| File: | gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_VdwNone_GeomW4W4_sse4_1_single.c |
| Location: | line 1517, column 5 |
| Description: | Value stored to 'j_coord_offsetD' 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_VdwNone_GeomW4W4_VF_sse4_1_single |
| 54 | * Electrostatics interaction: Ewald |
| 55 | * VdW interaction: None |
| 56 | * Geometry: Water4-Water4 |
| 57 | * Calculate force/pot: PotentialAndForce |
| 58 | */ |
| 59 | void |
| 60 | nb_kernel_ElecEwSh_VdwNone_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 vdwioffset1; |
| 86 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
| 87 | int vdwioffset2; |
| 88 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
| 89 | int vdwioffset3; |
| 90 | __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3; |
| 91 | int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D; |
| 92 | __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1; |
| 93 | int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D; |
| 94 | __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2; |
| 95 | int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D; |
| 96 | __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3; |
| 97 | __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11; |
| 98 | __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12; |
| 99 | __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13; |
| 100 | __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21; |
| 101 | __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22; |
| 102 | __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23; |
| 103 | __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31; |
| 104 | __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32; |
| 105 | __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33; |
| 106 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
| 107 | real *charge; |
| 108 | __m128i ewitab; |
| 109 | __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV; |
| 110 | real *ewtab; |
| 111 | __m128 dummy_mask,cutoff_mask; |
| 112 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
| 113 | __m128 one = _mm_set1_ps(1.0); |
| 114 | __m128 two = _mm_set1_ps(2.0); |
| 115 | x = xx[0]; |
| 116 | f = ff[0]; |
| 117 | |
| 118 | nri = nlist->nri; |
| 119 | iinr = nlist->iinr; |
| 120 | jindex = nlist->jindex; |
| 121 | jjnr = nlist->jjnr; |
| 122 | shiftidx = nlist->shift; |
| 123 | gid = nlist->gid; |
| 124 | shiftvec = fr->shift_vec[0]; |
| 125 | fshift = fr->fshift[0]; |
| 126 | facel = _mm_set1_ps(fr->epsfac); |
| 127 | charge = mdatoms->chargeA; |
| 128 | |
| 129 | sh_ewald = _mm_set1_ps(fr->ic->sh_ewald); |
| 130 | ewtab = fr->ic->tabq_coul_FDV0; |
| 131 | ewtabscale = _mm_set1_ps(fr->ic->tabq_scale); |
| 132 | ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale); |
| 133 | |
| 134 | /* Setup water-specific parameters */ |
| 135 | inr = nlist->iinr[0]; |
| 136 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
| 137 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
| 138 | iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3])); |
| 139 | |
| 140 | jq1 = _mm_set1_ps(charge[inr+1]); |
| 141 | jq2 = _mm_set1_ps(charge[inr+2]); |
| 142 | jq3 = _mm_set1_ps(charge[inr+3]); |
| 143 | qq11 = _mm_mul_ps(iq1,jq1); |
| 144 | qq12 = _mm_mul_ps(iq1,jq2); |
| 145 | qq13 = _mm_mul_ps(iq1,jq3); |
| 146 | qq21 = _mm_mul_ps(iq2,jq1); |
| 147 | qq22 = _mm_mul_ps(iq2,jq2); |
| 148 | qq23 = _mm_mul_ps(iq2,jq3); |
| 149 | qq31 = _mm_mul_ps(iq3,jq1); |
| 150 | qq32 = _mm_mul_ps(iq3,jq2); |
| 151 | qq33 = _mm_mul_ps(iq3,jq3); |
| 152 | |
| 153 | /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */ |
| 154 | rcutoff_scalar = fr->rcoulomb; |
| 155 | rcutoff = _mm_set1_ps(rcutoff_scalar); |
| 156 | rcutoff2 = _mm_mul_ps(rcutoff,rcutoff); |
| 157 | |
| 158 | /* Avoid stupid compiler warnings */ |
| 159 | jnrA = jnrB = jnrC = jnrD = 0; |
| 160 | j_coord_offsetA = 0; |
| 161 | j_coord_offsetB = 0; |
| 162 | j_coord_offsetC = 0; |
| 163 | j_coord_offsetD = 0; |
| 164 | |
| 165 | outeriter = 0; |
| 166 | inneriter = 0; |
| 167 | |
| 168 | for(iidx=0;iidx<4*DIM3;iidx++) |
| 169 | { |
| 170 | scratch[iidx] = 0.0; |
| 171 | } |
| 172 | |
| 173 | /* Start outer loop over neighborlists */ |
| 174 | for(iidx=0; iidx<nri; iidx++) |
| 175 | { |
| 176 | /* Load shift vector for this list */ |
| 177 | i_shift_offset = DIM3*shiftidx[iidx]; |
| 178 | |
| 179 | /* Load limits for loop over neighbors */ |
| 180 | j_index_start = jindex[iidx]; |
| 181 | j_index_end = jindex[iidx+1]; |
| 182 | |
| 183 | /* Get outer coordinate index */ |
| 184 | inr = iinr[iidx]; |
| 185 | i_coord_offset = DIM3*inr; |
| 186 | |
| 187 | /* Load i particle coords and add shift vector */ |
| 188 | gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM3, |
| 189 | &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3); |
| 190 | |
| 191 | fix1 = _mm_setzero_ps(); |
| 192 | fiy1 = _mm_setzero_ps(); |
| 193 | fiz1 = _mm_setzero_ps(); |
| 194 | fix2 = _mm_setzero_ps(); |
| 195 | fiy2 = _mm_setzero_ps(); |
| 196 | fiz2 = _mm_setzero_ps(); |
| 197 | fix3 = _mm_setzero_ps(); |
| 198 | fiy3 = _mm_setzero_ps(); |
| 199 | fiz3 = _mm_setzero_ps(); |
| 200 | |
| 201 | /* Reset potential sums */ |
| 202 | velecsum = _mm_setzero_ps(); |
| 203 | |
| 204 | /* Start inner kernel loop */ |
| 205 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
| 206 | { |
| 207 | |
| 208 | /* Get j neighbor index, and coordinate index */ |
| 209 | jnrA = jjnr[jidx]; |
| 210 | jnrB = jjnr[jidx+1]; |
| 211 | jnrC = jjnr[jidx+2]; |
| 212 | jnrD = jjnr[jidx+3]; |
| 213 | j_coord_offsetA = DIM3*jnrA; |
| 214 | j_coord_offsetB = DIM3*jnrB; |
| 215 | j_coord_offsetC = DIM3*jnrC; |
| 216 | j_coord_offsetD = DIM3*jnrD; |
| 217 | |
| 218 | /* load j atom coordinates */ |
| 219 | gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM3,x+j_coord_offsetB+DIM3, |
| 220 | x+j_coord_offsetC+DIM3,x+j_coord_offsetD+DIM3, |
| 221 | &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3); |
| 222 | |
| 223 | /* Calculate displacement vector */ |
| 224 | dx11 = _mm_sub_ps(ix1,jx1); |
| 225 | dy11 = _mm_sub_ps(iy1,jy1); |
| 226 | dz11 = _mm_sub_ps(iz1,jz1); |
| 227 | dx12 = _mm_sub_ps(ix1,jx2); |
| 228 | dy12 = _mm_sub_ps(iy1,jy2); |
| 229 | dz12 = _mm_sub_ps(iz1,jz2); |
| 230 | dx13 = _mm_sub_ps(ix1,jx3); |
| 231 | dy13 = _mm_sub_ps(iy1,jy3); |
| 232 | dz13 = _mm_sub_ps(iz1,jz3); |
| 233 | dx21 = _mm_sub_ps(ix2,jx1); |
| 234 | dy21 = _mm_sub_ps(iy2,jy1); |
| 235 | dz21 = _mm_sub_ps(iz2,jz1); |
| 236 | dx22 = _mm_sub_ps(ix2,jx2); |
| 237 | dy22 = _mm_sub_ps(iy2,jy2); |
| 238 | dz22 = _mm_sub_ps(iz2,jz2); |
| 239 | dx23 = _mm_sub_ps(ix2,jx3); |
| 240 | dy23 = _mm_sub_ps(iy2,jy3); |
| 241 | dz23 = _mm_sub_ps(iz2,jz3); |
| 242 | dx31 = _mm_sub_ps(ix3,jx1); |
| 243 | dy31 = _mm_sub_ps(iy3,jy1); |
| 244 | dz31 = _mm_sub_ps(iz3,jz1); |
| 245 | dx32 = _mm_sub_ps(ix3,jx2); |
| 246 | dy32 = _mm_sub_ps(iy3,jy2); |
| 247 | dz32 = _mm_sub_ps(iz3,jz2); |
| 248 | dx33 = _mm_sub_ps(ix3,jx3); |
| 249 | dy33 = _mm_sub_ps(iy3,jy3); |
| 250 | dz33 = _mm_sub_ps(iz3,jz3); |
| 251 | |
| 252 | /* Calculate squared distance and things based on it */ |
| 253 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
| 254 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
| 255 | rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13); |
| 256 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
| 257 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
| 258 | rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23); |
| 259 | rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31); |
| 260 | rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32); |
| 261 | rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33); |
| 262 | |
| 263 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
| 264 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
| 265 | rinv13 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq13); |
| 266 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
| 267 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
| 268 | rinv23 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq23); |
| 269 | rinv31 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq31); |
| 270 | rinv32 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq32); |
| 271 | rinv33 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq33); |
| 272 | |
| 273 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
| 274 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
| 275 | rinvsq13 = _mm_mul_ps(rinv13,rinv13); |
| 276 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
| 277 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
| 278 | rinvsq23 = _mm_mul_ps(rinv23,rinv23); |
| 279 | rinvsq31 = _mm_mul_ps(rinv31,rinv31); |
| 280 | rinvsq32 = _mm_mul_ps(rinv32,rinv32); |
| 281 | rinvsq33 = _mm_mul_ps(rinv33,rinv33); |
| 282 | |
| 283 | fjx1 = _mm_setzero_ps(); |
| 284 | fjy1 = _mm_setzero_ps(); |
| 285 | fjz1 = _mm_setzero_ps(); |
| 286 | fjx2 = _mm_setzero_ps(); |
| 287 | fjy2 = _mm_setzero_ps(); |
| 288 | fjz2 = _mm_setzero_ps(); |
| 289 | fjx3 = _mm_setzero_ps(); |
| 290 | fjy3 = _mm_setzero_ps(); |
| 291 | fjz3 = _mm_setzero_ps(); |
| 292 | |
| 293 | /************************** |
| 294 | * CALCULATE INTERACTIONS * |
| 295 | **************************/ |
| 296 | |
| 297 | if (gmx_mm_any_lt(rsq11,rcutoff2)) |
| 298 | { |
| 299 | |
| 300 | r11 = _mm_mul_ps(rsq11,rinv11); |
| 301 | |
| 302 | /* EWALD ELECTROSTATICS */ |
| 303 | |
| 304 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 305 | ewrt = _mm_mul_ps(r11,ewtabscale); |
| 306 | ewitab = _mm_cvttps_epi32(ewrt); |
| 307 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 308 | ewitab = _mm_slli_epi32(ewitab,2); |
| 309 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 310 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 311 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 312 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 313 | _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); |
| 314 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 315 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 316 | velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_sub_ps(rinv11,sh_ewald),velec)); |
| 317 | felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec)); |
| 318 | |
| 319 | cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2); |
| 320 | |
| 321 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 322 | velec = _mm_and_ps(velec,cutoff_mask); |
| 323 | velecsum = _mm_add_ps(velecsum,velec); |
| 324 | |
| 325 | fscal = felec; |
| 326 | |
| 327 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 328 | |
| 329 | /* Calculate temporary vectorial force */ |
| 330 | tx = _mm_mul_ps(fscal,dx11); |
| 331 | ty = _mm_mul_ps(fscal,dy11); |
| 332 | tz = _mm_mul_ps(fscal,dz11); |
| 333 | |
| 334 | /* Update vectorial force */ |
| 335 | fix1 = _mm_add_ps(fix1,tx); |
| 336 | fiy1 = _mm_add_ps(fiy1,ty); |
| 337 | fiz1 = _mm_add_ps(fiz1,tz); |
| 338 | |
| 339 | fjx1 = _mm_add_ps(fjx1,tx); |
| 340 | fjy1 = _mm_add_ps(fjy1,ty); |
| 341 | fjz1 = _mm_add_ps(fjz1,tz); |
| 342 | |
| 343 | } |
| 344 | |
| 345 | /************************** |
| 346 | * CALCULATE INTERACTIONS * |
| 347 | **************************/ |
| 348 | |
| 349 | if (gmx_mm_any_lt(rsq12,rcutoff2)) |
| 350 | { |
| 351 | |
| 352 | r12 = _mm_mul_ps(rsq12,rinv12); |
| 353 | |
| 354 | /* EWALD ELECTROSTATICS */ |
| 355 | |
| 356 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 357 | ewrt = _mm_mul_ps(r12,ewtabscale); |
| 358 | ewitab = _mm_cvttps_epi32(ewrt); |
| 359 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 360 | ewitab = _mm_slli_epi32(ewitab,2); |
| 361 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 362 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 363 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 364 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 365 | _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); |
