_SSE1     = gmx_mul_pr(beta2_SSE,gmx_and_pr(rsq_SSE1,wco_SSE1));
 571             brsq_SSE2     = gmx_mul_pr(beta2_SSE,gmx_and_pr(rsq_SSE2,wco_SSE2));
 572             brsq_SSE3     = gmx_mul_pr(beta2_SSE,gmx_and_pr(rsq_SSE3,wco_SSE3));
 573 #else
 574             /* Strangely, putting mul on a separate line is slower (icc 13) */
 575             brsq_SSE0     = gmx_mul_pr(beta2_SSE,gmx_blendv_pr(rsq_SSE0,zero_SSE,gmx_sub_pr(rc2_SSE,rsq_SSE0)));
 576             brsq_SSE1     = gmx_mul_pr(beta2_SSE,gmx_blendv_pr(rsq_SSE1,zero_SSE,gmx_sub_pr(rc2_SSE,rsq_SSE1)));
 577             brsq_SSE2     = gmx_mul_pr(beta2_SSE,gmx_blendv_pr(rsq_SSE2,zero_SSE,gmx_sub_pr(rc2_SSE,rsq_SSE2)));
 578             brsq_SSE3     = gmx_mul_pr(beta2_SSE,gmx_blendv_pr(rsq_SSE3,zero_SSE,gmx_sub_pr(rc2_SSE,rsq_SSE3)));
 579 #endif
 580             ewcorr_SSE0   = gmx_mul_pr(gmx_pmecorrF_pr(brsq_SSE0),beta_SSE);
 581             ewcorr_SSE1   = gmx_mul_pr(gmx_pmecorrF_pr(brsq_SSE1),beta_SSE);
 582             ewcorr_SSE2   = gmx_mul_pr(gmx_pmecorrF_pr(brsq_SSE2),beta_SSE);
 583             ewcorr_SSE3   = gmx_mul_pr(gmx_pmecorrF_pr(brsq_SSE3),beta_SSE);
 584             frcoul_SSE0   = gmx_mul_pr(qq_SSE0,gmx_add_pr(rinv_ex_SSE0,gmx_mul_pr(ewcorr_SSE0,brsq_SSE0)));
 585             frcoul_SSE1   = gmx_mul_pr(qq_SSE1,gmx_add_pr(rinv_ex_SSE1,gmx_mul_pr(ewcorr_SSE1,brsq_SSE1)));
 586             frcoul_SSE2   = gmx_mul_pr(qq_SSE2,gmx_add_pr(rinv_ex_SSE2,gmx_mul_pr(ewcorr_SSE2,brsq_SSE2)));
 587             frcoul_SSE3   = gmx_mul_pr(qq_SSE3,gmx_add_pr(rinv_ex_SSE3,gmx_mul_pr(ewcorr_SSE3,brsq_SSE3)));
 588
 589 #ifdef CALC_ENERGIES
 590             vc_sub_SSE0   = gmx_mul_pr(gmx_pmecorrV_pr(brsq_SSE0),beta_SSE);
 591             vc_sub_SSE1   = gmx_mul_pr(gmx_pmecorrV_pr(brsq_SSE1),beta_SSE);
 592             vc_sub_SSE2   = gmx_mul_pr(gmx_pmecorrV_pr(brsq_SSE2),beta_SSE);
 593             vc_sub_SSE3   = gmx_mul_pr(gmx_pmecorrV_pr(brsq_SSE3),beta_SSE);
 594 #endif
 595
 596 #endif /* CALC_COUL_EWALD */
 597
 598 #ifdef CALC_COUL_TAB
 599             /* Electrostatic interactions */
 600             r_SSE0        = gmx_mul_pr(rsq_SSE0,rinv_SSE0);
 601             r_SSE1        = gmx_mul_pr(rsq_SSE1,rinv_SSE1);
 602             r_SSE2        = gmx_mul_pr(rsq_SSE2,rinv_SSE2);
 603             r_SSE3        = gmx_mul_pr(rsq_SSE3,rinv_SSE3);
 604             /* Convert r to scaled table units */
 605             rs_SSE0       = gmx_mul_pr(r_SSE0,invtsp_SSE);
 606             rs_SSE1       = gmx_mul_pr(r_SSE1,invtsp_SSE);
 607             rs_SSE2       = gmx_mul_pr(r_SSE2,invtsp_SSE);
 608             rs_SSE3       = gmx_mul_pr(r_SSE3,invtsp_SSE);
 609             /* Truncate scaled r to an int */
 610             ti_SSE0       = gmx_cvttpr_epi32(rs_SSE0);
 611             ti_SSE1       = gmx_cvttpr_epi32(rs_SSE1);
 612             ti_SSE2       = gmx_cvttpr_epi32(rs_SSE2);
 613             ti_SSE3       = gmx_cvttpr_epi32(rs_SSE3);
