File: | gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_sse4_1_single.c |
Location: | line 126, column 22 |
Description: | Value stored to 'two' during its initialization is never read |
1 | /* |
2 | * This file is part of the GROMACS molecular simulation package. |
3 | * |
4 | * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by |
5 | * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl, |
6 | * and including many others, as listed in the AUTHORS file in the |
7 | * top-level source directory and at http://www.gromacs.org. |
8 | * |
9 | * GROMACS is free software; you can redistribute it and/or |
10 | * modify it under the terms of the GNU Lesser General Public License |
11 | * as published by the Free Software Foundation; either version 2.1 |
12 | * of the License, or (at your option) any later version. |
13 | * |
14 | * GROMACS is distributed in the hope that it will be useful, |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
17 | * Lesser General Public License for more details. |
18 | * |
19 | * You should have received a copy of the GNU Lesser General Public |
20 | * License along with GROMACS; if not, see |
21 | * http://www.gnu.org/licenses, or write to the Free Software Foundation, |
22 | * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. |
23 | * |
24 | * If you want to redistribute modifications to GROMACS, please |
25 | * consider that scientific software is very special. Version |
26 | * control is crucial - bugs must be traceable. We will be happy to |
27 | * consider code for inclusion in the official distribution, but |
28 | * derived work must not be called official GROMACS. Details are found |
29 | * in the README & COPYING files - if they are missing, get the |
30 | * official version at http://www.gromacs.org. |
31 | * |
32 | * To help us fund GROMACS development, we humbly ask that you cite |
33 | * the research papers on the package. Check out http://www.gromacs.org. |
34 | */ |
35 | /* |
36 | * Note: this file was generated by the GROMACS sse4_1_single kernel generator. |
37 | */ |
38 | #ifdef HAVE_CONFIG_H1 |
39 | #include <config.h> |
40 | #endif |
41 | |
42 | #include <math.h> |
43 | |
44 | #include "../nb_kernel.h" |
45 | #include "types/simple.h" |
46 | #include "gromacs/math/vec.h" |
47 | #include "nrnb.h" |
48 | |
49 | #include "gromacs/simd/math_x86_sse4_1_single.h" |
50 | #include "kernelutil_x86_sse4_1_single.h" |
51 | |
52 | /* |
53 | * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_VF_sse4_1_single |
54 | * Electrostatics interaction: ReactionField |
55 | * VdW interaction: CubicSplineTable |
56 | * Geometry: Water4-Water4 |
57 | * Calculate force/pot: PotentialAndForce |
58 | */ |
59 | void |
60 | nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_VF_sse4_1_single |
61 | (t_nblist * gmx_restrict nlist, |
62 | rvec * gmx_restrict xx, |
63 | rvec * gmx_restrict ff, |
64 | t_forcerec * gmx_restrict fr, |
65 | t_mdatoms * gmx_restrict mdatoms, |
66 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
67 | t_nrnb * gmx_restrict nrnb) |
68 | { |
69 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
70 | * just 0 for non-waters. |
71 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
72 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
73 | */ |
74 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
75 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
76 | int jnrA,jnrB,jnrC,jnrD; |
77 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
78 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
79 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
80 | real rcutoff_scalar; |
81 | real *shiftvec,*fshift,*x,*f; |
82 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
83 | real scratch[4*DIM3]; |
84 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
85 | int vdwioffset0; |
86 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
87 | int vdwioffset1; |
88 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
89 | int vdwioffset2; |
90 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
91 | int vdwioffset3; |
92 | __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3; |
93 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
94 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
95 | int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D; |
96 | __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1; |
97 | int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D; |
98 | __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2; |
99 | int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D; |
100 | __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3; |
101 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
102 | __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11; |
103 | __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12; |
104 | __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13; |
105 | __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21; |
106 | __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22; |
107 | __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23; |
108 | __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31; |
109 | __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32; |
110 | __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33; |
111 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
112 | real *charge; |
113 | int nvdwtype; |
114 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
115 | int *vdwtype; |
116 | real *vdwparam; |
117 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
118 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
119 | __m128i vfitab; |
120 | __m128i ifour = _mm_set1_epi32(4); |
121 | __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF; |
122 | real *vftab; |
123 | __m128 dummy_mask,cutoff_mask; |
124 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
125 | __m128 one = _mm_set1_ps(1.0); |
126 | __m128 two = _mm_set1_ps(2.0); |
Value stored to 'two' during its initialization is never read | |
127 | x = xx[0]; |
128 | f = ff[0]; |
129 | |
130 | nri = nlist->nri; |
131 | iinr = nlist->iinr; |
132 | jindex = nlist->jindex; |
133 | jjnr = nlist->jjnr; |
134 | shiftidx = nlist->shift; |
135 | gid = nlist->gid; |
136 | shiftvec = fr->shift_vec[0]; |
137 | fshift = fr->fshift[0]; |
138 | facel = _mm_set1_ps(fr->epsfac); |
139 | charge = mdatoms->chargeA; |
140 | krf = _mm_set1_ps(fr->ic->k_rf); |
141 | krf2 = _mm_set1_ps(fr->ic->k_rf*2.0); |
142 | crf = _mm_set1_ps(fr->ic->c_rf); |
143 | nvdwtype = fr->ntype; |
144 | vdwparam = fr->nbfp; |
145 | vdwtype = mdatoms->typeA; |
146 | |
147 | vftab = kernel_data->table_vdw->data; |
148 | vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale); |
149 | |
150 | /* Setup water-specific parameters */ |
151 | inr = nlist->iinr[0]; |
152 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
153 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
154 | iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3])); |
155 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
156 | |
157 | jq1 = _mm_set1_ps(charge[inr+1]); |
158 | jq2 = _mm_set1_ps(charge[inr+2]); |
159 | jq3 = _mm_set1_ps(charge[inr+3]); |
160 | vdwjidx0A = 2*vdwtype[inr+0]; |
161 | c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]); |
162 | c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]); |
163 | qq11 = _mm_mul_ps(iq1,jq1); |
164 | qq12 = _mm_mul_ps(iq1,jq2); |
165 | qq13 = _mm_mul_ps(iq1,jq3); |
166 | qq21 = _mm_mul_ps(iq2,jq1); |
167 | qq22 = _mm_mul_ps(iq2,jq2); |
168 | qq23 = _mm_mul_ps(iq2,jq3); |
169 | qq31 = _mm_mul_ps(iq3,jq1); |
170 | qq32 = _mm_mul_ps(iq3,jq2); |
171 | qq33 = _mm_mul_ps(iq3,jq3); |
172 | |
173 | /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */ |
174 | rcutoff_scalar = fr->rcoulomb; |
175 | rcutoff = _mm_set1_ps(rcutoff_scalar); |
176 | rcutoff2 = _mm_mul_ps(rcutoff,rcutoff); |
177 | |
178 | /* Avoid stupid compiler warnings */ |
179 | jnrA = jnrB = jnrC = jnrD = 0; |
180 | j_coord_offsetA = 0; |
181 | j_coord_offsetB = 0; |
182 | j_coord_offsetC = 0; |
183 | j_coord_offsetD = 0; |
184 | |
185 | outeriter = 0; |
186 | inneriter = 0; |
187 | |