| 366 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 367 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 368 | velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_sub_ps(rinv12,sh_ewald),velec)); |
| 369 | felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec)); |
| 370 | |
| 371 | cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2); |
| 372 | |
| 373 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 374 | velec = _mm_and_ps(velec,cutoff_mask); |
| 375 | velecsum = _mm_add_ps(velecsum,velec); |
| 376 | |
| 377 | fscal = felec; |
| 378 | |
| 379 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 380 | |
| 381 | /* Calculate temporary vectorial force */ |
| 382 | tx = _mm_mul_ps(fscal,dx12); |
| 383 | ty = _mm_mul_ps(fscal,dy12); |
| 384 | tz = _mm_mul_ps(fscal,dz12); |
| 385 | |
| 386 | /* Update vectorial force */ |
| 387 | fix1 = _mm_add_ps(fix1,tx); |
| 388 | fiy1 = _mm_add_ps(fiy1,ty); |
| 389 | fiz1 = _mm_add_ps(fiz1,tz); |
| 390 | |
| 391 | fjx2 = _mm_add_ps(fjx2,tx); |
| 392 | fjy2 = _mm_add_ps(fjy2,ty); |
| 393 | fjz2 = _mm_add_ps(fjz2,tz); |
| 394 | |
| 395 | } |
| 396 | |
| 397 | /************************** |
| 398 | * CALCULATE INTERACTIONS * |
| 399 | **************************/ |
| 400 | |
| 401 | if (gmx_mm_any_lt(rsq13,rcutoff2)) |
| 402 | { |
| 403 | |
| 404 | r13 = _mm_mul_ps(rsq13,rinv13); |
| 405 | |
| 406 | /* EWALD ELECTROSTATICS */ |
| 407 | |
| 408 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 409 | ewrt = _mm_mul_ps(r13,ewtabscale); |
| 410 | ewitab = _mm_cvttps_epi32(ewrt); |
| 411 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 412 | ewitab = _mm_slli_epi32(ewitab,2); |
| 413 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 414 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 415 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 416 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 417 | _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); |
| 418 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 419 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 420 | velec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_sub_ps(rinv13,sh_ewald),velec)); |
| 421 | felec = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec)); |
| 422 | |
| 423 | cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2); |
| 424 | |
| 425 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 426 | velec = _mm_and_ps(velec,cutoff_mask); |
| 427 | velecsum = _mm_add_ps(velecsum,velec); |
| 428 | |
| 429 | fscal = felec; |
| 430 | |
| 431 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 432 | |
| 433 | /* Calculate temporary vectorial force */ |
| 434 | tx = _mm_mul_ps(fscal,dx13); |
| 435 | ty = _mm_mul_ps(fscal,dy13); |
| 436 | tz = _mm_mul_ps(fscal,dz13); |
| 437 | |
| 438 | /* Update vectorial force */ |
| 439 | fix1 = _mm_add_ps(fix1,tx); |
| 440 | fiy1 = _mm_add_ps(fiy1,ty); |
| 441 | fiz1 = _mm_add_ps(fiz1,tz); |
| 442 | |
| 443 | fjx3 = _mm_add_ps(fjx3,tx); |
| 444 | fjy3 = _mm_add_ps(fjy3,ty); |
| 445 | fjz3 = _mm_add_ps(fjz3,tz); |
| 446 | |
| 447 | } |
| 448 | |
| 449 | /************************** |
| 450 | * CALCULATE INTERACTIONS * |
| 451 | **************************/ |
| 452 | |
| 453 | if (gmx_mm_any_lt(rsq21,rcutoff2)) |
| 454 | { |
| 455 | |
| 456 | r21 = _mm_mul_ps(rsq21,rinv21); |
| 457 | |
| 458 | /* EWALD ELECTROSTATICS */ |
| 459 | |
| 460 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 461 | ewrt = _mm_mul_ps(r21,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(qq21,_mm_sub_ps(_mm_sub_ps(rinv21,sh_ewald),velec)); |
| 473 | felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec)); |
| 474 | |
| 475 | cutoff_mask = _mm_cmplt_ps(rsq21,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,dx21); |
| 487 | ty = _mm_mul_ps(fscal,dy21); |
| 488 | tz = _mm_mul_ps(fscal,dz21); |
| 489 | |
| 490 | /* Update vectorial force */ |
| 491 | fix2 = _mm_add_ps(fix2,tx); |
| 492 | fiy2 = _mm_add_ps(fiy2,ty); |
| 493 | fiz2 = _mm_add_ps(fiz2,tz); |
| 494 | |
| 495 | fjx1 = _mm_add_ps(fjx1,tx); |
| 496 | fjy1 = _mm_add_ps(fjy1,ty); |
| 497 | fjz1 = _mm_add_ps(fjz1,tz); |
| 498 | |
| 499 | } |
| 500 | |
| 501 | /************************** |
| 502 | * CALCULATE INTERACTIONS * |
| 503 | **************************/ |
| 504 | |
| 505 | if (gmx_mm_any_lt(rsq22,rcutoff2)) |
| 506 | { |
| 507 | |
| 508 | r22 = _mm_mul_ps(rsq22,rinv22); |
| 509 | |
| 510 | /* EWALD ELECTROSTATICS */ |
| 511 | |
| 512 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 513 | ewrt = _mm_mul_ps(r22,ewtabscale); |
| 514 | ewitab = _mm_cvttps_epi32(ewrt); |
| 515 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 516 | ewitab = _mm_slli_epi32(ewitab,2); |
| 517 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 518 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 519 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 520 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 521 | _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); |
| 522 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 523 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 524 | velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_sub_ps(rinv22,sh_ewald),velec)); |
| 525 | felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec)); |
| 526 | |
| 527 | cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2); |
| 528 | |
| 529 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 530 | velec = _mm_and_ps(velec,cutoff_mask); |
| 531 | velecsum = _mm_add_ps(velecsum,velec); |
| 532 | |
| 533 | fscal = felec; |
| 534 | |
| 535 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 536 | |
| 537 | /* Calculate temporary vectorial force */ |
| 538 | tx = _mm_mul_ps(fscal,dx22); |
| 539 | ty = _mm_mul_ps(fscal,dy22); |
| 540 | tz = _mm_mul_ps(fscal,dz22); |
| 541 | |
| 542 | /* Update vectorial force */ |
| 543 | fix2 = _mm_add_ps(fix2,tx); |
| 544 | fiy2 = _mm_add_ps(fiy2,ty); |
| 545 | fiz2 = _mm_add_ps(fiz2,tz); |
| 546 | |
| 547 | fjx2 = _mm_add_ps(fjx2,tx); |
| 548 | fjy2 = _mm_add_ps(fjy2,ty); |
| 549 | fjz2 = _mm_add_ps(fjz2,tz); |
| 550 | |
| 551 | } |
| 552 | |
| 553 | /************************** |
| 554 | * CALCULATE INTERACTIONS * |
| 555 | **************************/ |
| 556 | |
| 557 | if (gmx_mm_any_lt(rsq23,rcutoff2)) |
| 558 | { |
| 559 | |
| 560 | r23 = _mm_mul_ps(rsq23,rinv23); |
| 561 | |
| 562 | /* EWALD ELECTROSTATICS */ |
| 563 | |
| 564 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 565 | ewrt = _mm_mul_ps(r23,ewtabscale); |
| 566 | ewitab = _mm_cvttps_epi32(ewrt); |
| 567 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 568 | ewitab = _mm_slli_epi32(ewitab,2); |
| 569 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 570 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 571 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 572 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 573 | _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); |
| 574 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 575 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 576 | velec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_sub_ps(rinv23,sh_ewald),velec)); |
| 577 | felec = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec)); |
| 578 | |
| 579 | cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2); |
| 580 | |
| 581 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 582 | velec = _mm_and_ps(velec,cutoff_mask); |
| 583 | velecsum = _mm_add_ps(velecsum,velec); |
| 584 | |
| 585 | fscal = felec; |
| 586 | |
| 587 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 588 | |
| 589 | /* Calculate temporary vectorial force */ |
| 590 | tx = _mm_mul_ps(fscal,dx23); |
| 591 | ty = _mm_mul_ps(fscal,dy23); |
| 592 | tz = _mm_mul_ps(fscal,dz23); |
| 593 | |
| 594 | /* Update vectorial force */ |
| 595 | fix2 = _mm_add_ps(fix2,tx); |
| 596 | fiy2 = _mm_add_ps(fiy2,ty); |
| 597 | fiz2 = _mm_add_ps(fiz2,tz); |
| 598 | |
| 599 | fjx3 = _mm_add_ps(fjx3,tx); |
| 600 | fjy3 = _mm_add_ps(fjy3,ty); |
| 601 | fjz3 = _mm_add_ps(fjz3,tz); |
| 602 | |
| 603 | } |
| 604 | |
| 605 | /************************** |
| 606 | * CALCULATE INTERACTIONS * |
| 607 | **************************/ |
| 608 | |
| 609 | if (gmx_mm_any_lt(rsq31,rcutoff2)) |
| 610 | { |
| 611 | |
| 612 | r31 = _mm_mul_ps(rsq31,rinv31); |
| 613 | |
| 614 | /* EWALD ELECTROSTATICS */ |
| 615 | |
| 616 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 617 | ewrt = _mm_mul_ps(r31,ewtabscale); |
| 618 | ewitab = _mm_cvttps_epi32(ewrt); |
| 619 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 620 | ewitab = _mm_slli_epi32(ewitab,2); |
| 621 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 622 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 623 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 624 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 625 | _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); |
| 626 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 627 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 628 | velec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_sub_ps(rinv31,sh_ewald),velec)); |
| 629 | felec = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec)); |
| 630 | |
| 631 | cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2); |
| 632 | |
| 633 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 634 | velec = _mm_and_ps(velec,cutoff_mask); |
| 635 | velecsum = _mm_add_ps(velecsum,velec); |
| 636 | |
| 637 | fscal = felec; |
| 638 | |
| 639 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 640 | |
| 641 | /* Calculate temporary vectorial force */ |
| 642 | tx = _mm_mul_ps(fscal,dx31); |
| 643 | ty = _mm_mul_ps(fscal,dy31); |
| 644 | tz = _mm_mul_ps(fscal,dz31); |
| 645 | |
| 646 | /* Update vectorial force */ |
| 647 | fix3 = _mm_add_ps(fix3,tx); |
| 648 | fiy3 = _mm_add_ps(fiy3,ty); |
| 649 | fiz3 = _mm_add_ps(fiz3,tz); |
| 650 | |
| 651 | fjx1 = _mm_add_ps(fjx1,tx); |
| 652 | fjy1 = _mm_add_ps(fjy1,ty); |
| 653 | fjz1 = _mm_add_ps(fjz1,tz); |
| 654 | |
| 655 | } |
| 656 | |
| 657 | /************************** |
| 658 | * CALCULATE INTERACTIONS * |
| 659 | **************************/ |
| 660 | |
| 661 | if (gmx_mm_any_lt(rsq32,rcutoff2)) |
| 662 | { |
| 663 | |
| 664 | r32 = _mm_mul_ps(rsq32,rinv32); |
| 665 | |
| 666 | /* EWALD ELECTROSTATICS */ |
| 667 | |
| 668 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 669 | ewrt = _mm_mul_ps(r32,ewtabscale); |
| 670 | ewitab = _mm_cvttps_epi32(ewrt); |
| 671 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 672 | ewitab = _mm_slli_epi32(ewitab,2); |
| 673 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 674 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 675 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 676 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 677 | _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); |
| 678 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 679 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 680 | velec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_sub_ps(rinv32,sh_ewald),velec)); |
| 681 | felec = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec)); |
| 682 | |
| 683 | cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2); |
| 684 | |
| 685 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 686 | velec = _mm_and_ps(velec,cutoff_mask); |
| 687 | velecsum = _mm_add_ps(velecsum,velec); |
| 688 | |
| 689 | fscal = felec; |
| 690 | |
| 691 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 692 | |
| 693 | /* Calculate temporary vectorial force */ |
| 694 | tx = _mm_mul_ps(fscal,dx32); |
| 695 | ty = _mm_mul_ps(fscal,dy32); |
| 696 | tz = _mm_mul_ps(fscal,dz32); |
| 697 | |
| 698 | /* Update vectorial force */ |
| 699 | fix3 = _mm_add_ps(fix3,tx); |
| 700 | fiy3 = _mm_add_ps(fiy3,ty); |
| 701 | fiz3 = _mm_add_ps(fiz3,tz); |
| 702 | |
| 703 | fjx2 = _mm_add_ps(fjx2,tx); |
| 704 | fjy2 = _mm_add_ps(fjy2,ty); |
| 705 | fjz2 = _mm_add_ps(fjz2,tz); |
| 706 | |
| 707 | } |
| 708 | |
| 709 | /************************** |
| 710 | * CALCULATE INTERACTIONS * |