 614 #ifdef GMX_X86_SSE4_1
 615             /* SSE4.1 floor is faster than gmx_cvtepi32_ps int->float cast */
 616             rf_SSE0       = gmx_floor_pr(rs_SSE0);
 617             rf_SSE1       = gmx_floor_pr(rs_SSE1);
 618             rf_SSE2       = gmx_floor_pr(rs_SSE2);
 619             rf_SSE3       = gmx_floor_pr(rs_SSE3);
 620 #else
 621             rf_SSE0       = gmx_cvtepi32_pr(ti_SSE0);
 622             rf_SSE1       = gmx_cvtepi32_pr(ti_SSE1);
 623             rf_SSE2       = gmx_cvtepi32_pr(ti_SSE2);
 624             rf_SSE3       = gmx_cvtepi32_pr(ti_SSE3);
 625 #endif
 626             frac_SSE0     = gmx_sub_pr(rs_SSE0,rf_SSE0);
 627             frac_SSE1     = gmx_sub_pr(rs_SSE1,rf_SSE1);
 628             frac_SSE2     = gmx_sub_pr(rs_SSE2,rf_SSE2);
 629             frac_SSE3     = gmx_sub_pr(rs_SSE3,rf_SSE3);
 630
 631             /* Load and interpolate table forces and possibly energies.
 632              * Force and energy can be combined in one table, stride 4: FDV0
 633              * or in two separate tables with stride 1: F and V
 634              * Currently single precision uses FDV0, double F and V.
 635              */
 636 #ifndef CALC_ENERGIES
 637             load_table_f(tab_coul_F,ti_SSE0,ti0,ctab0_SSE0,ctab1_SSE0);
 638             load_table_f(tab_coul_F,ti_SSE1,ti1,ctab0_SSE1,ctab1_SSE1);
 639             load_table_f(tab_coul_F,ti_SSE2,ti2,ctab0_SSE2,ctab1_SSE2);
 640             load_table_f(tab_coul_F,ti_SSE3,ti3,ctab0_SSE3,ctab1_SSE3);
 641 #else
 642 #ifdef TAB_FDV0
 643             load_table_f_v(tab_coul_F,ti_SSE0,ti0,ctab0_SSE0,ctab1_SSE0,ctabv_SSE0);
 644             load_table_f_v(tab_coul_F,ti_SSE1,ti1,ctab0_SSE1,ctab1_SSE1,ctabv_SSE1);
 645             load_table_f_v(tab_coul_F,ti_SSE2,ti2,ctab0_SSE2,ctab1_SSE2,ctabv_SSE2);
 646             load_table_f_v(tab_coul_F,ti_SSE3,ti3,ctab0_SSE3,ctab1_SSE3,ctabv_SSE3);
 647 #else
 648             load_table_f_v(tab_coul_F,tab_coul_V,ti_SSE0,ti0,ctab0_SSE0,ctab1_SSE0,ctabv_SSE0);
 649             load_table_f_v(tab_coul_F,tab_coul_V,ti_SSE1,ti1,ctab0_SSE1,ctab1_SSE1,ctabv_SSE1);
 650             load_table_f_v(tab_coul_F,tab_coul_V,ti_SSE2,ti2,ctab0_SSE2,ctab1_SSE2,ctabv_SSE2);
 651             load_table_f_v(tab_coul_F,tab_coul_V,ti_SSE3,ti3,ctab0_SSE3,ctab1_SSE3,ctabv_SSE3);
 652 #endif
 653 #endif
 654             fsub_SSE0     = gmx_add_pr(ctab0_SSE0,gmx_mul_pr(frac_SSE0,ctab1_SSE0));
 655             fsub_SSE1     = gmx_add_pr(ctab0_SSE1,gmx_mul_pr(frac_SSE1,ctab1_SSE1));
 656             fsub_SSE2     = gmx_add_pr(ctab0_SSE2,gmx_mul_pr(frac_SSE2,ctab1_SSE2));
 657             fsub_SSE3     = gmx_add_pr(ctab0_SSE3,gmx_mul_pr(frac_SSE3,ctab1_SSE3));
 658             frcoul_SSE0   = gmx_mul_pr(qq_SSE0,gmx_sub_pr(rinv_ex_SSE0,gmx_mul_pr(fsub_SSE0,r_SSE0)));
 659             frcoul_SSE1   = gmx_mul_pr(qq_SSE1,gmx_sub_pr(rinv_ex_SSE1,gmx_mul_pr(fsub_SSE1,r_SSE1)));