188 | for(iidx=0;iidx<4*DIM3;iidx++) |
189 | { |
190 | scratch[iidx] = 0.0; |
191 | } |
192 | |
193 | /* Start outer loop over neighborlists */ |
194 | for(iidx=0; iidx<nri; iidx++) |
195 | { |
196 | /* Load shift vector for this list */ |
197 | i_shift_offset = DIM3*shiftidx[iidx]; |
198 | |
199 | /* Load limits for loop over neighbors */ |
200 | j_index_start = jindex[iidx]; |
201 | j_index_end = jindex[iidx+1]; |
202 | |
203 | /* Get outer coordinate index */ |
204 | inr = iinr[iidx]; |
205 | i_coord_offset = DIM3*inr; |
206 | |
207 | /* Load i particle coords and add shift vector */ |
208 | gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
209 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3); |
210 | |
211 | fix0 = _mm_setzero_ps(); |
212 | fiy0 = _mm_setzero_ps(); |
213 | fiz0 = _mm_setzero_ps(); |
214 | fix1 = _mm_setzero_ps(); |
215 | fiy1 = _mm_setzero_ps(); |
216 | fiz1 = _mm_setzero_ps(); |
217 | fix2 = _mm_setzero_ps(); |
218 | fiy2 = _mm_setzero_ps(); |
219 | fiz2 = _mm_setzero_ps(); |
220 | fix3 = _mm_setzero_ps(); |
221 | fiy3 = _mm_setzero_ps(); |
222 | fiz3 = _mm_setzero_ps(); |
223 | |
224 | /* Reset potential sums */ |
225 | velecsum = _mm_setzero_ps(); |
226 | vvdwsum = _mm_setzero_ps(); |
227 | |
228 | /* Start inner kernel loop */ |
229 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
230 | { |
231 | |
232 | /* Get j neighbor index, and coordinate index */ |
233 | jnrA = jjnr[jidx]; |
234 | jnrB = jjnr[jidx+1]; |
235 | jnrC = jjnr[jidx+2]; |
236 | jnrD = jjnr[jidx+3]; |
237 | j_coord_offsetA = DIM3*jnrA; |
238 | j_coord_offsetB = DIM3*jnrB; |
239 | j_coord_offsetC = DIM3*jnrC; |
240 | j_coord_offsetD = DIM3*jnrD; |
241 | |
242 | /* load j atom coordinates */ |
243 | gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
244 | x+j_coord_offsetC,x+j_coord_offsetD, |
245 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2, |
246 | &jy2,&jz2,&jx3,&jy3,&jz3); |
247 | |
248 | /* Calculate displacement vector */ |
249 | dx00 = _mm_sub_ps(ix0,jx0); |
250 | dy00 = _mm_sub_ps(iy0,jy0); |
251 | dz00 = _mm_sub_ps(iz0,jz0); |
252 | dx11 = _mm_sub_ps(ix1,jx1); |
253 | dy11 = _mm_sub_ps(iy1,jy1); |
254 | dz11 = _mm_sub_ps(iz1,jz1); |
255 | dx12 = _mm_sub_ps(ix1,jx2); |
256 | dy12 = _mm_sub_ps(iy1,jy2); |
257 | dz12 = _mm_sub_ps(iz1,jz2); |
258 | dx13 = _mm_sub_ps(ix1,jx3); |
259 | dy13 = _mm_sub_ps(iy1,jy3); |
260 | dz13 = _mm_sub_ps(iz1,jz3); |
261 | dx21 = _mm_sub_ps(ix2,jx1); |
262 | dy21 = _mm_sub_ps(iy2,jy1); |
263 | dz21 = _mm_sub_ps(iz2,jz1); |
264 | dx22 = _mm_sub_ps(ix2,jx2); |
265 | dy22 = _mm_sub_ps(iy2,jy2); |
266 | dz22 = _mm_sub_ps(iz2,jz2); |
267 | dx23 = _mm_sub_ps(ix2,jx3); |
268 | dy23 = _mm_sub_ps(iy2,jy3); |
269 | dz23 = _mm_sub_ps(iz2,jz3); |
270 | dx31 = _mm_sub_ps(ix3,jx1); |
271 | dy31 = _mm_sub_ps(iy3,jy1); |
272 | dz31 = _mm_sub_ps(iz3,jz1); |
273 | dx32 = _mm_sub_ps(ix3,jx2); |
274 | dy32 = _mm_sub_ps(iy3,jy2); |
275 | dz32 = _mm_sub_ps(iz3,jz2); |
276 | dx33 = _mm_sub_ps(ix3,jx3); |
277 | dy33 = _mm_sub_ps(iy3,jy3); |
278 | dz33 = _mm_sub_ps(iz3,jz3); |
279 | |
280 | /* Calculate squared distance and things based on it */ |
281 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
282 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
283 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
284 | rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13); |
285 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
286 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
287 | rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23); |
288 | rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31); |
289 | rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32); |
290 | rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33); |
291 | |
292 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
293 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
294 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
295 | rinv13 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq13); |
296 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
297 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
298 | rinv23 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq23); |
299 | rinv31 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq31); |
300 | rinv32 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq32); |
301 | rinv33 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq33); |
302 | |
303 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
304 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
305 | rinvsq13 = _mm_mul_ps(rinv13,rinv13); |
306 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
307 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
308 | rinvsq23 = _mm_mul_ps(rinv23,rinv23); |
309 | rinvsq31 = _mm_mul_ps(rinv31,rinv31); |
310 | rinvsq32 = _mm_mul_ps(rinv32,rinv32); |
311 | rinvsq33 = _mm_mul_ps(rinv33,rinv33); |
312 | |
313 | fjx0 = _mm_setzero_ps(); |
314 | fjy0 = _mm_setzero_ps(); |
315 | fjz0 = _mm_setzero_ps(); |
316 | fjx1 = _mm_setzero_ps(); |
317 | fjy1 = _mm_setzero_ps(); |
318 | fjz1 = _mm_setzero_ps(); |
319 | fjx2 = _mm_setzero_ps(); |
320 | fjy2 = _mm_setzero_ps(); |
321 | fjz2 = _mm_setzero_ps(); |
322 | fjx3 = _mm_setzero_ps(); |
323 | fjy3 = _mm_setzero_ps(); |
324 | fjz3 = _mm_setzero_ps(); |
325 | |
326 | /************************** |
327 | * CALCULATE INTERACTIONS * |
328 | **************************/ |
329 | |
330 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
331 | { |
332 | |
333 | r00 = _mm_mul_ps(rsq00,rinv00); |
334 | |
335 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
336 | rt = _mm_mul_ps(r00,vftabscale); |
337 | vfitab = _mm_cvttps_epi32(rt); |
338 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
339 | vfitab = _mm_slli_epi32(vfitab,3); |
340 | |
341 | /* CUBIC SPLINE TABLE DISPERSION */ |
342 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
343 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
344 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
345 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
346 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
347 | Heps = _mm_mul_ps(vfeps,H); |
348 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
349 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
350 | vvdw6 = _mm_mul_ps(c6_00,VV); |
351 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
352 | fvdw6 = _mm_mul_ps(c6_00,FF); |
353 | |
354 | /* CUBIC SPLINE TABLE REPULSION */ |
355 | vfitab = _mm_add_epi32(vfitab,ifour); |
356 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
357 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
358 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
359 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
360 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
361 | Heps = _mm_mul_ps(vfeps,H); |
362 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
363 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
364 | vvdw12 = _mm_mul_ps(c12_00,VV); |
365 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
366 | fvdw12 = _mm_mul_ps(c12_00,FF); |
367 | vvdw = _mm_add_ps(vvdw12,vvdw6); |
368 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
369 | |
370 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
371 | |
372 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
373 | vvdw = _mm_and_ps(vvdw,cutoff_mask); |
374 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
375 | |
376 | fscal = fvdw; |
377 | |
378 | fscal = _mm_and_ps(fscal,cutoff_mask); |
379 | |
380 | /* Calculate temporary vectorial force */ |
381 | tx = _mm_mul_ps(fscal,dx00); |
382 | ty = _mm_mul_ps(fscal,dy00); |
383 | tz = _mm_mul_ps(fscal,dz00); |
384 | |
385 | /* Update vectorial force */ |
386 | fix0 = _mm_add_ps(fix0,tx); |
387 | fiy0 = _mm_add_ps(fiy0,ty); |
388 | fiz0 = _mm_add_ps(fiz0,tz); |
389 | |
390 | fjx0 = _mm_add_ps(fjx0,tx); |
391 | fjy0 = _mm_add_ps(fjy0,ty); |
392 | fjz0 = _mm_add_ps(fjz0,tz); |
393 | |
394 | } |
395 | |
396 | /************************** |
397 | * CALCULATE INTERACTIONS * |
398 | **************************/ |
399 | |
400 | if (gmx_mm_any_lt(rsq11,rcutoff2)) |
401 | { |
402 | |