| 711 | **************************/ |
| 712 | |
| 713 | if (gmx_mm_any_lt(rsq33,rcutoff2)) |
| 714 | { |
| 715 | |
| 716 | r33 = _mm_mul_ps(rsq33,rinv33); |
| 717 | |
| 718 | /* EWALD ELECTROSTATICS */ |
| 719 | |
| 720 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 721 | ewrt = _mm_mul_ps(r33,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(qq33,_mm_sub_ps(_mm_sub_ps(rinv33,sh_ewald),velec)); |
| 733 | felec = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec)); |
| 734 | |
| 735 | cutoff_mask = _mm_cmplt_ps(rsq33,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 | velecsum = _mm_add_ps(velecsum,velec); |
| 740 | |
| 741 | fscal = felec; |
| 742 | |
| 743 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 744 | |
| 745 | /* Calculate temporary vectorial force */ |
| 746 | tx = _mm_mul_ps(fscal,dx33); |
| 747 | ty = _mm_mul_ps(fscal,dy33); |
| 748 | tz = _mm_mul_ps(fscal,dz33); |
| 749 | |
| 750 | /* Update vectorial force */ |
| 751 | fix3 = _mm_add_ps(fix3,tx); |
| 752 | fiy3 = _mm_add_ps(fiy3,ty); |
| 753 | fiz3 = _mm_add_ps(fiz3,tz); |
| 754 | |
| 755 | fjx3 = _mm_add_ps(fjx3,tx); |
| 756 | fjy3 = _mm_add_ps(fjy3,ty); |
| 757 | fjz3 = _mm_add_ps(fjz3,tz); |
| 758 | |
| 759 | } |
| 760 | |
| 761 | fjptrA = f+j_coord_offsetA; |
| 762 | fjptrB = f+j_coord_offsetB; |
| 763 | fjptrC = f+j_coord_offsetC; |
| 764 | fjptrD = f+j_coord_offsetD; |
| 765 | |
| 766 | gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM3,fjptrB+DIM3,fjptrC+DIM3,fjptrD+DIM3, |
| 767 | fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3); |
| 768 | |
| 769 | /* Inner loop uses 414 flops */ |
| 770 | } |
| 771 | |
| 772 | if(jidx<j_index_end) |
| 773 | { |
| 774 | |
| 775 | /* Get j neighbor index, and coordinate index */ |
| 776 | jnrlistA = jjnr[jidx]; |
| 777 | jnrlistB = jjnr[jidx+1]; |
| 778 | jnrlistC = jjnr[jidx+2]; |
| 779 | jnrlistD = jjnr[jidx+3]; |
| 780 | /* Sign of each element will be negative for non-real atoms. |
| 781 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
| 782 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
| 783 | */ |
| 784 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
| 785 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
| 786 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
| 787 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
| 788 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
| 789 | j_coord_offsetA = DIM3*jnrA; |
| 790 | j_coord_offsetB = DIM3*jnrB; |
| 791 | j_coord_offsetC = DIM3*jnrC; |
| 792 | j_coord_offsetD = DIM3*jnrD; |
| 793 | |
| 794 | /* load j atom coordinates */ |
| 795 | gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM3,x+j_coord_offsetB+DIM3, |
| 796 | x+j_coord_offsetC+DIM3,x+j_coord_offsetD+DIM3, |
| 797 | &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3); |
| 798 | |
| 799 | /* Calculate displacement vector */ |
| 800 | dx11 = _mm_sub_ps(ix1,jx1); |
| 801 | dy11 = _mm_sub_ps(iy1,jy1); |
| 802 | dz11 = _mm_sub_ps(iz1,jz1); |
| 803 | dx12 = _mm_sub_ps(ix1,jx2); |
| 804 | dy12 = _mm_sub_ps(iy1,jy2); |
| 805 | dz12 = _mm_sub_ps(iz1,jz2); |
| 806 | dx13 = _mm_sub_ps(ix1,jx3); |
| 807 | dy13 = _mm_sub_ps(iy1,jy3); |
| 808 | dz13 = _mm_sub_ps(iz1,jz3); |
| 809 | dx21 = _mm_sub_ps(ix2,jx1); |
| 810 | dy21 = _mm_sub_ps(iy2,jy1); |
| 811 | dz21 = _mm_sub_ps(iz2,jz1); |
| 812 | dx22 = _mm_sub_ps(ix2,jx2); |
| 813 | dy22 = _mm_sub_ps(iy2,jy2); |
| 814 | dz22 = _mm_sub_ps(iz2,jz2); |
| 815 | dx23 = _mm_sub_ps(ix2,jx3); |
| 816 | dy23 = _mm_sub_ps(iy2,jy3); |
| 817 | dz23 = _mm_sub_ps(iz2,jz3); |
| 818 | dx31 = _mm_sub_ps(ix3,jx1); |
| 819 | dy31 = _mm_sub_ps(iy3,jy1); |
| 820 | dz31 = _mm_sub_ps(iz3,jz1); |
| 821 | dx32 = _mm_sub_ps(ix3,jx2); |
| 822 | dy32 = _mm_sub_ps(iy3,jy2); |
| 823 | dz32 = _mm_sub_ps(iz3,jz2); |
| 824 | dx33 = _mm_sub_ps(ix3,jx3); |
| 825 | dy33 = _mm_sub_ps(iy3,jy3); |
| 826 | dz33 = _mm_sub_ps(iz3,jz3); |
| 827 | |
| 828 | /* Calculate squared distance and things based on it */ |
| 829 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
| 830 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
| 831 | rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13); |
| 832 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
| 833 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
| 834 | rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23); |
| 835 | rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31); |
| 836 | rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32); |
| 837 | rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33); |
| 838 | |
| 839 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
| 840 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
| 841 | rinv13 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq13); |
| 842 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
| 843 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
| 844 | rinv23 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq23); |
| 845 | rinv31 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq31); |
| 846 | rinv32 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq32); |
| 847 | rinv33 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq33); |
| 848 | |
| 849 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
| 850 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
| 851 | rinvsq13 = _mm_mul_ps(rinv13,rinv13); |
| 852 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
| 853 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
| 854 | rinvsq23 = _mm_mul_ps(rinv23,rinv23); |
| 855 | rinvsq31 = _mm_mul_ps(rinv31,rinv31); |
| 856 | rinvsq32 = _mm_mul_ps(rinv32,rinv32); |
| 857 | rinvsq33 = _mm_mul_ps(rinv33,rinv33); |
| 858 | |
| 859 | fjx1 = _mm_setzero_ps(); |
| 860 | fjy1 = _mm_setzero_ps(); |
| 861 | fjz1 = _mm_setzero_ps(); |
| 862 | fjx2 = _mm_setzero_ps(); |
| 863 | fjy2 = _mm_setzero_ps(); |
| 864 | fjz2 = _mm_setzero_ps(); |
| 865 | fjx3 = _mm_setzero_ps(); |
| 866 | fjy3 = _mm_setzero_ps(); |
| 867 | fjz3 = _mm_setzero_ps(); |
| 868 | |
| 869 | /************************** |
| 870 | * CALCULATE INTERACTIONS * |
| 871 | **************************/ |
| 872 | |
| 873 | if (gmx_mm_any_lt(rsq11,rcutoff2)) |
| 874 | { |
| 875 | |
| 876 | r11 = _mm_mul_ps(rsq11,rinv11); |
| 877 | r11 = _mm_andnot_ps(dummy_mask,r11); |
| 878 | |
| 879 | /* EWALD ELECTROSTATICS */ |
| 880 | |
| 881 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 882 | ewrt = _mm_mul_ps(r11,ewtabscale); |
| 883 | ewitab = _mm_cvttps_epi32(ewrt); |
| 884 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 885 | ewitab = _mm_slli_epi32(ewitab,2); |
| 886 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 887 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 888 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 889 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 890 | _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); |
| 891 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 892 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 893 | velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_sub_ps(rinv11,sh_ewald),velec)); |
| 894 | felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec)); |
| 895 | |
| 896 | cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2); |
| 897 | |
| 898 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 899 | velec = _mm_and_ps(velec,cutoff_mask); |
| 900 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 901 | velecsum = _mm_add_ps(velecsum,velec); |
| 902 | |
| 903 | fscal = felec; |
| 904 | |
| 905 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 906 | |
| 907 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 908 | |
| 909 | /* Calculate temporary vectorial force */ |
| 910 | tx = _mm_mul_ps(fscal,dx11); |
| 911 | ty = _mm_mul_ps(fscal,dy11); |
| 912 | tz = _mm_mul_ps(fscal,dz11); |
| 913 | |
| 914 | /* Update vectorial force */ |
| 915 | fix1 = _mm_add_ps(fix1,tx); |
| 916 | fiy1 = _mm_add_ps(fiy1,ty); |
| 917 | fiz1 = _mm_add_ps(fiz1,tz); |
| 918 | |
| 919 | fjx1 = _mm_add_ps(fjx1,tx); |
| 920 | fjy1 = _mm_add_ps(fjy1,ty); |
| 921 | fjz1 = _mm_add_ps(fjz1,tz); |
| 922 | |
| 923 | } |
| 924 | |
| 925 | /************************** |
| 926 | * CALCULATE INTERACTIONS * |
| 927 | **************************/ |
| 928 | |
| 929 | if (gmx_mm_any_lt(rsq12,rcutoff2)) |
| 930 | { |
| 931 | |
| 932 | r12 = _mm_mul_ps(rsq12,rinv12); |
| 933 | r12 = _mm_andnot_ps(dummy_mask,r12); |
| 934 | |
| 935 | /* EWALD ELECTROSTATICS */ |
| 936 | |
| 937 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 938 | ewrt = _mm_mul_ps(r12,ewtabscale); |
| 939 | ewitab = _mm_cvttps_epi32(ewrt); |
| 940 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 941 | ewitab = _mm_slli_epi32(ewitab,2); |
| 942 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 943 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 944 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 945 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 946 | _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); |
| 947 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 948 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 949 | velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_sub_ps(rinv12,sh_ewald),velec)); |
| 950 | felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec)); |
| 951 | |
| 952 | cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2); |
| 953 | |
| 954 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 955 | velec = _mm_and_ps(velec,cutoff_mask); |
| 956 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 957 | velecsum = _mm_add_ps(velecsum,velec); |
| 958 | |
| 959 | fscal = felec; |
| 960 | |
| 961 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 962 | |
| 963 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 964 | |
| 965 | /* Calculate temporary vectorial force */ |
| 966 | tx = _mm_mul_ps(fscal,dx12); |
| 967 | ty = _mm_mul_ps(fscal,dy12); |
| 968 | tz = _mm_mul_ps(fscal,dz12); |
| 969 | |
| 970 | /* Update vectorial force */ |
| 971 | fix1 = _mm_add_ps(fix1,tx); |
| 972 | fiy1 = _mm_add_ps(fiy1,ty); |
| 973 | fiz1 = _mm_add_ps(fiz1,tz); |
| 974 | |
| 975 | fjx2 = _mm_add_ps(fjx2,tx); |
| 976 | fjy2 = _mm_add_ps(fjy2,ty); |
| 977 | fjz2 = _mm_add_ps(fjz2,tz); |
| 978 | |
| 979 | } |
| 980 | |
| 981 | /************************** |
| 982 | * CALCULATE INTERACTIONS * |
| 983 | **************************/ |
| 984 | |
| 985 | if (gmx_mm_any_lt(rsq13,rcutoff2)) |
| 986 | { |
| 987 | |
| 988 | r13 = _mm_mul_ps(rsq13,rinv13); |
| 989 | r13 = _mm_andnot_ps(dummy_mask,r13); |
| 990 | |
| 991 | /* EWALD ELECTROSTATICS */ |
| 992 | |
| 993 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 994 | ewrt = _mm_mul_ps(r13,ewtabscale); |
| 995 | ewitab = _mm_cvttps_epi32(ewrt); |
| 996 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 997 | ewitab = _mm_slli_epi32(ewitab,2); |
| 998 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 999 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 1000 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 1001 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 1002 | _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); |
| 1003 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 1004 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 1005 | velec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_sub_ps(rinv13,sh_ewald),velec)); |
| 1006 | felec = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec)); |
| 1007 | |
| 1008 | cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2); |
| 1009 | |
| 1010 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 1011 | velec = _mm_and_ps(velec,cutoff_mask); |
| 1012 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 1013 | velecsum = _mm_add_ps(velecsum,velec); |
| 1014 | |
| 1015 | fscal = felec; |
| 1016 | |
| 1017 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1018 | |
| 1019 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1020 | |
| 1021 | /* Calculate temporary vectorial force */ |
| 1022 | tx = _mm_mul_ps(fscal,dx13); |