 660             frcoul_SSE2   = gmx_mul_pr(qq_SSE2,gmx_sub_pr(rinv_ex_SSE2,gmx_mul_pr(fsub_SSE2,r_SSE2)));
 661             frcoul_SSE3   = gmx_mul_pr(qq_SSE3,gmx_sub_pr(rinv_ex_SSE3,gmx_mul_pr(fsub_SSE3,r_SSE3)));
 662
 663 #ifdef CALC_ENERGIES
 664             vc_sub_SSE0   = gmx_add_pr(ctabv_SSE0,gmx_mul_pr(gmx_mul_pr(mhalfsp_SSE,frac_SSE0),gmx_add_pr(ctab0_SSE0,fsub_SSE0)));
 665             vc_sub_SSE1   = gmx_add_pr(ctabv_SSE1,gmx_mul_pr(gmx_mul_pr(mhalfsp_SSE,frac_SSE1),gmx_add_pr(ctab0_SSE1,fsub_SSE1)));
 666             vc_sub_SSE2   = gmx_add_pr(ctabv_SSE2,gmx_mul_pr(gmx_mul_pr(mhalfsp_SSE,frac_SSE2),gmx_add_pr(ctab0_SSE2,fsub_SSE2)));
 667             vc_sub_SSE3   = gmx_add_pr(ctabv_SSE3,gmx_mul_pr(gmx_mul_pr(mhalfsp_SSE,frac_SSE3),gmx_add_pr(ctab0_SSE3,fsub_SSE3)));
 668 #endif
 669 #endif /* CALC_COUL_TAB */
 670
 671 #if defined CALC_ENERGIES && (defined CALC_COUL_EWALD || defined CALC_COUL_TAB)
 672 #ifndef NO_SHIFT_EWALD
 673             /* Add Ewald potential shift to vc_sub for convenience */
 674 #ifdef CHECK_EXCLS
 675             vc_sub_SSE0   = gmx_add_pr(vc_sub_SSE0,gmx_and_pr(sh_ewald_SSE,int_SSE0));
 676             vc_sub_SSE1   = gmx_add_pr(vc_sub_SSE1,gmx_and_pr(sh_ewald_SSE,int_SSE1));
 677             vc_sub_SSE2   = gmx_add_pr(vc_sub_SSE2,gmx_and_pr(sh_ewald_SSE,int_SSE2));
 678             vc_sub_SSE3   = gmx_add_pr(vc_sub_SSE3,gmx_and_pr(sh_ewald_SSE,int_SSE3));
 679 #else
 680             vc_sub_SSE0   = gmx_add_pr(vc_sub_SSE0,sh_ewald_SSE);
 681             vc_sub_SSE1   = gmx_add_pr(vc_sub_SSE1,sh_ewald_SSE);
 682             vc_sub_SSE2   = gmx_add_pr(vc_sub_SSE2,sh_ewald_SSE);
 683             vc_sub_SSE3   = gmx_add_pr(vc_sub_SSE3,sh_ewald_SSE);
 684 #endif
 685 #endif
 686
 687             vcoul_SSE0    = gmx_mul_pr(qq_SSE0,gmx_sub_pr(rinv_ex_SSE0,vc_sub_SSE0));
 688             vcoul_SSE1    = gmx_mul_pr(qq_SSE1,gmx_sub_pr(rinv_ex_SSE1,vc_sub_SSE1));
 689             vcoul_SSE2    = gmx_mul_pr(qq_SSE2,gmx_sub_pr(rinv_ex_SSE2,vc_sub_SSE2));
 690             vcoul_SSE3    = gmx_mul_pr(qq_SSE3,gmx_sub_pr(rinv_ex_SSE3,vc_sub_SSE3));
 691
 692 #endif
 693
 694 #ifdef CALC_ENERGIES
 695             /* Mask energy for cut-off and diagonal */
 696             vcoul_SSE0    = gmx_and_pr(vcoul_SSE0,wco_SSE0);
 697             vcoul_SSE1    = gmx_and_pr(vcoul_SSE1,wco_SSE1);
 698             vcoul_SSE2    = gmx_and_pr(vcoul_SSE2,wco_SSE2);
 699             vcoul_SSE3    = gmx_and_pr(vcoul_SSE3,wco_SSE3);
 700 #endif
 701
 702 #endif /* CALC_COULOMB */
 703
 704 #ifdef CALC_LJ
 705             /* Lennard-Jones interaction */
 706
 707 #ifdef VDW_CUTOFF_CHECK
 708             wco_vdw_SSE0  = gmx_cmplt_pr(rsq_SSE0,rcvdw2_SSE);
 709             wco_vdw_SSE1  = gmx_cmplt_pr(rsq_SSE1,rcvdw2_SSE);
 710 #ifndef HALF_LJ
 711             wco_vdw_SSE2  = gmx_cmplt_pr(rsq_SSE2,rcvdw2_SSE);
 712             wco_vdw_SSE3  = gmx_cmplt_pr(rsq_SSE3,rcvdw2_SSE);
 713 #endif
 714 #else
 715             /* Same cut-off for Coulomb and VdW, reuse the registers */