403 | /* REACTION-FIELD ELECTROSTATICS */ |
404 | velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_add_ps(rinv11,_mm_mul_ps(krf,rsq11)),crf)); |
405 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
406 | |
407 | cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2); |
408 | |
409 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
410 | velec = _mm_and_ps(velec,cutoff_mask); |
411 | velecsum = _mm_add_ps(velecsum,velec); |
412 | |
413 | fscal = felec; |
414 | |
415 | fscal = _mm_and_ps(fscal,cutoff_mask); |
416 | |
417 | /* Calculate temporary vectorial force */ |
418 | tx = _mm_mul_ps(fscal,dx11); |
419 | ty = _mm_mul_ps(fscal,dy11); |
420 | tz = _mm_mul_ps(fscal,dz11); |
421 | |
422 | /* Update vectorial force */ |
423 | fix1 = _mm_add_ps(fix1,tx); |
424 | fiy1 = _mm_add_ps(fiy1,ty); |
425 | fiz1 = _mm_add_ps(fiz1,tz); |
426 | |
427 | fjx1 = _mm_add_ps(fjx1,tx); |
428 | fjy1 = _mm_add_ps(fjy1,ty); |
429 | fjz1 = _mm_add_ps(fjz1,tz); |
430 | |
431 | } |
432 | |
433 | /************************** |
434 | * CALCULATE INTERACTIONS * |
435 | **************************/ |
436 | |
437 | if (gmx_mm_any_lt(rsq12,rcutoff2)) |
438 | { |
439 | |
440 | /* REACTION-FIELD ELECTROSTATICS */ |
441 | velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_add_ps(rinv12,_mm_mul_ps(krf,rsq12)),crf)); |
442 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
443 | |
444 | cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2); |
445 | |
446 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
447 | velec = _mm_and_ps(velec,cutoff_mask); |
448 | velecsum = _mm_add_ps(velecsum,velec); |
449 | |
450 | fscal = felec; |
451 | |
452 | fscal = _mm_and_ps(fscal,cutoff_mask); |
453 | |
454 | /* Calculate temporary vectorial force */ |
455 | tx = _mm_mul_ps(fscal,dx12); |
456 | ty = _mm_mul_ps(fscal,dy12); |
457 | tz = _mm_mul_ps(fscal,dz12); |
458 | |
459 | /* Update vectorial force */ |
460 | fix1 = _mm_add_ps(fix1,tx); |
461 | fiy1 = _mm_add_ps(fiy1,ty); |
462 | fiz1 = _mm_add_ps(fiz1,tz); |
463 | |
464 | fjx2 = _mm_add_ps(fjx2,tx); |
465 | fjy2 = _mm_add_ps(fjy2,ty); |
466 | fjz2 = _mm_add_ps(fjz2,tz); |
467 | |
468 | } |
469 | |
470 | /************************** |
471 | * CALCULATE INTERACTIONS * |
472 | **************************/ |
473 | |
474 | if (gmx_mm_any_lt(rsq13,rcutoff2)) |
475 | { |
476 | |
477 | /* REACTION-FIELD ELECTROSTATICS */ |
478 | velec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_add_ps(rinv13,_mm_mul_ps(krf,rsq13)),crf)); |
479 | felec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_mul_ps(rinv13,rinvsq13),krf2)); |
480 | |
481 | cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2); |
482 | |
483 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
484 | velec = _mm_and_ps(velec,cutoff_mask); |
485 | velecsum = _mm_add_ps(velecsum,velec); |
486 | |
487 | fscal = felec; |
488 | |
489 | fscal = _mm_and_ps(fscal,cutoff_mask); |
490 | |
491 | /* Calculate temporary vectorial force */ |
492 | tx = _mm_mul_ps(fscal,dx13); |
493 | ty = _mm_mul_ps(fscal,dy13); |
494 | tz = _mm_mul_ps(fscal,dz13); |
495 | |
496 | /* Update vectorial force */ |
497 | fix1 = _mm_add_ps(fix1,tx); |
498 | fiy1 = _mm_add_ps(fiy1,ty); |
499 | fiz1 = _mm_add_ps(fiz1,tz); |
500 | |
501 | fjx3 = _mm_add_ps(fjx3,tx); |
502 | fjy3 = _mm_add_ps(fjy3,ty); |
503 | fjz3 = _mm_add_ps(fjz3,tz); |
504 | |
505 | } |
506 | |
507 | /************************** |
508 | * CALCULATE INTERACTIONS * |
509 | **************************/ |
510 | |
511 | if (gmx_mm_any_lt(rsq21,rcutoff2)) |
512 | { |
513 | |
514 | /* REACTION-FIELD ELECTROSTATICS */ |
515 | velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_add_ps(rinv21,_mm_mul_ps(krf,rsq21)),crf)); |
516 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
517 | |
518 | cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2); |
519 | |
520 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
521 | velec = _mm_and_ps(velec,cutoff_mask); |
522 | velecsum = _mm_add_ps(velecsum,velec); |
523 | |
524 | fscal = felec; |
525 | |
526 | fscal = _mm_and_ps(fscal,cutoff_mask); |
527 | |
528 | /* Calculate temporary vectorial force */ |
529 | tx = _mm_mul_ps(fscal,dx21); |
530 | ty = _mm_mul_ps(fscal,dy21); |
531 | tz = _mm_mul_ps(fscal,dz21); |
532 | |
533 | /* Update vectorial force */ |
534 | fix2 = _mm_add_ps(fix2,tx); |
535 | fiy2 = _mm_add_ps(fiy2,ty); |
536 | fiz2 = _mm_add_ps(fiz2,tz); |
537 | |
538 | fjx1 = _mm_add_ps(fjx1,tx); |
539 | fjy1 = _mm_add_ps(fjy1,ty); |
540 | fjz1 = _mm_add_ps(fjz1,tz); |
541 | |
542 | } |
543 | |
544 | /************************** |
545 | * CALCULATE INTERACTIONS * |
546 | **************************/ |
547 | |
548 | if (gmx_mm_any_lt(rsq22,rcutoff2)) |
549 | { |
550 | |
551 | /* REACTION-FIELD ELECTROSTATICS */ |
552 | velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_add_ps(rinv22,_mm_mul_ps(krf,rsq22)),crf)); |
553 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
554 | |
555 | cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2); |
556 | |
557 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
558 | velec = _mm_and_ps(velec,cutoff_mask); |
559 | velecsum = _mm_add_ps(velecsum,velec); |
560 | |
561 | fscal = felec; |
562 | |
563 | fscal = _mm_and_ps(fscal,cutoff_mask); |
564 | |
565 | /* Calculate temporary vectorial force */ |
566 | tx = _mm_mul_ps(fscal,dx22); |
567 | ty = _mm_mul_ps(fscal,dy22); |
568 | tz = _mm_mul_ps(fscal,dz22); |
569 | |
570 | /* Update vectorial force */ |
571 | fix2 = _mm_add_ps(fix2,tx); |
572 | fiy2 = _mm_add_ps(fiy2,ty); |
573 | fiz2 = _mm_add_ps(fiz2,tz); |
574 | |
575 | fjx2 = _mm_add_ps(fjx2,tx); |
576 | fjy2 = _mm_add_ps(fjy2,ty); |
577 | fjz2 = _mm_add_ps(fjz2,tz); |
578 | |
579 | } |
580 | |
581 | /************************** |
582 | * CALCULATE INTERACTIONS * |
583 | **************************/ |
584 | |
585 | if (gmx_mm_any_lt(rsq23,rcutoff2)) |
586 | { |
587 | |
588 | /* REACTION-FIELD ELECTROSTATICS */ |
589 | velec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_add_ps(rinv23,_mm_mul_ps(krf,rsq23)),crf)); |
590 | felec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_mul_ps(rinv23,rinvsq23),krf2)); |
591 | |
592 | cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2); |
593 | |
594 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
595 | velec = _mm_and_ps(velec,cutoff_mask); |
596 | velecsum = _mm_add_ps(velecsum,velec); |
597 | |
598 | fscal = felec; |
599 | |
600 | fscal = _mm_and_ps(fscal,cutoff_mask); |
601 | |
602 | /* Calculate temporary vectorial force */ |
603 | tx = _mm_mul_ps(fscal,dx23); |
604 | ty = _mm_mul_ps(fscal,dy23); |
605 | tz = _mm_mul_ps(fscal,dz23); |
606 | |
607 | /* Update vectorial force */ |
608 | fix2 = _mm_add_ps(fix2,tx); |
609 | fiy2 = _mm_add_ps(fiy2,ty); |
610 | fiz2 = _mm_add_ps(fiz2,tz); |
611 | |
612 | fjx3 = _mm_add_ps(fjx3,tx); |
613 | fjy3 = _mm_add_ps(fjy3,ty); |
614 | fjz3 = _mm_add_ps(fjz3,tz); |
615 | |
616 | } |
617 | |
618 | /************************** |
619 | * CALCULATE INTERACTIONS * |
620 | **************************/ |
621 | |
622 | if (gmx_mm_any_lt(rsq31,rcutoff2)) |
623 | { |
624 | |
625 | /* REACTION-FIELD ELECTROSTATICS */ |
626 | velec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_add_ps(rinv31,_mm_mul_ps(krf,rsq31)),crf)); |
627 | felec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_mul_ps(rinv31,rinvsq31),krf2)); |
628 | |
629 | cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2); |
630 | |
631 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
632 | velec = _mm_and_ps(velec,cutoff_mask); |
633 | velecsum = _mm_add_ps(velecsum,velec); |
634 | |
635 | fscal = felec; |
636 | |
637 | fscal = _mm_and_ps(fscal,cutoff_mask); |
638 | |
639 | /* Calculate temporary vectorial force */ |
640 | tx = _mm_mul_ps(fscal,dx31); |
641 | ty = _mm_mul_ps(fscal,dy31); |
642 | tz = _mm_mul_ps(fscal,dz31); |
643 | |
644 | /* Update vectorial force */ |
645 | fix3 = _mm_add_ps(fix3,tx); |
646 | fiy3 = _mm_add_ps(fiy3,ty); |
647 | fiz3 = _mm_add_ps(fiz3,tz); |
648 | |
649 | fjx1 = _mm_add_ps(fjx1,tx); |
650 | fjy1 = _mm_add_ps(fjy1,ty); |
651 | fjz1 = _mm_add_ps(fjz1,tz); |
652 | |
653 | } |
654 | |
655 | /************************** |