| 1023 | ty = _mm_mul_ps(fscal,dy13); |
| 1024 | tz = _mm_mul_ps(fscal,dz13); |
| 1025 | |
| 1026 | /* Update vectorial force */ |
| 1027 | fix1 = _mm_add_ps(fix1,tx); |
| 1028 | fiy1 = _mm_add_ps(fiy1,ty); |
| 1029 | fiz1 = _mm_add_ps(fiz1,tz); |
| 1030 | |
| 1031 | fjx3 = _mm_add_ps(fjx3,tx); |
| 1032 | fjy3 = _mm_add_ps(fjy3,ty); |
| 1033 | fjz3 = _mm_add_ps(fjz3,tz); |
| 1034 | |
| 1035 | } |
| 1036 | |
| 1037 | /************************** |
| 1038 | * CALCULATE INTERACTIONS * |
| 1039 | **************************/ |
| 1040 | |
| 1041 | if (gmx_mm_any_lt(rsq21,rcutoff2)) |
| 1042 | { |
| 1043 | |
| 1044 | r21 = _mm_mul_ps(rsq21,rinv21); |
| 1045 | r21 = _mm_andnot_ps(dummy_mask,r21); |
| 1046 | |
| 1047 | /* EWALD ELECTROSTATICS */ |
| 1048 | |
| 1049 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1050 | ewrt = _mm_mul_ps(r21,ewtabscale); |
| 1051 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1052 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1053 | ewitab = _mm_slli_epi32(ewitab,2); |
| 1054 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 1055 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 1056 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 1057 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 1058 | _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); |
| 1059 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 1060 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 1061 | velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_sub_ps(rinv21,sh_ewald),velec)); |
| 1062 | felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec)); |
| 1063 | |
| 1064 | cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2); |
| 1065 | |
| 1066 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 1067 | velec = _mm_and_ps(velec,cutoff_mask); |
| 1068 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 1069 | velecsum = _mm_add_ps(velecsum,velec); |
| 1070 | |
| 1071 | fscal = felec; |
| 1072 | |
| 1073 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1074 | |
| 1075 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1076 | |
| 1077 | /* Calculate temporary vectorial force */ |
| 1078 | tx = _mm_mul_ps(fscal,dx21); |
| 1079 | ty = _mm_mul_ps(fscal,dy21); |
| 1080 | tz = _mm_mul_ps(fscal,dz21); |
| 1081 | |
| 1082 | /* Update vectorial force */ |
| 1083 | fix2 = _mm_add_ps(fix2,tx); |
| 1084 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1085 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1086 | |
| 1087 | fjx1 = _mm_add_ps(fjx1,tx); |
| 1088 | fjy1 = _mm_add_ps(fjy1,ty); |
| 1089 | fjz1 = _mm_add_ps(fjz1,tz); |
| 1090 | |
| 1091 | } |
| 1092 | |
| 1093 | /************************** |
| 1094 | * CALCULATE INTERACTIONS * |
| 1095 | **************************/ |
| 1096 | |
| 1097 | if (gmx_mm_any_lt(rsq22,rcutoff2)) |
| 1098 | { |
| 1099 | |
| 1100 | r22 = _mm_mul_ps(rsq22,rinv22); |
| 1101 | r22 = _mm_andnot_ps(dummy_mask,r22); |
| 1102 | |
| 1103 | /* EWALD ELECTROSTATICS */ |
| 1104 | |
| 1105 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1106 | ewrt = _mm_mul_ps(r22,ewtabscale); |
| 1107 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1108 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1109 | ewitab = _mm_slli_epi32(ewitab,2); |
| 1110 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 1111 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 1112 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 1113 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 1114 | _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); |
| 1115 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 1116 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 1117 | velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_sub_ps(rinv22,sh_ewald),velec)); |
| 1118 | felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec)); |
| 1119 | |
| 1120 | cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2); |
| 1121 | |
| 1122 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 1123 | velec = _mm_and_ps(velec,cutoff_mask); |
| 1124 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 1125 | velecsum = _mm_add_ps(velecsum,velec); |
| 1126 | |
| 1127 | fscal = felec; |
| 1128 | |
| 1129 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1130 | |
| 1131 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1132 | |
| 1133 | /* Calculate temporary vectorial force */ |
| 1134 | tx = _mm_mul_ps(fscal,dx22); |
| 1135 | ty = _mm_mul_ps(fscal,dy22); |
| 1136 | tz = _mm_mul_ps(fscal,dz22); |
| 1137 | |
| 1138 | /* Update vectorial force */ |
| 1139 | fix2 = _mm_add_ps(fix2,tx); |
| 1140 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1141 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1142 | |
| 1143 | fjx2 = _mm_add_ps(fjx2,tx); |
| 1144 | fjy2 = _mm_add_ps(fjy2,ty); |
| 1145 | fjz2 = _mm_add_ps(fjz2,tz); |
| 1146 | |
| 1147 | } |
| 1148 | |
| 1149 | /************************** |
| 1150 | * CALCULATE INTERACTIONS * |
| 1151 | **************************/ |
| 1152 | |
| 1153 | if (gmx_mm_any_lt(rsq23,rcutoff2)) |
| 1154 | { |
| 1155 | |
| 1156 | r23 = _mm_mul_ps(rsq23,rinv23); |
| 1157 | r23 = _mm_andnot_ps(dummy_mask,r23); |
| 1158 | |
| 1159 | /* EWALD ELECTROSTATICS */ |
| 1160 | |
| 1161 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1162 | ewrt = _mm_mul_ps(r23,ewtabscale); |
| 1163 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1164 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1165 | ewitab = _mm_slli_epi32(ewitab,2); |
| 1166 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 1167 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 1168 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 1169 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 1170 | _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); |
| 1171 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 1172 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 1173 | velec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_sub_ps(rinv23,sh_ewald),velec)); |
| 1174 | felec = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec)); |
| 1175 | |
| 1176 | cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2); |
| 1177 | |
| 1178 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 1179 | velec = _mm_and_ps(velec,cutoff_mask); |
| 1180 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 1181 | velecsum = _mm_add_ps(velecsum,velec); |
| 1182 | |
| 1183 | fscal = felec; |
| 1184 | |
| 1185 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1186 | |
| 1187 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1188 | |
| 1189 | /* Calculate temporary vectorial force */ |
| 1190 | tx = _mm_mul_ps(fscal,dx23); |
| 1191 | ty = _mm_mul_ps(fscal,dy23); |
| 1192 | tz = _mm_mul_ps(fscal,dz23); |
| 1193 | |
| 1194 | /* Update vectorial force */ |
| 1195 | fix2 = _mm_add_ps(fix2,tx); |
| 1196 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1197 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1198 | |
| 1199 | fjx3 = _mm_add_ps(fjx3,tx); |
| 1200 | fjy3 = _mm_add_ps(fjy3,ty); |
| 1201 | fjz3 = _mm_add_ps(fjz3,tz); |
| 1202 | |
| 1203 | } |
| 1204 | |
| 1205 | /************************** |
| 1206 | * CALCULATE INTERACTIONS * |
| 1207 | **************************/ |
| 1208 | |
| 1209 | if (gmx_mm_any_lt(rsq31,rcutoff2)) |
| 1210 | { |
| 1211 | |
| 1212 | r31 = _mm_mul_ps(rsq31,rinv31); |
| 1213 | r31 = _mm_andnot_ps(dummy_mask,r31); |
| 1214 | |
| 1215 | /* EWALD ELECTROSTATICS */ |
| 1216 | |
| 1217 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1218 | ewrt = _mm_mul_ps(r31,ewtabscale); |
| 1219 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1220 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1221 | ewitab = _mm_slli_epi32(ewitab,2); |
| 1222 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 1223 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 1224 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 1225 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 1226 | _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); |
| 1227 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 1228 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 1229 | velec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_sub_ps(rinv31,sh_ewald),velec)); |
| 1230 | felec = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec)); |
| 1231 | |
| 1232 | cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2); |
| 1233 | |
| 1234 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 1235 | velec = _mm_and_ps(velec,cutoff_mask); |
| 1236 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 1237 | velecsum = _mm_add_ps(velecsum,velec); |
| 1238 | |
| 1239 | fscal = felec; |
| 1240 | |
| 1241 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1242 | |
| 1243 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1244 | |
| 1245 | /* Calculate temporary vectorial force */ |
| 1246 | tx = _mm_mul_ps(fscal,dx31); |
| 1247 | ty = _mm_mul_ps(fscal,dy31); |
| 1248 | tz = _mm_mul_ps(fscal,dz31); |
| 1249 | |
| 1250 | /* Update vectorial force */ |
| 1251 | fix3 = _mm_add_ps(fix3,tx); |
| 1252 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1253 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1254 | |
| 1255 | fjx1 = _mm_add_ps(fjx1,tx); |
| 1256 | fjy1 = _mm_add_ps(fjy1,ty); |
| 1257 | fjz1 = _mm_add_ps(fjz1,tz); |
| 1258 | |
| 1259 | } |
| 1260 | |
| 1261 | /************************** |
| 1262 | * CALCULATE INTERACTIONS * |
| 1263 | **************************/ |
| 1264 | |
| 1265 | if (gmx_mm_any_lt(rsq32,rcutoff2)) |
| 1266 | { |
| 1267 | |
| 1268 | r32 = _mm_mul_ps(rsq32,rinv32); |
| 1269 | r32 = _mm_andnot_ps(dummy_mask,r32); |
| 1270 | |
| 1271 | /* EWALD ELECTROSTATICS */ |
| 1272 | |
| 1273 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1274 | ewrt = _mm_mul_ps(r32,ewtabscale); |
| 1275 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1276 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1277 | ewitab = _mm_slli_epi32(ewitab,2); |
| 1278 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 1279 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 1280 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 1281 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 1282 | _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); |
| 1283 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 1284 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 1285 | velec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_sub_ps(rinv32,sh_ewald),velec)); |
| 1286 | felec = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec)); |
| 1287 | |
| 1288 | cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2); |
| 1289 | |
| 1290 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 1291 | velec = _mm_and_ps(velec,cutoff_mask); |
| 1292 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 1293 | velecsum = _mm_add_ps(velecsum,velec); |
| 1294 | |
| 1295 | fscal = felec; |
| 1296 | |
| 1297 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1298 | |
| 1299 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1300 | |
| 1301 | /* Calculate temporary vectorial force */ |
| 1302 | tx = _mm_mul_ps(fscal,dx32); |
| 1303 | ty = _mm_mul_ps(fscal,dy32); |
| 1304 | tz = _mm_mul_ps(fscal,dz32); |
| 1305 | |
| 1306 | /* Update vectorial force */ |
| 1307 | fix3 = _mm_add_ps(fix3,tx); |
| 1308 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1309 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1310 | |
| 1311 | fjx2 = _mm_add_ps(fjx2,tx); |
| 1312 | fjy2 = _mm_add_ps(fjy2,ty); |
| 1313 | fjz2 = _mm_add_ps(fjz2,tz); |
| 1314 | |
| 1315 | } |
| 1316 | |
| 1317 | /************************** |
| 1318 | * CALCULATE INTERACTIONS * |
| 1319 | **************************/ |
| 1320 | |
| 1321 | if (gmx_mm_any_lt(rsq33,rcutoff2)) |
| 1322 | { |
| 1323 | |
| 1324 | r33 = _mm_mul_ps(rsq33,rinv33); |
| 1325 | r33 = _mm_andnot_ps(dummy_mask,r33); |
| 1326 | |
| 1327 | /* EWALD ELECTROSTATICS */ |
| 1328 | |
| 1329 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1330 | ewrt = _mm_mul_ps(r33,ewtabscale); |