 716 #define     wco_vdw_SSE0    wco_SSE0
 717 #define     wco_vdw_SSE1    wco_SSE1
 718 #define     wco_vdw_SSE2    wco_SSE2
 719 #define     wco_vdw_SSE3    wco_SSE3
 720 #endif
 721
 722 #ifndef LJ_COMB_LB
 723             rinvsix_SSE0  = gmx_mul_pr(rinvsq_SSE0,gmx_mul_pr(rinvsq_SSE0,rinvsq_SSE0));
 724             rinvsix_SSE1  = gmx_mul_pr(rinvsq_SSE1,gmx_mul_pr(rinvsq_SSE1,rinvsq_SSE1));
 725 #ifdef EXCL_FORCES
 726             rinvsix_SSE0  = gmx_and_pr(rinvsix_SSE0,int_SSE0);
 727             rinvsix_SSE1  = gmx_and_pr(rinvsix_SSE1,int_SSE1);
 728 #endif
 729 #ifndef HALF_LJ
 730             rinvsix_SSE2  = gmx_mul_pr(rinvsq_SSE2,gmx_mul_pr(rinvsq_SSE2,rinvsq_SSE2));
 731             rinvsix_SSE3  = gmx_mul_pr(rinvsq_SSE3,gmx_mul_pr(rinvsq_SSE3,rinvsq_SSE3));
 732 #ifdef EXCL_FORCES
 733             rinvsix_SSE2  = gmx_and_pr(rinvsix_SSE2,int_SSE2);
 734             rinvsix_SSE3  = gmx_and_pr(rinvsix_SSE3,int_SSE3);
 735 #endif
 736 #endif
 737 #ifdef VDW_CUTOFF_CHECK
 738             rinvsix_SSE0  = gmx_and_pr(rinvsix_SSE0,wco_vdw_SSE0);
 739             rinvsix_SSE1  = gmx_and_pr(rinvsix_SSE1,wco_vdw_SSE1);
 740 #ifndef HALF_LJ
 741             rinvsix_SSE2  = gmx_and_pr(rinvsix_SSE2,wco_vdw_SSE2);
 742             rinvsix_SSE3  = gmx_and_pr(rinvsix_SSE3,wco_vdw_SSE3);
 743 #endif
 744 #endif
 745             FrLJ6_SSE0    = gmx_mul_pr(c6_SSE0,rinvsix_SSE0);
 746             FrLJ6_SSE1    = gmx_mul_pr(c6_SSE1,rinvsix_SSE1);
 747 #ifndef HALF_LJ
 748             FrLJ6_SSE2    = gmx_mul_pr(c6_SSE2,rinvsix_SSE2);
 749             FrLJ6_SSE3    = gmx_mul_pr(c6_SSE3,rinvsix_SSE3);
 750 #endif
 751             FrLJ12_SSE0   = gmx_mul_pr(c12_SSE0,gmx_mul_pr(rinvsix_SSE0,rinvsix_SSE0));
 752             FrLJ12_SSE1   = gmx_mul_pr(c12_SSE1,gmx_mul_pr(rinvsix_SSE1,rinvsix_SSE1));
 753 #ifndef HALF_LJ
 754             FrLJ12_SSE2   = gmx_mul_pr(c12_SSE2,gmx_mul_pr(rinvsix_SSE2,rinvsix_SSE2));
 755             FrLJ12_SSE3   = gmx_mul_pr(c12_SSE3,gmx_mul_pr(rinvsix_SSE3,rinvsix_SSE3));
 756 #endif
 757 #endif /* not LJ_COMB_LB */
 758
 759 #ifdef LJ_COMB_LB
 760             sir_SSE0      = gmx_mul_pr(sig_SSE0,rinv_SSE0);
 761             sir_SSE1      = gmx_mul_pr(sig_SSE1,rinv_SSE1);
 762 #ifndef HALF_LJ
 763             sir_SSE2      = gmx_mul_pr(sig_SSE2,rinv_SSE2);
 764             sir_SSE3      = gmx_mul_pr(sig_SSE3,rinv_SSE3);
 765 #endif
 766             sir2_SSE0     = gmx_mul_pr(sir_SSE0,sir_SSE0);
 767             sir2_SSE1     = gmx_mul_pr(sir_SSE1,sir_SSE1);
 768 #ifndef HALF_LJ
 769             sir2_SSE2     = gmx_mul_pr(sir_SSE2,sir_SSE2);
 770             sir2_SSE3     = gmx_mul_pr(sir_SSE3,sir_SSE3);
 771 #endif
 772             sir6_SSE0     = gmx_mul_pr(sir2_SSE0,gmx_mul_pr(sir2_SSE0,sir2_SSE0));
 773             sir6_SSE1     = gmx_mul_pr(sir2_SSE1,gmx_mul_pr(sir2_SSE1,sir2_SSE1));
 774 #ifdef EXCL_FORCES
 775             sir6_SSE0     = gmx_and_pr(sir6_SSE0,int_SSE0);