656 | * CALCULATE INTERACTIONS * |
657 | **************************/ |
658 | |
659 | if (gmx_mm_any_lt(rsq32,rcutoff2)) |
660 | { |
661 | |
662 | /* REACTION-FIELD ELECTROSTATICS */ |
663 | velec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_add_ps(rinv32,_mm_mul_ps(krf,rsq32)),crf)); |
664 | felec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_mul_ps(rinv32,rinvsq32),krf2)); |
665 | |
666 | cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2); |
667 | |
668 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
669 | velec = _mm_and_ps(velec,cutoff_mask); |
670 | velecsum = _mm_add_ps(velecsum,velec); |
671 | |
672 | fscal = felec; |
673 | |
674 | fscal = _mm_and_ps(fscal,cutoff_mask); |
675 | |
676 | /* Calculate temporary vectorial force */ |
677 | tx = _mm_mul_ps(fscal,dx32); |
678 | ty = _mm_mul_ps(fscal,dy32); |
679 | tz = _mm_mul_ps(fscal,dz32); |
680 | |
681 | /* Update vectorial force */ |
682 | fix3 = _mm_add_ps(fix3,tx); |
683 | fiy3 = _mm_add_ps(fiy3,ty); |
684 | fiz3 = _mm_add_ps(fiz3,tz); |
685 | |
686 | fjx2 = _mm_add_ps(fjx2,tx); |
687 | fjy2 = _mm_add_ps(fjy2,ty); |
688 | fjz2 = _mm_add_ps(fjz2,tz); |
689 | |
690 | } |
691 | |
692 | /************************** |
693 | * CALCULATE INTERACTIONS * |
694 | **************************/ |
695 | |
696 | if (gmx_mm_any_lt(rsq33,rcutoff2)) |
697 | { |
698 | |
699 | /* REACTION-FIELD ELECTROSTATICS */ |
700 | velec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_add_ps(rinv33,_mm_mul_ps(krf,rsq33)),crf)); |
701 | felec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_mul_ps(rinv33,rinvsq33),krf2)); |
702 | |
703 | cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2); |
704 | |
705 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
706 | velec = _mm_and_ps(velec,cutoff_mask); |
707 | velecsum = _mm_add_ps(velecsum,velec); |
708 | |
709 | fscal = felec; |
710 | |
711 | fscal = _mm_and_ps(fscal,cutoff_mask); |
712 | |
713 | /* Calculate temporary vectorial force */ |
714 | tx = _mm_mul_ps(fscal,dx33); |
715 | ty = _mm_mul_ps(fscal,dy33); |
716 | tz = _mm_mul_ps(fscal,dz33); |
717 | |
718 | /* Update vectorial force */ |
719 | fix3 = _mm_add_ps(fix3,tx); |
720 | fiy3 = _mm_add_ps(fiy3,ty); |
721 | fiz3 = _mm_add_ps(fiz3,tz); |
722 | |
723 | fjx3 = _mm_add_ps(fjx3,tx); |
724 | fjy3 = _mm_add_ps(fjy3,ty); |
725 | fjz3 = _mm_add_ps(fjz3,tz); |
726 | |
727 | } |
728 | |
729 | fjptrA = f+j_coord_offsetA; |
730 | fjptrB = f+j_coord_offsetB; |
731 | fjptrC = f+j_coord_offsetC; |
732 | fjptrD = f+j_coord_offsetD; |
733 | |
734 | gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
735 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1, |
736 | fjx2,fjy2,fjz2,fjx3,fjy3,fjz3); |
737 | |
738 | /* Inner loop uses 387 flops */ |
739 | } |
740 | |
741 | if(jidx<j_index_end) |
742 | { |
743 | |
744 | /* Get j neighbor index, and coordinate index */ |
745 | jnrlistA = jjnr[jidx]; |
746 | jnrlistB = jjnr[jidx+1]; |
747 | jnrlistC = jjnr[jidx+2]; |
748 | jnrlistD = jjnr[jidx+3]; |
749 | /* Sign of each element will be negative for non-real atoms. |
750 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
751 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
752 | */ |
753 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
754 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
755 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
756 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
757 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
758 | j_coord_offsetA = DIM3*jnrA; |
759 | j_coord_offsetB = DIM3*jnrB; |
760 | j_coord_offsetC = DIM3*jnrC; |
761 | j_coord_offsetD = DIM3*jnrD; |
762 | |
763 | /* load j atom coordinates */ |
764 | gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
765 | x+j_coord_offsetC,x+j_coord_offsetD, |
766 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2, |
767 | &jy2,&jz2,&jx3,&jy3,&jz3); |
768 | |
769 | /* Calculate displacement vector */ |
770 | dx00 = _mm_sub_ps(ix0,jx0); |
771 | dy00 = _mm_sub_ps(iy0,jy0); |
772 | dz00 = _mm_sub_ps(iz0,jz0); |
773 | dx11 = _mm_sub_ps(ix1,jx1); |
774 | dy11 = _mm_sub_ps(iy1,jy1); |
775 | dz11 = _mm_sub_ps(iz1,jz1); |
776 | dx12 = _mm_sub_ps(ix1,jx2); |
777 | dy12 = _mm_sub_ps(iy1,jy2); |
778 | dz12 = _mm_sub_ps(iz1,jz2); |
779 | dx13 = _mm_sub_ps(ix1,jx3); |
780 | dy13 = _mm_sub_ps(iy1,jy3); |
781 | dz13 = _mm_sub_ps(iz1,jz3); |
782 | dx21 = _mm_sub_ps(ix2,jx1); |
783 | dy21 = _mm_sub_ps(iy2,jy1); |
784 | dz21 = _mm_sub_ps(iz2,jz1); |
785 | dx22 = _mm_sub_ps(ix2,jx2); |
786 | dy22 = _mm_sub_ps(iy2,jy2); |
787 | dz22 = _mm_sub_ps(iz2,jz2); |
788 | dx23 = _mm_sub_ps(ix2,jx3); |
789 | dy23 = _mm_sub_ps(iy2,jy3); |
790 | dz23 = _mm_sub_ps(iz2,jz3); |
791 | dx31 = _mm_sub_ps(ix3,jx1); |
792 | dy31 = _mm_sub_ps(iy3,jy1); |
793 | dz31 = _mm_sub_ps(iz3,jz1); |
794 | dx32 = _mm_sub_ps(ix3,jx2); |
795 | dy32 = _mm_sub_ps(iy3,jy2); |
796 | dz32 = _mm_sub_ps(iz3,jz2); |
797 | dx33 = _mm_sub_ps(ix3,jx3); |
798 | dy33 = _mm_sub_ps(iy3,jy3); |
799 | dz33 = _mm_sub_ps(iz3,jz3); |
800 | |
801 | /* Calculate squared distance and things based on it */ |
802 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
803 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
804 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
805 | rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13); |
806 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
807 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
808 | rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23); |
809 | rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31); |
810 | rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32); |
811 | rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33); |
812 | |
813 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
814 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
815 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
816 | rinv13 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq13); |
817 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
818 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
819 | rinv23 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq23); |
820 | rinv31 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq31); |
821 | rinv32 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq32); |
822 | rinv33 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq33); |
823 | |
824 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
825 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
826 | rinvsq13 = _mm_mul_ps(rinv13,rinv13); |
827 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
828 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
829 | rinvsq23 = _mm_mul_ps(rinv23,rinv23); |
830 | rinvsq31 = _mm_mul_ps(rinv31,rinv31); |
831 | rinvsq32 = _mm_mul_ps(rinv32,rinv32); |
832 | rinvsq33 = _mm_mul_ps(rinv33,rinv33); |
833 | |
834 | fjx0 = _mm_setzero_ps(); |
835 | fjy0 = _mm_setzero_ps(); |
836 | fjz0 = _mm_setzero_ps(); |
837 | fjx1 = _mm_setzero_ps(); |
838 | fjy1 = _mm_setzero_ps(); |
839 | fjz1 = _mm_setzero_ps(); |
840 | fjx2 = _mm_setzero_ps(); |
841 | fjy2 = _mm_setzero_ps(); |
842 | fjz2 = _mm_setzero_ps(); |
843 | fjx3 = _mm_setzero_ps(); |
844 | fjy3 = _mm_setzero_ps(); |
845 | fjz3 = _mm_setzero_ps(); |
846 | |
847 | /************************** |
848 | * CALCULATE INTERACTIONS * |
849 | **************************/ |
850 | |
851 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
852 | { |
853 | |
854 | r00 = _mm_mul_ps(rsq00,rinv00); |
855 | r00 = _mm_andnot_ps(dummy_mask,r00); |
856 | |
857 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
858 | rt = _mm_mul_ps(r00,vftabscale); |
859 | vfitab = _mm_cvttps_epi32(rt); |
860 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
861 | vfitab = _mm_slli_epi32(vfitab,3); |
862 | |
863 | /* CUBIC SPLINE TABLE DISPERSION */ |
864 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
865 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
866 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
867 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
868 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
869 | Heps = _mm_mul_ps(vfeps,H); |