| 1331 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1332 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1333 | ewitab = _mm_slli_epi32(ewitab,2); |
| 1334 | ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(0) & 3];})) ); |
| 1335 | ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(1) & 3];})) ); |
| 1336 | ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(2) & 3];})) ); |
| 1337 | ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3)(__extension__ ({ __v4si __a = (__v4si)(ewitab); __a[(3) & 3];})) ); |
| 1338 | _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); |
| 1339 | felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD)); |
| 1340 | velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec))); |
| 1341 | velec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_sub_ps(rinv33,sh_ewald),velec)); |
| 1342 | felec = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec)); |
| 1343 | |
| 1344 | cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2); |
| 1345 | |
| 1346 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
| 1347 | velec = _mm_and_ps(velec,cutoff_mask); |
| 1348 | velec = _mm_andnot_ps(dummy_mask,velec); |
| 1349 | velecsum = _mm_add_ps(velecsum,velec); |
| 1350 | |
| 1351 | fscal = felec; |
| 1352 | |
| 1353 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1354 | |
| 1355 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 1356 | |
| 1357 | /* Calculate temporary vectorial force */ |
| 1358 | tx = _mm_mul_ps(fscal,dx33); |
| 1359 | ty = _mm_mul_ps(fscal,dy33); |
| 1360 | tz = _mm_mul_ps(fscal,dz33); |
| 1361 | |
| 1362 | /* Update vectorial force */ |
| 1363 | fix3 = _mm_add_ps(fix3,tx); |
| 1364 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1365 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1366 | |
| 1367 | fjx3 = _mm_add_ps(fjx3,tx); |
| 1368 | fjy3 = _mm_add_ps(fjy3,ty); |
| 1369 | fjz3 = _mm_add_ps(fjz3,tz); |
| 1370 | |
| 1371 | } |
| 1372 | |
| 1373 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
| 1374 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
| 1375 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
| 1376 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
| 1377 | |
| 1378 | gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM3,fjptrB+DIM3,fjptrC+DIM3,fjptrD+DIM3, |
| 1379 | fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3); |
| 1380 | |
| 1381 | /* Inner loop uses 423 flops */ |
| 1382 | } |
| 1383 | |
| 1384 | /* End of innermost loop */ |
| 1385 | |
| 1386 | gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3, |
| 1387 | f+i_coord_offset+DIM3,fshift+i_shift_offset); |
| 1388 | |
| 1389 | ggid = gid[iidx]; |
| 1390 | /* Update potential energies */ |
| 1391 | gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid); |
| 1392 | |
| 1393 | /* Increment number of inner iterations */ |
| 1394 | inneriter += j_index_end - j_index_start; |
| 1395 | |
| 1396 | /* Outer loop uses 19 flops */ |
| 1397 | } |
| 1398 | |
| 1399 | /* Increment number of outer iterations */ |
| 1400 | outeriter += nri; |
| 1401 | |
| 1402 | /* Update outer/inner flops */ |
| 1403 | |
| 1404 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*423)(nrnb)->n[eNR_NBKERNEL_ELEC_W4W4_VF] += outeriter*19 + inneriter *423; |
| 1405 | } |
| 1406 | /* |
| 1407 | * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW4W4_F_sse4_1_single |
| 1408 | * Electrostatics interaction: Ewald |
| 1409 | * VdW interaction: None |
| 1410 | * Geometry: Water4-Water4 |
| 1411 | * Calculate force/pot: Force |
| 1412 | */ |
| 1413 | void |
| 1414 | nb_kernel_ElecEwSh_VdwNone_GeomW4W4_F_sse4_1_single |
| 1415 | (t_nblist * gmx_restrict nlist, |
| 1416 | rvec * gmx_restrict xx, |
| 1417 | rvec * gmx_restrict ff, |
| 1418 | t_forcerec * gmx_restrict fr, |
| 1419 | t_mdatoms * gmx_restrict mdatoms, |
| 1420 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
| 1421 | t_nrnb * gmx_restrict nrnb) |
| 1422 | { |
| 1423 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
| 1424 | * just 0 for non-waters. |
| 1425 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
| 1426 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
| 1427 | */ |
| 1428 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
| 1429 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
| 1430 | int jnrA,jnrB,jnrC,jnrD; |
| 1431 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
| 1432 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
| 1433 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
| 1434 | real rcutoff_scalar; |
| 1435 | real *shiftvec,*fshift,*x,*f; |
| 1436 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
| 1437 | real scratch[4*DIM3]; |
| 1438 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
| 1439 | int vdwioffset1; |
| 1440 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
| 1441 | int vdwioffset2; |
| 1442 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
| 1443 | int vdwioffset3; |
| 1444 | __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3; |
| 1445 | int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D; |
| 1446 | __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1; |
| 1447 | int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D; |
| 1448 | __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2; |
| 1449 | int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D; |
| 1450 | __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3; |
| 1451 | __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11; |
| 1452 | __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12; |
| 1453 | __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13; |
| 1454 | __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21; |
| 1455 | __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22; |
| 1456 | __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23; |
| 1457 | __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31; |
| 1458 | __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32; |
| 1459 | __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33; |
| 1460 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
| 1461 | real *charge; |
| 1462 | __m128i ewitab; |
| 1463 | __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV; |
| 1464 | real *ewtab; |
| 1465 | __m128 dummy_mask,cutoff_mask; |
| 1466 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
| 1467 | __m128 one = _mm_set1_ps(1.0); |
| 1468 | __m128 two = _mm_set1_ps(2.0); |
| 1469 | x = xx[0]; |
| 1470 | f = ff[0]; |
| 1471 | |
| 1472 | nri = nlist->nri; |
| 1473 | iinr = nlist->iinr; |
| 1474 | jindex = nlist->jindex; |
| 1475 | jjnr = nlist->jjnr; |
| 1476 | shiftidx = nlist->shift; |
| 1477 | gid = nlist->gid; |
| 1478 | shiftvec = fr->shift_vec[0]; |
| 1479 | fshift = fr->fshift[0]; |
| 1480 | facel = _mm_set1_ps(fr->epsfac); |
| 1481 | charge = mdatoms->chargeA; |
| 1482 | |
| 1483 | sh_ewald = _mm_set1_ps(fr->ic->sh_ewald); |
| 1484 | ewtab = fr->ic->tabq_coul_F; |
| 1485 | ewtabscale = _mm_set1_ps(fr->ic->tabq_scale); |
| 1486 | ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale); |
| 1487 | |
| 1488 | /* Setup water-specific parameters */ |
| 1489 | inr = nlist->iinr[0]; |
| 1490 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
| 1491 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
| 1492 | iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3])); |
| 1493 | |
| 1494 | jq1 = _mm_set1_ps(charge[inr+1]); |
| 1495 | jq2 = _mm_set1_ps(charge[inr+2]); |
| 1496 | jq3 = _mm_set1_ps(charge[inr+3]); |
| 1497 | qq11 = _mm_mul_ps(iq1,jq1); |
| 1498 | qq12 = _mm_mul_ps(iq1,jq2); |
| 1499 | qq13 = _mm_mul_ps(iq1,jq3); |
| 1500 | qq21 = _mm_mul_ps(iq2,jq1); |
| 1501 | qq22 = _mm_mul_ps(iq2,jq2); |
| 1502 | qq23 = _mm_mul_ps(iq2,jq3); |
| 1503 | qq31 = _mm_mul_ps(iq3,jq1); |
| 1504 | qq32 = _mm_mul_ps(iq3,jq2); |
| 1505 | qq33 = _mm_mul_ps(iq3,jq3); |
| 1506 | |
| 1507 | /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */ |
| 1508 | rcutoff_scalar = fr->rcoulomb; |
| 1509 | rcutoff = _mm_set1_ps(rcutoff_scalar); |
| 1510 | rcutoff2 = _mm_mul_ps(rcutoff,rcutoff); |
| 1511 | |
| 1512 | /* Avoid stupid compiler warnings */ |
| 1513 | jnrA = jnrB = jnrC = jnrD = 0; |
| 1514 | j_coord_offsetA = 0; |
| 1515 | j_coord_offsetB = 0; |
| 1516 | j_coord_offsetC = 0; |
| 1517 | j_coord_offsetD = 0; |
Value stored to 'j_coord_offsetD' is never read | |
| 1518 | |
| 1519 | outeriter = 0; |
| 1520 | inneriter = 0; |
| 1521 | |
| 1522 | for(iidx=0;iidx<4*DIM3;iidx++) |
| 1523 | { |
| 1524 | scratch[iidx] = 0.0; |
| 1525 | } |
| 1526 | |
| 1527 | /* Start outer loop over neighborlists */ |
| 1528 | for(iidx=0; iidx<nri; iidx++) |
| 1529 | { |
| 1530 | /* Load shift vector for this list */ |
| 1531 | i_shift_offset = DIM3*shiftidx[iidx]; |
| 1532 | |
| 1533 | /* Load limits for loop over neighbors */ |
| 1534 | j_index_start = jindex[iidx]; |
| 1535 | j_index_end = jindex[iidx+1]; |
| 1536 | |
| 1537 | /* Get outer coordinate index */ |
| 1538 | inr = iinr[iidx]; |
| 1539 | i_coord_offset = DIM3*inr; |
| 1540 | |
| 1541 | /* Load i particle coords and add shift vector */ |
| 1542 | gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM3, |
| 1543 | &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3); |
| 1544 | |
| 1545 | fix1 = _mm_setzero_ps(); |
| 1546 | fiy1 = _mm_setzero_ps(); |
| 1547 | fiz1 = _mm_setzero_ps(); |
| 1548 | fix2 = _mm_setzero_ps(); |
| 1549 | fiy2 = _mm_setzero_ps(); |
| 1550 | fiz2 = _mm_setzero_ps(); |
| 1551 | fix3 = _mm_setzero_ps(); |
| 1552 | fiy3 = _mm_setzero_ps(); |
| 1553 | fiz3 = _mm_setzero_ps(); |
| 1554 | |
| 1555 | /* Start inner kernel loop */ |
| 1556 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
| 1557 | { |
| 1558 | |
| 1559 | /* Get j neighbor index, and coordinate index */ |
| 1560 | jnrA = jjnr[jidx]; |
| 1561 | jnrB = jjnr[jidx+1]; |
| 1562 | jnrC = jjnr[jidx+2]; |
| 1563 | jnrD = jjnr[jidx+3]; |
| 1564 | j_coord_offsetA = DIM3*jnrA; |
| 1565 | j_coord_offsetB = DIM3*jnrB; |
| 1566 | j_coord_offsetC = DIM3*jnrC; |
| 1567 | j_coord_offsetD = DIM3*jnrD; |
| 1568 | |
| 1569 | /* load j atom coordinates */ |
| 1570 | gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM3,x+j_coord_offsetB+DIM3, |
| 1571 | x+j_coord_offsetC+DIM3,x+j_coord_offsetD+DIM3, |
| 1572 | &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3); |
| 1573 | |
| 1574 | /* Calculate displacement vector */ |
| 1575 | dx11 = _mm_sub_ps(ix1,jx1); |
| 1576 | dy11 = _mm_sub_ps(iy1,jy1); |
| 1577 | dz11 = _mm_sub_ps(iz1,jz1); |
| 1578 | dx12 = _mm_sub_ps(ix1,jx2); |
| 1579 | dy12 = _mm_sub_ps(iy1,jy2); |
| 1580 | dz12 = _mm_sub_ps(iz1,jz2); |
| 1581 | dx13 = _mm_sub_ps(ix1,jx3); |
| 1582 | dy13 = _mm_sub_ps(iy1,jy3); |
| 1583 | dz13 = _mm_sub_ps(iz1,jz3); |
| 1584 | dx21 = _mm_sub_ps(ix2,jx1); |
| 1585 | dy21 = _mm_sub_ps(iy2,jy1); |
| 1586 | dz21 = _mm_sub_ps(iz2,jz1); |
| 1587 | dx22 = _mm_sub_ps(ix2,jx2); |
| 1588 | dy22 = _mm_sub_ps(iy2,jy2); |
| 1589 | dz22 = _mm_sub_ps(iz2,jz2); |
| 1590 | dx23 = _mm_sub_ps(ix2,jx3); |
| 1591 | dy23 = _mm_sub_ps(iy2,jy3); |
| 1592 | dz23 = _mm_sub_ps(iz2,jz3); |
| 1593 | dx31 = _mm_sub_ps(ix3,jx1); |
| 1594 | dy31 = _mm_sub_ps(iy3,jy1); |
| 1595 | dz31 = _mm_sub_ps(iz3,jz1); |
| 1596 | dx32 = _mm_sub_ps(ix3,jx2); |
| 1597 | dy32 = _mm_sub_ps(iy3,jy2); |
| 1598 | dz32 = _mm_sub_ps(iz3,jz2); |
| 1599 | dx33 = _mm_sub_ps(ix3,jx3); |
| 1600 | dy33 = _mm_sub_ps(iy3,jy3); |
| 1601 | dz33 = _mm_sub_ps(iz3,jz3); |
| 1602 | |
| 1603 | /* Calculate squared distance and things based on it */ |
| 1604 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
| 1605 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
| 1606 | rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13); |
| 1607 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
| 1608 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
| 1609 | rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23); |
| 1610 | rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31); |
| 1611 | rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32); |
