| File: | gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRF_VdwCSTab_GeomW4W4_sse4_1_single.c |
| Location: | line 125, column 22 |
| Description: | Value stored to 'one' during its initialization 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_ElecRF_VdwCSTab_GeomW4W4_VF_sse4_1_single |
| 54 | * Electrostatics interaction: ReactionField |
| 55 | * VdW interaction: CubicSplineTable |
| 56 | * Geometry: Water4-Water4 |
| 57 | * Calculate force/pot: PotentialAndForce |
| 58 | */ |
| 59 | void |
| 60 | nb_kernel_ElecRF_VdwCSTab_GeomW4W4_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 | int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D; |
| 96 | __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1; |
| 97 | int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D; |
| 98 | __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2; |
| 99 | int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D; |
| 100 | __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3; |
| 101 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
| 102 | __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11; |
| 103 | __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12; |
| 104 | __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13; |
| 105 | __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21; |
| 106 | __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22; |
| 107 | __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23; |
| 108 | __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31; |
| 109 | __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32; |
| 110 | __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33; |
| 111 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
| 112 | real *charge; |
| 113 | int nvdwtype; |
| 114 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
| 115 | int *vdwtype; |
| 116 | real *vdwparam; |
| 117 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
| 118 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
| 119 | __m128i vfitab; |
| 120 | __m128i ifour = _mm_set1_epi32(4); |
| 121 | __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF; |
| 122 | real *vftab; |
| 123 | __m128 dummy_mask,cutoff_mask; |
| 124 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
| 125 | __m128 one = _mm_set1_ps(1.0); |
Value stored to 'one' during its initialization is never read | |
| 126 | __m128 two = _mm_set1_ps(2.0); |
| 127 | x = xx[0]; |
| 128 | f = ff[0]; |
| 129 | |
| 130 | nri = nlist->nri; |
| 131 | iinr = nlist->iinr; |
| 132 | jindex = nlist->jindex; |
| 133 | jjnr = nlist->jjnr; |
| 134 | shiftidx = nlist->shift; |
| 135 | gid = nlist->gid; |
| 136 | shiftvec = fr->shift_vec[0]; |
| 137 | fshift = fr->fshift[0]; |
| 138 | facel = _mm_set1_ps(fr->epsfac); |
| 139 | charge = mdatoms->chargeA; |
| 140 | krf = _mm_set1_ps(fr->ic->k_rf); |
| 141 | krf2 = _mm_set1_ps(fr->ic->k_rf*2.0); |
| 142 | crf = _mm_set1_ps(fr->ic->c_rf); |
| 143 | nvdwtype = fr->ntype; |
| 144 | vdwparam = fr->nbfp; |
| 145 | vdwtype = mdatoms->typeA; |
| 146 | |
| 147 | vftab = kernel_data->table_vdw->data; |
| 148 | vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale); |
| 149 | |
| 150 | /* Setup water-specific parameters */ |
| 151 | inr = nlist->iinr[0]; |
| 152 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
| 153 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
| 154 | iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3])); |
| 155 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
| 156 | |
| 157 | jq1 = _mm_set1_ps(charge[inr+1]); |
| 158 | jq2 = _mm_set1_ps(charge[inr+2]); |
| 159 | jq3 = _mm_set1_ps(charge[inr+3]); |
| 160 | vdwjidx0A = 2*vdwtype[inr+0]; |
| 161 | c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]); |
| 162 | c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]); |
| 163 | qq11 = _mm_mul_ps(iq1,jq1); |
| 164 | qq12 = _mm_mul_ps(iq1,jq2); |
| 165 | qq13 = _mm_mul_ps(iq1,jq3); |
| 166 | qq21 = _mm_mul_ps(iq2,jq1); |
| 167 | qq22 = _mm_mul_ps(iq2,jq2); |
| 168 | qq23 = _mm_mul_ps(iq2,jq3); |
| 169 | qq31 = _mm_mul_ps(iq3,jq1); |
| 170 | qq32 = _mm_mul_ps(iq3,jq2); |
| 171 | qq33 = _mm_mul_ps(iq3,jq3); |
| 172 | |
| 173 | /* Avoid stupid compiler warnings */ |
| 174 | jnrA = jnrB = jnrC = jnrD = 0; |
| 175 | j_coord_offsetA = 0; |
| 176 | j_coord_offsetB = 0; |
| 177 | j_coord_offsetC = 0; |
| 178 | j_coord_offsetD = 0; |
| 179 | |
| 180 | outeriter = 0; |
| 181 | inneriter = 0; |
| 182 | |
| 183 | for(iidx=0;iidx<4*DIM3;iidx++) |
| 184 | { |
| 185 | scratch[iidx] = 0.0; |
| 186 | } |
| 187 | |
| 188 | /* Start outer loop over neighborlists */ |
| 189 | for(iidx=0; iidx<nri; iidx++) |
| 190 | { |
| 191 | /* Load shift vector for this list */ |
| 192 | i_shift_offset = DIM3*shiftidx[iidx]; |
| 193 | |
| 194 | /* Load limits for loop over neighbors */ |
| 195 | j_index_start = jindex[iidx]; |
| 196 | j_index_end = jindex[iidx+1]; |
| 197 | |
| 198 | /* Get outer coordinate index */ |
| 199 | inr = iinr[iidx]; |
| 200 | i_coord_offset = DIM3*inr; |
| 201 | |
| 202 | /* Load i particle coords and add shift vector */ |
| 203 | gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
| 204 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3); |
| 205 | |
| 206 | fix0 = _mm_setzero_ps(); |
| 207 | fiy0 = _mm_setzero_ps(); |
| 208 | fiz0 = _mm_setzero_ps(); |
| 209 | fix1 = _mm_setzero_ps(); |
| 210 | fiy1 = _mm_setzero_ps(); |
| 211 | fiz1 = _mm_setzero_ps(); |
| 212 | fix2 = _mm_setzero_ps(); |
| 213 | fiy2 = _mm_setzero_ps(); |
| 214 | fiz2 = _mm_setzero_ps(); |
| 215 | fix3 = _mm_setzero_ps(); |
| 216 | fiy3 = _mm_setzero_ps(); |
| 217 | fiz3 = _mm_setzero_ps(); |
| 218 | |
| 219 | /* Reset potential sums */ |
| 220 | velecsum = _mm_setzero_ps(); |
| 221 | vvdwsum = _mm_setzero_ps(); |
| 222 | |
| 223 | /* Start inner kernel loop */ |
| 224 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
| 225 | { |
| 226 | |
| 227 | /* Get j neighbor index, and coordinate index */ |
| 228 | jnrA = jjnr[jidx]; |
| 229 | jnrB = jjnr[jidx+1]; |
| 230 | jnrC = jjnr[jidx+2]; |
| 231 | jnrD = jjnr[jidx+3]; |
| 232 | j_coord_offsetA = DIM3*jnrA; |
| 233 | j_coord_offsetB = DIM3*jnrB; |
| 234 | j_coord_offsetC = DIM3*jnrC; |
| 235 | j_coord_offsetD = DIM3*jnrD; |
| 236 | |
| 237 | /* load j atom coordinates */ |
| 238 | gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
| 239 | x+j_coord_offsetC,x+j_coord_offsetD, |
| 240 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2, |
| 241 | &jy2,&jz2,&jx3,&jy3,&jz3); |
| 242 | |
| 243 | /* Calculate displacement vector */ |
| 244 | dx00 = _mm_sub_ps(ix0,jx0); |
| 245 | dy00 = _mm_sub_ps(iy0,jy0); |
| 246 | dz00 = _mm_sub_ps(iz0,jz0); |
| 247 | dx11 = _mm_sub_ps(ix1,jx1); |
| 248 | dy11 = _mm_sub_ps(iy1,jy1); |
| 249 | dz11 = _mm_sub_ps(iz1,jz1); |
| 250 | dx12 = _mm_sub_ps(ix1,jx2); |
| 251 | dy12 = _mm_sub_ps(iy1,jy2); |
| 252 | dz12 = _mm_sub_ps(iz1,jz2); |
| 253 | dx13 = _mm_sub_ps(ix1,jx3); |
| 254 | dy13 = _mm_sub_ps(iy1,jy3); |
| 255 | dz13 = _mm_sub_ps(iz1,jz3); |
| 256 | dx21 = _mm_sub_ps(ix2,jx1); |
| 257 | dy21 = _mm_sub_ps(iy2,jy1); |
| 258 | dz21 = _mm_sub_ps(iz2,jz1); |
| 259 | dx22 = _mm_sub_ps(ix2,jx2); |
| 260 | dy22 = _mm_sub_ps(iy2,jy2); |
| 261 | dz22 = _mm_sub_ps(iz2,jz2); |
| 262 | dx23 = _mm_sub_ps(ix2,jx3); |
| 263 | dy23 = _mm_sub_ps(iy2,jy3); |
| 264 | dz23 = _mm_sub_ps(iz2,jz3); |
| 265 | dx31 = _mm_sub_ps(ix3,jx1); |
| 266 | dy31 = _mm_sub_ps(iy3,jy1); |
| 267 | dz31 = _mm_sub_ps(iz3,jz1); |
| 268 | dx32 = _mm_sub_ps(ix3,jx2); |
| 269 | dy32 = _mm_sub_ps(iy3,jy2); |
| 270 | dz32 = _mm_sub_ps(iz3,jz2); |
| 271 | dx33 = _mm_sub_ps(ix3,jx3); |
| 272 | dy33 = _mm_sub_ps(iy3,jy3); |
| 273 | dz33 = _mm_sub_ps(iz3,jz3); |
| 274 | |
| 275 | /* Calculate squared distance and things based on it */ |
| 276 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
| 277 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
| 278 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
| 279 | rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13); |
| 280 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
| 281 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
| 282 | rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23); |
| 283 | rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31); |
| 284 | rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32); |
| 285 | rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33); |
| 286 | |
| 287 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
| 288 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
| 289 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
| 290 | rinv13 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq13); |
| 291 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
| 292 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
| 293 | rinv23 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq23); |
| 294 | rinv31 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq31); |
| 295 | rinv32 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq32); |
| 296 | rinv33 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq33); |
| 297 | |
| 298 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
| 299 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
| 300 | rinvsq13 = _mm_mul_ps(rinv13,rinv13); |
| 301 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
| 302 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
| 303 | rinvsq23 = _mm_mul_ps(rinv23,rinv23); |
| 304 | rinvsq31 = _mm_mul_ps(rinv31,rinv31); |
| 305 | rinvsq32 = _mm_mul_ps(rinv32,rinv32); |
| 306 | rinvsq33 = _mm_mul_ps(rinv33,rinv33); |
| 307 | |
| 308 | fjx0 = _mm_setzero_ps(); |
| 309 | fjy0 = _mm_setzero_ps(); |
| 310 | fjz0 = _mm_setzero_ps(); |
| 311 | fjx1 = _mm_setzero_ps(); |
| 312 | fjy1 = _mm_setzero_ps(); |
| 313 | fjz1 = _mm_setzero_ps(); |