 776             sir6_SSE1     = gmx_and_pr(sir6_SSE1,int_SSE1);
 777 #endif
 778 #ifndef HALF_LJ
 779             sir6_SSE2     = gmx_mul_pr(sir2_SSE2,gmx_mul_pr(sir2_SSE2,sir2_SSE2));
 780             sir6_SSE3     = gmx_mul_pr(sir2_SSE3,gmx_mul_pr(sir2_SSE3,sir2_SSE3));
 781 #ifdef EXCL_FORCES
 782             sir6_SSE2     = gmx_and_pr(sir6_SSE2,int_SSE2);
 783             sir6_SSE3     = gmx_and_pr(sir6_SSE3,int_SSE3);
 784 #endif
 785 #endif
 786 #ifdef VDW_CUTOFF_CHECK
 787             sir6_SSE0     = gmx_and_pr(sir6_SSE0,wco_vdw_SSE0);
 788             sir6_SSE1     = gmx_and_pr(sir6_SSE1,wco_vdw_SSE1);
 789 #ifndef HALF_LJ
 790             sir6_SSE2     = gmx_and_pr(sir6_SSE2,wco_vdw_SSE2);
 791             sir6_SSE3     = gmx_and_pr(sir6_SSE3,wco_vdw_SSE3);
 792 #endif
 793 #endif
 794             FrLJ6_SSE0    = gmx_mul_pr(eps_SSE0,sir6_SSE0);
 795             FrLJ6_SSE1    = gmx_mul_pr(eps_SSE1,sir6_SSE1);
 796 #ifndef HALF_LJ
 797             FrLJ6_SSE2    = gmx_mul_pr(eps_SSE2,sir6_SSE2);
 798             FrLJ6_SSE3    = gmx_mul_pr(eps_SSE3,sir6_SSE3);
 799 #endif
 800             FrLJ12_SSE0   = gmx_mul_pr(FrLJ6_SSE0,sir6_SSE0);
 801             FrLJ12_SSE1   = gmx_mul_pr(FrLJ6_SSE1,sir6_SSE1);
 802 #ifndef HALF_LJ
 803             FrLJ12_SSE2   = gmx_mul_pr(FrLJ6_SSE2,sir6_SSE2);
 804             FrLJ12_SSE3   = gmx_mul_pr(FrLJ6_SSE3,sir6_SSE3);
 805 #endif
 806 #if defined CALC_ENERGIES
 807             /* We need C6 and C12 to calculate the LJ potential shift */
 808             sig2_SSE0     = gmx_mul_pr(sig_SSE0,sig_SSE0);
 809             sig2_SSE1     = gmx_mul_pr(sig_SSE1,sig_SSE1);
 810 #ifndef HALF_LJ
 811             sig2_SSE2     = gmx_mul_pr(sig_SSE2,sig_SSE2);
 812             sig2_SSE3     = gmx_mul_pr(sig_SSE3,sig_SSE3);
 813 #endif
 814             sig6_SSE0     = gmx_mul_pr(sig2_SSE0,gmx_mul_pr(sig2_SSE0,sig2_SSE0));
 815             sig6_SSE1     = gmx_mul_pr(sig2_SSE1,gmx_mul_pr(sig2_SSE1,sig2_SSE1));
 816 #ifndef HALF_LJ
 817             sig6_SSE2     = gmx_mul_pr(sig2_SSE2,gmx_mul_pr(sig2_SSE2,sig2_SSE2));
 818             sig6_SSE3     = gmx_mul_pr(sig2_SSE3,gmx_mul_pr(sig2_SSE3,sig2_SSE3));
 819 #endif
 820             c6_SSE0       = gmx_mul_pr(eps_SSE0,sig6_SSE0);
 821             c6_SSE1       = gmx_mul_pr(eps_SSE1,sig6_SSE1);
 822 #ifndef HALF_LJ
 823             c6_SSE2       = gmx_mul_pr(eps_SSE2,sig6_SSE2);
 824             c6_SSE3       = gmx_mul_pr(eps_SSE3,sig6_SSE3);
 825 #endif
 826             c12_SSE0      = gmx_mul_pr(c6_SSE0,sig6_SSE0);
 827             c12_SSE1      = gmx_mul_pr(c6_SSE1,sig6_SSE1);
 828 #ifndef HALF_LJ
 829             c12_SSE2      = gmx_mul_pr(c6_SSE2,sig6_SSE2);
 830             c12_SSE3      = gmx_mul_pr(c6_SSE3,sig6_SSE3);
 831 #endif
 832 #endif
 833 #endif /* LJ_COMB_LB */
 834
 835 #endif /* CALC_LJ */
 836
 837 #ifdef CALC_ENERGIES
 838 #ifdef ENERGY_GROUPS
 839             /* Extract the group pair index per j pair.