870 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
871 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
872 | vvdw6 = _mm_mul_ps(c6_00,VV); |
873 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
874 | fvdw6 = _mm_mul_ps(c6_00,FF); |
875 | |
876 | /* CUBIC SPLINE TABLE REPULSION */ |
877 | vfitab = _mm_add_epi32(vfitab,ifour); |
878 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
879 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
880 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
881 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
882 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
883 | Heps = _mm_mul_ps(vfeps,H); |
884 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
885 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
886 | vvdw12 = _mm_mul_ps(c12_00,VV); |
887 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
888 | fvdw12 = _mm_mul_ps(c12_00,FF); |
889 | vvdw = _mm_add_ps(vvdw12,vvdw6); |
890 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
891 | |
892 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
893 | |
894 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
895 | vvdw = _mm_and_ps(vvdw,cutoff_mask); |
896 | vvdw = _mm_andnot_ps(dummy_mask,vvdw); |
897 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
898 | |
899 | fscal = fvdw; |
900 | |
901 | fscal = _mm_and_ps(fscal,cutoff_mask); |
902 | |
903 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
904 | |
905 | /* Calculate temporary vectorial force */ |
906 | tx = _mm_mul_ps(fscal,dx00); |
907 | ty = _mm_mul_ps(fscal,dy00); |
908 | tz = _mm_mul_ps(fscal,dz00); |
909 | |
910 | /* Update vectorial force */ |
911 | fix0 = _mm_add_ps(fix0,tx); |
912 | fiy0 = _mm_add_ps(fiy0,ty); |
913 | fiz0 = _mm_add_ps(fiz0,tz); |
914 | |
915 | fjx0 = _mm_add_ps(fjx0,tx); |
916 | fjy0 = _mm_add_ps(fjy0,ty); |
917 | fjz0 = _mm_add_ps(fjz0,tz); |
918 | |
919 | } |
920 | |
921 | /************************** |
922 | * CALCULATE INTERACTIONS * |
923 | **************************/ |
924 | |
925 | if (gmx_mm_any_lt(rsq11,rcutoff2)) |
926 | { |
927 | |
928 | /* REACTION-FIELD ELECTROSTATICS */ |
929 | velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_add_ps(rinv11,_mm_mul_ps(krf,rsq11)),crf)); |
930 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
931 | |
932 | cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2); |
933 | |
934 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
935 | velec = _mm_and_ps(velec,cutoff_mask); |
936 | velec = _mm_andnot_ps(dummy_mask,velec); |
937 | velecsum = _mm_add_ps(velecsum,velec); |
938 | |
939 | fscal = felec; |
940 | |
941 | fscal = _mm_and_ps(fscal,cutoff_mask); |
942 | |
943 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
944 | |
945 | /* Calculate temporary vectorial force */ |
946 | tx = _mm_mul_ps(fscal,dx11); |
947 | ty = _mm_mul_ps(fscal,dy11); |
948 | tz = _mm_mul_ps(fscal,dz11); |
949 | |
950 | /* Update vectorial force */ |
951 | fix1 = _mm_add_ps(fix1,tx); |
952 | fiy1 = _mm_add_ps(fiy1,ty); |
953 | fiz1 = _mm_add_ps(fiz1,tz); |
954 | |
955 | fjx1 = _mm_add_ps(fjx1,tx); |
956 | fjy1 = _mm_add_ps(fjy1,ty); |
957 | fjz1 = _mm_add_ps(fjz1,tz); |
958 | |
959 | } |
960 | |
961 | /************************** |
962 | * CALCULATE INTERACTIONS * |
963 | **************************/ |
964 | |
965 | if (gmx_mm_any_lt(rsq12,rcutoff2)) |
966 | { |
967 | |
968 | /* REACTION-FIELD ELECTROSTATICS */ |
969 | velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_add_ps(rinv12,_mm_mul_ps(krf,rsq12)),crf)); |
970 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
971 | |
972 | cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2); |
973 | |
974 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
975 | velec = _mm_and_ps(velec,cutoff_mask); |
976 | velec = _mm_andnot_ps(dummy_mask,velec); |
977 | velecsum = _mm_add_ps(velecsum,velec); |
978 | |
979 | fscal = felec; |
980 | |
981 | fscal = _mm_and_ps(fscal,cutoff_mask); |
982 | |
983 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
984 | |
985 | /* Calculate temporary vectorial force */ |
986 | tx = _mm_mul_ps(fscal,dx12); |
987 | ty = _mm_mul_ps(fscal,dy12); |
988 | tz = _mm_mul_ps(fscal,dz12); |
989 | |
990 | /* Update vectorial force */ |
991 | fix1 = _mm_add_ps(fix1,tx); |
992 | fiy1 = _mm_add_ps(fiy1,ty); |
993 | fiz1 = _mm_add_ps(fiz1,tz); |
994 | |
995 | fjx2 = _mm_add_ps(fjx2,tx); |
996 | fjy2 = _mm_add_ps(fjy2,ty); |
997 | fjz2 = _mm_add_ps(fjz2,tz); |
998 | |
999 | } |
1000 | |
1001 | /************************** |
1002 | * CALCULATE INTERACTIONS * |
1003 | **************************/ |
1004 | |
1005 | if (gmx_mm_any_lt(rsq13,rcutoff2)) |
1006 | { |
1007 | |
1008 | /* REACTION-FIELD ELECTROSTATICS */ |
1009 | velec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_add_ps(rinv13,_mm_mul_ps(krf,rsq13)),crf)); |
1010 | felec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_mul_ps(rinv13,rinvsq13),krf2)); |
1011 | |
1012 | cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2); |
1013 | |
1014 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
1015 | velec = _mm_and_ps(velec,cutoff_mask); |
1016 | velec = _mm_andnot_ps(dummy_mask,velec); |
1017 | velecsum = _mm_add_ps(velecsum,velec); |
1018 | |
1019 | fscal = felec; |
1020 | |
1021 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1022 | |
1023 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1024 | |
1025 | /* Calculate temporary vectorial force */ |
1026 | tx = _mm_mul_ps(fscal,dx13); |
1027 | ty = _mm_mul_ps(fscal,dy13); |
1028 | tz = _mm_mul_ps(fscal,dz13); |
1029 | |
1030 | /* Update vectorial force */ |
1031 | fix1 = _mm_add_ps(fix1,tx); |
1032 | fiy1 = _mm_add_ps(fiy1,ty); |
1033 | fiz1 = _mm_add_ps(fiz1,tz); |
1034 | |
1035 | fjx3 = _mm_add_ps(fjx3,tx); |
1036 | fjy3 = _mm_add_ps(fjy3,ty); |
1037 | fjz3 = _mm_add_ps(fjz3,tz); |
1038 | |
1039 | } |
1040 | |
1041 | /************************** |
1042 | * CALCULATE INTERACTIONS * |
1043 | **************************/ |
1044 | |
1045 | if (gmx_mm_any_lt(rsq21,rcutoff2)) |
1046 | { |
1047 | |
1048 | /* REACTION-FIELD ELECTROSTATICS */ |
1049 | velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_add_ps(rinv21,_mm_mul_ps(krf,rsq21)),crf)); |
1050 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
1051 | |
1052 | cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2); |
1053 | |
1054 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
1055 | velec = _mm_and_ps(velec,cutoff_mask); |
1056 | velec = _mm_andnot_ps(dummy_mask,velec); |
1057 | velecsum = _mm_add_ps(velecsum,velec); |
1058 | |
1059 | fscal = felec; |
1060 | |
1061 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1062 | |
1063 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1064 | |
1065 | /* Calculate temporary vectorial force */ |
1066 | tx = _mm_mul_ps(fscal,dx21); |
1067 | ty = _mm_mul_ps(fscal,dy21); |
1068 | tz = _mm_mul_ps(fscal,dz21); |
1069 | |
1070 | /* Update vectorial force */ |
1071 | fix2 = _mm_add_ps(fix2,tx); |
1072 | fiy2 = _mm_add_ps(fiy2,ty); |
1073 | fiz2 = _mm_add_ps(fiz2,tz); |
1074 | |
1075 | fjx1 = _mm_add_ps(fjx1,tx); |
1076 | fjy1 = _mm_add_ps(fjy1,ty); |
1077 | fjz1 = _mm_add_ps(fjz1,tz); |
1078 | |
1079 | } |
1080 | |
1081 | /************************** |
1082 | * CALCULATE INTERACTIONS * |
1083 | **************************/ |
1084 | |
1085 | if (gmx_mm_any_lt(rsq22,rcutoff2)) |
1086 | { |
1087 | |
1088 | /* REACTION-FIELD ELECTROSTATICS */ |
1089 | velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_add_ps(rinv22,_mm_mul_ps(krf,rsq22)),crf)); |
1090 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
1091 | |
1092 | cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2); |
1093 | |
1094 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
1095 | velec = _mm_and_ps(velec,cutoff_mask); |
1096 | velec = _mm_andnot_ps(dummy_mask,velec); |
1097 | velecsum = _mm_add_ps(velecsum,velec); |
1098 | |
1099 | fscal = felec; |
1100 | |
1101 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1102 | |
1103 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1104 | |
1105 | /* Calculate temporary vectorial force */ |
1106 | tx = _mm_mul_ps(fscal,dx22); |