| 314 | fjx2 = _mm_setzero_ps(); |
| 315 | fjy2 = _mm_setzero_ps(); |
| 316 | fjz2 = _mm_setzero_ps(); |
| 317 | fjx3 = _mm_setzero_ps(); |
| 318 | fjy3 = _mm_setzero_ps(); |
| 319 | fjz3 = _mm_setzero_ps(); |
| 320 | |
| 321 | /************************** |
| 322 | * CALCULATE INTERACTIONS * |
| 323 | **************************/ |
| 324 | |
| 325 | r00 = _mm_mul_ps(rsq00,rinv00); |
| 326 | |
| 327 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
| 328 | rt = _mm_mul_ps(r00,vftabscale); |
| 329 | vfitab = _mm_cvttps_epi32(rt); |
| 330 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 331 | vfitab = _mm_slli_epi32(vfitab,3); |
| 332 | |
| 333 | /* CUBIC SPLINE TABLE DISPERSION */ |
| 334 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
| 335 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
| 336 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
| 337 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
| 338 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
| 339 | Heps = _mm_mul_ps(vfeps,H); |
| 340 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
| 341 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
| 342 | vvdw6 = _mm_mul_ps(c6_00,VV); |
| 343 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
| 344 | fvdw6 = _mm_mul_ps(c6_00,FF); |
| 345 | |
| 346 | /* CUBIC SPLINE TABLE REPULSION */ |
| 347 | vfitab = _mm_add_epi32(vfitab,ifour); |
| 348 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
| 349 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
| 350 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
| 351 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
| 352 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
| 353 | Heps = _mm_mul_ps(vfeps,H); |
| 354 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
| 355 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
| 356 | vvdw12 = _mm_mul_ps(c12_00,VV); |
| 357 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
| 358 | fvdw12 = _mm_mul_ps(c12_00,FF); |
| 359 | vvdw = _mm_add_ps(vvdw12,vvdw6); |
| 360 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
| 361 | |
| 362 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 363 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
| 364 | |
| 365 | fscal = fvdw; |
| 366 | |
| 367 | /* Calculate temporary vectorial force */ |
| 368 | tx = _mm_mul_ps(fscal,dx00); |
| 369 | ty = _mm_mul_ps(fscal,dy00); |
| 370 | tz = _mm_mul_ps(fscal,dz00); |
| 371 | |
| 372 | /* Update vectorial force */ |
| 373 | fix0 = _mm_add_ps(fix0,tx); |
| 374 | fiy0 = _mm_add_ps(fiy0,ty); |
| 375 | fiz0 = _mm_add_ps(fiz0,tz); |
| 376 | |
| 377 | fjx0 = _mm_add_ps(fjx0,tx); |
| 378 | fjy0 = _mm_add_ps(fjy0,ty); |
| 379 | fjz0 = _mm_add_ps(fjz0,tz); |
| 380 | |
| 381 | /************************** |
| 382 | * CALCULATE INTERACTIONS * |
| 383 | **************************/ |
| 384 | |
| 385 | /* REACTION-FIELD ELECTROSTATICS */ |
| 386 | velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_add_ps(rinv11,_mm_mul_ps(krf,rsq11)),crf)); |
| 387 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
| 388 | |
| 389 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 390 | velecsum = _mm_add_ps(velecsum,velec); |
| 391 | |
| 392 | fscal = felec; |
| 393 | |
| 394 | /* Calculate temporary vectorial force */ |
| 395 | tx = _mm_mul_ps(fscal,dx11); |
| 396 | ty = _mm_mul_ps(fscal,dy11); |
| 397 | tz = _mm_mul_ps(fscal,dz11); |
| 398 | |
| 399 | /* Update vectorial force */ |
| 400 | fix1 = _mm_add_ps(fix1,tx); |
| 401 | fiy1 = _mm_add_ps(fiy1,ty); |
| 402 | fiz1 = _mm_add_ps(fiz1,tz); |
| 403 | |
| 404 | fjx1 = _mm_add_ps(fjx1,tx); |
| 405 | fjy1 = _mm_add_ps(fjy1,ty); |
| 406 | fjz1 = _mm_add_ps(fjz1,tz); |
| 407 | |
| 408 | /************************** |
| 409 | * CALCULATE INTERACTIONS * |
| 410 | **************************/ |
| 411 | |
| 412 | /* REACTION-FIELD ELECTROSTATICS */ |
| 413 | velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_add_ps(rinv12,_mm_mul_ps(krf,rsq12)),crf)); |
| 414 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
| 415 | |
| 416 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 417 | velecsum = _mm_add_ps(velecsum,velec); |
| 418 | |
| 419 | fscal = felec; |
| 420 | |
| 421 | /* Calculate temporary vectorial force */ |
| 422 | tx = _mm_mul_ps(fscal,dx12); |
| 423 | ty = _mm_mul_ps(fscal,dy12); |
| 424 | tz = _mm_mul_ps(fscal,dz12); |
| 425 | |
| 426 | /* Update vectorial force */ |
| 427 | fix1 = _mm_add_ps(fix1,tx); |
| 428 | fiy1 = _mm_add_ps(fiy1,ty); |
| 429 | fiz1 = _mm_add_ps(fiz1,tz); |
| 430 | |
| 431 | fjx2 = _mm_add_ps(fjx2,tx); |
| 432 | fjy2 = _mm_add_ps(fjy2,ty); |
| 433 | fjz2 = _mm_add_ps(fjz2,tz); |
| 434 | |
| 435 | /************************** |
| 436 | * CALCULATE INTERACTIONS * |
| 437 | **************************/ |
| 438 | |
| 439 | /* REACTION-FIELD ELECTROSTATICS */ |
| 440 | velec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_add_ps(rinv13,_mm_mul_ps(krf,rsq13)),crf)); |
| 441 | felec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_mul_ps(rinv13,rinvsq13),krf2)); |
| 442 | |
| 443 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 444 | velecsum = _mm_add_ps(velecsum,velec); |
| 445 | |
| 446 | fscal = felec; |
| 447 | |
| 448 | /* Calculate temporary vectorial force */ |
| 449 | tx = _mm_mul_ps(fscal,dx13); |
| 450 | ty = _mm_mul_ps(fscal,dy13); |
| 451 | tz = _mm_mul_ps(fscal,dz13); |
| 452 | |
| 453 | /* Update vectorial force */ |
| 454 | fix1 = _mm_add_ps(fix1,tx); |
| 455 | fiy1 = _mm_add_ps(fiy1,ty); |
| 456 | fiz1 = _mm_add_ps(fiz1,tz); |
| 457 | |
| 458 | fjx3 = _mm_add_ps(fjx3,tx); |
| 459 | fjy3 = _mm_add_ps(fjy3,ty); |
| 460 | fjz3 = _mm_add_ps(fjz3,tz); |
| 461 | |
| 462 | /************************** |
| 463 | * CALCULATE INTERACTIONS * |
| 464 | **************************/ |
| 465 | |
| 466 | /* REACTION-FIELD ELECTROSTATICS */ |
| 467 | velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_add_ps(rinv21,_mm_mul_ps(krf,rsq21)),crf)); |
| 468 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
| 469 | |
| 470 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 471 | velecsum = _mm_add_ps(velecsum,velec); |
| 472 | |
| 473 | fscal = felec; |
| 474 | |
| 475 | /* Calculate temporary vectorial force */ |
| 476 | tx = _mm_mul_ps(fscal,dx21); |
| 477 | ty = _mm_mul_ps(fscal,dy21); |
| 478 | tz = _mm_mul_ps(fscal,dz21); |
| 479 | |
| 480 | /* Update vectorial force */ |
| 481 | fix2 = _mm_add_ps(fix2,tx); |
| 482 | fiy2 = _mm_add_ps(fiy2,ty); |
| 483 | fiz2 = _mm_add_ps(fiz2,tz); |
| 484 | |
| 485 | fjx1 = _mm_add_ps(fjx1,tx); |
| 486 | fjy1 = _mm_add_ps(fjy1,ty); |
| 487 | fjz1 = _mm_add_ps(fjz1,tz); |
| 488 | |
| 489 | /************************** |
| 490 | * CALCULATE INTERACTIONS * |
| 491 | **************************/ |
| 492 | |
| 493 | /* REACTION-FIELD ELECTROSTATICS */ |
| 494 | velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_add_ps(rinv22,_mm_mul_ps(krf,rsq22)),crf)); |
| 495 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
| 496 | |
| 497 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 498 | velecsum = _mm_add_ps(velecsum,velec); |
| 499 | |
| 500 | fscal = felec; |
| 501 | |
| 502 | /* Calculate temporary vectorial force */ |
| 503 | tx = _mm_mul_ps(fscal,dx22); |
| 504 | ty = _mm_mul_ps(fscal,dy22); |
| 505 | tz = _mm_mul_ps(fscal,dz22); |
| 506 | |
| 507 | /* Update vectorial force */ |
| 508 | fix2 = _mm_add_ps(fix2,tx); |
| 509 | fiy2 = _mm_add_ps(fiy2,ty); |
| 510 | fiz2 = _mm_add_ps(fiz2,tz); |
| 511 | |
| 512 | fjx2 = _mm_add_ps(fjx2,tx); |
| 513 | fjy2 = _mm_add_ps(fjy2,ty); |
| 514 | fjz2 = _mm_add_ps(fjz2,tz); |
| 515 | |
| 516 | /************************** |
| 517 | * CALCULATE INTERACTIONS * |
| 518 | **************************/ |
| 519 | |
| 520 | /* REACTION-FIELD ELECTROSTATICS */ |
| 521 | velec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_add_ps(rinv23,_mm_mul_ps(krf,rsq23)),crf)); |
| 522 | felec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_mul_ps(rinv23,rinvsq23),krf2)); |
| 523 | |
| 524 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 525 | velecsum = _mm_add_ps(velecsum,velec); |
| 526 | |
| 527 | fscal = felec; |
| 528 | |
| 529 | /* Calculate temporary vectorial force */ |
| 530 | tx = _mm_mul_ps(fscal,dx23); |
| 531 | ty = _mm_mul_ps(fscal,dy23); |
| 532 | tz = _mm_mul_ps(fscal,dz23); |
| 533 | |
| 534 | /* Update vectorial force */ |
| 535 | fix2 = _mm_add_ps(fix2,tx); |
| 536 | fiy2 = _mm_add_ps(fiy2,ty); |
| 537 | fiz2 = _mm_add_ps(fiz2,tz); |
| 538 | |
| 539 | fjx3 = _mm_add_ps(fjx3,tx); |
| 540 | fjy3 = _mm_add_ps(fjy3,ty); |
| 541 | fjz3 = _mm_add_ps(fjz3,tz); |
| 542 | |
| 543 | /************************** |
| 544 | * CALCULATE INTERACTIONS * |
| 545 | **************************/ |
| 546 | |
| 547 | /* REACTION-FIELD ELECTROSTATICS */ |
| 548 | velec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_add_ps(rinv31,_mm_mul_ps(krf,rsq31)),crf)); |
| 549 | felec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_mul_ps(rinv31,rinvsq31),krf2)); |
| 550 | |
| 551 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 552 | velecsum = _mm_add_ps(velecsum,velec); |
| 553 | |
| 554 | fscal = felec; |
| 555 | |
| 556 | /* Calculate temporary vectorial force */ |
| 557 | tx = _mm_mul_ps(fscal,dx31); |
| 558 | ty = _mm_mul_ps(fscal,dy31); |
| 559 | tz = _mm_mul_ps(fscal,dz31); |
| 560 | |
| 561 | /* Update vectorial force */ |
| 562 | fix3 = _mm_add_ps(fix3,tx); |
| 563 | fiy3 = _mm_add_ps(fiy3,ty); |
| 564 | fiz3 = _mm_add_ps(fiz3,tz); |
| 565 | |
| 566 | fjx1 = _mm_add_ps(fjx1,tx); |
| 567 | fjy1 = _mm_add_ps(fjy1,ty); |
| 568 | fjz1 = _mm_add_ps(fjz1,tz); |
| 569 | |
| 570 | /************************** |
| 571 | * CALCULATE INTERACTIONS * |
| 572 | **************************/ |
| 573 | |
| 574 | /* REACTION-FIELD ELECTROSTATICS */ |