| 1612 | rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33); |
| 1613 | |
| 1614 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
| 1615 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
| 1616 | rinv13 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq13); |
| 1617 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
| 1618 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
| 1619 | rinv23 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq23); |
| 1620 | rinv31 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq31); |
| 1621 | rinv32 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq32); |
| 1622 | rinv33 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq33); |
| 1623 | |
| 1624 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
| 1625 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
| 1626 | rinvsq13 = _mm_mul_ps(rinv13,rinv13); |
| 1627 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
| 1628 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
| 1629 | rinvsq23 = _mm_mul_ps(rinv23,rinv23); |
| 1630 | rinvsq31 = _mm_mul_ps(rinv31,rinv31); |
| 1631 | rinvsq32 = _mm_mul_ps(rinv32,rinv32); |
| 1632 | rinvsq33 = _mm_mul_ps(rinv33,rinv33); |
| 1633 | |
| 1634 | fjx1 = _mm_setzero_ps(); |
| 1635 | fjy1 = _mm_setzero_ps(); |
| 1636 | fjz1 = _mm_setzero_ps(); |
| 1637 | fjx2 = _mm_setzero_ps(); |
| 1638 | fjy2 = _mm_setzero_ps(); |
| 1639 | fjz2 = _mm_setzero_ps(); |
| 1640 | fjx3 = _mm_setzero_ps(); |
| 1641 | fjy3 = _mm_setzero_ps(); |
| 1642 | fjz3 = _mm_setzero_ps(); |
| 1643 | |
| 1644 | /************************** |
| 1645 | * CALCULATE INTERACTIONS * |
| 1646 | **************************/ |
| 1647 | |
| 1648 | if (gmx_mm_any_lt(rsq11,rcutoff2)) |
| 1649 | { |
| 1650 | |
| 1651 | r11 = _mm_mul_ps(rsq11,rinv11); |
| 1652 | |
| 1653 | /* EWALD ELECTROSTATICS */ |
| 1654 | |
| 1655 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1656 | ewrt = _mm_mul_ps(r11,ewtabscale); |
| 1657 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1658 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1659 | 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];})), |
| 1660 | 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];})), |
| 1661 | &ewtabF,&ewtabFn); |
| 1662 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1663 | felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec)); |
| 1664 | |
| 1665 | cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2); |
| 1666 | |
| 1667 | fscal = felec; |
| 1668 | |
| 1669 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1670 | |
| 1671 | /* Calculate temporary vectorial force */ |
| 1672 | tx = _mm_mul_ps(fscal,dx11); |
| 1673 | ty = _mm_mul_ps(fscal,dy11); |
| 1674 | tz = _mm_mul_ps(fscal,dz11); |
| 1675 | |
| 1676 | /* Update vectorial force */ |
| 1677 | fix1 = _mm_add_ps(fix1,tx); |
| 1678 | fiy1 = _mm_add_ps(fiy1,ty); |
| 1679 | fiz1 = _mm_add_ps(fiz1,tz); |
| 1680 | |
| 1681 | fjx1 = _mm_add_ps(fjx1,tx); |
| 1682 | fjy1 = _mm_add_ps(fjy1,ty); |
| 1683 | fjz1 = _mm_add_ps(fjz1,tz); |
| 1684 | |
| 1685 | } |
| 1686 | |
| 1687 | /************************** |
| 1688 | * CALCULATE INTERACTIONS * |
| 1689 | **************************/ |
| 1690 | |
| 1691 | if (gmx_mm_any_lt(rsq12,rcutoff2)) |
| 1692 | { |
| 1693 | |
| 1694 | r12 = _mm_mul_ps(rsq12,rinv12); |
| 1695 | |
| 1696 | /* EWALD ELECTROSTATICS */ |
| 1697 | |
| 1698 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1699 | ewrt = _mm_mul_ps(r12,ewtabscale); |
| 1700 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1701 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1702 | 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];})), |
| 1703 | 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];})), |
| 1704 | &ewtabF,&ewtabFn); |
| 1705 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1706 | felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec)); |
| 1707 | |
| 1708 | cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2); |
| 1709 | |
| 1710 | fscal = felec; |
| 1711 | |
| 1712 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1713 | |
| 1714 | /* Calculate temporary vectorial force */ |
| 1715 | tx = _mm_mul_ps(fscal,dx12); |
| 1716 | ty = _mm_mul_ps(fscal,dy12); |
| 1717 | tz = _mm_mul_ps(fscal,dz12); |
| 1718 | |
| 1719 | /* Update vectorial force */ |
| 1720 | fix1 = _mm_add_ps(fix1,tx); |
| 1721 | fiy1 = _mm_add_ps(fiy1,ty); |
| 1722 | fiz1 = _mm_add_ps(fiz1,tz); |
| 1723 | |
| 1724 | fjx2 = _mm_add_ps(fjx2,tx); |
| 1725 | fjy2 = _mm_add_ps(fjy2,ty); |
| 1726 | fjz2 = _mm_add_ps(fjz2,tz); |
| 1727 | |
| 1728 | } |
| 1729 | |
| 1730 | /************************** |
| 1731 | * CALCULATE INTERACTIONS * |
| 1732 | **************************/ |
| 1733 | |
| 1734 | if (gmx_mm_any_lt(rsq13,rcutoff2)) |
| 1735 | { |
| 1736 | |
| 1737 | r13 = _mm_mul_ps(rsq13,rinv13); |
| 1738 | |
| 1739 | /* EWALD ELECTROSTATICS */ |
| 1740 | |
| 1741 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1742 | ewrt = _mm_mul_ps(r13,ewtabscale); |
| 1743 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1744 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1745 | 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];})), |
| 1746 | 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];})), |
| 1747 | &ewtabF,&ewtabFn); |
| 1748 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1749 | felec = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec)); |
| 1750 | |
| 1751 | cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2); |
| 1752 | |
| 1753 | fscal = felec; |
| 1754 | |
| 1755 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1756 | |
| 1757 | /* Calculate temporary vectorial force */ |
| 1758 | tx = _mm_mul_ps(fscal,dx13); |
| 1759 | ty = _mm_mul_ps(fscal,dy13); |
| 1760 | tz = _mm_mul_ps(fscal,dz13); |
| 1761 | |
| 1762 | /* Update vectorial force */ |
| 1763 | fix1 = _mm_add_ps(fix1,tx); |
| 1764 | fiy1 = _mm_add_ps(fiy1,ty); |
| 1765 | fiz1 = _mm_add_ps(fiz1,tz); |
| 1766 | |
| 1767 | fjx3 = _mm_add_ps(fjx3,tx); |
| 1768 | fjy3 = _mm_add_ps(fjy3,ty); |
| 1769 | fjz3 = _mm_add_ps(fjz3,tz); |
| 1770 | |
| 1771 | } |
| 1772 | |
| 1773 | /************************** |
| 1774 | * CALCULATE INTERACTIONS * |
| 1775 | **************************/ |
| 1776 | |
| 1777 | if (gmx_mm_any_lt(rsq21,rcutoff2)) |
| 1778 | { |
| 1779 | |
| 1780 | r21 = _mm_mul_ps(rsq21,rinv21); |
| 1781 | |
| 1782 | /* EWALD ELECTROSTATICS */ |
| 1783 | |
| 1784 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1785 | ewrt = _mm_mul_ps(r21,ewtabscale); |
| 1786 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1787 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1788 | 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];})), |
| 1789 | 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];})), |
| 1790 | &ewtabF,&ewtabFn); |
| 1791 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1792 | felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec)); |
| 1793 | |
| 1794 | cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2); |
| 1795 | |
| 1796 | fscal = felec; |
| 1797 | |
| 1798 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1799 | |
| 1800 | /* Calculate temporary vectorial force */ |
| 1801 | tx = _mm_mul_ps(fscal,dx21); |
| 1802 | ty = _mm_mul_ps(fscal,dy21); |
| 1803 | tz = _mm_mul_ps(fscal,dz21); |
| 1804 | |
| 1805 | /* Update vectorial force */ |
| 1806 | fix2 = _mm_add_ps(fix2,tx); |
| 1807 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1808 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1809 | |
| 1810 | fjx1 = _mm_add_ps(fjx1,tx); |
| 1811 | fjy1 = _mm_add_ps(fjy1,ty); |
| 1812 | fjz1 = _mm_add_ps(fjz1,tz); |
| 1813 | |
| 1814 | } |
| 1815 | |
| 1816 | /************************** |
| 1817 | * CALCULATE INTERACTIONS * |
| 1818 | **************************/ |
| 1819 | |
| 1820 | if (gmx_mm_any_lt(rsq22,rcutoff2)) |
| 1821 | { |
| 1822 | |
| 1823 | r22 = _mm_mul_ps(rsq22,rinv22); |
| 1824 | |
| 1825 | /* EWALD ELECTROSTATICS */ |
| 1826 | |
| 1827 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1828 | ewrt = _mm_mul_ps(r22,ewtabscale); |
| 1829 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1830 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1831 | 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];})), |
| 1832 | 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];})), |
| 1833 | &ewtabF,&ewtabFn); |
| 1834 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1835 | felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec)); |
| 1836 | |
| 1837 | cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2); |
| 1838 | |
| 1839 | fscal = felec; |
| 1840 | |
| 1841 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1842 | |
| 1843 | /* Calculate temporary vectorial force */ |
| 1844 | tx = _mm_mul_ps(fscal,dx22); |
| 1845 | ty = _mm_mul_ps(fscal,dy22); |
| 1846 | tz = _mm_mul_ps(fscal,dz22); |
| 1847 | |
| 1848 | /* Update vectorial force */ |
| 1849 | fix2 = _mm_add_ps(fix2,tx); |
| 1850 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1851 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1852 | |
| 1853 | fjx2 = _mm_add_ps(fjx2,tx); |
| 1854 | fjy2 = _mm_add_ps(fjy2,ty); |
| 1855 | fjz2 = _mm_add_ps(fjz2,tz); |
| 1856 | |
| 1857 | } |
| 1858 | |
| 1859 | /************************** |
| 1860 | * CALCULATE INTERACTIONS * |
| 1861 | **************************/ |
| 1862 | |
| 1863 | if (gmx_mm_any_lt(rsq23,rcutoff2)) |
| 1864 | { |
| 1865 | |
| 1866 | r23 = _mm_mul_ps(rsq23,rinv23); |
| 1867 | |
| 1868 | /* EWALD ELECTROSTATICS */ |
| 1869 | |
| 1870 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1871 | ewrt = _mm_mul_ps(r23,ewtabscale); |
| 1872 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1873 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1874 | 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];})), |
| 1875 | 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];})), |
| 1876 | &ewtabF,&ewtabFn); |
| 1877 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1878 | felec = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec)); |
| 1879 | |
| 1880 | cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2); |
| 1881 | |
| 1882 | fscal = felec; |
| 1883 | |
| 1884 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1885 | |
| 1886 | /* Calculate temporary vectorial force */ |
| 1887 | tx = _mm_mul_ps(fscal,dx23); |
| 1888 | ty = _mm_mul_ps(fscal,dy23); |
| 1889 | tz = _mm_mul_ps(fscal,dz23); |
| 1890 | |
| 1891 | /* Update vectorial force */ |
| 1892 | fix2 = _mm_add_ps(fix2,tx); |
| 1893 | fiy2 = _mm_add_ps(fiy2,ty); |
| 1894 | fiz2 = _mm_add_ps(fiz2,tz); |
| 1895 | |
| 1896 | fjx3 = _mm_add_ps(fjx3,tx); |
| 1897 | fjy3 = _mm_add_ps(fjy3,ty); |
| 1898 | fjz3 = _mm_add_ps(fjz3,tz); |
| 1899 | |
| 1900 | } |
| 1901 | |
| 1902 | /************************** |
| 1903 | * CALCULATE INTERACTIONS * |
| 1904 | **************************/ |
| 1905 | |
| 1906 | if (gmx_mm_any_lt(rsq31,rcutoff2)) |
| 1907 | { |
| 1908 | |
| 1909 | r31 = _mm_mul_ps(rsq31,rinv31); |
| 1910 | |
| 1911 | /* EWALD ELECTROSTATICS */ |
| 1912 | |
| 1913 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1914 | ewrt = _mm_mul_ps(r31,ewtabscale); |
| 1915 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1916 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1917 | 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];})), |
| 1918 | 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];})), |
| 1919 | &ewtabF,&ewtabFn); |
| 1920 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1921 | felec = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec)); |
| 1922 | |
| 1923 | cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2); |
| 1924 | |
| 1925 | fscal = felec; |
| 1926 | |
| 1927 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1928 | |
| 1929 | /* Calculate temporary vectorial force */ |
| 1930 | tx = _mm_mul_ps(fscal,dx31); |
| 1931 | ty = _mm_mul_ps(fscal,dy31); |
| 1932 | tz = _mm_mul_ps(fscal,dz31); |
| 1933 | |
| 1934 | /* Update vectorial force */ |
| 1935 | fix3 = _mm_add_ps(fix3,tx); |