1107 | ty = _mm_mul_ps(fscal,dy22); |
1108 | tz = _mm_mul_ps(fscal,dz22); |
1109 | |
1110 | /* Update vectorial force */ |
1111 | fix2 = _mm_add_ps(fix2,tx); |
1112 | fiy2 = _mm_add_ps(fiy2,ty); |
1113 | fiz2 = _mm_add_ps(fiz2,tz); |
1114 | |
1115 | fjx2 = _mm_add_ps(fjx2,tx); |
1116 | fjy2 = _mm_add_ps(fjy2,ty); |
1117 | fjz2 = _mm_add_ps(fjz2,tz); |
1118 | |
1119 | } |
1120 | |
1121 | /************************** |
1122 | * CALCULATE INTERACTIONS * |
1123 | **************************/ |
1124 | |
1125 | if (gmx_mm_any_lt(rsq23,rcutoff2)) |
1126 | { |
1127 | |
1128 | /* REACTION-FIELD ELECTROSTATICS */ |
1129 | velec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_add_ps(rinv23,_mm_mul_ps(krf,rsq23)),crf)); |
1130 | felec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_mul_ps(rinv23,rinvsq23),krf2)); |
1131 | |
1132 | cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2); |
1133 | |
1134 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
1135 | velec = _mm_and_ps(velec,cutoff_mask); |
1136 | velec = _mm_andnot_ps(dummy_mask,velec); |
1137 | velecsum = _mm_add_ps(velecsum,velec); |
1138 | |
1139 | fscal = felec; |
1140 | |
1141 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1142 | |
1143 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1144 | |
1145 | /* Calculate temporary vectorial force */ |
1146 | tx = _mm_mul_ps(fscal,dx23); |
1147 | ty = _mm_mul_ps(fscal,dy23); |
1148 | tz = _mm_mul_ps(fscal,dz23); |
1149 | |
1150 | /* Update vectorial force */ |
1151 | fix2 = _mm_add_ps(fix2,tx); |
1152 | fiy2 = _mm_add_ps(fiy2,ty); |
1153 | fiz2 = _mm_add_ps(fiz2,tz); |
1154 | |
1155 | fjx3 = _mm_add_ps(fjx3,tx); |
1156 | fjy3 = _mm_add_ps(fjy3,ty); |
1157 | fjz3 = _mm_add_ps(fjz3,tz); |
1158 | |
1159 | } |
1160 | |
1161 | /************************** |
1162 | * CALCULATE INTERACTIONS * |
1163 | **************************/ |
1164 | |
1165 | if (gmx_mm_any_lt(rsq31,rcutoff2)) |
1166 | { |
1167 | |
1168 | /* REACTION-FIELD ELECTROSTATICS */ |
1169 | velec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_add_ps(rinv31,_mm_mul_ps(krf,rsq31)),crf)); |
1170 | felec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_mul_ps(rinv31,rinvsq31),krf2)); |
1171 | |
1172 | cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2); |
1173 | |
1174 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
1175 | velec = _mm_and_ps(velec,cutoff_mask); |
1176 | velec = _mm_andnot_ps(dummy_mask,velec); |
1177 | velecsum = _mm_add_ps(velecsum,velec); |
1178 | |
1179 | fscal = felec; |
1180 | |
1181 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1182 | |
1183 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1184 | |
1185 | /* Calculate temporary vectorial force */ |
1186 | tx = _mm_mul_ps(fscal,dx31); |
1187 | ty = _mm_mul_ps(fscal,dy31); |
1188 | tz = _mm_mul_ps(fscal,dz31); |
1189 | |
1190 | /* Update vectorial force */ |
1191 | fix3 = _mm_add_ps(fix3,tx); |
1192 | fiy3 = _mm_add_ps(fiy3,ty); |
1193 | fiz3 = _mm_add_ps(fiz3,tz); |
1194 | |
1195 | fjx1 = _mm_add_ps(fjx1,tx); |
1196 | fjy1 = _mm_add_ps(fjy1,ty); |
1197 | fjz1 = _mm_add_ps(fjz1,tz); |
1198 | |
1199 | } |
1200 | |
1201 | /************************** |
1202 | * CALCULATE INTERACTIONS * |
1203 | **************************/ |
1204 | |
1205 | if (gmx_mm_any_lt(rsq32,rcutoff2)) |
1206 | { |
1207 | |
1208 | /* REACTION-FIELD ELECTROSTATICS */ |
1209 | velec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_add_ps(rinv32,_mm_mul_ps(krf,rsq32)),crf)); |
1210 | felec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_mul_ps(rinv32,rinvsq32),krf2)); |
1211 | |
1212 | cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2); |
1213 | |
1214 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
1215 | velec = _mm_and_ps(velec,cutoff_mask); |
1216 | velec = _mm_andnot_ps(dummy_mask,velec); |
1217 | velecsum = _mm_add_ps(velecsum,velec); |
1218 | |
1219 | fscal = felec; |
1220 | |
1221 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1222 | |
1223 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1224 | |
1225 | /* Calculate temporary vectorial force */ |
1226 | tx = _mm_mul_ps(fscal,dx32); |
1227 | ty = _mm_mul_ps(fscal,dy32); |
1228 | tz = _mm_mul_ps(fscal,dz32); |
1229 | |
1230 | /* Update vectorial force */ |
1231 | fix3 = _mm_add_ps(fix3,tx); |
1232 | fiy3 = _mm_add_ps(fiy3,ty); |
1233 | fiz3 = _mm_add_ps(fiz3,tz); |
1234 | |
1235 | fjx2 = _mm_add_ps(fjx2,tx); |
1236 | fjy2 = _mm_add_ps(fjy2,ty); |
1237 | fjz2 = _mm_add_ps(fjz2,tz); |
1238 | |
1239 | } |
1240 | |
1241 | /************************** |
1242 | * CALCULATE INTERACTIONS * |
1243 | **************************/ |
1244 | |
1245 | if (gmx_mm_any_lt(rsq33,rcutoff2)) |
1246 | { |
1247 | |
1248 | /* REACTION-FIELD ELECTROSTATICS */ |
1249 | velec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_add_ps(rinv33,_mm_mul_ps(krf,rsq33)),crf)); |
1250 | felec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_mul_ps(rinv33,rinvsq33),krf2)); |
1251 | |
1252 | cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2); |
1253 | |
1254 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
1255 | velec = _mm_and_ps(velec,cutoff_mask); |
1256 | velec = _mm_andnot_ps(dummy_mask,velec); |
1257 | velecsum = _mm_add_ps(velecsum,velec); |
1258 | |
1259 | fscal = felec; |
1260 | |
1261 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1262 | |
1263 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1264 | |
1265 | /* Calculate temporary vectorial force */ |
1266 | tx = _mm_mul_ps(fscal,dx33); |
1267 | ty = _mm_mul_ps(fscal,dy33); |
1268 | tz = _mm_mul_ps(fscal,dz33); |
1269 | |
1270 | /* Update vectorial force */ |
1271 | fix3 = _mm_add_ps(fix3,tx); |
1272 | fiy3 = _mm_add_ps(fiy3,ty); |
1273 | fiz3 = _mm_add_ps(fiz3,tz); |
1274 | |
1275 | fjx3 = _mm_add_ps(fjx3,tx); |
1276 | fjy3 = _mm_add_ps(fjy3,ty); |
1277 | fjz3 = _mm_add_ps(fjz3,tz); |
1278 | |
1279 | } |
1280 | |
1281 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
1282 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
1283 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
1284 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
1285 | |
1286 | gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
1287 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1, |
1288 | fjx2,fjy2,fjz2,fjx3,fjy3,fjz3); |
1289 | |
1290 | /* Inner loop uses 388 flops */ |
1291 | } |
1292 | |
1293 | /* End of innermost loop */ |
1294 | |
1295 | gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3, |
1296 | f+i_coord_offset,fshift+i_shift_offset); |
1297 | |
1298 | ggid = gid[iidx]; |
1299 | /* Update potential energies */ |
1300 | gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid); |
1301 | gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid); |
1302 | |
1303 | /* Increment number of inner iterations */ |
1304 | inneriter += j_index_end - j_index_start; |
1305 | |
1306 | /* Outer loop uses 26 flops */ |
1307 | } |
1308 | |
1309 | /* Increment number of outer iterations */ |
1310 | outeriter += nri; |
1311 | |
1312 | /* Update outer/inner flops */ |
1313 | |
1314 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*388)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W4W4_VF] += outeriter*26 + inneriter*388; |
1315 | } |
1316 | /* |
1317 | * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_F_sse4_1_single |
1318 | * Electrostatics interaction: ReactionField |
1319 | * VdW interaction: CubicSplineTable |
1320 | * Geometry: Water4-Water4 |
1321 | * Calculate force/pot: Force |
1322 | */ |
1323 | void |
1324 | nb_kernel_ElecRFCut_VdwCSTab_GeomW4W4_F_sse4_1_single |
1325 | (t_nblist * gmx_restrict nlist, |
1326 | rvec * gmx_restrict xx, |
1327 | rvec * gmx_restrict ff, |
1328 | t_forcerec * gmx_restrict fr, |
1329 | t_mdatoms * gmx_restrict mdatoms, |
1330 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
1331 | t_nrnb * gmx_restrict nrnb) |
1332 | { |
1333 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
1334 | * just 0 for non-waters. |
1335 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
1336 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
1337 | */ |
1338 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