| 575 | velec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_add_ps(rinv32,_mm_mul_ps(krf,rsq32)),crf)); |
| 576 | felec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_mul_ps(rinv32,rinvsq32),krf2)); |
| 577 | |
| 578 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 579 | velecsum = _mm_add_ps(velecsum,velec); |
| 580 | |
| 581 | fscal = felec; |
| 582 | |
| 583 | /* Calculate temporary vectorial force */ |
| 584 | tx = _mm_mul_ps(fscal,dx32); |
| 585 | ty = _mm_mul_ps(fscal,dy32); |
| 586 | tz = _mm_mul_ps(fscal,dz32); |
| 587 | |
| 588 | /* Update vectorial force */ |
| 589 | fix3 = _mm_add_ps(fix3,tx); |
| 590 | fiy3 = _mm_add_ps(fiy3,ty); |
| 591 | fiz3 = _mm_add_ps(fiz3,tz); |
| 592 | |
| 593 | fjx2 = _mm_add_ps(fjx2,tx); |
| 594 | fjy2 = _mm_add_ps(fjy2,ty); |
| 595 | fjz2 = _mm_add_ps(fjz2,tz); |
| 596 | |
| 597 | /************************** |
| 598 | * CALCULATE INTERACTIONS * |
| 599 | **************************/ |
| 600 | |
| 601 | /* REACTION-FIELD ELECTROSTATICS */ |
| 602 | velec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_add_ps(rinv33,_mm_mul_ps(krf,rsq33)),crf)); |
| 603 | felec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_mul_ps(rinv33,rinvsq33),krf2)); |
| 604 | |
| 605 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 606 | velecsum = _mm_add_ps(velecsum,velec); |
| 607 | |
| 608 | fscal = felec; |
| 609 | |
| 610 | /* Calculate temporary vectorial force */ |
| 611 | tx = _mm_mul_ps(fscal,dx33); |
| 612 | ty = _mm_mul_ps(fscal,dy33); |
| 613 | tz = _mm_mul_ps(fscal,dz33); |
| 614 | |
| 615 | /* Update vectorial force */ |
| 616 | fix3 = _mm_add_ps(fix3,tx); |
| 617 | fiy3 = _mm_add_ps(fiy3,ty); |
| 618 | fiz3 = _mm_add_ps(fiz3,tz); |
| 619 | |
| 620 | fjx3 = _mm_add_ps(fjx3,tx); |
| 621 | fjy3 = _mm_add_ps(fjy3,ty); |
| 622 | fjz3 = _mm_add_ps(fjz3,tz); |
| 623 | |
| 624 | fjptrA = f+j_coord_offsetA; |
| 625 | fjptrB = f+j_coord_offsetB; |
| 626 | fjptrC = f+j_coord_offsetC; |
| 627 | fjptrD = f+j_coord_offsetD; |
| 628 | |
| 629 | gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
| 630 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1, |
| 631 | fjx2,fjy2,fjz2,fjx3,fjy3,fjz3); |
| 632 | |
| 633 | /* Inner loop uses 347 flops */ |
| 634 | } |
| 635 | |
| 636 | if(jidx<j_index_end) |
| 637 | { |
| 638 | |
| 639 | /* Get j neighbor index, and coordinate index */ |
| 640 | jnrlistA = jjnr[jidx]; |
| 641 | jnrlistB = jjnr[jidx+1]; |
| 642 | jnrlistC = jjnr[jidx+2]; |
| 643 | jnrlistD = jjnr[jidx+3]; |
| 644 | /* Sign of each element will be negative for non-real atoms. |
| 645 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
| 646 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
| 647 | */ |
| 648 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
| 649 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
| 650 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
| 651 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
| 652 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
| 653 | j_coord_offsetA = DIM3*jnrA; |
| 654 | j_coord_offsetB = DIM3*jnrB; |
| 655 | j_coord_offsetC = DIM3*jnrC; |
| 656 | j_coord_offsetD = DIM3*jnrD; |
| 657 | |
| 658 | /* load j atom coordinates */ |
| 659 | gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
| 660 | x+j_coord_offsetC,x+j_coord_offsetD, |
| 661 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2, |
| 662 | &jy2,&jz2,&jx3,&jy3,&jz3); |
| 663 | |
| 664 | /* Calculate displacement vector */ |
| 665 | dx00 = _mm_sub_ps(ix0,jx0); |
| 666 | dy00 = _mm_sub_ps(iy0,jy0); |
| 667 | dz00 = _mm_sub_ps(iz0,jz0); |
| 668 | dx11 = _mm_sub_ps(ix1,jx1); |
| 669 | dy11 = _mm_sub_ps(iy1,jy1); |
| 670 | dz11 = _mm_sub_ps(iz1,jz1); |
| 671 | dx12 = _mm_sub_ps(ix1,jx2); |
| 672 | dy12 = _mm_sub_ps(iy1,jy2); |
| 673 | dz12 = _mm_sub_ps(iz1,jz2); |
| 674 | dx13 = _mm_sub_ps(ix1,jx3); |
| 675 | dy13 = _mm_sub_ps(iy1,jy3); |
| 676 | dz13 = _mm_sub_ps(iz1,jz3); |
| 677 | dx21 = _mm_sub_ps(ix2,jx1); |
| 678 | dy21 = _mm_sub_ps(iy2,jy1); |
| 679 | dz21 = _mm_sub_ps(iz2,jz1); |
| 680 | dx22 = _mm_sub_ps(ix2,jx2); |
| 681 | dy22 = _mm_sub_ps(iy2,jy2); |
| 682 | dz22 = _mm_sub_ps(iz2,jz2); |
| 683 | dx23 = _mm_sub_ps(ix2,jx3); |
| 684 | dy23 = _mm_sub_ps(iy2,jy3); |
| 685 | dz23 = _mm_sub_ps(iz2,jz3); |
| 686 | dx31 = _mm_sub_ps(ix3,jx1); |
| 687 | dy31 = _mm_sub_ps(iy3,jy1); |
| 688 | dz31 = _mm_sub_ps(iz3,jz1); |
| 689 | dx32 = _mm_sub_ps(ix3,jx2); |
| 690 | dy32 = _mm_sub_ps(iy3,jy2); |
| 691 | dz32 = _mm_sub_ps(iz3,jz2); |
| 692 | dx33 = _mm_sub_ps(ix3,jx3); |
| 693 | dy33 = _mm_sub_ps(iy3,jy3); |
| 694 | dz33 = _mm_sub_ps(iz3,jz3); |
| 695 | |
| 696 | /* Calculate squared distance and things based on it */ |
| 697 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
| 698 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
| 699 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
| 700 | rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13); |
| 701 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
| 702 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
| 703 | rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23); |
| 704 | rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31); |
| 705 | rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32); |
| 706 | rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33); |
| 707 | |
| 708 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
| 709 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
| 710 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
| 711 | rinv13 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq13); |
| 712 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
| 713 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
| 714 | rinv23 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq23); |
| 715 | rinv31 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq31); |
| 716 | rinv32 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq32); |
| 717 | rinv33 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq33); |
| 718 | |
| 719 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
| 720 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
| 721 | rinvsq13 = _mm_mul_ps(rinv13,rinv13); |
| 722 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
| 723 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
| 724 | rinvsq23 = _mm_mul_ps(rinv23,rinv23); |
| 725 | rinvsq31 = _mm_mul_ps(rinv31,rinv31); |
| 726 | rinvsq32 = _mm_mul_ps(rinv32,rinv32); |
| 727 | rinvsq33 = _mm_mul_ps(rinv33,rinv33); |
| 728 | |
| 729 | fjx0 = _mm_setzero_ps(); |
| 730 | fjy0 = _mm_setzero_ps(); |
| 731 | fjz0 = _mm_setzero_ps(); |
| 732 | fjx1 = _mm_setzero_ps(); |
| 733 | fjy1 = _mm_setzero_ps(); |
| 734 | fjz1 = _mm_setzero_ps(); |
| 735 | fjx2 = _mm_setzero_ps(); |
| 736 | fjy2 = _mm_setzero_ps(); |
| 737 | fjz2 = _mm_setzero_ps(); |
| 738 | fjx3 = _mm_setzero_ps(); |
| 739 | fjy3 = _mm_setzero_ps(); |
| 740 | fjz3 = _mm_setzero_ps(); |
| 741 | |
| 742 | /************************** |
| 743 | * CALCULATE INTERACTIONS * |
| 744 | **************************/ |
| 745 | |
| 746 | r00 = _mm_mul_ps(rsq00,rinv00); |
| 747 | r00 = _mm_andnot_ps(dummy_mask,r00); |
| 748 | |
| 749 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
| 750 | rt = _mm_mul_ps(r00,vftabscale); |
| 751 | vfitab = _mm_cvttps_epi32(rt); |
| 752 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 753 | vfitab = _mm_slli_epi32(vfitab,3); |
| 754 | |
| 755 | /* CUBIC SPLINE TABLE DISPERSION */ |
| 756 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
| 757 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
| 758 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
| 759 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
| 760 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
| 761 | Heps = _mm_mul_ps(vfeps,H); |
| 762 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
| 763 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
| 764 | vvdw6 = _mm_mul_ps(c6_00,VV); |
| 765 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
| 766 | fvdw6 = _mm_mul_ps(c6_00,FF); |
| 767 | |
| 768 | /* CUBIC SPLINE TABLE REPULSION */ |
| 769 | vfitab = _mm_add_epi32(vfitab,ifour); |
| 770 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
| 771 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
| 772 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
| 773 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
| 774 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
| 775 | Heps = _mm_mul_ps(vfeps,H); |
| 776 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
| 777 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
| 778 | vvdw12 = _mm_mul_ps(c12_00,VV); |
| 779 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
| 780 | fvdw12 = _mm_mul_ps(c12_00,FF); |
| 781 | vvdw = _mm_add_ps(vvdw12,vvdw6); |
| 782 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
| 783 | |
| 784 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 785 | vvdw = _mm_andnot_ps(dummy_mask,vvdw); |
| 786 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
| 787 | |
| 788 | fscal = fvdw; |
| 789 | |
| 790 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 791 | |
| 792 | /* Calculate temporary vectorial force */ |
| 793 | tx = _mm_mul_ps(fscal,dx00); |
| 794 | ty = _mm_mul_ps(fscal,dy00); |
| 795 | tz = _mm_mul_ps(fscal,dz00); |
| 796 | |
| 797 | /* Update vectorial force */ |
| 798 | fix0 = _mm_add_ps(fix0,tx); |
| 799 | fiy0 = _mm_add_ps(fiy0,ty); |
| 800 | fiz0 = _mm_add_ps(fiz0,tz); |
| 801 | |
| 802 | fjx0 = _mm_add_ps(fjx0,tx); |
| 803 | fjy0 = _mm_add_ps(fjy0,ty); |
| 804 | fjz0 = _mm_add_ps(fjz0,tz); |
| 805 | |
| 806 | /************************** |
| 807 | * CALCULATE INTERACTIONS * |
| 808 | **************************/ |
| 809 | |
| 810 | /* REACTION-FIELD ELECTROSTATICS */ |