 840              * Energy groups are stored per i-cluster, so things get
 841              * complicated when the i- and j-cluster size don't match.
 842              */
 843             {
 844                 int egps_j;
 845 #if UNROLLJ == 2
 846                 egps_j    = nbat->energrp[cj>>1];
 847                 egp_jj[0] = ((egps_j >> ((cj & 1)*egps_jshift)) & egps_jmask)*egps_jstride;
 848 #else
 849                 /* We assume UNROLLI <= UNROLLJ */
 850                 int jdi;
 851                 for(jdi=0; jdi<UNROLLJ/UNROLLI; jdi++)
 852                 {
 853                     int jj;
 854                     egps_j = nbat->energrp[cj*(UNROLLJ/UNROLLI)+jdi];
 855                     for(jj=0; jj<(UNROLLI/2); jj++)
 856                     {
 857                         egp_jj[jdi*(UNROLLI/2)+jj] = ((egps_j >> (jj*egps_jshift)) & egps_jmask)*egps_jstride;
 858                     }
 859                 }
 860 #endif
 861             }
 862 #endif
 863
 864 #ifdef CALC_COULOMB
 865 #ifndef ENERGY_GROUPS
 866             vctotSSE      = gmx_add_pr(vctotSSE, gmx_sum4_pr(vcoul_SSE0,vcoul_SSE1,vcoul_SSE2,vcoul_SSE3));
 867 #else
 868             add_ener_grp(vcoul_SSE0,vctp[0],egp_jj);
 869             add_ener_grp(vcoul_SSE1,vctp[1],egp_jj);
 870             add_ener_grp(vcoul_SSE2,vctp[2],egp_jj);
 871             add_ener_grp(vcoul_SSE3,vctp[3],egp_jj);
 872 #endif
 873 #endif
 874
 875 #ifdef CALC_LJ
 876             /* Calculate the LJ energies */
 877             VLJ6_SSE0     = gmx_mul_pr(sixthSSE,gmx_sub_pr(FrLJ6_SSE0,gmx_mul_pr(c6_SSE0,sh_invrc6_SSE)));
 878             VLJ6_SSE1     = gmx_mul_pr(sixthSSE,gmx_sub_pr(FrLJ6_SSE1,gmx_mul_pr(c6_SSE1,sh_invrc6_SSE)));
 879 #ifndef HALF_LJ
 880             VLJ6_SSE2     = gmx_mul_pr(sixthSSE,gmx_sub_pr(FrLJ6_SSE2,gmx_mul_pr(c6_SSE2,sh_invrc6_SSE)));
 881             VLJ6_SSE3     = gmx_mul_pr(sixthSSE,gmx_sub_pr(FrLJ6_SSE3,gmx_mul_pr(c6_SSE3,sh_invrc6_SSE)));
 882 #endif
 883             VLJ12_SSE0    = gmx_mul_pr(twelvethSSE,gmx_sub_pr(FrLJ12_SSE0,gmx_mul_pr(c12_SSE0,sh_invrc12_SSE)));
 884             VLJ12_SSE1    = gmx_mul_pr(twelvethSSE,gmx_sub_pr(FrLJ12_SSE1,gmx_mul_pr(c12_SSE1,sh_invrc12_SSE)));
 885 #ifndef HALF_LJ
 886             VLJ12_SSE2    = gmx_mul_pr(twelvethSSE,gmx_sub_pr(FrLJ12_SSE2,gmx_mul_pr(c12_SSE2,sh_invrc12_SSE)));
 887             VLJ12_SSE3    = gmx_mul_pr(twelvethSSE,gmx_sub_pr(FrLJ12_SSE3,gmx_mul_pr(c12_SSE3,sh_invrc12_SSE)));
 888 #endif
 889
 890             VLJ_SSE0      = gmx_sub_pr(VLJ12_SSE0,VLJ6_SSE0);
 891             VLJ_SSE1      = gmx_sub_pr(VLJ12_SSE1,VLJ6_SSE1);
 892 #ifndef HALF_LJ
 893             VLJ_SSE2      = gmx_sub_pr(VLJ12_SSE2,VLJ6_SSE2);
 894             VLJ_SSE3      = gmx_sub_pr(VLJ12_SSE3,VLJ6_SSE3);
 895 #endif
 896             /* The potential shift should be removed for pairs beyond cut-off */
