| File: | gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_sse4_1_single.c |
| Location: | line 281, column 13 |
| Description: | Value stored to 'r00' is never read |
| 1 | /* |
| 2 | * This file is part of the GROMACS molecular simulation package. |
| 3 | * |
| 4 | * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by |
| 5 | * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl, |
| 6 | * and including many others, as listed in the AUTHORS file in the |
| 7 | * top-level source directory and at http://www.gromacs.org. |
| 8 | * |
| 9 | * GROMACS is free software; you can redistribute it and/or |
| 10 | * modify it under the terms of the GNU Lesser General Public License |
| 11 | * as published by the Free Software Foundation; either version 2.1 |
| 12 | * of the License, or (at your option) any later version. |
| 13 | * |
| 14 | * GROMACS is distributed in the hope that it will be useful, |
| 15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 17 | * Lesser General Public License for more details. |
| 18 | * |
| 19 | * You should have received a copy of the GNU Lesser General Public |
| 20 | * License along with GROMACS; if not, see |
| 21 | * http://www.gnu.org/licenses, or write to the Free Software Foundation, |
| 22 | * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. |
| 23 | * |
| 24 | * If you want to redistribute modifications to GROMACS, please |
| 25 | * consider that scientific software is very special. Version |
| 26 | * control is crucial - bugs must be traceable. We will be happy to |
| 27 | * consider code for inclusion in the official distribution, but |
| 28 | * derived work must not be called official GROMACS. Details are found |
| 29 | * in the README & COPYING files - if they are missing, get the |
| 30 | * official version at http://www.gromacs.org. |
| 31 | * |
| 32 | * To help us fund GROMACS development, we humbly ask that you cite |
| 33 | * the research papers on the package. Check out http://www.gromacs.org. |
| 34 | */ |
| 35 | /* |
| 36 | * Note: this file was generated by the GROMACS sse4_1_single kernel generator. |
| 37 | */ |
| 38 | #ifdef HAVE_CONFIG_H1 |
| 39 | #include <config.h> |
| 40 | #endif |
| 41 | |
| 42 | #include <math.h> |
| 43 | |
| 44 | #include "../nb_kernel.h" |
| 45 | #include "types/simple.h" |
| 46 | #include "gromacs/math/vec.h" |
| 47 | #include "nrnb.h" |
| 48 | |
| 49 | #include "gromacs/simd/math_x86_sse4_1_single.h" |
| 50 | #include "kernelutil_x86_sse4_1_single.h" |
| 51 | |
| 52 | /* |
| 53 | * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse4_1_single |
| 54 | * Electrostatics interaction: Ewald |
| 55 | * VdW interaction: LJEwald |
| 56 | * Geometry: Water4-Particle |
| 57 | * Calculate force/pot: PotentialAndForce |
| 58 | */ |
| 59 | void |
| 60 | nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_sse4_1_single |
| 61 | (t_nblist * gmx_restrict nlist, |
| 62 | rvec * gmx_restrict xx, |
| 63 | rvec * gmx_restrict ff, |
| 64 | t_forcerec * gmx_restrict fr, |
| 65 | t_mdatoms * gmx_restrict mdatoms, |
| 66 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
| 67 | t_nrnb * gmx_restrict nrnb) |
| 68 | { |
| 69 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
| 70 | * just 0 for non-waters. |
| 71 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
| 72 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
| 73 | */ |
| 74 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
| 75 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
| 76 | int jnrA,jnrB,jnrC,jnrD; |
| 77 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
| 78 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
| 79 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
| 80 | real rcutoff_scalar; |
| 81 | real *shiftvec,*fshift,*x,*f; |
| 82 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
| 83 | real scratch[4*DIM3]; |
| 84 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
| 85 | int vdwioffset0; |
| 86 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
| 87 | int vdwioffset1; |
| 88 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
| 89 | int vdwioffset2; |
| 90 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
| 91 | int vdwioffset3; |
| 92 | __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3; |
| 93 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
| 94 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
| 95 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
| 96 | __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10; |
| 97 | __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20; |
| 98 | __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30; |
| 99 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
| 100 | real *charge; |
| 101 | int nvdwtype; |
| 102 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
| 103 | int *vdwtype; |
| 104 | real *vdwparam; |
| 105 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
| 106 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
| 107 | __m128 c6grid_00; |
| 108 | __m128 c6grid_10; |
| 109 | __m128 c6grid_20; |
| 110 | __m128 c6grid_30; |
| 111 | __m128 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald; |
| 112 | real *vdwgridparam; |
| 113 | __m128 one_half = _mm_set1_ps(0.5); |
| 114 | __m128 minus_one = _mm_set1_ps(-1.0); |
| 115 | __m128i ewitab; |
| 116 | __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV; |
| 117 | real *ewtab; |
| 118 | __m128 dummy_mask,cutoff_mask; |
| 119 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
| 120 | __m128 one = _mm_set1_ps(1.0); |
| 121 | __m128 two = _mm_set1_ps(2.0); |
| 122 | x = xx[0]; |
| 123 | f = ff[0]; |
| 124 | |
| 125 | nri = nlist->nri; |
| 126 | iinr = nlist->iinr; |
| 127 | jindex = nlist->jindex; |
| 128 | jjnr = nlist->jjnr; |
| 129 | shiftidx = nlist->shift; |
| 130 | gid = nlist->gid; |
| 131 | shiftvec = fr->shift_vec[0]; |
| 132 | fshift = fr->fshift[0]; |
| 133 | facel = _mm_set1_ps(fr->epsfac); |
| 134 | charge = mdatoms->chargeA; |
| 135 | nvdwtype = fr->ntype; |
| 136 | vdwparam = fr->nbfp; |
| 137 | vdwtype = mdatoms->typeA; |
| 138 | vdwgridparam = fr->ljpme_c6grid; |
| 139 | sh_lj_ewald = _mm_set1_ps(fr->ic->sh_lj_ewald); |
| 140 | ewclj = _mm_set1_ps(fr->ewaldcoeff_lj); |
| 141 | ewclj2 = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj)); |
| 142 | |
| 143 | sh_ewald = _mm_set1_ps(fr->ic->sh_ewald); |
| 144 | ewtab = fr->ic->tabq_coul_FDV0; |
| 145 | ewtabscale = _mm_set1_ps(fr->ic->tabq_scale); |
| 146 | ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale); |
| 147 | |
| 148 | /* Setup water-specific parameters */ |
| 149 | inr = nlist->iinr[0]; |
| 150 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
| 151 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
| 152 | iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3])); |
| 153 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
| 154 | |
| 155 | /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */ |
| 156 | rcutoff_scalar = fr->rcoulomb; |
| 157 | rcutoff = _mm_set1_ps(rcutoff_scalar); |
| 158 | rcutoff2 = _mm_mul_ps(rcutoff,rcutoff); |
| 159 | |
| 160 | sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6); |
| 161 | rvdw = _mm_set1_ps(fr->rvdw); |
| 162 | |
| 163 | /* Avoid stupid compiler warnings */ |
| 164 | jnrA = jnrB = jnrC = jnrD = 0; |
| 165 | j_coord_offsetA = 0; |
| 166 | j_coord_offsetB = 0; |
| 167 | j_coord_offsetC = 0; |
| 168 | j_coord_offsetD = 0; |
| 169 | |
| 170 | outeriter = 0; |
| 171 | inneriter = 0; |
| 172 | |
| 173 | for(iidx=0;iidx<4*DIM3;iidx++) |
| 174 | { |
| 175 | scratch[iidx] = 0.0; |
| 176 | } |
| 177 | |
| 178 | /* Start outer loop over neighborlists */ |
| 179 | for(iidx=0; iidx<nri; iidx++) |
| 180 | { |
| 181 | /* Load shift vector for this list */ |
| 182 | i_shift_offset = DIM3*shiftidx[iidx]; |
| 183 | |
| 184 | /* Load limits for loop over neighbors */ |
| 185 | j_index_start = jindex[iidx]; |
| 186 | j_index_end = jindex[iidx+1]; |
| 187 | |
| 188 | /* Get outer coordinate index */ |
| 189 | inr = iinr[iidx]; |
| 190 | i_coord_offset = DIM3*inr; |
| 191 | |