| 811 | velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_add_ps(rinv11,_mm_mul_ps(krf,rsq11)),crf)); |
| 812 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
| 813 | |
| 814 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 815 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 816 | velecsum = _mm_add_ps(velecsum,velec); |
| 817 | |
| 818 | fscal = felec; |
| 819 | |
| 820 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 821 | |
| 822 | /* Calculate temporary vectorial force */ |
| 823 | tx = _mm_mul_ps(fscal,dx11); |
| 824 | ty = _mm_mul_ps(fscal,dy11); |
| 825 | tz = _mm_mul_ps(fscal,dz11); |
| 826 | |
| 827 | /* Update vectorial force */ |
| 828 | fix1 = _mm_add_ps(fix1,tx); |
| 829 | fiy1 = _mm_add_ps(fiy1,ty); |
| 830 | fiz1 = _mm_add_ps(fiz1,tz); |
| 831 | |
| 832 | fjx1 = _mm_add_ps(fjx1,tx); |
| 833 | fjy1 = _mm_add_ps(fjy1,ty); |
| 834 | fjz1 = _mm_add_ps(fjz1,tz); |
| 835 | |
| 836 | /************************** |
| 837 | * CALCULATE INTERACTIONS * |
| 838 | **************************/ |
| 839 | |
| 840 | /* REACTION-FIELD ELECTROSTATICS */ |
| 841 | velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_add_ps(rinv12,_mm_mul_ps(krf,rsq12)),crf)); |
| 842 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
| 843 | |
| 844 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 845 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 846 | velecsum = _mm_add_ps(velecsum,velec); |
| 847 | |
| 848 | fscal = felec; |
| 849 | |
| 850 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 851 | |
| 852 | /* Calculate temporary vectorial force */ |
| 853 | tx = _mm_mul_ps(fscal,dx12); |
| 854 | ty = _mm_mul_ps(fscal,dy12); |
| 855 | tz = _mm_mul_ps(fscal,dz12); |
| 856 | |
| 857 | /* Update vectorial force */ |
| 858 | fix1 = _mm_add_ps(fix1,tx); |
| 859 | fiy1 = _mm_add_ps(fiy1,ty); |
| 860 | fiz1 = _mm_add_ps(fiz1,tz); |
| 861 | |
| 862 | fjx2 = _mm_add_ps(fjx2,tx); |
| 863 | fjy2 = _mm_add_ps(fjy2,ty); |
| 864 | fjz2 = _mm_add_ps(fjz2,tz); |
| 865 | |
| 866 | /************************** |
| 867 | * CALCULATE INTERACTIONS * |
| 868 | **************************/ |
| 869 | |
| 870 | /* REACTION-FIELD ELECTROSTATICS */ |
| 871 | velec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_add_ps(rinv13,_mm_mul_ps(krf,rsq13)),crf)); |
| 872 | felec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_mul_ps(rinv13,rinvsq13),krf2)); |
| 873 | |
| 874 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 875 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 876 | velecsum = _mm_add_ps(velecsum,velec); |
| 877 | |
| 878 | fscal = felec; |
| 879 | |
| 880 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 881 | |
| 882 | /* Calculate temporary vectorial force */ |
| 883 | tx = _mm_mul_ps(fscal,dx13); |
| 884 | ty = _mm_mul_ps(fscal,dy13); |
| 885 | tz = _mm_mul_ps(fscal,dz13); |
| 886 | |
| 887 | /* Update vectorial force */ |
| 888 | fix1 = _mm_add_ps(fix1,tx); |
| 889 | fiy1 = _mm_add_ps(fiy1,ty); |
| 890 | fiz1 = _mm_add_ps(fiz1,tz); |
| 891 | |
| 892 | fjx3 = _mm_add_ps(fjx3,tx); |
| 893 | fjy3 = _mm_add_ps(fjy3,ty); |
| 894 | fjz3 = _mm_add_ps(fjz3,tz); |
| 895 | |
| 896 | /************************** |
| 897 | * CALCULATE INTERACTIONS * |
| 898 | **************************/ |
| 899 | |
| 900 | /* REACTION-FIELD ELECTROSTATICS */ |
| 901 | velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_add_ps(rinv21,_mm_mul_ps(krf,rsq21)),crf)); |
| 902 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
| 903 | |
| 904 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 905 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 906 | velecsum = _mm_add_ps(velecsum,velec); |
| 907 | |
| 908 | fscal = felec; |
| 909 | |
| 910 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 911 | |
| 912 | /* Calculate temporary vectorial force */ |
| 913 | tx = _mm_mul_ps(fscal,dx21); |
| 914 | ty = _mm_mul_ps(fscal,dy21); |
| 915 | tz = _mm_mul_ps(fscal,dz21); |
| 916 | |
| 917 | /* Update vectorial force */ |
| 918 | fix2 = _mm_add_ps(fix2,tx); |
| 919 | fiy2 = _mm_add_ps(fiy2,ty); |
| 920 | fiz2 = _mm_add_ps(fiz2,tz); |
| 921 | |
| 922 | fjx1 = _mm_add_ps(fjx1,tx); |
| 923 | fjy1 = _mm_add_ps(fjy1,ty); |
| 924 | fjz1 = _mm_add_ps(fjz1,tz); |
| 925 | |
| 926 | /************************** |
| 927 | * CALCULATE INTERACTIONS * |
| 928 | **************************/ |
| 929 | |
| 930 | /* REACTION-FIELD ELECTROSTATICS */ |
| 931 | velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_add_ps(rinv22,_mm_mul_ps(krf,rsq22)),crf)); |
| 932 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
| 933 | |
| 934 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 935 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 936 | velecsum = _mm_add_ps(velecsum,velec); |
| 937 | |
| 938 | fscal = felec; |
| 939 | |
| 940 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 941 | |
| 942 | /* Calculate temporary vectorial force */ |
| 943 | tx = _mm_mul_ps(fscal,dx22); |
| 944 | ty = _mm_mul_ps(fscal,dy22); |
| 945 | tz = _mm_mul_ps(fscal,dz22); |
| 946 | |
| 947 | /* Update vectorial force */ |
| 948 | fix2 = _mm_add_ps(fix2,tx); |
| 949 | fiy2 = _mm_add_ps(fiy2,ty); |
| 950 | fiz2 = _mm_add_ps(fiz2,tz); |
| 951 | |
| 952 | fjx2 = _mm_add_ps(fjx2,tx); |
| 953 | fjy2 = _mm_add_ps(fjy2,ty); |
| 954 | fjz2 = _mm_add_ps(fjz2,tz); |
| 955 | |
| 956 | /************************** |
| 957 | * CALCULATE INTERACTIONS * |
| 958 | **************************/ |
| 959 | |
| 960 | /* REACTION-FIELD ELECTROSTATICS */ |
| 961 | velec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_add_ps(rinv23,_mm_mul_ps(krf,rsq23)),crf)); |
| 962 | felec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_mul_ps(rinv23,rinvsq23),krf2)); |
| 963 | |
| 964 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 965 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 966 | velecsum = _mm_add_ps(velecsum,velec); |
| 967 | |
| 968 | fscal = felec; |
| 969 | |
| 970 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 971 | |
| 972 | /* Calculate temporary vectorial force */ |
| 973 | tx = _mm_mul_ps(fscal,dx23); |
| 974 | ty = _mm_mul_ps(fscal,dy23); |
| 975 | tz = _mm_mul_ps(fscal,dz23); |
| 976 | |
| 977 | /* Update vectorial force */ |
| 978 | fix2 = _mm_add_ps(fix2,tx); |
| 979 | fiy2 = _mm_add_ps(fiy2,ty); |
| 980 | fiz2 = _mm_add_ps(fiz2,tz); |
| 981 | |
| 982 | fjx3 = _mm_add_ps(fjx3,tx); |
| 983 | fjy3 = _mm_add_ps(fjy3,ty); |
| 984 | fjz3 = _mm_add_ps(fjz3,tz); |
| 985 | |
| 986 | /************************** |
| 987 | * CALCULATE INTERACTIONS * |
| 988 | **************************/ |
| 989 | |
| 990 | /* REACTION-FIELD ELECTROSTATICS */ |
| 991 | velec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_add_ps(rinv31,_mm_mul_ps(krf,rsq31)),crf)); |
| 992 | felec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_mul_ps(rinv31,rinvsq31),krf2)); |
| 993 | |
| 994 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 995 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 996 | velecsum = _mm_add_ps(velecsum,velec); |
| 997 | |
| 998 | fscal = felec; |
| 999 | |
| 1000 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1001 | |
| 1002 | /* Calculate temporary vectorial force */ |
| 1003 | tx = _mm_mul_ps(fscal,dx31); |
| 1004 | ty = _mm_mul_ps(fscal,dy31); |
| 1005 | tz = _mm_mul_ps(fscal,dz31); |
| 1006 | |
| 1007 | /* Update vectorial force */ |
| 1008 | fix3 = _mm_add_ps(fix3,tx); |
| 1009 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1010 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1011 | |
| 1012 | fjx1 = _mm_add_ps(fjx1,tx); |
| 1013 | fjy1 = _mm_add_ps(fjy1,ty); |
| 1014 | fjz1 = _mm_add_ps(fjz1,tz); |
| 1015 | |
| 1016 | /************************** |
| 1017 | * CALCULATE INTERACTIONS * |
| 1018 | **************************/ |
| 1019 | |
| 1020 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1021 | velec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_add_ps(rinv32,_mm_mul_ps(krf,rsq32)),crf)); |
| 1022 | felec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_mul_ps(rinv32,rinvsq32),krf2)); |
| 1023 | |
| 1024 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 1025 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 1026 | velecsum = _mm_add_ps(velecsum,velec); |
| 1027 | |
| 1028 | fscal = felec; |
| 1029 | |
| 1030 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1031 | |
| 1032 | /* Calculate temporary vectorial force */ |
| 1033 | tx = _mm_mul_ps(fscal,dx32); |
| 1034 | ty = _mm_mul_ps(fscal,dy32); |
| 1035 | tz = _mm_mul_ps(fscal,dz32); |
| 1036 | |
| 1037 | /* Update vectorial force */ |
| 1038 | fix3 = _mm_add_ps(fix3,tx); |
| 1039 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1040 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1041 | |
| 1042 | fjx2 = _mm_add_ps(fjx2,tx); |
| 1043 | fjy2 = _mm_add_ps(fjy2,ty); |
| 1044 | fjz2 = _mm_add_ps(fjz2,tz); |
| 1045 | |
| 1046 | /************************** |
| 1047 | * CALCULATE INTERACTIONS * |
| 1048 | **************************/ |
| 1049 | |
| 1050 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1051 | velec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_add_ps(rinv33,_mm_mul_ps(krf,rsq33)),crf)); |
| 1052 | felec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_mul_ps(rinv33,rinvsq33),krf2)); |
| 1053 | |
| 1054 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 1055 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 1056 | velecsum = _mm_add_ps(velecsum,velec); |
| 1057 | |
| 1058 | fscal = felec; |
| 1059 | |
| 1060 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1061 | |
| 1062 | /* Calculate temporary vectorial force */ |
| 1063 | tx = _mm_mul_ps(fscal,dx33); |
| 1064 | ty = _mm_mul_ps(fscal,dy33); |
| 1065 | tz = _mm_mul_ps(fscal,dz33); |
| 1066 | |
| 1067 | /* Update vectorial force */ |
| 1068 | fix3 = _mm_add_ps(fix3,tx); |