| 1936 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1937 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1938 | |
| 1939 | fjx1 = _mm_add_ps(fjx1,tx); |
| 1940 | fjy1 = _mm_add_ps(fjy1,ty); |
| 1941 | fjz1 = _mm_add_ps(fjz1,tz); |
| 1942 | |
| 1943 | } |
| 1944 | |
| 1945 | /************************** |
| 1946 | * CALCULATE INTERACTIONS * |
| 1947 | **************************/ |
| 1948 | |
| 1949 | if (gmx_mm_any_lt(rsq32,rcutoff2)) |
| 1950 | { |
| 1951 | |
| 1952 | r32 = _mm_mul_ps(rsq32,rinv32); |
| 1953 | |
| 1954 | /* EWALD ELECTROSTATICS */ |
| 1955 | |
| 1956 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 1957 | ewrt = _mm_mul_ps(r32,ewtabscale); |
| 1958 | ewitab = _mm_cvttps_epi32(ewrt); |
| 1959 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 1960 | 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];})), |
| 1961 | 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];})), |
| 1962 | &ewtabF,&ewtabFn); |
| 1963 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 1964 | felec = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec)); |
| 1965 | |
| 1966 | cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2); |
| 1967 | |
| 1968 | fscal = felec; |
| 1969 | |
| 1970 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 1971 | |
| 1972 | /* Calculate temporary vectorial force */ |
| 1973 | tx = _mm_mul_ps(fscal,dx32); |
| 1974 | ty = _mm_mul_ps(fscal,dy32); |
| 1975 | tz = _mm_mul_ps(fscal,dz32); |
| 1976 | |
| 1977 | /* Update vectorial force */ |
| 1978 | fix3 = _mm_add_ps(fix3,tx); |
| 1979 | fiy3 = _mm_add_ps(fiy3,ty); |
| 1980 | fiz3 = _mm_add_ps(fiz3,tz); |
| 1981 | |
| 1982 | fjx2 = _mm_add_ps(fjx2,tx); |
| 1983 | fjy2 = _mm_add_ps(fjy2,ty); |
| 1984 | fjz2 = _mm_add_ps(fjz2,tz); |
| 1985 | |
| 1986 | } |
| 1987 | |
| 1988 | /************************** |
| 1989 | * CALCULATE INTERACTIONS * |
| 1990 | **************************/ |
| 1991 | |
| 1992 | if (gmx_mm_any_lt(rsq33,rcutoff2)) |
| 1993 | { |
| 1994 | |
| 1995 | r33 = _mm_mul_ps(rsq33,rinv33); |
| 1996 | |
| 1997 | /* EWALD ELECTROSTATICS */ |
| 1998 | |
| 1999 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 2000 | ewrt = _mm_mul_ps(r33,ewtabscale); |
| 2001 | ewitab = _mm_cvttps_epi32(ewrt); |
| 2002 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 2003 | 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];})), |
| 2004 | 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];})), |
| 2005 | &ewtabF,&ewtabFn); |
| 2006 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 2007 | felec = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec)); |
| 2008 | |
| 2009 | cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2); |
| 2010 | |
| 2011 | fscal = felec; |
| 2012 | |
| 2013 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 2014 | |
| 2015 | /* Calculate temporary vectorial force */ |
| 2016 | tx = _mm_mul_ps(fscal,dx33); |
| 2017 | ty = _mm_mul_ps(fscal,dy33); |
| 2018 | tz = _mm_mul_ps(fscal,dz33); |
| 2019 | |
| 2020 | /* Update vectorial force */ |
| 2021 | fix3 = _mm_add_ps(fix3,tx); |
| 2022 | fiy3 = _mm_add_ps(fiy3,ty); |
| 2023 | fiz3 = _mm_add_ps(fiz3,tz); |
| 2024 | |
| 2025 | fjx3 = _mm_add_ps(fjx3,tx); |
| 2026 | fjy3 = _mm_add_ps(fjy3,ty); |
| 2027 | fjz3 = _mm_add_ps(fjz3,tz); |
| 2028 | |
| 2029 | } |
| 2030 | |
| 2031 | fjptrA = f+j_coord_offsetA; |
| 2032 | fjptrB = f+j_coord_offsetB; |
| 2033 | fjptrC = f+j_coord_offsetC; |
| 2034 | fjptrD = f+j_coord_offsetD; |
| 2035 | |
| 2036 | gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM3,fjptrB+DIM3,fjptrC+DIM3,fjptrD+DIM3, |
| 2037 | fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3); |
| 2038 | |
| 2039 | /* Inner loop uses 351 flops */ |
| 2040 | } |
| 2041 | |
| 2042 | if(jidx<j_index_end) |
| 2043 | { |
| 2044 | |
| 2045 | /* Get j neighbor index, and coordinate index */ |
| 2046 | jnrlistA = jjnr[jidx]; |
| 2047 | jnrlistB = jjnr[jidx+1]; |
| 2048 | jnrlistC = jjnr[jidx+2]; |
| 2049 | jnrlistD = jjnr[jidx+3]; |
| 2050 | /* Sign of each element will be negative for non-real atoms. |
| 2051 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
| 2052 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
| 2053 | */ |
| 2054 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
| 2055 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
| 2056 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
| 2057 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
| 2058 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
| 2059 | j_coord_offsetA = DIM3*jnrA; |
| 2060 | j_coord_offsetB = DIM3*jnrB; |
| 2061 | j_coord_offsetC = DIM3*jnrC; |
| 2062 | j_coord_offsetD = DIM3*jnrD; |
| 2063 | |
| 2064 | /* load j atom coordinates */ |
| 2065 | gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM3,x+j_coord_offsetB+DIM3, |
| 2066 | x+j_coord_offsetC+DIM3,x+j_coord_offsetD+DIM3, |
| 2067 | &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3); |
| 2068 | |
| 2069 | /* Calculate displacement vector */ |
| 2070 | dx11 = _mm_sub_ps(ix1,jx1); |
| 2071 | dy11 = _mm_sub_ps(iy1,jy1); |
| 2072 | dz11 = _mm_sub_ps(iz1,jz1); |
| 2073 | dx12 = _mm_sub_ps(ix1,jx2); |
| 2074 | dy12 = _mm_sub_ps(iy1,jy2); |
| 2075 | dz12 = _mm_sub_ps(iz1,jz2); |
| 2076 | dx13 = _mm_sub_ps(ix1,jx3); |
| 2077 | dy13 = _mm_sub_ps(iy1,jy3); |
| 2078 | dz13 = _mm_sub_ps(iz1,jz3); |
| 2079 | dx21 = _mm_sub_ps(ix2,jx1); |
| 2080 | dy21 = _mm_sub_ps(iy2,jy1); |
| 2081 | dz21 = _mm_sub_ps(iz2,jz1); |
| 2082 | dx22 = _mm_sub_ps(ix2,jx2); |
| 2083 | dy22 = _mm_sub_ps(iy2,jy2); |
| 2084 | dz22 = _mm_sub_ps(iz2,jz2); |
| 2085 | dx23 = _mm_sub_ps(ix2,jx3); |
| 2086 | dy23 = _mm_sub_ps(iy2,jy3); |
| 2087 | dz23 = _mm_sub_ps(iz2,jz3); |
| 2088 | dx31 = _mm_sub_ps(ix3,jx1); |
| 2089 | dy31 = _mm_sub_ps(iy3,jy1); |
| 2090 | dz31 = _mm_sub_ps(iz3,jz1); |
| 2091 | dx32 = _mm_sub_ps(ix3,jx2); |
| 2092 | dy32 = _mm_sub_ps(iy3,jy2); |
| 2093 | dz32 = _mm_sub_ps(iz3,jz2); |
| 2094 | dx33 = _mm_sub_ps(ix3,jx3); |
| 2095 | dy33 = _mm_sub_ps(iy3,jy3); |
| 2096 | dz33 = _mm_sub_ps(iz3,jz3); |
| 2097 | |
| 2098 | /* Calculate squared distance and things based on it */ |
| 2099 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
| 2100 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
| 2101 | rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13); |
| 2102 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
| 2103 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
| 2104 | rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23); |
| 2105 | rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31); |
| 2106 | rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32); |
| 2107 | rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33); |
| 2108 | |
| 2109 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
| 2110 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
| 2111 | rinv13 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq13); |
| 2112 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
| 2113 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
| 2114 | rinv23 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq23); |
| 2115 | rinv31 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq31); |
| 2116 | rinv32 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq32); |
| 2117 | rinv33 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq33); |
| 2118 | |
| 2119 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
| 2120 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
| 2121 | rinvsq13 = _mm_mul_ps(rinv13,rinv13); |
| 2122 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
| 2123 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
| 2124 | rinvsq23 = _mm_mul_ps(rinv23,rinv23); |
| 2125 | rinvsq31 = _mm_mul_ps(rinv31,rinv31); |
| 2126 | rinvsq32 = _mm_mul_ps(rinv32,rinv32); |
| 2127 | rinvsq33 = _mm_mul_ps(rinv33,rinv33); |
| 2128 | |
| 2129 | fjx1 = _mm_setzero_ps(); |
| 2130 | fjy1 = _mm_setzero_ps(); |
| 2131 | fjz1 = _mm_setzero_ps(); |
| 2132 | fjx2 = _mm_setzero_ps(); |
| 2133 | fjy2 = _mm_setzero_ps(); |
| 2134 | fjz2 = _mm_setzero_ps(); |
| 2135 | fjx3 = _mm_setzero_ps(); |
| 2136 | fjy3 = _mm_setzero_ps(); |
| 2137 | fjz3 = _mm_setzero_ps(); |
| 2138 | |
| 2139 | /************************** |
| 2140 | * CALCULATE INTERACTIONS * |
| 2141 | **************************/ |
| 2142 | |
| 2143 | if (gmx_mm_any_lt(rsq11,rcutoff2)) |
| 2144 | { |
| 2145 | |
| 2146 | r11 = _mm_mul_ps(rsq11,rinv11); |
| 2147 | r11 = _mm_andnot_ps(dummy_mask,r11); |
| 2148 | |
| 2149 | /* EWALD ELECTROSTATICS */ |
| 2150 | |
| 2151 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 2152 | ewrt = _mm_mul_ps(r11,ewtabscale); |
| 2153 | ewitab = _mm_cvttps_epi32(ewrt); |
| 2154 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 2155 | 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];})), |
| 2156 | 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];})), |
| 2157 | &ewtabF,&ewtabFn); |
| 2158 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 2159 | felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec)); |
| 2160 | |
| 2161 | cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2); |
| 2162 | |
| 2163 | fscal = felec; |
| 2164 | |
| 2165 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 2166 | |
| 2167 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 2168 | |
| 2169 | /* Calculate temporary vectorial force */ |
| 2170 | tx = _mm_mul_ps(fscal,dx11); |
| 2171 | ty = _mm_mul_ps(fscal,dy11); |
| 2172 | tz = _mm_mul_ps(fscal,dz11); |
| 2173 | |
| 2174 | /* Update vectorial force */ |
| 2175 | fix1 = _mm_add_ps(fix1,tx); |
| 2176 | fiy1 = _mm_add_ps(fiy1,ty); |
| 2177 | fiz1 = _mm_add_ps(fiz1,tz); |
| 2178 | |
| 2179 | fjx1 = _mm_add_ps(fjx1,tx); |
| 2180 | fjy1 = _mm_add_ps(fjy1,ty); |
| 2181 | fjz1 = _mm_add_ps(fjz1,tz); |
| 2182 | |
| 2183 | } |
| 2184 | |
| 2185 | /************************** |
| 2186 | * CALCULATE INTERACTIONS * |
| 2187 | **************************/ |
| 2188 | |
| 2189 | if (gmx_mm_any_lt(rsq12,rcutoff2)) |
| 2190 | { |
| 2191 | |
| 2192 | r12 = _mm_mul_ps(rsq12,rinv12); |
| 2193 | r12 = _mm_andnot_ps(dummy_mask,r12); |
| 2194 | |
| 2195 | /* EWALD ELECTROSTATICS */ |
| 2196 | |
| 2197 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 2198 | ewrt = _mm_mul_ps(r12,ewtabscale); |
| 2199 | ewitab = _mm_cvttps_epi32(ewrt); |
| 2200 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 2201 | 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];})), |
| 2202 | 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];})), |
| 2203 | &ewtabF,&ewtabFn); |
| 2204 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 2205 | felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec)); |
| 2206 | |
| 2207 | cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2); |
| 2208 | |
| 2209 | fscal = felec; |
| 2210 | |
| 2211 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 2212 | |
| 2213 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 2214 | |
| 2215 | /* Calculate temporary vectorial force */ |
| 2216 | tx = _mm_mul_ps(fscal,dx12); |
| 2217 | ty = _mm_mul_ps(fscal,dy12); |
| 2218 | tz = _mm_mul_ps(fscal,dz12); |
| 2219 | |
| 2220 | /* Update vectorial force */ |
| 2221 | fix1 = _mm_add_ps(fix1,tx); |
| 2222 | fiy1 = _mm_add_ps(fiy1,ty); |
| 2223 | fiz1 = _mm_add_ps(fiz1,tz); |
| 2224 | |
| 2225 | fjx2 = _mm_add_ps(fjx2,tx); |
| 2226 | fjy2 = _mm_add_ps(fjy2,ty); |
| 2227 | fjz2 = _mm_add_ps(fjz2,tz); |
| 2228 | |
| 2229 | } |
| 2230 | |
| 2231 | /************************** |
| 2232 | * CALCULATE INTERACTIONS * |
| 2233 | **************************/ |
| 2234 | |
| 2235 | if (gmx_mm_any_lt(rsq13,rcutoff2)) |
| 2236 | { |
| 2237 | |
| 2238 | r13 = _mm_mul_ps(rsq13,rinv13); |
| 2239 | r13 = _mm_andnot_ps(dummy_mask,r13); |
| 2240 | |
| 2241 | /* EWALD ELECTROSTATICS */ |
| 2242 | |
| 2243 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 2244 | ewrt = _mm_mul_ps(r13,ewtabscale); |
| 2245 | ewitab = _mm_cvttps_epi32(ewrt); |
| 2246 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 2247 | 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];})), |
| 2248 | 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];})), |