1339 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
1340 | int jnrA,jnrB,jnrC,jnrD; |
1341 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
1342 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
1343 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
1344 | real rcutoff_scalar; |
1345 | real *shiftvec,*fshift,*x,*f; |
1346 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
1347 | real scratch[4*DIM3]; |
1348 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
1349 | int vdwioffset0; |
1350 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
1351 | int vdwioffset1; |
1352 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
1353 | int vdwioffset2; |
1354 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
1355 | int vdwioffset3; |
1356 | __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3; |
1357 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
1358 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
1359 | int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D; |
1360 | __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1; |
1361 | int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D; |
1362 | __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2; |
1363 | int vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D; |
1364 | __m128 jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3; |
1365 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
1366 | __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11; |
1367 | __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12; |
1368 | __m128 dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13; |
1369 | __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21; |
1370 | __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22; |
1371 | __m128 dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23; |
1372 | __m128 dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31; |
1373 | __m128 dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32; |
1374 | __m128 dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33; |
1375 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
1376 | real *charge; |
1377 | int nvdwtype; |
1378 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
1379 | int *vdwtype; |
1380 | real *vdwparam; |
1381 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
1382 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
1383 | __m128i vfitab; |
1384 | __m128i ifour = _mm_set1_epi32(4); |
1385 | __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF; |
1386 | real *vftab; |
1387 | __m128 dummy_mask,cutoff_mask; |
1388 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
1389 | __m128 one = _mm_set1_ps(1.0); |
1390 | __m128 two = _mm_set1_ps(2.0); |
1391 | x = xx[0]; |
1392 | f = ff[0]; |
1393 | |
1394 | nri = nlist->nri; |
1395 | iinr = nlist->iinr; |
1396 | jindex = nlist->jindex; |
1397 | jjnr = nlist->jjnr; |
1398 | shiftidx = nlist->shift; |
1399 | gid = nlist->gid; |
1400 | shiftvec = fr->shift_vec[0]; |
1401 | fshift = fr->fshift[0]; |
1402 | facel = _mm_set1_ps(fr->epsfac); |
1403 | charge = mdatoms->chargeA; |
1404 | krf = _mm_set1_ps(fr->ic->k_rf); |
1405 | krf2 = _mm_set1_ps(fr->ic->k_rf*2.0); |
1406 | crf = _mm_set1_ps(fr->ic->c_rf); |
1407 | nvdwtype = fr->ntype; |
1408 | vdwparam = fr->nbfp; |
1409 | vdwtype = mdatoms->typeA; |
1410 | |
1411 | vftab = kernel_data->table_vdw->data; |
1412 | vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale); |
1413 | |
1414 | /* Setup water-specific parameters */ |
1415 | inr = nlist->iinr[0]; |
1416 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
1417 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
1418 | iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3])); |
1419 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
1420 | |
1421 | jq1 = _mm_set1_ps(charge[inr+1]); |
1422 | jq2 = _mm_set1_ps(charge[inr+2]); |
1423 | jq3 = _mm_set1_ps(charge[inr+3]); |
1424 | vdwjidx0A = 2*vdwtype[inr+0]; |
1425 | c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]); |
1426 | c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]); |
1427 | qq11 = _mm_mul_ps(iq1,jq1); |
1428 | qq12 = _mm_mul_ps(iq1,jq2); |
1429 | qq13 = _mm_mul_ps(iq1,jq3); |
1430 | qq21 = _mm_mul_ps(iq2,jq1); |
1431 | qq22 = _mm_mul_ps(iq2,jq2); |
1432 | qq23 = _mm_mul_ps(iq2,jq3); |
1433 | qq31 = _mm_mul_ps(iq3,jq1); |
1434 | qq32 = _mm_mul_ps(iq3,jq2); |
1435 | qq33 = _mm_mul_ps(iq3,jq3); |
1436 | |
1437 | /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */ |
1438 | rcutoff_scalar = fr->rcoulomb; |
1439 | rcutoff = _mm_set1_ps(rcutoff_scalar); |
1440 | rcutoff2 = _mm_mul_ps(rcutoff,rcutoff); |
1441 | |
1442 | /* Avoid stupid compiler warnings */ |
1443 | jnrA = jnrB = jnrC = jnrD = 0; |
1444 | j_coord_offsetA = 0; |
1445 | j_coord_offsetB = 0; |
1446 | j_coord_offsetC = 0; |
1447 | j_coord_offsetD = 0; |
1448 | |
1449 | outeriter = 0; |
1450 | inneriter = 0; |
1451 | |
1452 | for(iidx=0;iidx<4*DIM3;iidx++) |
1453 | { |
1454 | scratch[iidx] = 0.0; |
1455 | } |
1456 | |
1457 | /* Start outer loop over neighborlists */ |
1458 | for(iidx=0; iidx<nri; iidx++) |
1459 | { |
1460 | /* Load shift vector for this list */ |
1461 | i_shift_offset = DIM3*shiftidx[iidx]; |
1462 | |
1463 | /* Load limits for loop over neighbors */ |
1464 | j_index_start = jindex[iidx]; |
1465 | j_index_end = jindex[iidx+1]; |
1466 | |
1467 | /* Get outer coordinate index */ |
1468 | inr = iinr[iidx]; |
1469 | i_coord_offset = DIM3*inr; |
1470 | |
1471 | /* Load i particle coords and add shift vector */ |
1472 | gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
1473 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3); |
1474 | |
1475 | fix0 = _mm_setzero_ps(); |
1476 | fiy0 = _mm_setzero_ps(); |
1477 | fiz0 = _mm_setzero_ps(); |
1478 | fix1 = _mm_setzero_ps(); |
1479 | fiy1 = _mm_setzero_ps(); |
1480 | fiz1 = _mm_setzero_ps(); |
1481 | fix2 = _mm_setzero_ps(); |
1482 | fiy2 = _mm_setzero_ps(); |
1483 | fiz2 = _mm_setzero_ps(); |
1484 | fix3 = _mm_setzero_ps(); |
1485 | fiy3 = _mm_setzero_ps(); |
1486 | fiz3 = _mm_setzero_ps(); |
1487 | |
1488 | /* Start inner kernel loop */ |
1489 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
1490 | { |
1491 | |
1492 | /* Get j neighbor index, and coordinate index */ |
1493 | jnrA = jjnr[jidx]; |
1494 | jnrB = jjnr[jidx+1]; |
1495 | jnrC = jjnr[jidx+2]; |
1496 | jnrD = jjnr[jidx+3]; |
1497 | j_coord_offsetA = DIM3*jnrA; |
1498 | j_coord_offsetB = DIM3*jnrB; |
1499 | j_coord_offsetC = DIM3*jnrC; |
1500 | j_coord_offsetD = DIM3*jnrD; |
1501 | |
1502 | /* load j atom coordinates */ |
1503 | gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
1504 | x+j_coord_offsetC,x+j_coord_offsetD, |
1505 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2, |
1506 | &jy2,&jz2,&jx3,&jy3,&jz3); |
1507 | |
1508 | /* Calculate displacement vector */ |
1509 | dx00 = _mm_sub_ps(ix0,jx0); |
1510 | dy00 = _mm_sub_ps(iy0,jy0); |
1511 | dz00 = _mm_sub_ps(iz0,jz0); |
1512 | dx11 = _mm_sub_ps(ix1,jx1); |
1513 | dy11 = _mm_sub_ps(iy1,jy1); |
1514 | dz11 = _mm_sub_ps(iz1,jz1); |
1515 | dx12 = _mm_sub_ps(ix1,jx2); |
1516 | dy12 = _mm_sub_ps(iy1,jy2); |
1517 | dz12 = _mm_sub_ps(iz1,jz2); |
1518 | dx13 = _mm_sub_ps(ix1,jx3); |
1519 | dy13 = _mm_sub_ps(iy1,jy3); |
1520 | dz13 = _mm_sub_ps(iz1,jz3); |
1521 | dx21 = _mm_sub_ps(ix2,jx1); |
1522 | dy21 = _mm_sub_ps(iy2,jy1); |
1523 | dz21 = _mm_sub_ps(iz2,jz1); |
1524 | dx22 = _mm_sub_ps(ix2,jx2); |
1525 | dy22 = _mm_sub_ps(iy2,jy2); |
1526 | dz22 = _mm_sub_ps(iz2,jz2); |
1527 | dx23 = _mm_sub_ps(ix2,jx3); |
1528 | dy23 = _mm_sub_ps(iy2,jy3); |
1529 | dz23 = _mm_sub_ps(iz2,jz3); |
1530 | dx31 = _mm_sub_ps(ix3,jx1); |
1531 | dy31 = _mm_sub_ps(iy3,jy1); |
1532 | dz31 = _mm_sub_ps(iz3,jz1); |
1533 | dx32 = _mm_sub_ps(ix3,jx2); |
1534 | dy32 = _mm_sub_ps(iy3,jy2); |
1535 | dz32 = _mm_sub_ps(iz3,jz2); |
1536 | dx33 = _mm_sub_ps(ix3,jx3); |
1537 | dy33 = _mm_sub_ps(iy3,jy3); |
1538 | dz33 = _mm_sub_ps(iz3,jz3); |
1539 | |
1540 | /* Calculate squared distance and things based on it */ |
1541 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
1542 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