| 1069 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1070 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1071 | |
| 1072 | fjx3 = _mm_add_ps(fjx3,tx); |
| 1073 | fjy3 = _mm_add_ps(fjy3,ty); |
| 1074 | fjz3 = _mm_add_ps(fjz3,tz); |
| 1075 | |
| 1076 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
| 1077 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
| 1078 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
| 1079 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
| 1080 | |
| 1081 | gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
| 1082 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1, |
| 1083 | fjx2,fjy2,fjz2,fjx3,fjy3,fjz3); |
| 1084 | |
| 1085 | /* Inner loop uses 348 flops */ |
| 1086 | } |
| 1087 | |
| 1088 | /* End of innermost loop */ |
| 1089 | |
| 1090 | gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3, |
| 1091 | f+i_coord_offset,fshift+i_shift_offset); |
| 1092 | |
| 1093 | ggid = gid[iidx]; |
| 1094 | /* Update potential energies */ |
| 1095 | gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid); |
| 1096 | gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid); |
| 1097 | |
| 1098 | /* Increment number of inner iterations */ |
| 1099 | inneriter += j_index_end - j_index_start; |
| 1100 | |
| 1101 | /* Outer loop uses 26 flops */ |
| 1102 | } |
| 1103 | |
| 1104 | /* Increment number of outer iterations */ |
| 1105 | outeriter += nri; |
| 1106 | |
| 1107 | /* Update outer/inner flops */ |
| 1108 | |
| 1109 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*348)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W4W4_VF] += outeriter*26 + inneriter*348; |
| 1110 | } |
| 1111 | /* |
| 1112 | * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW4W4_F_sse4_1_single |
| 1113 | * Electrostatics interaction: ReactionField |
| 1114 | * VdW interaction: CubicSplineTable |
| 1115 | * Geometry: Water4-Water4 |
| 1116 | * Calculate force/pot: Force |
| 1117 | */ |
| 1118 | void |
| 1119 | nb_kernel_ElecRF_VdwCSTab_GeomW4W4_F_sse4_1_single |
| 1120 | (t_nblist * gmx_restrict nlist, |
| 1121 | rvec * gmx_restrict xx, |
| 1122 | rvec * gmx_restrict ff, |
| 1123 | t_forcerec * gmx_restrict fr, |
| 1124 | t_mdatoms * gmx_restrict mdatoms, |
| 1125 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
| 1126 | t_nrnb * gmx_restrict nrnb) |
| 1127 | { |
| 1128 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
| 1129 | * just 0 for non-waters. |
| 1130 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
| 1131 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
| 1132 | */ |
| 1133 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
| 1134 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
| 1135 | int jnrA,jnrB,jnrC,jnrD; |
| 1136 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
| 1137 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
| 1138 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
| 1139 | real rcutoff_scalar; |
| 1140 | real *shiftvec,*fshift,*x,*f; |
| 1141 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
| 1142 | real scratch[4*DIM3]; |
| 1143 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
| 1144 | int vdwioffset0; |
| 1145 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
| 1146 | int vdwioffset1; |
| 1147 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
| 1148 | int vdwioffset2; |
| 1149 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
| 1150 | int vdwioffset3; |
| 1151 | __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3; |
| 1152 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
| 1153 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
| 1154 | int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D; |
| 1155 | __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1; |
| 1156 | int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D; |
| 1157 | __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2; |
| 1158 | int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D; |
| 1159 | __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3; |
| 1160 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
| 1161 | __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11; |
| 1162 | __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12; |
| 1163 | __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13; |
| 1164 | __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21; |
| 1165 | __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22; |
| 1166 | __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23; |
| 1167 | __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31; |
| 1168 | __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32; |
| 1169 | __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33; |
| 1170 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
| 1171 | real *charge; |
| 1172 | int nvdwtype; |
| 1173 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
| 1174 | int *vdwtype; |
| 1175 | real *vdwparam; |
| 1176 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
| 1177 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
| 1178 | __m128i vfitab; |
| 1179 | __m128i ifour = _mm_set1_epi32(4); |
| 1180 | __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF; |
| 1181 | real *vftab; |
| 1182 | __m128 dummy_mask,cutoff_mask; |
| 1183 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
| 1184 | __m128 one = _mm_set1_ps(1.0); |
| 1185 | __m128 two = _mm_set1_ps(2.0); |
| 1186 | x = xx[0]; |
| 1187 | f = ff[0]; |
| 1188 | |
| 1189 | nri = nlist->nri; |
| 1190 | iinr = nlist->iinr; |
| 1191 | jindex = nlist->jindex; |
| 1192 | jjnr = nlist->jjnr; |
| 1193 | shiftidx = nlist->shift; |
| 1194 | gid = nlist->gid; |
| 1195 | shiftvec = fr->shift_vec[0]; |
| 1196 | fshift = fr->fshift[0]; |
| 1197 | facel = _mm_set1_ps(fr->epsfac); |
| 1198 | charge = mdatoms->chargeA; |
| 1199 | krf = _mm_set1_ps(fr->ic->k_rf); |
| 1200 | krf2 = _mm_set1_ps(fr->ic->k_rf*2.0); |
| 1201 | crf = _mm_set1_ps(fr->ic->c_rf); |
| 1202 | nvdwtype = fr->ntype; |
| 1203 | vdwparam = fr->nbfp; |
| 1204 | vdwtype = mdatoms->typeA; |
| 1205 | |
| 1206 | vftab = kernel_data->table_vdw->data; |
| 1207 | vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale); |
| 1208 | |
| 1209 | /* Setup water-specific parameters */ |
| 1210 | inr = nlist->iinr[0]; |
| 1211 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
| 1212 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
| 1213 | iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3])); |
| 1214 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
| 1215 | |
| 1216 | jq1 = _mm_set1_ps(charge[inr+1]); |
| 1217 | jq2 = _mm_set1_ps(charge[inr+2]); |
| 1218 | jq3 = _mm_set1_ps(charge[inr+3]); |
| 1219 | vdwjidx0A = 2*vdwtype[inr+0]; |
| 1220 | c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]); |
| 1221 | c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]); |
| 1222 | qq11 = _mm_mul_ps(iq1,jq1); |
| 1223 | qq12 = _mm_mul_ps(iq1,jq2); |
| 1224 | qq13 = _mm_mul_ps(iq1,jq3); |
| 1225 | qq21 = _mm_mul_ps(iq2,jq1); |
| 1226 | qq22 = _mm_mul_ps(iq2,jq2); |
| 1227 | qq23 = _mm_mul_ps(iq2,jq3); |
| 1228 | qq31 = _mm_mul_ps(iq3,jq1); |
| 1229 | qq32 = _mm_mul_ps(iq3,jq2); |
| 1230 | qq33 = _mm_mul_ps(iq3,jq3); |
| 1231 | |
| 1232 | /* Avoid stupid compiler warnings */ |
| 1233 | jnrA = jnrB = jnrC = jnrD = 0; |
| 1234 | j_coord_offsetA = 0; |
| 1235 | j_coord_offsetB = 0; |
| 1236 | j_coord_offsetC = 0; |
| 1237 | j_coord_offsetD = 0; |
| 1238 | |
| 1239 | outeriter = 0; |
| 1240 | inneriter = 0; |
| 1241 | |
| 1242 | for(iidx=0;iidx<4*DIM3;iidx++) |
| 1243 | { |
| 1244 | scratch[iidx] = 0.0; |
| 1245 | } |
| 1246 | |
| 1247 | /* Start outer loop over neighborlists */ |
| 1248 | for(iidx=0; iidx<nri; iidx++) |
| 1249 | { |
| 1250 | /* Load shift vector for this list */ |
| 1251 | i_shift_offset = DIM3*shiftidx[iidx]; |
| 1252 | |
| 1253 | /* Load limits for loop over neighbors */ |
| 1254 | j_index_start = jindex[iidx]; |
| 1255 | j_index_end = jindex[iidx+1]; |
| 1256 | |
| 1257 | /* Get outer coordinate index */ |
| 1258 | inr = iinr[iidx]; |
| 1259 | i_coord_offset = DIM3*inr; |
| 1260 | |
| 1261 | /* Load i particle coords and add shift vector */ |
| 1262 | gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
| 1263 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3); |
| 1264 | |
| 1265 | fix0 = _mm_setzero_ps(); |
| 1266 | fiy0 = _mm_setzero_ps(); |
| 1267 | fiz0 = _mm_setzero_ps(); |
| 1268 | fix1 = _mm_setzero_ps(); |
| 1269 | fiy1 = _mm_setzero_ps(); |
| 1270 | fiz1 = _mm_setzero_ps(); |
| 1271 | fix2 = _mm_setzero_ps(); |
| 1272 | fiy2 = _mm_setzero_ps(); |
| 1273 | fiz2 = _mm_setzero_ps(); |
| 1274 | fix3 = _mm_setzero_ps(); |
| 1275 | fiy3 = _mm_setzero_ps(); |
| 1276 | fiz3 = _mm_setzero_ps(); |
| 1277 | |
| 1278 | /* Start inner kernel loop */ |
| 1279 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
| 1280 | { |
| 1281 | |
| 1282 | /* Get j neighbor index, and coordinate index */ |
| 1283 | jnrA = jjnr[jidx]; |
| 1284 | jnrB = jjnr[jidx+1]; |
| 1285 | jnrC = jjnr[jidx+2]; |
| 1286 | jnrD = jjnr[jidx+3]; |
| 1287 | j_coord_offsetA = DIM3*jnrA; |
| 1288 | j_coord_offsetB = DIM3*jnrB; |
| 1289 | j_coord_offsetC = DIM3*jnrC; |
| 1290 | j_coord_offsetD = DIM3*jnrD; |
| 1291 | |
| 1292 | /* load j atom coordinates */ |
| 1293 | gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
| 1294 | x+j_coord_offsetC,x+j_coord_offsetD, |
| 1295 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2, |
| 1296 | &jy2,&jz2,&jx3,&jy3,&jz3); |
| 1297 | |
| 1298 | /* Calculate displacement vector */ |
| 1299 | dx00 = _mm_sub_ps(ix0,jx0); |
| 1300 | dy00 = _mm_sub_ps(iy0,jy0); |
| 1301 | dz00 = _mm_sub_ps(iz0,jz0); |
| 1302 | dx11 = _mm_sub_ps(ix1,jx1); |
| 1303 | dy11 = _mm_sub_ps(iy1,jy1); |
| 1304 | dz11 = _mm_sub_ps(iz1,jz1); |
| 1305 | dx12 = _mm_sub_ps(ix1,jx2); |
| 1306 | dy12 = _mm_sub_ps(iy1,jy2); |