| 192 | /* Load i particle coords and add shift vector */ |
| 193 | gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
| 194 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3); |
| 195 | |
| 196 | fix0 = _mm_setzero_ps(); |
| 197 | fiy0 = _mm_setzero_ps(); |
| 198 | fiz0 = _mm_setzero_ps(); |
| 199 | fix1 = _mm_setzero_ps(); |
| 200 | fiy1 = _mm_setzero_ps(); |
| 201 | fiz1 = _mm_setzero_ps(); |
| 202 | fix2 = _mm_setzero_ps(); |
| 203 | fiy2 = _mm_setzero_ps(); |
| 204 | fiz2 = _mm_setzero_ps(); |
| 205 | fix3 = _mm_setzero_ps(); |
| 206 | fiy3 = _mm_setzero_ps(); |
| 207 | fiz3 = _mm_setzero_ps(); |
| 208 | |
| 209 | /* Reset potential sums */ |
| 210 | velecsum = _mm_setzero_ps(); |
| 211 | vvdwsum = _mm_setzero_ps(); |
| 212 | |
| 213 | /* Start inner kernel loop */ |
| 214 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
| 215 | { |
| 216 | |
| 217 | /* Get j neighbor index, and coordinate index */ |
| 218 | jnrA = jjnr[jidx]; |
| 219 | jnrB = jjnr[jidx+1]; |
| 220 | jnrC = jjnr[jidx+2]; |
| 221 | jnrD = jjnr[jidx+3]; |
| 222 | j_coord_offsetA = DIM3*jnrA; |
| 223 | j_coord_offsetB = DIM3*jnrB; |
| 224 | j_coord_offsetC = DIM3*jnrC; |
| 225 | j_coord_offsetD = DIM3*jnrD; |
| 226 | |
| 227 | /* load j atom coordinates */ |
| 228 | gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
| 229 | x+j_coord_offsetC,x+j_coord_offsetD, |
| 230 | &jx0,&jy0,&jz0); |
| 231 | |
| 232 | /* Calculate displacement vector */ |
| 233 | dx00 = _mm_sub_ps(ix0,jx0); |
| 234 | dy00 = _mm_sub_ps(iy0,jy0); |
| 235 | dz00 = _mm_sub_ps(iz0,jz0); |
| 236 | dx10 = _mm_sub_ps(ix1,jx0); |
| 237 | dy10 = _mm_sub_ps(iy1,jy0); |
| 238 | dz10 = _mm_sub_ps(iz1,jz0); |
| 239 | dx20 = _mm_sub_ps(ix2,jx0); |
| 240 | dy20 = _mm_sub_ps(iy2,jy0); |
| 241 | dz20 = _mm_sub_ps(iz2,jz0); |
| 242 | dx30 = _mm_sub_ps(ix3,jx0); |
| 243 | dy30 = _mm_sub_ps(iy3,jy0); |
| 244 | dz30 = _mm_sub_ps(iz3,jz0); |
| 245 | |
| 246 | /* Calculate squared distance and things based on it */ |
| 247 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
| 248 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
| 249 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
| 250 | rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30); |
| 251 | |
| 252 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
| 253 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
| 254 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
| 255 | rinv30 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq30); |
| 256 | |
| 257 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
| 258 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
| 259 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
| 260 | rinvsq30 = _mm_mul_ps(rinv30,rinv30); |
| 261 | |
| 262 | /* Load parameters for j particles */ |
| 263 | jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0, |
| 264 | charge+jnrC+0,charge+jnrD+0); |
| 265 | vdwjidx0A = 2*vdwtype[jnrA+0]; |
| 266 | vdwjidx0B = 2*vdwtype[jnrB+0]; |
| 267 | vdwjidx0C = 2*vdwtype[jnrC+0]; |
| 268 | vdwjidx0D = 2*vdwtype[jnrD+0]; |
| 269 | |
| 270 | fjx0 = _mm_setzero_ps(); |
| 271 | fjy0 = _mm_setzero_ps(); |
| 272 | fjz0 = _mm_setzero_ps(); |
| 273 | |
| 274 | /************************** |
| 275 | * CALCULATE INTERACTIONS * |
| 276 | **************************/ |
| 277 | |
| 278 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
| 279 | { |
| 280 | |
| 281 | r00 = _mm_mul_ps(rsq00,rinv00); |
Value stored to 'r00' is never read | |
| 282 | |
| 283 | /* Compute parameters for interactions between i and j atoms */ |
| 284 | gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A, |
| 285 | vdwparam+vdwioffset0+vdwjidx0B, |
| 286 | vdwparam+vdwioffset0+vdwjidx0C, |
| 287 | vdwparam+vdwioffset0+vdwjidx0D, |
| 288 | &c6_00,&c12_00); |
| 289 | |
| 290 | c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A, |
| 291 | vdwgridparam+vdwioffset0+vdwjidx0B, |
| 292 | vdwgridparam+vdwioffset0+vdwjidx0C, |
| 293 | vdwgridparam+vdwioffset0+vdwjidx0D); |
| 294 | |
| 295 | /* Analytical LJ-PME */ |
| 296 | rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00); |
| 297 | ewcljrsq = _mm_mul_ps(ewclj2,rsq00); |
| 298 | ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2)); |
| 299 | exponent = gmx_simd_exp_rgmx_simd_exp_f(ewcljrsq); |
| 300 | /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */ |
| 301 | poly = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half))); |
| 302 | /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */ |
| 303 | vvdw6 = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix); |
| 304 | vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix)); |
| 305 | vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth), |
| 306 | _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth)); |
| 307 | /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */ |
| 308 | fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00); |
| 309 | |
| 310 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
| 311 | |
| 312 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 313 | vvdw = _mm_and_ps(vvdw,cutoff_mask); |
| 314 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
| 315 | |
| 316 | fscal = fvdw; |
| 317 | |
| 318 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 319 | |
| 320 | /* Calculate temporary vectorial force */ |
| 321 | tx = _mm_mul_ps(fscal,dx00); |
| 322 | ty = _mm_mul_ps(fscal,dy00); |
| 323 | tz = _mm_mul_ps(fscal,dz00); |
| 324 | |
| 325 | /* Update vectorial force */ |
| 326 | fix0 = _mm_add_ps(fix0,tx); |
| 327 | fiy0 = _mm_add_ps(fiy0,ty); |
| 328 | fiz0 = _mm_add_ps(fiz0,tz); |
| 329 | |
| 330 | fjx0 = _mm_add_ps(fjx0,tx); |
| 331 | fjy0 = _mm_add_ps(fjy0,ty); |
| 332 | fjz0 = _mm_add_ps(fjz0,tz); |
| 333 | |
| 334 | } |
| 335 | |
| 336 | /************************** |
| 337 | * CALCULATE INTERACTIONS * |
| 338 | **************************/ |
| 339 | |
| 340 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
| 341 | { |
| 342 | |
| 343 | r10 = _mm_mul_ps(rsq10,rinv10); |
| 344 | |
| 345 | /* Compute parameters for interactions between i and j atoms */ |
| 346 | qq10 = _mm_mul_ps(iq1,jq0); |
| 347 | |
| 348 | /* EWALD ELECTROSTATICS */ |
| 349 | |
| 350 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 351 | ewrt = _mm_mul_ps(r10,ewtabscale); |
| 352 | ewitab = _mm_cvttps_epi32(ewrt); |
| 353 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 354 | ewitab = _mm_slli_epi32(ewitab,2); |
| 355 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 356 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 357 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 358 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 359 | _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((ewtabF ), (ewtabD)); tmp2 = _mm_unpacklo_ps((ewtabV), (ewtabFn)); tmp1 = _mm_unpackhi_ps((ewtabF), (ewtabD)); tmp3 = _mm_unpackhi_ps ((ewtabV), (ewtabFn)); (ewtabF) = _mm_movelh_ps(tmp0, tmp2); ( ewtabD) = _mm_movehl_ps(tmp2, tmp0); (ewtabV) = _mm_movelh_ps (tmp1, tmp3); (ewtabFn) = _mm_movehl_ps(tmp3, tmp1); } while ( 0); |
| 360 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 361 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 362 | velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec)); |
| 363 | felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec)); |
| 364 | |
| 365 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
| 366 | |
| 367 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 368 | velec = _mm_and_ps(velec,cutoff_mask); |
| 369 | velecsum = _mm_add_ps(velecsum,velec); |
| 370 | |
| 371 | fscal = felec; |
| 372 | |
| 373 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 374 | |
| 375 | /* Calculate temporary vectorial force */ |
| 376 | tx = _mm_mul_ps(fscal,dx10); |
| 377 | ty = _mm_mul_ps(fscal,dy10); |
| 378 | tz = _mm_mul_ps(fscal,dz10); |
| 379 | |
| 380 | /* Update vectorial force */ |
| 381 | fix1 = _mm_add_ps(fix1,tx); |
| 382 | fiy1 = _mm_add_ps(fiy1,ty); |
| 383 | fiz1 = _mm_add_ps(fiz1,tz); |
| 384 | |
| 385 | fjx0 = _mm_add_ps(fjx0,tx); |
| 386 | fjy0 = _mm_add_ps(fjy0,ty); |
| 387 | fjz0 = _mm_add_ps(fjz0,tz); |