| 2249 | &ewtabF,&ewtabFn); |
| 2250 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 2251 | felec = _mm_mul_ps(_mm_mul_ps(qq13,rinv13),_mm_sub_ps(rinvsq13,felec)); |
| 2252 | |
| 2253 | cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2); |
| 2254 | |
| 2255 | fscal = felec; |
| 2256 | |
| 2257 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 2258 | |
| 2259 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 2260 | |
| 2261 | /* Calculate temporary vectorial force */ |
| 2262 | tx = _mm_mul_ps(fscal,dx13); |
| 2263 | ty = _mm_mul_ps(fscal,dy13); |
| 2264 | tz = _mm_mul_ps(fscal,dz13); |
| 2265 | |
| 2266 | /* Update vectorial force */ |
| 2267 | fix1 = _mm_add_ps(fix1,tx); |
| 2268 | fiy1 = _mm_add_ps(fiy1,ty); |
| 2269 | fiz1 = _mm_add_ps(fiz1,tz); |
| 2270 | |
| 2271 | fjx3 = _mm_add_ps(fjx3,tx); |
| 2272 | fjy3 = _mm_add_ps(fjy3,ty); |
| 2273 | fjz3 = _mm_add_ps(fjz3,tz); |
| 2274 | |
| 2275 | } |
| 2276 | |
| 2277 | /************************** |
| 2278 | * CALCULATE INTERACTIONS * |
| 2279 | **************************/ |
| 2280 | |
| 2281 | if (gmx_mm_any_lt(rsq21,rcutoff2)) |
| 2282 | { |
| 2283 | |
| 2284 | r21 = _mm_mul_ps(rsq21,rinv21); |
| 2285 | r21 = _mm_andnot_ps(dummy_mask,r21); |
| 2286 | |
| 2287 | /* EWALD ELECTROSTATICS */ |
| 2288 | |
| 2289 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 2290 | ewrt = _mm_mul_ps(r21,ewtabscale); |
| 2291 | ewitab = _mm_cvttps_epi32(ewrt); |
| 2292 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 2293 | 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];})), |
| 2294 | 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];})), |
| 2295 | &ewtabF,&ewtabFn); |
| 2296 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 2297 | felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec)); |
| 2298 | |
| 2299 | cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2); |
| 2300 | |
| 2301 | fscal = felec; |
| 2302 | |
| 2303 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 2304 | |
| 2305 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 2306 | |
| 2307 | /* Calculate temporary vectorial force */ |
| 2308 | tx = _mm_mul_ps(fscal,dx21); |
| 2309 | ty = _mm_mul_ps(fscal,dy21); |
| 2310 | tz = _mm_mul_ps(fscal,dz21); |
| 2311 | |
| 2312 | /* Update vectorial force */ |
| 2313 | fix2 = _mm_add_ps(fix2,tx); |
| 2314 | fiy2 = _mm_add_ps(fiy2,ty); |
| 2315 | fiz2 = _mm_add_ps(fiz2,tz); |
| 2316 | |
| 2317 | fjx1 = _mm_add_ps(fjx1,tx); |
| 2318 | fjy1 = _mm_add_ps(fjy1,ty); |
| 2319 | fjz1 = _mm_add_ps(fjz1,tz); |
| 2320 | |
| 2321 | } |
| 2322 | |
| 2323 | /************************** |
| 2324 | * CALCULATE INTERACTIONS * |
| 2325 | **************************/ |
| 2326 | |
| 2327 | if (gmx_mm_any_lt(rsq22,rcutoff2)) |
| 2328 | { |
| 2329 | |
| 2330 | r22 = _mm_mul_ps(rsq22,rinv22); |
| 2331 | r22 = _mm_andnot_ps(dummy_mask,r22); |
| 2332 | |
| 2333 | /* EWALD ELECTROSTATICS */ |
| 2334 | |
| 2335 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 2336 | ewrt = _mm_mul_ps(r22,ewtabscale); |
| 2337 | ewitab = _mm_cvttps_epi32(ewrt); |
| 2338 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 2339 | 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];})), |
| 2340 | 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];})), |
| 2341 | &ewtabF,&ewtabFn); |
| 2342 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 2343 | felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec)); |
| 2344 | |
| 2345 | cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2); |
| 2346 | |
| 2347 | fscal = felec; |
| 2348 | |
| 2349 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 2350 | |
| 2351 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 2352 | |
| 2353 | /* Calculate temporary vectorial force */ |
| 2354 | tx = _mm_mul_ps(fscal,dx22); |
| 2355 | ty = _mm_mul_ps(fscal,dy22); |
| 2356 | tz = _mm_mul_ps(fscal,dz22); |
| 2357 | |
| 2358 | /* Update vectorial force */ |
| 2359 | fix2 = _mm_add_ps(fix2,tx); |
| 2360 | fiy2 = _mm_add_ps(fiy2,ty); |
| 2361 | fiz2 = _mm_add_ps(fiz2,tz); |
| 2362 | |
| 2363 | fjx2 = _mm_add_ps(fjx2,tx); |
| 2364 | fjy2 = _mm_add_ps(fjy2,ty); |
| 2365 | fjz2 = _mm_add_ps(fjz2,tz); |
| 2366 | |
| 2367 | } |
| 2368 | |
| 2369 | /************************** |
| 2370 | * CALCULATE INTERACTIONS * |
| 2371 | **************************/ |
| 2372 | |
| 2373 | if (gmx_mm_any_lt(rsq23,rcutoff2)) |
| 2374 | { |
| 2375 | |
| 2376 | r23 = _mm_mul_ps(rsq23,rinv23); |
| 2377 | r23 = _mm_andnot_ps(dummy_mask,r23); |
| 2378 | |
| 2379 | /* EWALD ELECTROSTATICS */ |
| 2380 | |
| 2381 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 2382 | ewrt = _mm_mul_ps(r23,ewtabscale); |
| 2383 | ewitab = _mm_cvttps_epi32(ewrt); |
| 2384 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 2385 | 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];})), |
| 2386 | 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];})), |
| 2387 | &ewtabF,&ewtabFn); |
| 2388 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 2389 | felec = _mm_mul_ps(_mm_mul_ps(qq23,rinv23),_mm_sub_ps(rinvsq23,felec)); |
| 2390 | |
| 2391 | cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2); |
| 2392 | |
| 2393 | fscal = felec; |
| 2394 | |
| 2395 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 2396 | |
| 2397 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 2398 | |
| 2399 | /* Calculate temporary vectorial force */ |
| 2400 | tx = _mm_mul_ps(fscal,dx23); |
| 2401 | ty = _mm_mul_ps(fscal,dy23); |
| 2402 | tz = _mm_mul_ps(fscal,dz23); |
| 2403 | |
| 2404 | /* Update vectorial force */ |
| 2405 | fix2 = _mm_add_ps(fix2,tx); |
| 2406 | fiy2 = _mm_add_ps(fiy2,ty); |
| 2407 | fiz2 = _mm_add_ps(fiz2,tz); |
| 2408 | |
| 2409 | fjx3 = _mm_add_ps(fjx3,tx); |
| 2410 | fjy3 = _mm_add_ps(fjy3,ty); |
| 2411 | fjz3 = _mm_add_ps(fjz3,tz); |
| 2412 | |
| 2413 | } |
| 2414 | |
| 2415 | /************************** |
| 2416 | * CALCULATE INTERACTIONS * |
| 2417 | **************************/ |
| 2418 | |
| 2419 | if (gmx_mm_any_lt(rsq31,rcutoff2)) |
| 2420 | { |
| 2421 | |
| 2422 | r31 = _mm_mul_ps(rsq31,rinv31); |
| 2423 | r31 = _mm_andnot_ps(dummy_mask,r31); |
| 2424 | |
| 2425 | /* EWALD ELECTROSTATICS */ |
| 2426 | |
| 2427 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 2428 | ewrt = _mm_mul_ps(r31,ewtabscale); |
| 2429 | ewitab = _mm_cvttps_epi32(ewrt); |
| 2430 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 2431 | 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];})), |
| 2432 | 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];})), |
| 2433 | &ewtabF,&ewtabFn); |
| 2434 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 2435 | felec = _mm_mul_ps(_mm_mul_ps(qq31,rinv31),_mm_sub_ps(rinvsq31,felec)); |
| 2436 | |
| 2437 | cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2); |
| 2438 | |
| 2439 | fscal = felec; |
| 2440 | |
| 2441 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 2442 | |
| 2443 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 2444 | |
| 2445 | /* Calculate temporary vectorial force */ |
| 2446 | tx = _mm_mul_ps(fscal,dx31); |
| 2447 | ty = _mm_mul_ps(fscal,dy31); |
| 2448 | tz = _mm_mul_ps(fscal,dz31); |
| 2449 | |
| 2450 | /* Update vectorial force */ |
| 2451 | fix3 = _mm_add_ps(fix3,tx); |
| 2452 | fiy3 = _mm_add_ps(fiy3,ty); |
| 2453 | fiz3 = _mm_add_ps(fiz3,tz); |
| 2454 | |
| 2455 | fjx1 = _mm_add_ps(fjx1,tx); |
| 2456 | fjy1 = _mm_add_ps(fjy1,ty); |
| 2457 | fjz1 = _mm_add_ps(fjz1,tz); |
| 2458 | |
| 2459 | } |
| 2460 | |
| 2461 | /************************** |
| 2462 | * CALCULATE INTERACTIONS * |
| 2463 | **************************/ |
| 2464 | |
| 2465 | if (gmx_mm_any_lt(rsq32,rcutoff2)) |
| 2466 | { |
| 2467 | |
| 2468 | r32 = _mm_mul_ps(rsq32,rinv32); |
| 2469 | r32 = _mm_andnot_ps(dummy_mask,r32); |
| 2470 | |
| 2471 | /* EWALD ELECTROSTATICS */ |
| 2472 | |
| 2473 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 2474 | ewrt = _mm_mul_ps(r32,ewtabscale); |
| 2475 | ewitab = _mm_cvttps_epi32(ewrt); |
| 2476 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 2477 | 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];})), |
| 2478 | 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];})), |
| 2479 | &ewtabF,&ewtabFn); |
| 2480 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 2481 | felec = _mm_mul_ps(_mm_mul_ps(qq32,rinv32),_mm_sub_ps(rinvsq32,felec)); |
| 2482 | |
| 2483 | cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2); |
| 2484 | |
| 2485 | fscal = felec; |
| 2486 | |
| 2487 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 2488 | |
| 2489 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 2490 | |
| 2491 | /* Calculate temporary vectorial force */ |
| 2492 | tx = _mm_mul_ps(fscal,dx32); |
| 2493 | ty = _mm_mul_ps(fscal,dy32); |
| 2494 | tz = _mm_mul_ps(fscal,dz32); |
| 2495 | |
| 2496 | /* Update vectorial force */ |
| 2497 | fix3 = _mm_add_ps(fix3,tx); |
| 2498 | fiy3 = _mm_add_ps(fiy3,ty); |
| 2499 | fiz3 = _mm_add_ps(fiz3,tz); |
| 2500 | |
| 2501 | fjx2 = _mm_add_ps(fjx2,tx); |
| 2502 | fjy2 = _mm_add_ps(fjy2,ty); |
| 2503 | fjz2 = _mm_add_ps(fjz2,tz); |
| 2504 | |
| 2505 | } |
| 2506 | |
| 2507 | /************************** |
| 2508 | * CALCULATE INTERACTIONS * |
| 2509 | **************************/ |
| 2510 | |
| 2511 | if (gmx_mm_any_lt(rsq33,rcutoff2)) |
| 2512 | { |
| 2513 | |
| 2514 | r33 = _mm_mul_ps(rsq33,rinv33); |
| 2515 | r33 = _mm_andnot_ps(dummy_mask,r33); |
| 2516 | |
| 2517 | /* EWALD ELECTROSTATICS */ |
| 2518 | |
| 2519 | /* Calculate Ewald table index by multiplying r with scale and truncate to integer */ |
| 2520 | ewrt = _mm_mul_ps(r33,ewtabscale); |
| 2521 | ewitab = _mm_cvttps_epi32(ewrt); |
| 2522 | eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (ewrt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
| 2523 | 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];})), |
| 2524 | 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];})), |
| 2525 | &ewtabF,&ewtabFn); |
| 2526 | felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn)); |
| 2527 | felec = _mm_mul_ps(_mm_mul_ps(qq33,rinv33),_mm_sub_ps(rinvsq33,felec)); |
| 2528 | |
| 2529 | cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2); |
| 2530 | |
| 2531 | fscal = felec; |
| 2532 | |
| 2533 | fscal = _mm_and_ps(fscal,cutoff_mask); |
| 2534 | |
| 2535 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
| 2536 | |
| 2537 | /* Calculate temporary vectorial force */ |
| 2538 | tx = _mm_mul_ps(fscal,dx33); |
| 2539 | ty = _mm_mul_ps(fscal,dy33); |
| 2540 | tz = _mm_mul_ps(fscal,dz33); |
| 2541 | |
| 2542 | /* Update vectorial force */ |
| 2543 | fix3 = _mm_add_ps(fix3,tx); |
| 2544 | fiy3 = _mm_add_ps(fiy3,ty); |
| 2545 | fiz3 = _mm_add_ps(fiz3,tz); |
| 2546 | |
| 2547 | fjx3 = _mm_add_ps(fjx3,tx); |
| 2548 | fjy3 = _mm_add_ps(fjy3,ty); |
| 2549 | fjz3 = _mm_add_ps(fjz3,tz); |
| 2550 | |
| 2551 | } |
| 2552 | |
| 2553 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
| 2554 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
| 2555 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
| 2556 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
| 2557 | |
| 2558 | gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM3,fjptrB+DIM3,fjptrC+DIM3,fjptrD+DIM3, |
| 2559 | fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3); |
| 2560 | |
| 2561 | /* Inner loop uses 360 flops */ |
| 2562 | } |
| 2563 | |
| 2564 | /* End of innermost loop */ |
| 2565 | |
| 2566 | gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3, |
| 2567 | f+i_coord_offset+DIM3,fshift+i_shift_offset); |
| 2568 | |
| 2569 | /* Increment number of inner iterations */ |
| 2570 | inneriter += j_index_end - j_index_start; |
| 2571 | |
| 2572 | /* Outer loop uses 18 flops */ |
| 2573 | } |
| 2574 | |
| 2575 | /* Increment number of outer iterations */ |
| 2576 | outeriter += nri; |
| 2577 | |
| 2578 | /* Update outer/inner flops */ |
| 2579 | |
| 2580 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*360)(nrnb)->n[eNR_NBKERNEL_ELEC_W4W4_F] += outeriter*18 + inneriter *360; |
| 2581 | } |