| 1307 | dz12 = _mm_sub_ps(iz1,jz2); |
| 1308 | dx13 = _mm_sub_ps(ix1,jx3); |
| 1309 | dy13 = _mm_sub_ps(iy1,jy3); |
| 1310 | dz13 = _mm_sub_ps(iz1,jz3); |
| 1311 | dx21 = _mm_sub_ps(ix2,jx1); |
| 1312 | dy21 = _mm_sub_ps(iy2,jy1); |
| 1313 | dz21 = _mm_sub_ps(iz2,jz1); |
| 1314 | dx22 = _mm_sub_ps(ix2,jx2); |
| 1315 | dy22 = _mm_sub_ps(iy2,jy2); |
| 1316 | dz22 = _mm_sub_ps(iz2,jz2); |
| 1317 | dx23 = _mm_sub_ps(ix2,jx3); |
| 1318 | dy23 = _mm_sub_ps(iy2,jy3); |
| 1319 | dz23 = _mm_sub_ps(iz2,jz3); |
| 1320 | dx31 = _mm_sub_ps(ix3,jx1); |
| 1321 | dy31 = _mm_sub_ps(iy3,jy1); |
| 1322 | dz31 = _mm_sub_ps(iz3,jz1); |
| 1323 | dx32 = _mm_sub_ps(ix3,jx2); |
| 1324 | dy32 = _mm_sub_ps(iy3,jy2); |
| 1325 | dz32 = _mm_sub_ps(iz3,jz2); |
| 1326 | dx33 = _mm_sub_ps(ix3,jx3); |
| 1327 | dy33 = _mm_sub_ps(iy3,jy3); |
| 1328 | dz33 = _mm_sub_ps(iz3,jz3); |
| 1329 | |
| 1330 | /* Calculate squared distance and things based on it */ |
| 1331 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
| 1332 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
| 1333 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
| 1334 | rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13); |
| 1335 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
| 1336 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
| 1337 | rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23); |
| 1338 | rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31); |
| 1339 | rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32); |
| 1340 | rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33); |
| 1341 | |
| 1342 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
| 1343 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
| 1344 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
| 1345 | rinv13 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq13); |
| 1346 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
| 1347 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
| 1348 | rinv23 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq23); |
| 1349 | rinv31 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq31); |
| 1350 | rinv32 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq32); |
| 1351 | rinv33 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq33); |
| 1352 | |
| 1353 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
| 1354 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
| 1355 | rinvsq13 = _mm_mul_ps(rinv13,rinv13); |
| 1356 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
| 1357 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
| 1358 | rinvsq23 = _mm_mul_ps(rinv23,rinv23); |
| 1359 | rinvsq31 = _mm_mul_ps(rinv31,rinv31); |
| 1360 | rinvsq32 = _mm_mul_ps(rinv32,rinv32); |
| 1361 | rinvsq33 = _mm_mul_ps(rinv33,rinv33); |
| 1362 | |
| 1363 | fjx0 = _mm_setzero_ps(); |
| 1364 | fjy0 = _mm_setzero_ps(); |
| 1365 | fjz0 = _mm_setzero_ps(); |
| 1366 | fjx1 = _mm_setzero_ps(); |
| 1367 | fjy1 = _mm_setzero_ps(); |
| 1368 | fjz1 = _mm_setzero_ps(); |
| 1369 | fjx2 = _mm_setzero_ps(); |
| 1370 | fjy2 = _mm_setzero_ps(); |
| 1371 | fjz2 = _mm_setzero_ps(); |
| 1372 | fjx3 = _mm_setzero_ps(); |
| 1373 | fjy3 = _mm_setzero_ps(); |
| 1374 | fjz3 = _mm_setzero_ps(); |
| 1375 | |
| 1376 | /************************** |
| 1377 | * CALCULATE INTERACTIONS * |
| 1378 | **************************/ |
| 1379 | |
| 1380 | r00 = _mm_mul_ps(rsq00,rinv00); |
| 1381 | |
| 1382 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
| 1383 | rt = _mm_mul_ps(r00,vftabscale); |
| 1384 | vfitab = _mm_cvttps_epi32(rt); |
| 1385 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1386 | vfitab = _mm_slli_epi32(vfitab,3); |
| 1387 | |
| 1388 | /* CUBIC SPLINE TABLE DISPERSION */ |
| 1389 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
| 1390 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
| 1391 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
| 1392 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
| 1393 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
| 1394 | Heps = _mm_mul_ps(vfeps,H); |
| 1395 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
| 1396 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
| 1397 | fvdw6 = _mm_mul_ps(c6_00,FF); |
| 1398 | |
| 1399 | /* CUBIC SPLINE TABLE REPULSION */ |
| 1400 | vfitab = _mm_add_epi32(vfitab,ifour); |
| 1401 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
| 1402 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
| 1403 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
| 1404 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
| 1405 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
| 1406 | Heps = _mm_mul_ps(vfeps,H); |
| 1407 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
| 1408 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
| 1409 | fvdw12 = _mm_mul_ps(c12_00,FF); |
| 1410 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
| 1411 | |
| 1412 | fscal = fvdw; |
| 1413 | |
| 1414 | /* Calculate temporary vectorial force */ |
| 1415 | tx = _mm_mul_ps(fscal,dx00); |
| 1416 | ty = _mm_mul_ps(fscal,dy00); |
| 1417 | tz = _mm_mul_ps(fscal,dz00); |
| 1418 | |
| 1419 | /* Update vectorial force */ |
| 1420 | fix0 = _mm_add_ps(fix0,tx); |
| 1421 | fiy0 = _mm_add_ps(fiy0,ty); |
| 1422 | fiz0 = _mm_add_ps(fiz0,tz); |
| 1423 | |
| 1424 | fjx0 = _mm_add_ps(fjx0,tx); |
| 1425 | fjy0 = _mm_add_ps(fjy0,ty); |
| 1426 | fjz0 = _mm_add_ps(fjz0,tz); |
| 1427 | |
| 1428 | /************************** |
| 1429 | * CALCULATE INTERACTIONS * |
| 1430 | **************************/ |
| 1431 | |
| 1432 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1433 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
| 1434 | |
| 1435 | fscal = felec; |
| 1436 | |
| 1437 | /* Calculate temporary vectorial force */ |
| 1438 | tx = _mm_mul_ps(fscal,dx11); |
| 1439 | ty = _mm_mul_ps(fscal,dy11); |
| 1440 | tz = _mm_mul_ps(fscal,dz11); |
| 1441 | |
| 1442 | /* Update vectorial force */ |
| 1443 | fix1 = _mm_add_ps(fix1,tx); |
| 1444 | fiy1 = _mm_add_ps(fiy1,ty); |
| 1445 | fiz1 = _mm_add_ps(fiz1,tz); |
| 1446 | |
| 1447 | fjx1 = _mm_add_ps(fjx1,tx); |
| 1448 | fjy1 = _mm_add_ps(fjy1,ty); |
| 1449 | fjz1 = _mm_add_ps(fjz1,tz); |
| 1450 | |
| 1451 | /************************** |
| 1452 | * CALCULATE INTERACTIONS * |
| 1453 | **************************/ |
| 1454 | |
| 1455 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1456 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
| 1457 | |
| 1458 | fscal = felec; |
| 1459 | |
| 1460 | /* Calculate temporary vectorial force */ |
| 1461 | tx = _mm_mul_ps(fscal,dx12); |
| 1462 | ty = _mm_mul_ps(fscal,dy12); |
| 1463 | tz = _mm_mul_ps(fscal,dz12); |
| 1464 | |
| 1465 | /* Update vectorial force */ |
| 1466 | fix1 = _mm_add_ps(fix1,tx); |
| 1467 | fiy1 = _mm_add_ps(fiy1,ty); |
| 1468 | fiz1 = _mm_add_ps(fiz1,tz); |
| 1469 | |
| 1470 | fjx2 = _mm_add_ps(fjx2,tx); |
| 1471 | fjy2 = _mm_add_ps(fjy2,ty); |
| 1472 | fjz2 = _mm_add_ps(fjz2,tz); |
| 1473 | |
| 1474 | /************************** |
| 1475 | * CALCULATE INTERACTIONS * |
| 1476 | **************************/ |
| 1477 | |
| 1478 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1479 | felec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_mul_ps(rinv13,rinvsq13),krf2)); |
| 1480 | |
| 1481 | fscal = felec; |
| 1482 | |
| 1483 | /* Calculate temporary vectorial force */ |
| 1484 | tx = _mm_mul_ps(fscal,dx13); |
| 1485 | ty = _mm_mul_ps(fscal,dy13); |
| 1486 | tz = _mm_mul_ps(fscal,dz13); |
| 1487 | |
| 1488 | /* Update vectorial force */ |
| 1489 | fix1 = _mm_add_ps(fix1,tx); |
| 1490 | fiy1 = _mm_add_ps(fiy1,ty); |
| 1491 | fiz1 = _mm_add_ps(fiz1,tz); |
| 1492 | |
| 1493 | fjx3 = _mm_add_ps(fjx3,tx); |
| 1494 | fjy3 = _mm_add_ps(fjy3,ty); |
| 1495 | fjz3 = _mm_add_ps(fjz3,tz); |
| 1496 | |
| 1497 | /************************** |
| 1498 | * CALCULATE INTERACTIONS * |
| 1499 | **************************/ |
| 1500 | |
| 1501 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1502 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
| 1503 | |
| 1504 | fscal = felec; |
| 1505 | |
| 1506 | /* Calculate temporary vectorial force */ |
| 1507 | tx = _mm_mul_ps(fscal,dx21); |
| 1508 | ty = _mm_mul_ps(fscal,dy21); |
| 1509 | tz = _mm_mul_ps(fscal,dz21); |
| 1510 | |
| 1511 | /* Update vectorial force */ |
| 1512 | fix2 = _mm_add_ps(fix2,tx); |
| 1513 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1514 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1515 | |
| 1516 | fjx1 = _mm_add_ps(fjx1,tx); |
| 1517 | fjy1 = _mm_add_ps(fjy1,ty); |
| 1518 | fjz1 = _mm_add_ps(fjz1,tz); |
| 1519 | |
| 1520 | /************************** |
| 1521 | * CALCULATE INTERACTIONS * |
| 1522 | **************************/ |
| 1523 | |
| 1524 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1525 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
| 1526 | |
| 1527 | fscal = felec; |
| 1528 | |
| 1529 | /* Calculate temporary vectorial force */ |
| 1530 | tx = _mm_mul_ps(fscal,dx22); |
| 1531 | ty = _mm_mul_ps(fscal,dy22); |
| 1532 | tz = _mm_mul_ps(fscal,dz22); |
| 1533 | |
| 1534 | /* Update vectorial force */ |
| 1535 | fix2 = _mm_add_ps(fix2,tx); |
| 1536 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1537 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1538 | |
| 1539 | fjx2 = _mm_add_ps(fjx2,tx); |
| 1540 | fjy2 = _mm_add_ps(fjy2,ty); |
| 1541 | fjz2 = _mm_add_ps(fjz2,tz); |
| 1542 | |
| 1543 | /************************** |
| 1544 | * CALCULATE INTERACTIONS * |
| 1545 | **************************/ |
| 1546 | |
| 1547 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1548 | felec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_mul_ps(rinv23,rinvsq23),krf2)); |
| 1549 | |
| 1550 | fscal = felec; |
| 1551 | |
| 1552 | /* Calculate temporary vectorial force */ |