| 388 | |
| 389 | } |
| 390 | |
| 391 | /************************** |
| 392 | * CALCULATE INTERACTIONS * |
| 393 | **************************/ |
| 394 | |
| 395 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
| 396 | { |
| 397 | |
| 398 | r20 = _mm_mul_ps(rsq20,rinv20); |
| 399 | |
| 400 | /* Compute parameters for interactions between i and j atoms */ |
| 401 | qq20 = _mm_mul_ps(iq2,jq0); |
| 402 | |
| 403 | /* EWALD ELECTROSTATICS */ |
| 404 | |
| 405 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 406 | ewrt = _mm_mul_ps(r20,ewtabscale); |
| 407 | ewitab = _mm_cvttps_epi32(ewrt); |
| 408 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 409 | ewitab = _mm_slli_epi32(ewitab,2); |
| 410 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 411 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 412 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 413 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 414 | _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((ewtabF ), (ewtabD)); tmp2 = _mm_unpacklo_ps((ewtabV), (ewtabFn)); tmp1 = _mm_unpackhi_ps((ewtabF), (ewtabD)); tmp3 = _mm_unpackhi_ps ((ewtabV), (ewtabFn)); (ewtabF) = _mm_movelh_ps(tmp0, tmp2); ( ewtabD) = _mm_movehl_ps(tmp2, tmp0); (ewtabV) = _mm_movelh_ps (tmp1, tmp3); (ewtabFn) = _mm_movehl_ps(tmp3, tmp1); } while ( 0); |
| 415 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 416 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 417 | velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec)); |
| 418 | felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec)); |
| 419 | |
| 420 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
| 421 | |
| 422 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 423 | velec = _mm_and_ps(velec,cutoff_mask); |
| 424 | velecsum = _mm_add_ps(velecsum,velec); |
| 425 | |
| 426 | fscal = felec; |
| 427 | |
| 428 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 429 | |
| 430 | /* Calculate temporary vectorial force */ |
| 431 | tx = _mm_mul_ps(fscal,dx20); |
| 432 | ty = _mm_mul_ps(fscal,dy20); |
| 433 | tz = _mm_mul_ps(fscal,dz20); |
| 434 | |
| 435 | /* Update vectorial force */ |
| 436 | fix2 = _mm_add_ps(fix2,tx); |
| 437 | fiy2 = _mm_add_ps(fiy2,ty); |
| 438 | fiz2 = _mm_add_ps(fiz2,tz); |
| 439 | |
| 440 | fjx0 = _mm_add_ps(fjx0,tx); |
| 441 | fjy0 = _mm_add_ps(fjy0,ty); |
| 442 | fjz0 = _mm_add_ps(fjz0,tz); |
| 443 | |
| 444 | } |
| 445 | |
| 446 | /************************** |
| 447 | * CALCULATE INTERACTIONS * |
| 448 | **************************/ |
| 449 | |
| 450 | if (gmx_mm_any_lt(rsq30,rcutoff2)) |
| 451 | { |
| 452 | |
| 453 | r30 = _mm_mul_ps(rsq30,rinv30); |
| 454 | |
| 455 | /* Compute parameters for interactions between i and j atoms */ |
| 456 | qq30 = _mm_mul_ps(iq3,jq0); |
| 457 | |
| 458 | /* EWALD ELECTROSTATICS */ |
| 459 | |
| 460 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 461 | ewrt = _mm_mul_ps(r30,ewtabscale); |
| 462 | ewitab = _mm_cvttps_epi32(ewrt); |
| 463 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 464 | ewitab = _mm_slli_epi32(ewitab,2); |
| 465 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 466 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 467 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 468 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 469 | _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((ewtabF ), (ewtabD)); tmp2 = _mm_unpacklo_ps((ewtabV), (ewtabFn)); tmp1 = _mm_unpackhi_ps((ewtabF), (ewtabD)); tmp3 = _mm_unpackhi_ps ((ewtabV), (ewtabFn)); (ewtabF) = _mm_movelh_ps(tmp0, tmp2); ( ewtabD) = _mm_movehl_ps(tmp2, tmp0); (ewtabV) = _mm_movelh_ps (tmp1, tmp3); (ewtabFn) = _mm_movehl_ps(tmp3, tmp1); } while ( 0); |
| 470 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 471 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 472 | velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec)); |
| 473 | felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec)); |
| 474 | |
| 475 | cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2); |
| 476 | |
| 477 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 478 | velec = _mm_and_ps(velec,cutoff_mask); |
| 479 | velecsum = _mm_add_ps(velecsum,velec); |
| 480 | |
| 481 | fscal = felec; |
| 482 | |
| 483 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 484 | |
| 485 | /* Calculate temporary vectorial force */ |
| 486 | tx = _mm_mul_ps(fscal,dx30); |
| 487 | ty = _mm_mul_ps(fscal,dy30); |
| 488 | tz = _mm_mul_ps(fscal,dz30); |
| 489 | |
| 490 | /* Update vectorial force */ |
| 491 | fix3 = _mm_add_ps(fix3,tx); |
| 492 | fiy3 = _mm_add_ps(fiy3,ty); |
| 493 | fiz3 = _mm_add_ps(fiz3,tz); |
| 494 | |
| 495 | fjx0 = _mm_add_ps(fjx0,tx); |
| 496 | fjy0 = _mm_add_ps(fjy0,ty); |
| 497 | fjz0 = _mm_add_ps(fjz0,tz); |
| 498 | |
| 499 | } |
| 500 | |
| 501 | fjptrA = f+j_coord_offsetA; |
| 502 | fjptrB = f+j_coord_offsetB; |
| 503 | fjptrC = f+j_coord_offsetC; |
| 504 | fjptrD = f+j_coord_offsetD; |
| 505 | |
| 506 | gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0); |
| 507 | |
| 508 | /* Inner loop uses 200 flops */ |
| 509 | } |
| 510 | |
| 511 | if(jidx<j_index_end) |
| 512 | { |
| 513 | |
| 514 | /* Get j neighbor index, and coordinate index */ |
| 515 | jnrlistA = jjnr[jidx]; |
| 516 | jnrlistB = jjnr[jidx+1]; |
| 517 | jnrlistC = jjnr[jidx+2]; |
| 518 | jnrlistD = jjnr[jidx+3]; |
| 519 | /* Sign of each element will be negative for non-real atoms. |
| 520 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
| 521 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
| 522 | */ |
| 523 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
| 524 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
| 525 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
| 526 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
| 527 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
| 528 | j_coord_offsetA = DIM3*jnrA; |
| 529 | j_coord_offsetB = DIM3*jnrB; |
| 530 | j_coord_offsetC = DIM3*jnrC; |
| 531 | j_coord_offsetD = DIM3*jnrD; |
| 532 | |
| 533 | /* load j atom coordinates */ |
| 534 | gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
| 535 | x+j_coord_offsetC,x+j_coord_offsetD, |
| 536 | &jx0,&jy0,&jz0); |
| 537 | |
| 538 | /* Calculate displacement vector */ |
| 539 | dx00 = _mm_sub_ps(ix0,jx0); |
| 540 | dy00 = _mm_sub_ps(iy0,jy0); |
| 541 | dz00 = _mm_sub_ps(iz0,jz0); |
| 542 | dx10 = _mm_sub_ps(ix1,jx0); |
| 543 | dy10 = _mm_sub_ps(iy1,jy0); |
| 544 | dz10 = _mm_sub_ps(iz1,jz0); |
| 545 | dx20 = _mm_sub_ps(ix2,jx0); |
| 546 | dy20 = _mm_sub_ps(iy2,jy0); |
| 547 | dz20 = _mm_sub_ps(iz2,jz0); |
| 548 | dx30 = _mm_sub_ps(ix3,jx0); |
| 549 | dy30 = _mm_sub_ps(iy3,jy0); |
| 550 | dz30 = _mm_sub_ps(iz3,jz0); |
| 551 | |
| 552 | /* Calculate squared distance and things based on it */ |
| 553 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
| 554 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
| 555 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
| 556 | rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30); |
| 557 | |
| 558 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
| 559 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
| 560 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
| 561 | rinv30 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq30); |
| 562 | |
| 563 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
| 564 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
| 565 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
| 566 | rinvsq30 = _mm_mul_ps(rinv30,rinv30); |
| 567 | |
| 568 | /* Load parameters for j particles */ |
| 569 | jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0, |
| 570 | charge+jnrC+0,charge+jnrD+0); |
| 571 | vdwjidx0A = 2*vdwtype[jnrA+0]; |
| 572 | vdwjidx0B = 2*vdwtype[jnrB+0]; |
| 573 | vdwjidx0C = 2*vdwtype[jnrC+0]; |
| 574 | vdwjidx0D = 2*vdwtype[jnrD+0]; |
| 575 | |
| 576 | fjx0 = _mm_setzero_ps(); |
| 577 | fjy0 = _mm_setzero_ps(); |
| 578 | fjz0 = _mm_setzero_ps(); |
| 579 | |
| 580 | /************************** |
| 581 | * CALCULATE INTERACTIONS * |