 897             VLJ_SSE0      = gmx_and_pr(VLJ_SSE0,wco_vdw_SSE0);
 898             VLJ_SSE1      = gmx_and_pr(VLJ_SSE1,wco_vdw_SSE1);
 899 #ifndef HALF_LJ
 900             VLJ_SSE2      = gmx_and_pr(VLJ_SSE2,wco_vdw_SSE2);
 901             VLJ_SSE3      = gmx_and_pr(VLJ_SSE3,wco_vdw_SSE3);
 902 #endif
 903 #ifdef CHECK_EXCLS
 904             /* The potential shift should be removed for excluded pairs */
 905             VLJ_SSE0      = gmx_and_pr(VLJ_SSE0,int_SSE0);
 906             VLJ_SSE1      = gmx_and_pr(VLJ_SSE1,int_SSE1);
 907 #ifndef HALF_LJ
 908             VLJ_SSE2      = gmx_and_pr(VLJ_SSE2,int_SSE2);
 909             VLJ_SSE3      = gmx_and_pr(VLJ_SSE3,int_SSE3);
 910 #endif
 911 #endif
 912 #ifndef ENERGY_GROUPS
 913             VvdwtotSSE    = gmx_add_pr(VvdwtotSSE,
 914 #ifndef HALF_LJ
 915                                        gmx_sum4_pr(VLJ_SSE0,VLJ_SSE1,VLJ_SSE2,VLJ_SSE3)
 916 #else
 917                                        gmx_add_pr(VLJ_SSE0,VLJ_SSE1)
 918 #endif
 919                                       );
 920 #else
 921             add_ener_grp(VLJ_SSE0,vvdwtp[0],egp_jj);
 922             add_ener_grp(VLJ_SSE1,vvdwtp[1],egp_jj);
 923 #ifndef HALF_LJ
 924             add_ener_grp(VLJ_SSE2,vvdwtp[2],egp_jj);
 925             add_ener_grp(VLJ_SSE3,vvdwtp[3],egp_jj);
 926 #endif
 927 #endif
 928 #endif /* CALC_LJ */
 929 #endif /* CALC_ENERGIES */
 930
 931 #ifdef CALC_LJ
 932             fscal_SSE0    = gmx_mul_pr(rinvsq_SSE0,
 933 #ifdef CALC_COULOMB
 934                                                    gmx_add_pr(frcoul_SSE0,
 935 #else
 936                                                    (
 937 #endif
 938                                                     gmx_sub_pr(FrLJ12_SSE0,FrLJ6_SSE0)));
 939             fscal_SSE1    = gmx_mul_pr(rinvsq_SSE1,
 940 #ifdef CALC_COULOMB
 941                                                    gmx_add_pr(frcoul_SSE1,
 942 #else
 943                                                    (
 944 #endif
 945                                                     gmx_sub_pr(FrLJ12_SSE1,FrLJ6_SSE1)));
 946 #else
 947             fscal_SSE0    = gmx_mul_pr(rinvsq_SSE0,frcoul_SSE0);
 948             fscal_SSE1    = gmx_mul_pr(rinvsq_SSE1,frcoul_SSE1);
 949 #endif /* CALC_LJ */
 950 #if defined CALC_LJ && !defined HALF_LJ
 951             fscal_SSE2    = gmx_mul_pr(rinvsq_SSE2,
 952 #ifdef CALC_COULOMB
 953                                                    gmx_add_pr(frcoul_SSE2,
 954 #else
 955                                                    (
 956 #endif
 957                                                     gmx_sub_pr(FrLJ12_SSE2,FrLJ6_SSE2)));
 958             fscal_SSE3    = gmx_mul_pr(rinvsq_SSE3,
 959 #ifdef CALC_COULOMB
 960                                                    gmx_add_pr(frcoul_SSE3,
 961 #else