| 1553 | tx = _mm_mul_ps(fscal,dx23); |
| 1554 | ty = _mm_mul_ps(fscal,dy23); |
| 1555 | tz = _mm_mul_ps(fscal,dz23); |
| 1556 | |
| 1557 | /* Update vectorial force */ |
| 1558 | fix2 = _mm_add_ps(fix2,tx); |
| 1559 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1560 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1561 | |
| 1562 | fjx3 = _mm_add_ps(fjx3,tx); |
| 1563 | fjy3 = _mm_add_ps(fjy3,ty); |
| 1564 | fjz3 = _mm_add_ps(fjz3,tz); |
| 1565 | |
| 1566 | /************************** |
| 1567 | * CALCULATE INTERACTIONS * |
| 1568 | **************************/ |
| 1569 | |
| 1570 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1571 | felec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_mul_ps(rinv31,rinvsq31),krf2)); |
| 1572 | |
| 1573 | fscal = felec; |
| 1574 | |
| 1575 | /* Calculate temporary vectorial force */ |
| 1576 | tx = _mm_mul_ps(fscal,dx31); |
| 1577 | ty = _mm_mul_ps(fscal,dy31); |
| 1578 | tz = _mm_mul_ps(fscal,dz31); |
| 1579 | |
| 1580 | /* Update vectorial force */ |
| 1581 | fix3 = _mm_add_ps(fix3,tx); |
| 1582 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1583 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1584 | |
| 1585 | fjx1 = _mm_add_ps(fjx1,tx); |
| 1586 | fjy1 = _mm_add_ps(fjy1,ty); |
| 1587 | fjz1 = _mm_add_ps(fjz1,tz); |
| 1588 | |
| 1589 | /************************** |
| 1590 | * CALCULATE INTERACTIONS * |
| 1591 | **************************/ |
| 1592 | |
| 1593 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1594 | felec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_mul_ps(rinv32,rinvsq32),krf2)); |
| 1595 | |
| 1596 | fscal = felec; |
| 1597 | |
| 1598 | /* Calculate temporary vectorial force */ |
| 1599 | tx = _mm_mul_ps(fscal,dx32); |
| 1600 | ty = _mm_mul_ps(fscal,dy32); |
| 1601 | tz = _mm_mul_ps(fscal,dz32); |
| 1602 | |
| 1603 | /* Update vectorial force */ |
| 1604 | fix3 = _mm_add_ps(fix3,tx); |
| 1605 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1606 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1607 | |
| 1608 | fjx2 = _mm_add_ps(fjx2,tx); |
| 1609 | fjy2 = _mm_add_ps(fjy2,ty); |
| 1610 | fjz2 = _mm_add_ps(fjz2,tz); |
| 1611 | |
| 1612 | /************************** |
| 1613 | * CALCULATE INTERACTIONS * |
| 1614 | **************************/ |
| 1615 | |
| 1616 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1617 | felec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_mul_ps(rinv33,rinvsq33),krf2)); |
| 1618 | |
| 1619 | fscal = felec; |
| 1620 | |
| 1621 | /* Calculate temporary vectorial force */ |
| 1622 | tx = _mm_mul_ps(fscal,dx33); |
| 1623 | ty = _mm_mul_ps(fscal,dy33); |
| 1624 | tz = _mm_mul_ps(fscal,dz33); |
| 1625 | |
| 1626 | /* Update vectorial force */ |
| 1627 | fix3 = _mm_add_ps(fix3,tx); |
| 1628 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1629 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1630 | |
| 1631 | fjx3 = _mm_add_ps(fjx3,tx); |
| 1632 | fjy3 = _mm_add_ps(fjy3,ty); |
| 1633 | fjz3 = _mm_add_ps(fjz3,tz); |
| 1634 | |
| 1635 | fjptrA = f+j_coord_offsetA; |
| 1636 | fjptrB = f+j_coord_offsetB; |
| 1637 | fjptrC = f+j_coord_offsetC; |
| 1638 | fjptrD = f+j_coord_offsetD; |
| 1639 | |
| 1640 | gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
| 1641 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1, |
| 1642 | fjx2,fjy2,fjz2,fjx3,fjy3,fjz3); |
| 1643 | |
| 1644 | /* Inner loop uses 294 flops */ |
| 1645 | } |
| 1646 | |
| 1647 | if(jidx<j_index_end) |
| 1648 | { |
| 1649 | |
| 1650 | /* Get j neighbor index, and coordinate index */ |
| 1651 | jnrlistA = jjnr[jidx]; |
| 1652 | jnrlistB = jjnr[jidx+1]; |
| 1653 | jnrlistC = jjnr[jidx+2]; |
| 1654 | jnrlistD = jjnr[jidx+3]; |
| 1655 | /* Sign of each element will be negative for non-real atoms. |
| 1656 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
| 1657 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
| 1658 | */ |
| 1659 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
| 1660 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
| 1661 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
| 1662 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
| 1663 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
| 1664 | j_coord_offsetA = DIM3*jnrA; |
| 1665 | j_coord_offsetB = DIM3*jnrB; |
| 1666 | j_coord_offsetC = DIM3*jnrC; |
| 1667 | j_coord_offsetD = DIM3*jnrD; |
| 1668 | |
| 1669 | /* load j atom coordinates */ |
| 1670 | gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
| 1671 | x+j_coord_offsetC,x+j_coord_offsetD, |
| 1672 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2, |
| 1673 | &jy2,&jz2,&jx3,&jy3,&jz3); |
| 1674 | |
| 1675 | /* Calculate displacement vector */ |
| 1676 | dx00 = _mm_sub_ps(ix0,jx0); |
| 1677 | dy00 = _mm_sub_ps(iy0,jy0); |
| 1678 | dz00 = _mm_sub_ps(iz0,jz0); |
| 1679 | dx11 = _mm_sub_ps(ix1,jx1); |
| 1680 | dy11 = _mm_sub_ps(iy1,jy1); |
| 1681 | dz11 = _mm_sub_ps(iz1,jz1); |
| 1682 | dx12 = _mm_sub_ps(ix1,jx2); |
| 1683 | dy12 = _mm_sub_ps(iy1,jy2); |
| 1684 | dz12 = _mm_sub_ps(iz1,jz2); |
| 1685 | dx13 = _mm_sub_ps(ix1,jx3); |
| 1686 | dy13 = _mm_sub_ps(iy1,jy3); |
| 1687 | dz13 = _mm_sub_ps(iz1,jz3); |
| 1688 | dx21 = _mm_sub_ps(ix2,jx1); |
| 1689 | dy21 = _mm_sub_ps(iy2,jy1); |
| 1690 | dz21 = _mm_sub_ps(iz2,jz1); |
| 1691 | dx22 = _mm_sub_ps(ix2,jx2); |
| 1692 | dy22 = _mm_sub_ps(iy2,jy2); |
| 1693 | dz22 = _mm_sub_ps(iz2,jz2); |
| 1694 | dx23 = _mm_sub_ps(ix2,jx3); |
| 1695 | dy23 = _mm_sub_ps(iy2,jy3); |
| 1696 | dz23 = _mm_sub_ps(iz2,jz3); |
| 1697 | dx31 = _mm_sub_ps(ix3,jx1); |
| 1698 | dy31 = _mm_sub_ps(iy3,jy1); |
| 1699 | dz31 = _mm_sub_ps(iz3,jz1); |
| 1700 | dx32 = _mm_sub_ps(ix3,jx2); |
| 1701 | dy32 = _mm_sub_ps(iy3,jy2); |
| 1702 | dz32 = _mm_sub_ps(iz3,jz2); |
| 1703 | dx33 = _mm_sub_ps(ix3,jx3); |
| 1704 | dy33 = _mm_sub_ps(iy3,jy3); |
| 1705 | dz33 = _mm_sub_ps(iz3,jz3); |
| 1706 | |
| 1707 | /* Calculate squared distance and things based on it */ |
| 1708 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
| 1709 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
| 1710 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
| 1711 | rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13); |
| 1712 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
| 1713 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
| 1714 | rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23); |
| 1715 | rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31); |
| 1716 | rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32); |
| 1717 | rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33); |
| 1718 | |
| 1719 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
| 1720 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
| 1721 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
| 1722 | rinv13 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq13); |
| 1723 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
| 1724 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
| 1725 | rinv23 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq23); |
| 1726 | rinv31 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq31); |
| 1727 | rinv32 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq32); |
| 1728 | rinv33 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq33); |
| 1729 | |
| 1730 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
| 1731 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
| 1732 | rinvsq13 = _mm_mul_ps(rinv13,rinv13); |
| 1733 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
| 1734 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
| 1735 | rinvsq23 = _mm_mul_ps(rinv23,rinv23); |
| 1736 | rinvsq31 = _mm_mul_ps(rinv31,rinv31); |
| 1737 | rinvsq32 = _mm_mul_ps(rinv32,rinv32); |
| 1738 | rinvsq33 = _mm_mul_ps(rinv33,rinv33); |
| 1739 | |
| 1740 | fjx0 = _mm_setzero_ps(); |
| 1741 | fjy0 = _mm_setzero_ps(); |
| 1742 | fjz0 = _mm_setzero_ps(); |
| 1743 | fjx1 = _mm_setzero_ps(); |
| 1744 | fjy1 = _mm_setzero_ps(); |
| 1745 | fjz1 = _mm_setzero_ps(); |
| 1746 | fjx2 = _mm_setzero_ps(); |
| 1747 | fjy2 = _mm_setzero_ps(); |
| 1748 | fjz2 = _mm_setzero_ps(); |
| 1749 | fjx3 = _mm_setzero_ps(); |
| 1750 | fjy3 = _mm_setzero_ps(); |
| 1751 | fjz3 = _mm_setzero_ps(); |
| 1752 | |
| 1753 | /************************** |
| 1754 | * CALCULATE INTERACTIONS * |
| 1755 | **************************/ |
| 1756 | |
| 1757 | r00 = _mm_mul_ps(rsq00,rinv00); |
| 1758 | r00 = _mm_andnot_ps(dummy_mask,r00); |
| 1759 | |
| 1760 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
| 1761 | rt = _mm_mul_ps(r00,vftabscale); |
| 1762 | vfitab = _mm_cvttps_epi32(rt); |
| 1763 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1764 | vfitab = _mm_slli_epi32(vfitab,3); |
| 1765 | |
| 1766 | /* CUBIC SPLINE TABLE DISPERSION */ |
| 1767 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
| 1768 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
| 1769 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
| 1770 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
| 1771 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
| 1772 | Heps = _mm_mul_ps(vfeps,H); |
| 1773 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
| 1774 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
| 1775 | fvdw6 = _mm_mul_ps(c6_00,FF); |
| 1776 | |
| 1777 | /* CUBIC SPLINE TABLE REPULSION */ |
| 1778 | vfitab = _mm_add_epi32(vfitab,ifour); |
| 1779 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
| 1780 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
| 1781 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
| 1782 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
| 1783 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
| 1784 | Heps = _mm_mul_ps(vfeps,H); |
| 1785 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
| 1786 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