| 582 | **************************/ |
| 583 | |
| 584 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
| 585 | { |
| 586 | |
| 587 | r00 = _mm_mul_ps(rsq00,rinv00); |
| 588 | r00 = _mm_andnot_ps(dummy_mask,r00); |
| 589 | |
| 590 | /* Compute parameters for interactions between i and j atoms */ |
| 591 | gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A, |
| 592 | vdwparam+vdwioffset0+vdwjidx0B, |
| 593 | vdwparam+vdwioffset0+vdwjidx0C, |
| 594 | vdwparam+vdwioffset0+vdwjidx0D, |
| 595 | &c6_00,&c12_00); |
| 596 | |
| 597 | c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A, |
| 598 | vdwgridparam+vdwioffset0+vdwjidx0B, |
| 599 | vdwgridparam+vdwioffset0+vdwjidx0C, |
| 600 | vdwgridparam+vdwioffset0+vdwjidx0D); |
| 601 | |
| 602 | /* Analytical LJ-PME */ |
| 603 | rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00); |
| 604 | ewcljrsq = _mm_mul_ps(ewclj2,rsq00); |
| 605 | ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2)); |
| 606 | exponent = gmx_simd_exp_rgmx_simd_exp_f(ewcljrsq); |
| 607 | /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */ |
| 608 | poly = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half))); |
| 609 | /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */ |
| 610 | vvdw6 = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix); |
| 611 | vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix)); |
| 612 | vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth), |
| 613 | _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth)); |
| 614 | /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */ |
| 615 | fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00); |
| 616 | |
| 617 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
| 618 | |
| 619 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 620 | vvdw = _mm_and_ps(vvdw,cutoff_mask); |
| 621 | vvdw = _mm_andnot_ps(dummy_mask,vvdw); |
| 622 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
| 623 | |
| 624 | fscal = fvdw; |
| 625 | |
| 626 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 627 | |
| 628 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 629 | |
| 630 | /* Calculate temporary vectorial force */ |
| 631 | tx = _mm_mul_ps(fscal,dx00); |
| 632 | ty = _mm_mul_ps(fscal,dy00); |
| 633 | tz = _mm_mul_ps(fscal,dz00); |
| 634 | |
| 635 | /* Update vectorial force */ |
| 636 | fix0 = _mm_add_ps(fix0,tx); |
| 637 | fiy0 = _mm_add_ps(fiy0,ty); |
| 638 | fiz0 = _mm_add_ps(fiz0,tz); |
| 639 | |
| 640 | fjx0 = _mm_add_ps(fjx0,tx); |
| 641 | fjy0 = _mm_add_ps(fjy0,ty); |
| 642 | fjz0 = _mm_add_ps(fjz0,tz); |
| 643 | |
| 644 | } |
| 645 | |
| 646 | /************************** |
| 647 | * CALCULATE INTERACTIONS * |
| 648 | **************************/ |
| 649 | |
| 650 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
| 651 | { |
| 652 | |
| 653 | r10 = _mm_mul_ps(rsq10,rinv10); |
| 654 | r10 = _mm_andnot_ps(dummy_mask,r10); |
| 655 | |
| 656 | /* Compute parameters for interactions between i and j atoms */ |
| 657 | qq10 = _mm_mul_ps(iq1,jq0); |
| 658 | |
| 659 | /* EWALD ELECTROSTATICS */ |
| 660 | |
| 661 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 662 | ewrt = _mm_mul_ps(r10,ewtabscale); |
| 663 | ewitab = _mm_cvttps_epi32(ewrt); |
| 664 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 665 | ewitab = _mm_slli_epi32(ewitab,2); |
| 666 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 667 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 668 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 669 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 670 | _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((ewtabF ), (ewtabD)); tmp2 = _mm_unpacklo_ps((ewtabV), (ewtabFn)); tmp1 = _mm_unpackhi_ps((ewtabF), (ewtabD)); tmp3 = _mm_unpackhi_ps ((ewtabV), (ewtabFn)); (ewtabF) = _mm_movelh_ps(tmp0, tmp2); ( ewtabD) = _mm_movehl_ps(tmp2, tmp0); (ewtabV) = _mm_movelh_ps (tmp1, tmp3); (ewtabFn) = _mm_movehl_ps(tmp3, tmp1); } while ( 0); |
| 671 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 672 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 673 | velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec)); |
| 674 | felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec)); |
| 675 | |
| 676 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
| 677 | |
| 678 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 679 | velec = _mm_and_ps(velec,cutoff_mask); |
| 680 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 681 | velecsum = _mm_add_ps(velecsum,velec); |
| 682 | |
| 683 | fscal = felec; |
| 684 | |
| 685 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 686 | |
| 687 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 688 | |
| 689 | /* Calculate temporary vectorial force */ |
| 690 | tx = _mm_mul_ps(fscal,dx10); |
| 691 | ty = _mm_mul_ps(fscal,dy10); |
| 692 | tz = _mm_mul_ps(fscal,dz10); |
| 693 | |
| 694 | /* Update vectorial force */ |
| 695 | fix1 = _mm_add_ps(fix1,tx); |
| 696 | fiy1 = _mm_add_ps(fiy1,ty); |
| 697 | fiz1 = _mm_add_ps(fiz1,tz); |
| 698 | |
| 699 | fjx0 = _mm_add_ps(fjx0,tx); |
| 700 | fjy0 = _mm_add_ps(fjy0,ty); |
| 701 | fjz0 = _mm_add_ps(fjz0,tz); |
| 702 | |
| 703 | } |
| 704 | |
| 705 | /************************** |
| 706 | * CALCULATE INTERACTIONS * |
| 707 | **************************/ |
| 708 | |
| 709 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
| 710 | { |
| 711 | |
| 712 | r20 = _mm_mul_ps(rsq20,rinv20); |
| 713 | r20 = _mm_andnot_ps(dummy_mask,r20); |
| 714 | |
| 715 | /* Compute parameters for interactions between i and j atoms */ |
| 716 | qq20 = _mm_mul_ps(iq2,jq0); |
| 717 | |
| 718 | /* EWALD ELECTROSTATICS */ |
| 719 | |
| 720 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 721 | ewrt = _mm_mul_ps(r20,ewtabscale); |
| 722 | ewitab = _mm_cvttps_epi32(ewrt); |
| 723 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 724 | ewitab = _mm_slli_epi32(ewitab,2); |
| 725 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 726 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 727 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 728 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 729 | _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((ewtabF ), (ewtabD)); tmp2 = _mm_unpacklo_ps((ewtabV), (ewtabFn)); tmp1 = _mm_unpackhi_ps((ewtabF), (ewtabD)); tmp3 = _mm_unpackhi_ps ((ewtabV), (ewtabFn)); (ewtabF) = _mm_movelh_ps(tmp0, tmp2); ( ewtabD) = _mm_movehl_ps(tmp2, tmp0); (ewtabV) = _mm_movelh_ps (tmp1, tmp3); (ewtabFn) = _mm_movehl_ps(tmp3, tmp1); } while ( 0); |
| 730 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 731 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 732 | velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec)); |
| 733 | felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec)); |
| 734 | |
| 735 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
| 736 | |
| 737 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 738 | velec = _mm_and_ps(velec,cutoff_mask); |
| 739 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 740 | velecsum = _mm_add_ps(velecsum,velec); |
| 741 | |
| 742 | fscal = felec; |
| 743 | |
| 744 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 745 | |
| 746 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 747 | |
| 748 | /* Calculate temporary vectorial force */ |
| 749 | tx = _mm_mul_ps(fscal,dx20); |
| 750 | ty = _mm_mul_ps(fscal,dy20); |
| 751 | tz = _mm_mul_ps(fscal,dz20); |
| 752 | |
| 753 | /* Update vectorial force */ |
| 754 | fix2 = _mm_add_ps(fix2,tx); |
| 755 | fiy2 = _mm_add_ps(fiy2,ty); |
| 756 | fiz2 = _mm_add_ps(fiz2,tz); |
| 757 | |
| 758 | fjx0 = _mm_add_ps(fjx0,tx); |
| 759 | fjy0 = _mm_add_ps(fjy0,ty); |
| 760 | fjz0 = _mm_add_ps(fjz0,tz); |
| 761 | |
| 762 | } |
| 763 | |
| 764 | /************************** |
| 765 | * CALCULATE INTERACTIONS * |
| 766 | **************************/ |
| 767 | |
| 768 | if (gmx_mm_any_lt(rsq30,rcutoff2)) |
| 769 | { |
| 770 | |
| 771 | r30 = _mm_mul_ps(rsq30,rinv30); |
| 772 | r30 = _mm_andnot_ps(dummy_mask,r30); |
| 773 | |
| 774 | /* Compute parameters for interactions between i and j atoms */ |