 962                                                    (
 963 #endif
 964                                                     gmx_sub_pr(FrLJ12_SSE3,FrLJ6_SSE3)));
 965 #else
 966             /* Atom 2 and 3 don't have LJ, so only add Coulomb forces */
 967             fscal_SSE2    = gmx_mul_pr(rinvsq_SSE2,frcoul_SSE2);
 968             fscal_SSE3    = gmx_mul_pr(rinvsq_SSE3,frcoul_SSE3);
 969 #endif
 970
 971             /* Calculate temporary vectorial force */
 972             tx_SSE0       = gmx_mul_pr(fscal_SSE0,dx_SSE0);
 973             tx_SSE1       = gmx_mul_pr(fscal_SSE1,dx_SSE1);
 974             tx_SSE2       = gmx_mul_pr(fscal_SSE2,dx_SSE2);
 975             tx_SSE3       = gmx_mul_pr(fscal_SSE3,dx_SSE3);
 976             ty_SSE0       = gmx_mul_pr(fscal_SSE0,dy_SSE0);
 977             ty_SSE1       = gmx_mul_pr(fscal_SSE1,dy_SSE1);
 978             ty_SSE2       = gmx_mul_pr(fscal_SSE2,dy_SSE2);
 979             ty_SSE3       = gmx_mul_pr(fscal_SSE3,dy_SSE3);
 980             tz_SSE0       = gmx_mul_pr(fscal_SSE0,dz_SSE0);
 981             tz_SSE1       = gmx_mul_pr(fscal_SSE1,dz_SSE1);
 982             tz_SSE2       = gmx_mul_pr(fscal_SSE2,dz_SSE2);
 983             tz_SSE3       = gmx_mul_pr(fscal_SSE3,dz_SSE3);
 984
 985             /* Increment i atom force */
 986             fix_SSE0      = gmx_add_pr(fix_SSE0,tx_SSE0);
 987             fix_SSE1      = gmx_add_pr(fix_SSE1,tx_SSE1);
 988             fix_SSE2      = gmx_add_pr(fix_SSE2,tx_SSE2);
 989             fix_SSE3      = gmx_add_pr(fix_SSE3,tx_SSE3);
 990             fiy_SSE0      = gmx_add_pr(fiy_SSE0,ty_SSE0);
 991             fiy_SSE1      = gmx_add_pr(fiy_SSE1,ty_SSE1);
 992             fiy_SSE2      = gmx_add_pr(fiy_SSE2,ty_SSE2);
 993             fiy_SSE3      = gmx_add_pr(fiy_SSE3,ty_SSE3);
 994             fiz_SSE0      = gmx_add_pr(fiz_SSE0,tz_SSE0);
 995             fiz_SSE1      = gmx_add_pr(fiz_SSE1,tz_SSE1);
 996             fiz_SSE2      = gmx_add_pr(fiz_SSE2,tz_SSE2);
 997             fiz_SSE3      = gmx_add_pr(fiz_SSE3,tz_SSE3);
 998
 999             /* Decrement j atom force */
1000             gmx_store_pr(f+ajx,
1001                          gmx_sub_pr( gmx_load_pr(f+ajx), gmx_sum4_pr(tx_SSE0,tx_SSE1,tx_SSE2,tx_SSE3) ));
1002             gmx_store_pr(f+ajy,
1003                          gmx_sub_pr( gmx_load_pr(f+ajy), gmx_sum4_pr(ty_SSE0,ty_SSE1,ty_SSE2,ty_SSE3) ));
1004             gmx_store_pr(f+ajz,
1005                          gmx_sub_pr( gmx_load_pr(f+ajz), gmx_sum4_pr(tz_SSE0,tz_SSE1,tz_SSE2,tz_SSE3) ));
1006         }
1007
1008 #undef  rinv_ex_SSE0
1009 #undef  rinv_ex_SSE1
1010 #undef  rinv_ex_SSE2
1011 #undef  rinv_ex_SSE3
1012
1013 #undef  wco_vdw_SSE0
1014 #undef  wco_vdw_SSE1
1015 #undef  wco_vdw_SSE2
1016 #undef  wco_vdw_SSE3
1017
1018 #undef  CUTOFF_BLENDV
1019
1020 #undef  EXCL_FORCES