| 1787 | fvdw12 = _mm_mul_ps(c12_00,FF); |
| 1788 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
| 1789 | |
| 1790 | fscal = fvdw; |
| 1791 | |
| 1792 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1793 | |
| 1794 | /* Calculate temporary vectorial force */ |
| 1795 | tx = _mm_mul_ps(fscal,dx00); |
| 1796 | ty = _mm_mul_ps(fscal,dy00); |
| 1797 | tz = _mm_mul_ps(fscal,dz00); |
| 1798 | |
| 1799 | /* Update vectorial force */ |
| 1800 | fix0 = _mm_add_ps(fix0,tx); |
| 1801 | fiy0 = _mm_add_ps(fiy0,ty); |
| 1802 | fiz0 = _mm_add_ps(fiz0,tz); |
| 1803 | |
| 1804 | fjx0 = _mm_add_ps(fjx0,tx); |
| 1805 | fjy0 = _mm_add_ps(fjy0,ty); |
| 1806 | fjz0 = _mm_add_ps(fjz0,tz); |
| 1807 | |
| 1808 | /************************** |
| 1809 | * CALCULATE INTERACTIONS * |
| 1810 | **************************/ |
| 1811 | |
| 1812 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1813 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
| 1814 | |
| 1815 | fscal = felec; |
| 1816 | |
| 1817 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1818 | |
| 1819 | /* Calculate temporary vectorial force */ |
| 1820 | tx = _mm_mul_ps(fscal,dx11); |
| 1821 | ty = _mm_mul_ps(fscal,dy11); |
| 1822 | tz = _mm_mul_ps(fscal,dz11); |
| 1823 | |
| 1824 | /* Update vectorial force */ |
| 1825 | fix1 = _mm_add_ps(fix1,tx); |
| 1826 | fiy1 = _mm_add_ps(fiy1,ty); |
| 1827 | fiz1 = _mm_add_ps(fiz1,tz); |
| 1828 | |
| 1829 | fjx1 = _mm_add_ps(fjx1,tx); |
| 1830 | fjy1 = _mm_add_ps(fjy1,ty); |
| 1831 | fjz1 = _mm_add_ps(fjz1,tz); |
| 1832 | |
| 1833 | /************************** |
| 1834 | * CALCULATE INTERACTIONS * |
| 1835 | **************************/ |
| 1836 | |
| 1837 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1838 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
| 1839 | |
| 1840 | fscal = felec; |
| 1841 | |
| 1842 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1843 | |
| 1844 | /* Calculate temporary vectorial force */ |
| 1845 | tx = _mm_mul_ps(fscal,dx12); |
| 1846 | ty = _mm_mul_ps(fscal,dy12); |
| 1847 | tz = _mm_mul_ps(fscal,dz12); |
| 1848 | |
| 1849 | /* Update vectorial force */ |
| 1850 | fix1 = _mm_add_ps(fix1,tx); |
| 1851 | fiy1 = _mm_add_ps(fiy1,ty); |
| 1852 | fiz1 = _mm_add_ps(fiz1,tz); |
| 1853 | |
| 1854 | fjx2 = _mm_add_ps(fjx2,tx); |
| 1855 | fjy2 = _mm_add_ps(fjy2,ty); |
| 1856 | fjz2 = _mm_add_ps(fjz2,tz); |
| 1857 | |
| 1858 | /************************** |
| 1859 | * CALCULATE INTERACTIONS * |
| 1860 | **************************/ |
| 1861 | |
| 1862 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1863 | felec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_mul_ps(rinv13,rinvsq13),krf2)); |
| 1864 | |
| 1865 | fscal = felec; |
| 1866 | |
| 1867 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1868 | |
| 1869 | /* Calculate temporary vectorial force */ |
| 1870 | tx = _mm_mul_ps(fscal,dx13); |
| 1871 | ty = _mm_mul_ps(fscal,dy13); |
| 1872 | tz = _mm_mul_ps(fscal,dz13); |
| 1873 | |
| 1874 | /* Update vectorial force */ |
| 1875 | fix1 = _mm_add_ps(fix1,tx); |
| 1876 | fiy1 = _mm_add_ps(fiy1,ty); |
| 1877 | fiz1 = _mm_add_ps(fiz1,tz); |
| 1878 | |
| 1879 | fjx3 = _mm_add_ps(fjx3,tx); |
| 1880 | fjy3 = _mm_add_ps(fjy3,ty); |
| 1881 | fjz3 = _mm_add_ps(fjz3,tz); |
| 1882 | |
| 1883 | /************************** |
| 1884 | * CALCULATE INTERACTIONS * |
| 1885 | **************************/ |
| 1886 | |
| 1887 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1888 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
| 1889 | |
| 1890 | fscal = felec; |
| 1891 | |
| 1892 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1893 | |
| 1894 | /* Calculate temporary vectorial force */ |
| 1895 | tx = _mm_mul_ps(fscal,dx21); |
| 1896 | ty = _mm_mul_ps(fscal,dy21); |
| 1897 | tz = _mm_mul_ps(fscal,dz21); |
| 1898 | |
| 1899 | /* Update vectorial force */ |
| 1900 | fix2 = _mm_add_ps(fix2,tx); |
| 1901 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1902 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1903 | |
| 1904 | fjx1 = _mm_add_ps(fjx1,tx); |
| 1905 | fjy1 = _mm_add_ps(fjy1,ty); |
| 1906 | fjz1 = _mm_add_ps(fjz1,tz); |
| 1907 | |
| 1908 | /************************** |
| 1909 | * CALCULATE INTERACTIONS * |
| 1910 | **************************/ |
| 1911 | |
| 1912 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1913 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
| 1914 | |
| 1915 | fscal = felec; |
| 1916 | |
| 1917 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1918 | |
| 1919 | /* Calculate temporary vectorial force */ |
| 1920 | tx = _mm_mul_ps(fscal,dx22); |
| 1921 | ty = _mm_mul_ps(fscal,dy22); |
| 1922 | tz = _mm_mul_ps(fscal,dz22); |
| 1923 | |
| 1924 | /* Update vectorial force */ |
| 1925 | fix2 = _mm_add_ps(fix2,tx); |
| 1926 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1927 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1928 | |
| 1929 | fjx2 = _mm_add_ps(fjx2,tx); |
| 1930 | fjy2 = _mm_add_ps(fjy2,ty); |
| 1931 | fjz2 = _mm_add_ps(fjz2,tz); |
| 1932 | |
| 1933 | /************************** |
| 1934 | * CALCULATE INTERACTIONS * |
| 1935 | **************************/ |
| 1936 | |
| 1937 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1938 | felec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_mul_ps(rinv23,rinvsq23),krf2)); |
| 1939 | |
| 1940 | fscal = felec; |
| 1941 | |
| 1942 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1943 | |
| 1944 | /* Calculate temporary vectorial force */ |
| 1945 | tx = _mm_mul_ps(fscal,dx23); |
| 1946 | ty = _mm_mul_ps(fscal,dy23); |
| 1947 | tz = _mm_mul_ps(fscal,dz23); |
| 1948 | |
| 1949 | /* Update vectorial force */ |
| 1950 | fix2 = _mm_add_ps(fix2,tx); |
| 1951 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1952 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1953 | |
| 1954 | fjx3 = _mm_add_ps(fjx3,tx); |
| 1955 | fjy3 = _mm_add_ps(fjy3,ty); |
| 1956 | fjz3 = _mm_add_ps(fjz3,tz); |
| 1957 | |
| 1958 | /************************** |
| 1959 | * CALCULATE INTERACTIONS * |
| 1960 | **************************/ |
| 1961 | |
| 1962 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1963 | felec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_mul_ps(rinv31,rinvsq31),krf2)); |
| 1964 | |
| 1965 | fscal = felec; |
| 1966 | |
| 1967 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1968 | |
| 1969 | /* Calculate temporary vectorial force */ |
| 1970 | tx = _mm_mul_ps(fscal,dx31); |
| 1971 | ty = _mm_mul_ps(fscal,dy31); |
| 1972 | tz = _mm_mul_ps(fscal,dz31); |
| 1973 | |
| 1974 | /* Update vectorial force */ |
| 1975 | fix3 = _mm_add_ps(fix3,tx); |
| 1976 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1977 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1978 | |
| 1979 | fjx1 = _mm_add_ps(fjx1,tx); |
| 1980 | fjy1 = _mm_add_ps(fjy1,ty); |
| 1981 | fjz1 = _mm_add_ps(fjz1,tz); |
| 1982 | |
| 1983 | /************************** |
| 1984 | * CALCULATE INTERACTIONS * |
| 1985 | **************************/ |
| 1986 | |
| 1987 | /* REACTION-FIELD ELECTROSTATICS */ |
| 1988 | felec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_mul_ps(rinv32,rinvsq32),krf2)); |
| 1989 | |
| 1990 | fscal = felec; |
| 1991 | |
| 1992 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1993 | |
| 1994 | /* Calculate temporary vectorial force */ |
| 1995 | tx = _mm_mul_ps(fscal,dx32); |
| 1996 | ty = _mm_mul_ps(fscal,dy32); |
| 1997 | tz = _mm_mul_ps(fscal,dz32); |
| 1998 | |
| 1999 | /* Update vectorial force */ |
| 2000 | fix3 = _mm_add_ps(fix3,tx); |
| 2001 | fiy3 = _mm_add_ps(fiy3,ty); |
| 2002 | fiz3 = _mm_add_ps(fiz3,tz); |
| 2003 | |
| 2004 | fjx2 = _mm_add_ps(fjx2,tx); |
| 2005 | fjy2 = _mm_add_ps(fjy2,ty); |
| 2006 | fjz2 = _mm_add_ps(fjz2,tz); |
| 2007 | |
| 2008 | /************************** |
| 2009 | * CALCULATE INTERACTIONS * |
| 2010 | **************************/ |
| 2011 | |
| 2012 | /* REACTION-FIELD ELECTROSTATICS */ |
| 2013 | felec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_mul_ps(rinv33,rinvsq33),krf2)); |
| 2014 | |
| 2015 | fscal = felec; |
| 2016 | |
| 2017 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 2018 | |
| 2019 | /* Calculate temporary vectorial force */ |
| 2020 | tx = _mm_mul_ps(fscal,dx33); |
| 2021 | ty = _mm_mul_ps(fscal,dy33); |
| 2022 | tz = _mm_mul_ps(fscal,dz33); |
| 2023 | |
| 2024 | /* Update vectorial force */ |
| 2025 | fix3 = _mm_add_ps(fix3,tx); |
| 2026 | fiy3 = _mm_add_ps(fiy3,ty); |
| 2027 | fiz3 = _mm_add_ps(fiz3,tz); |
| 2028 | |
| 2029 | fjx3 = _mm_add_ps(fjx3,tx); |
| 2030 | fjy3 = _mm_add_ps(fjy3,ty); |
| 2031 | fjz3 = _mm_add_ps(fjz3,tz); |
| 2032 | |
| 2033 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
| 2034 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
| 2035 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
| 2036 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
| 2037 | |
| 2038 | gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
| 2039 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1, |
| 2040 | fjx2,fjy2,fjz2,fjx3,fjy3,fjz3); |
| 2041 | |
| 2042 | /* Inner loop uses 295 flops */ |
| 2043 | } |
| 2044 | |
| 2045 | /* End of innermost loop */ |
| 2046 | |
| 2047 | gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3, |
| 2048 | f+i_coord_offset,fshift+i_shift_offset); |
| 2049 | |
| 2050 | /* Increment number of inner iterations */ |
| 2051 | inneriter += j_index_end - j_index_start; |
| 2052 | |
| 2053 | /* Outer loop uses 24 flops */ |
| 2054 | } |
| 2055 | |
| 2056 | /* Increment number of outer iterations */ |
| 2057 | outeriter += nri; |
| 2058 | |
| 2059 | /* Update outer/inner flops */ |
| 2060 | |
| 2061 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*295)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W4W4_F] += outeriter*24 + inneriter *295; |
| 2062 | } |