| 775 | qq30 = _mm_mul_ps(iq3,jq0); |
| 776 | |
| 777 | /* EWALD ELECTROSTATICS */ |
| 778 | |
| 779 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 780 | ewrt = _mm_mul_ps(r30,ewtabscale); |
| 781 | ewitab = _mm_cvttps_epi32(ewrt); |
| 782 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 783 | ewitab = _mm_slli_epi32(ewitab,2); |
| 784 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 785 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 786 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 787 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 788 | _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((ewtabF ), (ewtabD)); tmp2 = _mm_unpacklo_ps((ewtabV), (ewtabFn)); tmp1 = _mm_unpackhi_ps((ewtabF), (ewtabD)); tmp3 = _mm_unpackhi_ps ((ewtabV), (ewtabFn)); (ewtabF) = _mm_movelh_ps(tmp0, tmp2); ( ewtabD) = _mm_movehl_ps(tmp2, tmp0); (ewtabV) = _mm_movelh_ps (tmp1, tmp3); (ewtabFn) = _mm_movehl_ps(tmp3, tmp1); } while ( 0); |
| 789 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 790 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 791 | velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec)); |
| 792 | felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec)); |
| 793 | |
| 794 | cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2); |
| 795 | |
| 796 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 797 | velec = _mm_and_ps(velec,cutoff_mask); |
| 798 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 799 | velecsum = _mm_add_ps(velecsum,velec); |
| 800 | |
| 801 | fscal = felec; |
| 802 | |
| 803 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 804 | |
| 805 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 806 | |
| 807 | /* Calculate temporary vectorial force */ |
| 808 | tx = _mm_mul_ps(fscal,dx30); |
| 809 | ty = _mm_mul_ps(fscal,dy30); |
| 810 | tz = _mm_mul_ps(fscal,dz30); |
| 811 | |
| 812 | /* Update vectorial force */ |
| 813 | fix3 = _mm_add_ps(fix3,tx); |
| 814 | fiy3 = _mm_add_ps(fiy3,ty); |
| 815 | fiz3 = _mm_add_ps(fiz3,tz); |
| 816 | |
| 817 | fjx0 = _mm_add_ps(fjx0,tx); |
| 818 | fjy0 = _mm_add_ps(fjy0,ty); |
| 819 | fjz0 = _mm_add_ps(fjz0,tz); |
| 820 | |
| 821 | } |
| 822 | |
| 823 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
| 824 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
| 825 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
| 826 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
| 827 | |
| 828 | gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0); |
| 829 | |
| 830 | /* Inner loop uses 204 flops */ |
| 831 | } |
| 832 | |
| 833 | /* End of innermost loop */ |
| 834 | |
| 835 | gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3, |
| 836 | f+i_coord_offset,fshift+i_shift_offset); |
| 837 | |
| 838 | ggid = gid[iidx]; |
| 839 | /* Update potential energies */ |
| 840 | gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid); |
| 841 | gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid); |
| 842 | |
| 843 | /* Increment number of inner iterations */ |
| 844 | inneriter += j_index_end - j_index_start; |
| 845 | |
| 846 | /* Outer loop uses 26 flops */ |
| 847 | } |
| 848 | |
| 849 | /* Increment number of outer iterations */ |
| 850 | outeriter += nri; |
| 851 | |
| 852 | /* Update outer/inner flops */ |
| 853 | |
| 854 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*204)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W4_VF] += outeriter*26 + inneriter *204; |
| 855 | } |
| 856 | /* |
| 857 | * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse4_1_single |
| 858 | * Electrostatics interaction: Ewald |
| 859 | * VdW interaction: LJEwald |
| 860 | * Geometry: Water4-Particle |
| 861 | * Calculate force/pot: Force |
| 862 | */ |
| 863 | void |
| 864 | nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_sse4_1_single |
| 865 | (t_nblist * gmx_restrict nlist, |
| 866 | rvec * gmx_restrict xx, |
| 867 | rvec * gmx_restrict ff, |
| 868 | t_forcerec * gmx_restrict fr, |
| 869 | t_mdatoms * gmx_restrict mdatoms, |
| 870 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
| 871 | t_nrnb * gmx_restrict nrnb) |
| 872 | { |
| 873 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
| 874 | * just 0 for non-waters. |
| 875 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
| 876 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
| 877 | */ |
| 878 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
| 879 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
| 880 | int jnrA,jnrB,jnrC,jnrD; |
| 881 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
| 882 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
| 883 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
| 884 | real rcutoff_scalar; |
| 885 | real *shiftvec,*fshift,*x,*f; |
| 886 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
| 887 | real scratch[4*DIM3]; |
| 888 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
| 889 | int vdwioffset0; |
| 890 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
| 891 | int vdwioffset1; |
| 892 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
| 893 | int vdwioffset2; |
| 894 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
| 895 | int vdwioffset3; |
| 896 | __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3; |
| 897 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
| 898 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
| 899 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
| 900 | __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10; |
| 901 | __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20; |
| 902 | __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30; |
| 903 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
| 904 | real *charge; |
| 905 | int nvdwtype; |
| 906 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
| 907 | int *vdwtype; |
| 908 | real *vdwparam; |
| 909 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
| 910 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
| 911 | __m128 c6grid_00; |
| 912 | __m128 c6grid_10; |
| 913 | __m128 c6grid_20; |
| 914 | __m128 c6grid_30; |
| 915 | __m128 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald; |
| 916 | real *vdwgridparam; |
| 917 | __m128 one_half = _mm_set1_ps(0.5); |
| 918 | __m128 minus_one = _mm_set1_ps(-1.0); |
| 919 | __m128i ewitab; |
| 920 | __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV; |
| 921 | real *ewtab; |
| 922 | __m128 dummy_mask,cutoff_mask; |
| 923 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
| 924 | __m128 one = _mm_set1_ps(1.0); |
| 925 | __m128 two = _mm_set1_ps(2.0); |
| 926 | x = xx[0]; |
| 927 | f = ff[0]; |
| 928 | |
| 929 | nri = nlist->nri; |
| 930 | iinr = nlist->iinr; |
| 931 | jindex = nlist->jindex; |
| 932 | jjnr = nlist->jjnr; |
| 933 | shiftidx = nlist->shift; |
| 934 | gid = nlist->gid; |
| 935 | shiftvec = fr->shift_vec[0]; |
| 936 | fshift = fr->fshift[0]; |
| 937 | facel = _mm_set1_ps(fr->epsfac); |
| 938 | charge = mdatoms->chargeA; |
| 939 | nvdwtype = fr->ntype; |
| 940 | vdwparam = fr->nbfp; |
| 941 | vdwtype = mdatoms->typeA; |
| 942 | vdwgridparam = fr->ljpme_c6grid; |
| 943 | sh_lj_ewald = _mm_set1_ps(fr->ic->sh_lj_ewald); |
| 944 | ewclj = _mm_set1_ps(fr->ewaldcoeff_lj); |
| 945 | ewclj2 = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj)); |
| 946 | |
| 947 | sh_ewald = _mm_set1_ps(fr->ic->sh_ewald); |
| 948 | ewtab = fr->ic->tabq_coul_F; |
| 949 | ewtabscale = _mm_set1_ps(fr->ic->tabq_scale); |
| 950 | ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale); |
| 951 | |
| 952 | /* Setup water-specific parameters */ |
| 953 | inr = nlist->iinr[0]; |
| 954 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
| 955 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
| 956 | iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3])); |
| 957 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
| 958 | |
| 959 | /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */ |
| 960 | rcutoff_scalar = fr->rcoulomb; |
| 961 | rcutoff = _mm_set1_ps(rcutoff_scalar); |
| 962 | rcutoff2 = _mm_mul_ps(rcutoff,rcutoff); |
| 963 | |
| 964 | sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6); |
| 965 | rvdw = _mm_set1_ps(fr->rvdw); |
| 966 | |
| 967 | /* Avoid stupid compiler warnings */ |
| 968 | jnrA = jnrB = jnrC = jnrD = 0; |
| 969 | j_coord_offsetA = 0; |
| 970 | j_coord_offsetB = 0; |
| 971 | j_coord_offsetC = 0; |
| 972 | j_coord_offsetD = 0; |
| 973 | |
| 974 | outeriter = 0; |
| 975 | inneriter = 0; |
| 976 | |
| 977 | for(iidx=0;iidx<4*DIM3;iidx++) |
| 978 | { |
| 979 | scratch[iidx] = 0.0; |
| 980 | } |
| 981 | |
| 982 | /* Start outer loop over neighborlists */ |
| 983 | for(iidx=0; iidx<nri; iidx++) |
| 984 | { |
| 985 | /* Load shift vector for this list */ |
| 986 | i_shift_offset = DIM3*shiftidx[iidx]; |
| 987 | |
| 988 | /* Load limits for loop over neighbors */ |
| 989 | j_index_start = jindex[iidx]; |
| 990 | j_index_end = jindex[iidx+1]; |
| 991 | |
| 992 | /* Get outer coordinate index */ |
| 993 | inr = iinr[iidx]; |
| 994 | i_coord_offset = DIM3*inr; |
| 995 | |
| 996 | /* Load i particle coords and add shift vector */ |
| 997 | gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
| 998 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3); |
| 999 | |
| 1000 | fix0 = _mm_setzero_ps(); |
| 1001 | fiy0 = _mm_setzero_ps(); |
| 1002 | fiz0 = _mm_setzero_ps(); |
| 1003 | fix1 = _mm_setzero_ps(); |
| 1004 | fiy1 = _mm_setzero_ps(); |
| 1005 | fiz1 = _mm_setzero_ps(); |
| 1006 | fix2 = _mm_setzero_ps(); |
| 1007 | fiy2 = _mm_setzero_ps(); |
| 1008 | fiz2 = _mm_setzero_ps(); |
| 1009 | fix3 = _mm_setzero_ps(); |
| 1010 | fiy3 = _mm_setzero_ps(); |
| 1011 | fiz3 = _mm_setzero_ps(); |
| 1012 | |
| 1013 | /* Start inner kernel loop */ |
| 1014 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
| 1015 | { |
| 1016 | |
| 1017 | /* Get j neighbor index, and coordinate index */ |
| 1018 | jnrA = jjnr[jidx]; |
| 1019 | jnrB = jjnr[jidx+1]; |
| 1020 | jnrC = jjnr[jidx+2]; |
| 1021 | jnrD = jjnr[jidx+3]; |
| 1022 | j_coord_offsetA = DIM3*jnrA; |
| 1023 | j_coord_offsetB = DIM3*jnrB; |
| 1024 | j_coord_offsetC = DIM3*jnrC; |
| 1025 | j_coord_offsetD = DIM3*jnrD; |
| 1026 | |
| 1027 | /* load j atom coordinates */ |
| 1028 | gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
| 1029 | x+j_coord_offsetC,x+j_coord_offsetD, |
| 1030 | &jx0,&jy0,&jz0); |
| 1031 | |
| 1032 | /* Calculate displacement vector */ |
| 1033 | dx00 = _mm_sub_ps(ix0,jx0); |
| 1034 | dy00 = _mm_sub_ps(iy0,jy0); |
| 1035 | dz00 = _mm_sub_ps(iz0,jz0); |
| 1036 | dx10 = _mm_sub_ps(ix1,jx0); |
| 1037 | dy10 = _mm_sub_ps(iy1,jy0); |
| 1038 | dz10 = _mm_sub_ps(iz1,jz0); |
| 1039 | dx20 = _mm_sub_ps(ix2,jx0); |
| 1040 | dy20 = _mm_sub_ps(iy2,jy0); |
| 1041 | dz20 = _mm_sub_ps(iz2,jz0); |
| 1042 | dx30 = _mm_sub_ps(ix3,jx0); |
| 1043 | dy30 = _mm_sub_ps(iy3,jy0); |
| 1044 | dz30 = _mm_sub_ps(iz3,jz0); |
| 1045 | |
| 1046 | /* Calculate squared distance and things based on it */ |
| 1047 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
| 1048 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
| 1049 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
| 1050 | rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30); |
| 1051 | |
| 1052 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
| 1053 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
| 1054 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
| 1055 | rinv30 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq30); |
| 1056 | |
| 1057 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
| 1058 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
| 1059 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
| 1060 | rinvsq30 = _mm_mul_ps(rinv30,rinv30); |
| 1061 | |
| 1062 | /* Load parameters for j particles */ |
| 1063 | jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0, |
| 1064 | charge+jnrC+0,charge+jnrD+0); |
| 1065 | vdwjidx0A = 2*vdwtype[jnrA+0]; |
| 1066 | vdwjidx0B = 2*vdwtype[jnrB+0]; |
| 1067 | vdwjidx0C = 2*vdwtype[jnrC+0]; |
| 1068 | vdwjidx0D = 2*vdwtype[jnrD+0]; |
| 1069 | |
| 1070 | fjx0 = _mm_setzero_ps(); |
| 1071 | fjy0 = _mm_setzero_ps(); |
| 1072 | fjz0 = _mm_setzero_ps(); |
| 1073 | |
| 1074 | /************************** |
| 1075 | * CALCULATE INTERACTIONS * |
| 1076 | **************************/ |
| 1077 | |
| 1078 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
| 1079 | { |
| 1080 | |
| 1081 | r00 = _mm_mul_ps(rsq00,rinv00); |
| 1082 | |
| 1083 | /* Compute parameters for interactions between i and j atoms */ |
| 1084 | gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A, |
| 1085 | vdwparam+vdwioffset0+vdwjidx0B, |
| 1086 | vdwparam+vdwioffset0+vdwjidx0C, |
| 1087 | vdwparam+vdwioffset0+vdwjidx0D, |
| 1088 | &c6_00,&c12_00); |
| 1089 | |
| 1090 | c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A, |
| 1091 | vdwgridparam+vdwioffset0+vdwjidx0B, |
| 1092 | vdwgridparam+vdwioffset0+vdwjidx0C, |
| 1093 | vdwgridparam+vdwioffset0+vdwjidx0D); |
| 1094 | |
| 1095 | /* Analytical LJ-PME */ |
| 1096 | rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00); |
| 1097 | ewcljrsq = _mm_mul_ps(ewclj2,rsq00); |
| 1098 | ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2)); |
| 1099 | exponent = gmx_simd_exp_rgmx_simd_exp_f(ewcljrsq); |
| 1100 | /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */ |
| 1101 | poly = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half))); |
| 1102 | /* f6A = 6 * C6grid * (1 - poly) */ |
| 1103 | f6A = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly)); |
| 1104 | /* f6B = C6grid * exponent * beta^6 */ |
| 1105 | f6B = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)); |
| 1106 | /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */ |
| 1107 | fvdw = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00); |
| 1108 | |
| 1109 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
| 1110 | |
| 1111 | fscal = fvdw; |
| 1112 | |
| 1113 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1114 | |
| 1115 | /* Calculate temporary vectorial force */ |
| 1116 | tx = _mm_mul_ps(fscal,dx00); |
| 1117 | ty = _mm_mul_ps(fscal,dy00); |
| 1118 | tz = _mm_mul_ps(fscal,dz00); |
| 1119 | |
| 1120 | /* Update vectorial force */ |
| 1121 | fix0 = _mm_add_ps(fix0,tx); |
| 1122 | fiy0 = _mm_add_ps(fiy0,ty); |
| 1123 | fiz0 = _mm_add_ps(fiz0,tz); |
| 1124 | |
| 1125 | fjx0 = _mm_add_ps(fjx0,tx); |
| 1126 | fjy0 = _mm_add_ps(fjy0,ty); |
| 1127 | fjz0 = _mm_add_ps(fjz0,tz); |
| 1128 | |
| 1129 | } |
| 1130 | |
| 1131 | /************************** |
| 1132 | * CALCULATE INTERACTIONS * |
| 1133 | **************************/ |
| 1134 | |
| 1135 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
| 1136 | { |
| 1137 | |
| 1138 | r10 = _mm_mul_ps(rsq10,rinv10); |
| 1139 | |
| 1140 | /* Compute parameters for interactions between i and j atoms */ |
| 1141 | qq10 = _mm_mul_ps(iq1,jq0); |
| 1142 | |
| 1143 | /* EWALD ELECTROSTATICS */ |
| 1144 | |
| 1145 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1146 | ewrt = _mm_mul_ps(r10,ewtabscale); |
| 1147 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1148 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1149 | gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})),ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})), |
| 1150 | ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})),ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})), |
| 1151 | &ewtabF,&ewtabFn); |
| 1152 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1153 | felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec)); |
| 1154 | |
| 1155 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
| 1156 | |
| 1157 | fscal = felec; |
| 1158 | |
| 1159 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1160 | |
| 1161 | /* Calculate temporary vectorial force */ |
| 1162 | tx = _mm_mul_ps(fscal,dx10); |
| 1163 | ty = _mm_mul_ps(fscal,dy10); |
| 1164 | tz = _mm_mul_ps(fscal,dz10); |
| 1165 | |
| 1166 | /* Update vectorial force */ |
| 1167 | fix1 = _mm_add_ps(fix1,tx); |
| 1168 | fiy1 = _mm_add_ps(fiy1,ty); |
| 1169 | fiz1 = _mm_add_ps(fiz1,tz); |
| 1170 | |
| 1171 | fjx0 = _mm_add_ps(fjx0,tx); |
| 1172 | fjy0 = _mm_add_ps(fjy0,ty); |
| 1173 | fjz0 = _mm_add_ps(fjz0,tz); |
| 1174 | |
| 1175 | } |
| 1176 | |
| 1177 | /************************** |
| 1178 | * CALCULATE INTERACTIONS * |
| 1179 | **************************/ |
| 1180 | |
| 1181 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
| 1182 | { |
| 1183 | |
| 1184 | r20 = _mm_mul_ps(rsq20,rinv20); |
| 1185 | |
| 1186 | /* Compute parameters for interactions between i and j atoms */ |
| 1187 | qq20 = _mm_mul_ps(iq2,jq0); |
| 1188 | |
| 1189 | /* EWALD ELECTROSTATICS */ |
| 1190 | |
| 1191 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1192 | ewrt = _mm_mul_ps(r20,ewtabscale); |
| 1193 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1194 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1195 | gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})),ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})), |
| 1196 | ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})),ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})), |
| 1197 | &ewtabF,&ewtabFn); |
| 1198 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1199 | felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec)); |
| 1200 | |
| 1201 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
| 1202 | |
| 1203 | fscal = felec; |
| 1204 | |
| 1205 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1206 | |
| 1207 | /* Calculate temporary vectorial force */ |
| 1208 | tx = _mm_mul_ps(fscal,dx20); |
| 1209 | ty = _mm_mul_ps(fscal,dy20); |
| 1210 | tz = _mm_mul_ps(fscal,dz20); |
| 1211 | |
| 1212 | /* Update vectorial force */ |
| 1213 | fix2 = _mm_add_ps(fix2,tx); |
| 1214 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1215 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1216 | |
| 1217 | fjx0 = _mm_add_ps(fjx0,tx); |
| 1218 | fjy0 = _mm_add_ps(fjy0,ty); |
| 1219 | fjz0 = _mm_add_ps(fjz0,tz); |
| 1220 | |
| 1221 | } |
| 1222 | |
| 1223 | /************************** |
| 1224 | * CALCULATE INTERACTIONS * |
| 1225 | **************************/ |
| 1226 | |
| 1227 | if (gmx_mm_any_lt(rsq30,rcutoff2)) |
| 1228 | { |
| 1229 | |
| 1230 | r30 = _mm_mul_ps(rsq30,rinv30); |
| 1231 | |
| 1232 | /* Compute parameters for interactions between i and j atoms */ |
| 1233 | qq30 = _mm_mul_ps(iq3,jq0); |
| 1234 | |
| 1235 | /* EWALD ELECTROSTATICS */ |
| 1236 | |
| 1237 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1238 | ewrt = _mm_mul_ps(r30,ewtabscale); |
| 1239 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1240 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1241 | gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})),ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})), |
| 1242 | ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})),ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})), |
| 1243 | &ewtabF,&ewtabFn); |
| 1244 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1245 | felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec)); |
| 1246 | |
| 1247 | cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2); |
| 1248 | |
| 1249 | fscal = felec; |
| 1250 | |
| 1251 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1252 | |
| 1253 | /* Calculate temporary vectorial force */ |
| 1254 | tx = _mm_mul_ps(fscal,dx30); |
| 1255 | ty = _mm_mul_ps(fscal,dy30); |
| 1256 | tz = _mm_mul_ps(fscal,dz30); |
| 1257 | |
| 1258 | /* Update vectorial force */ |
| 1259 | fix3 = _mm_add_ps(fix3,tx); |
| 1260 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1261 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1262 | |
| 1263 | fjx0 = _mm_add_ps(fjx0,tx); |
| 1264 | fjy0 = _mm_add_ps(fjy0,ty); |
| 1265 | fjz0 = _mm_add_ps(fjz0,tz); |
| 1266 | |
| 1267 | } |
| 1268 | |
| 1269 | fjptrA = f+j_coord_offsetA; |
| 1270 | fjptrB = f+j_coord_offsetB; |
| 1271 | fjptrC = f+j_coord_offsetC; |
| 1272 | fjptrD = f+j_coord_offsetD; |
| 1273 | |
| 1274 | gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0); |
| 1275 | |
| 1276 | /* Inner loop uses 166 flops */ |
| 1277 | } |
| 1278 | |
| 1279 | if(jidx<j_index_end) |
| 1280 | { |
| 1281 | |
| 1282 | /* Get j neighbor index, and coordinate index */ |
| 1283 | jnrlistA = jjnr[jidx]; |
| 1284 | jnrlistB = jjnr[jidx+1]; |
| 1285 | jnrlistC = jjnr[jidx+2]; |
| 1286 | jnrlistD = jjnr[jidx+3]; |
| 1287 | /* Sign of each element will be negative for non-real atoms. |
| 1288 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
| 1289 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
| 1290 | */ |
| 1291 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
| 1292 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
| 1293 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
| 1294 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
| 1295 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
| 1296 | j_coord_offsetA = DIM3*jnrA; |
| 1297 | j_coord_offsetB = DIM3*jnrB; |
| 1298 | j_coord_offsetC = DIM3*jnrC; |
| 1299 | j_coord_offsetD = DIM3*jnrD; |
| 1300 | |
| 1301 | /* load j atom coordinates */ |
| 1302 | gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
| 1303 | x+j_coord_offsetC,x+j_coord_offsetD, |
| 1304 | &jx0,&jy0,&jz0); |
| 1305 | |
| 1306 | /* Calculate displacement vector */ |
| 1307 | dx00 = _mm_sub_ps(ix0,jx0); |
| 1308 | dy00 = _mm_sub_ps(iy0,jy0); |
| 1309 | dz00 = _mm_sub_ps(iz0,jz0); |
| 1310 | dx10 = _mm_sub_ps(ix1,jx0); |
| 1311 | dy10 = _mm_sub_ps(iy1,jy0); |
| 1312 | dz10 = _mm_sub_ps(iz1,jz0); |
| 1313 | dx20 = _mm_sub_ps(ix2,jx0); |
| 1314 | dy20 = _mm_sub_ps(iy2,jy0); |
| 1315 | dz20 = _mm_sub_ps(iz2,jz0); |
| 1316 | dx30 = _mm_sub_ps(ix3,jx0); |
| 1317 | dy30 = _mm_sub_ps(iy3,jy0); |
| 1318 | dz30 = _mm_sub_ps(iz3,jz0); |
| 1319 | |
| 1320 | /* Calculate squared distance and things based on it */ |
| 1321 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
| 1322 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
| 1323 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
| 1324 | rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30); |
| 1325 | |
| 1326 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
| 1327 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
| 1328 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
| 1329 | rinv30 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq30); |
| 1330 | |
| 1331 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
| 1332 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
| 1333 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
| 1334 | rinvsq30 = _mm_mul_ps(rinv30,rinv30); |
| 1335 | |
| 1336 | /* Load parameters for j particles */ |
| 1337 | jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0, |
| 1338 | charge+jnrC+0,charge+jnrD+0); |
| 1339 | vdwjidx0A = 2*vdwtype[jnrA+0]; |
| 1340 | vdwjidx0B = 2*vdwtype[jnrB+0]; |
| 1341 | vdwjidx0C = 2*vdwtype[jnrC+0]; |
| 1342 | vdwjidx0D = 2*vdwtype[jnrD+0]; |
| 1343 | |
| 1344 | fjx0 = _mm_setzero_ps(); |
| 1345 | fjy0 = _mm_setzero_ps(); |
| 1346 | fjz0 = _mm_setzero_ps(); |
| 1347 | |
| 1348 | /************************** |
| 1349 | * CALCULATE INTERACTIONS * |
| 1350 | **************************/ |
| 1351 | |
| 1352 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
| 1353 | { |
| 1354 | |
| 1355 | r00 = _mm_mul_ps(rsq00,rinv00); |
| 1356 | r00 = _mm_andnot_ps(dummy_mask,r00); |
| 1357 | |
| 1358 | /* Compute parameters for interactions between i and j atoms */ |
| 1359 | gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A, |
| 1360 | vdwparam+vdwioffset0+vdwjidx0B, |
| 1361 | vdwparam+vdwioffset0+vdwjidx0C, |
| 1362 | vdwparam+vdwioffset0+vdwjidx0D, |
| 1363 | &c6_00,&c12_00); |
| 1364 | |
| 1365 | c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A, |
| 1366 | vdwgridparam+vdwioffset0+vdwjidx0B, |
| 1367 | vdwgridparam+vdwioffset0+vdwjidx0C, |
| 1368 | vdwgridparam+vdwioffset0+vdwjidx0D); |
| 1369 | |
| 1370 | /* Analytical LJ-PME */ |
| 1371 | rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00); |
| 1372 | ewcljrsq = _mm_mul_ps(ewclj2,rsq00); |
| 1373 | ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2)); |
| 1374 | exponent = gmx_simd_exp_rgmx_simd_exp_f(ewcljrsq); |
| 1375 | /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */ |
| 1376 | poly = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half))); |
| 1377 | /* f6A = 6 * C6grid * (1 - poly) */ |
| 1378 | f6A = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly)); |
| 1379 | /* f6B = C6grid * exponent * beta^6 */ |
| 1380 | f6B = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)); |
| 1381 | /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */ |
| 1382 | fvdw = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00); |
| 1383 | |
| 1384 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
| 1385 | |
| 1386 | fscal = fvdw; |
| 1387 | |
| 1388 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1389 | |
| 1390 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1391 | |
| 1392 | /* Calculate temporary vectorial force */ |
| 1393 | tx = _mm_mul_ps(fscal,dx00); |
| 1394 | ty = _mm_mul_ps(fscal,dy00); |
| 1395 | tz = _mm_mul_ps(fscal,dz00); |
| 1396 | |
| 1397 | /* Update vectorial force */ |
| 1398 | fix0 = _mm_add_ps(fix0,tx); |
| 1399 | fiy0 = _mm_add_ps(fiy0,ty); |
| 1400 | fiz0 = _mm_add_ps(fiz0,tz); |
| 1401 | |
| 1402 | fjx0 = _mm_add_ps(fjx0,tx); |
| 1403 | fjy0 = _mm_add_ps(fjy0,ty); |
| 1404 | fjz0 = _mm_add_ps(fjz0,tz); |
| 1405 | |
| 1406 | } |
| 1407 | |
| 1408 | /************************** |
| 1409 | * CALCULATE INTERACTIONS * |
| 1410 | **************************/ |
| 1411 | |
| 1412 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
| 1413 | { |
| 1414 | |
| 1415 | r10 = _mm_mul_ps(rsq10,rinv10); |
| 1416 | r10 = _mm_andnot_ps(dummy_mask,r10); |
| 1417 | |
| 1418 | /* Compute parameters for interactions between i and j atoms */ |
| 1419 | qq10 = _mm_mul_ps(iq1,jq0); |
| 1420 | |
| 1421 | /* EWALD ELECTROSTATICS */ |
| 1422 | |
| 1423 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1424 | ewrt = _mm_mul_ps(r10,ewtabscale); |
| 1425 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1426 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1427 | gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})),ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})), |
| 1428 | ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})),ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})), |
| 1429 | &ewtabF,&ewtabFn); |
| 1430 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1431 | felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec)); |
| 1432 | |
| 1433 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
| 1434 | |
| 1435 | fscal = felec; |
| 1436 | |
| 1437 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1438 | |
| 1439 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1440 | |
| 1441 | /* Calculate temporary vectorial force */ |
| 1442 | tx = _mm_mul_ps(fscal,dx10); |
| 1443 | ty = _mm_mul_ps(fscal,dy10); |
| 1444 | tz = _mm_mul_ps(fscal,dz10); |
| 1445 | |
| 1446 | /* Update vectorial force */ |
| 1447 | fix1 = _mm_add_ps(fix1,tx); |
| 1448 | fiy1 = _mm_add_ps(fiy1,ty); |
| 1449 | fiz1 = _mm_add_ps(fiz1,tz); |
| 1450 | |
| 1451 | fjx0 = _mm_add_ps(fjx0,tx); |
| 1452 | fjy0 = _mm_add_ps(fjy0,ty); |
| 1453 | fjz0 = _mm_add_ps(fjz0,tz); |
| 1454 | |
| 1455 | } |
| 1456 | |
| 1457 | /************************** |
| 1458 | * CALCULATE INTERACTIONS * |
| 1459 | **************************/ |
| 1460 | |
| 1461 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
| 1462 | { |
| 1463 | |
| 1464 | r20 = _mm_mul_ps(rsq20,rinv20); |
| 1465 | r20 = _mm_andnot_ps(dummy_mask,r20); |
| 1466 | |
| 1467 | /* Compute parameters for interactions between i and j atoms */ |
| 1468 | qq20 = _mm_mul_ps(iq2,jq0); |
| 1469 | |
| 1470 | /* EWALD ELECTROSTATICS */ |
| 1471 | |
| 1472 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1473 | ewrt = _mm_mul_ps(r20,ewtabscale); |
| 1474 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1475 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1476 | gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})),ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})), |
| 1477 | ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})),ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})), |
| 1478 | &ewtabF,&ewtabFn); |
| 1479 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1480 | felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec)); |
| 1481 | |
| 1482 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
| 1483 | |
| 1484 | fscal = felec; |
| 1485 | |
| 1486 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1487 | |
| 1488 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1489 | |
| 1490 | /* Calculate temporary vectorial force */ |
| 1491 | tx = _mm_mul_ps(fscal,dx20); |
| 1492 | ty = _mm_mul_ps(fscal,dy20); |
| 1493 | tz = _mm_mul_ps(fscal,dz20); |
| 1494 | |
| 1495 | /* Update vectorial force */ |
| 1496 | fix2 = _mm_add_ps(fix2,tx); |
| 1497 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1498 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1499 | |
| 1500 | fjx0 = _mm_add_ps(fjx0,tx); |
| 1501 | fjy0 = _mm_add_ps(fjy0,ty); |
| 1502 | fjz0 = _mm_add_ps(fjz0,tz); |
| 1503 | |
| 1504 | } |
| 1505 | |
| 1506 | /************************** |
| 1507 | * CALCULATE INTERACTIONS * |
| 1508 | **************************/ |
| 1509 | |
| 1510 | if (gmx_mm_any_lt(rsq30,rcutoff2)) |
| 1511 | { |
| 1512 | |
| 1513 | r30 = _mm_mul_ps(rsq30,rinv30); |
| 1514 | r30 = _mm_andnot_ps(dummy_mask,r30); |
| 1515 | |
| 1516 | /* Compute parameters for interactions between i and j atoms */ |
| 1517 | qq30 = _mm_mul_ps(iq3,jq0); |
| 1518 | |
| 1519 | /* EWALD ELECTROSTATICS */ |
| 1520 | |
| 1521 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1522 | ewrt = _mm_mul_ps(r30,ewtabscale); |
| 1523 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1524 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1525 | gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})),ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})), |
| 1526 | ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})),ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})), |
| 1527 | &ewtabF,&ewtabFn); |
| 1528 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1529 | felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec)); |
| 1530 | |
| 1531 | cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2); |
| 1532 | |
| 1533 | fscal = felec; |
| 1534 | |
| 1535 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1536 | |
| 1537 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1538 | |
| 1539 | /* Calculate temporary vectorial force */ |
| 1540 | tx = _mm_mul_ps(fscal,dx30); |
| 1541 | ty = _mm_mul_ps(fscal,dy30); |
| 1542 | tz = _mm_mul_ps(fscal,dz30); |
| 1543 | |
| 1544 | /* Update vectorial force */ |
| 1545 | fix3 = _mm_add_ps(fix3,tx); |
| 1546 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1547 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1548 | |
| 1549 | fjx0 = _mm_add_ps(fjx0,tx); |
| 1550 | fjy0 = _mm_add_ps(fjy0,ty); |
| 1551 | fjz0 = _mm_add_ps(fjz0,tz); |
| 1552 | |
| 1553 | } |
| 1554 | |
| 1555 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
| 1556 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
| 1557 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
| 1558 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
| 1559 | |
| 1560 | gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0); |
| 1561 | |
| 1562 | /* Inner loop uses 170 flops */ |
| 1563 | } |
| 1564 | |
| 1565 | /* End of innermost loop */ |
| 1566 | |
| 1567 | gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3, |
| 1568 | f+i_coord_offset,fshift+i_shift_offset); |
| 1569 | |
| 1570 | /* Increment number of inner iterations */ |
| 1571 | inneriter += j_index_end - j_index_start; |
| 1572 | |
| 1573 | /* Outer loop uses 24 flops */ |
| 1574 | } |
| 1575 | |
| 1576 | /* Increment number of outer iterations */ |
| 1577 | outeriter += nri; |
| 1578 | |
| 1579 | /* Update outer/inner flops */ |
| 1580 | |
| 1581 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*170)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W4_F] += outeriter*24 + inneriter *170; |
| 1582 | } |