1543 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
1544 | rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13); |
1545 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
1546 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
1547 | rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23); |
1548 | rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31); |
1549 | rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32); |
1550 | rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33); |
1551 | |
1552 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
1553 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
1554 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
1555 | rinv13 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq13); |
1556 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
1557 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
1558 | rinv23 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq23); |
1559 | rinv31 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq31); |
1560 | rinv32 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq32); |
1561 | rinv33 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq33); |
1562 | |
1563 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
1564 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
1565 | rinvsq13 = _mm_mul_ps(rinv13,rinv13); |
1566 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
1567 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
1568 | rinvsq23 = _mm_mul_ps(rinv23,rinv23); |
1569 | rinvsq31 = _mm_mul_ps(rinv31,rinv31); |
1570 | rinvsq32 = _mm_mul_ps(rinv32,rinv32); |
1571 | rinvsq33 = _mm_mul_ps(rinv33,rinv33); |
1572 | |
1573 | fjx0 = _mm_setzero_ps(); |
1574 | fjy0 = _mm_setzero_ps(); |
1575 | fjz0 = _mm_setzero_ps(); |
1576 | fjx1 = _mm_setzero_ps(); |
1577 | fjy1 = _mm_setzero_ps(); |
1578 | fjz1 = _mm_setzero_ps(); |
1579 | fjx2 = _mm_setzero_ps(); |
1580 | fjy2 = _mm_setzero_ps(); |
1581 | fjz2 = _mm_setzero_ps(); |
1582 | fjx3 = _mm_setzero_ps(); |
1583 | fjy3 = _mm_setzero_ps(); |
1584 | fjz3 = _mm_setzero_ps(); |
1585 | |
1586 | /************************** |
1587 | * CALCULATE INTERACTIONS * |
1588 | **************************/ |
1589 | |
1590 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
1591 | { |
1592 | |
1593 | r00 = _mm_mul_ps(rsq00,rinv00); |
1594 | |
1595 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
1596 | rt = _mm_mul_ps(r00,vftabscale); |
1597 | vfitab = _mm_cvttps_epi32(rt); |
1598 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
1599 | vfitab = _mm_slli_epi32(vfitab,3); |
1600 | |
1601 | /* CUBIC SPLINE TABLE DISPERSION */ |
1602 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
1603 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
1604 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
1605 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
1606 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
1607 | Heps = _mm_mul_ps(vfeps,H); |
1608 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1609 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1610 | fvdw6 = _mm_mul_ps(c6_00,FF); |
1611 | |
1612 | /* CUBIC SPLINE TABLE REPULSION */ |
1613 | vfitab = _mm_add_epi32(vfitab,ifour); |
1614 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
1615 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
1616 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
1617 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
1618 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
1619 | Heps = _mm_mul_ps(vfeps,H); |
1620 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1621 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1622 | fvdw12 = _mm_mul_ps(c12_00,FF); |
1623 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
1624 | |
1625 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
1626 | |
1627 | fscal = fvdw; |
1628 | |
1629 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1630 | |
1631 | /* Calculate temporary vectorial force */ |
1632 | tx = _mm_mul_ps(fscal,dx00); |
1633 | ty = _mm_mul_ps(fscal,dy00); |
1634 | tz = _mm_mul_ps(fscal,dz00); |
1635 | |
1636 | /* Update vectorial force */ |
1637 | fix0 = _mm_add_ps(fix0,tx); |
1638 | fiy0 = _mm_add_ps(fiy0,ty); |
1639 | fiz0 = _mm_add_ps(fiz0,tz); |
1640 | |
1641 | fjx0 = _mm_add_ps(fjx0,tx); |
1642 | fjy0 = _mm_add_ps(fjy0,ty); |
1643 | fjz0 = _mm_add_ps(fjz0,tz); |
1644 | |
1645 | } |
1646 | |
1647 | /************************** |
1648 | * CALCULATE INTERACTIONS * |
1649 | **************************/ |
1650 | |
1651 | if (gmx_mm_any_lt(rsq11,rcutoff2)) |
1652 | { |
1653 | |
1654 | /* REACTION-FIELD ELECTROSTATICS */ |
1655 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
1656 | |
1657 | cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2); |
1658 | |
1659 | fscal = felec; |
1660 | |
1661 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1662 | |
1663 | /* Calculate temporary vectorial force */ |
1664 | tx = _mm_mul_ps(fscal,dx11); |
1665 | ty = _mm_mul_ps(fscal,dy11); |
1666 | tz = _mm_mul_ps(fscal,dz11); |
1667 | |
1668 | /* Update vectorial force */ |
1669 | fix1 = _mm_add_ps(fix1,tx); |
1670 | fiy1 = _mm_add_ps(fiy1,ty); |
1671 | fiz1 = _mm_add_ps(fiz1,tz); |
1672 | |
1673 | fjx1 = _mm_add_ps(fjx1,tx); |
1674 | fjy1 = _mm_add_ps(fjy1,ty); |
1675 | fjz1 = _mm_add_ps(fjz1,tz); |
1676 | |
1677 | } |
1678 | |
1679 | /************************** |
1680 | * CALCULATE INTERACTIONS * |
1681 | **************************/ |
1682 | |
1683 | if (gmx_mm_any_lt(rsq12,rcutoff2)) |
1684 | { |
1685 | |
1686 | /* REACTION-FIELD ELECTROSTATICS */ |
1687 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
1688 | |
1689 | cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2); |
1690 | |
1691 | fscal = felec; |
1692 | |
1693 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1694 | |
1695 | /* Calculate temporary vectorial force */ |
1696 | tx = _mm_mul_ps(fscal,dx12); |
1697 | ty = _mm_mul_ps(fscal,dy12); |
1698 | tz = _mm_mul_ps(fscal,dz12); |
1699 | |
1700 | /* Update vectorial force */ |
1701 | fix1 = _mm_add_ps(fix1,tx); |
1702 | fiy1 = _mm_add_ps(fiy1,ty); |
1703 | fiz1 = _mm_add_ps(fiz1,tz); |
1704 | |
1705 | fjx2 = _mm_add_ps(fjx2,tx); |
1706 | fjy2 = _mm_add_ps(fjy2,ty); |
1707 | fjz2 = _mm_add_ps(fjz2,tz); |
1708 | |
1709 | } |
1710 | |
1711 | /************************** |
1712 | * CALCULATE INTERACTIONS * |
1713 | **************************/ |
1714 | |
1715 | if (gmx_mm_any_lt(rsq13,rcutoff2)) |
1716 | { |
1717 | |
1718 | /* REACTION-FIELD ELECTROSTATICS */ |
1719 | felec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_mul_ps(rinv13,rinvsq13),krf2)); |
1720 | |
1721 | cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2); |
1722 | |
1723 | fscal = felec; |
1724 | |
1725 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1726 | |
1727 | /* Calculate temporary vectorial force */ |
1728 | tx = _mm_mul_ps(fscal,dx13); |
1729 | ty = _mm_mul_ps(fscal,dy13); |
1730 | tz = _mm_mul_ps(fscal,dz13); |
1731 | |
1732 | /* Update vectorial force */ |
1733 | fix1 = _mm_add_ps(fix1,tx); |
1734 | fiy1 = _mm_add_ps(fiy1,ty); |
1735 | fiz1 = _mm_add_ps(fiz1,tz); |
1736 | |
1737 | fjx3 = _mm_add_ps(fjx3,tx); |
1738 | fjy3 = _mm_add_ps(fjy3,ty); |
1739 | fjz3 = _mm_add_ps(fjz3,tz); |
1740 | |
1741 | } |
1742 | |
1743 | /************************** |
1744 | * CALCULATE INTERACTIONS * |
1745 | **************************/ |
1746 | |
1747 | if (gmx_mm_any_lt(rsq21,rcutoff2)) |
1748 | { |
1749 | |
1750 | /* REACTION-FIELD ELECTROSTATICS */ |
1751 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
1752 | |
1753 | cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2); |
1754 | |
1755 | fscal = felec; |
1756 | |
1757 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1758 | |
1759 | /* Calculate temporary vectorial force */ |
1760 | tx = _mm_mul_ps(fscal,dx21); |
1761 | ty = _mm_mul_ps(fscal,dy21); |
1762 | tz = _mm_mul_ps(fscal,dz21); |
1763 | |
1764 | /* Update vectorial force */ |
1765 | fix2 = _mm_add_ps(fix2,tx); |
1766 | fiy2 = _mm_add_ps(fiy2,ty); |
1767 | fiz2 = _mm_add_ps(fiz2,tz); |
1768 | |
1769 | fjx1 = _mm_add_ps(fjx1,tx); |
1770 | fjy1 = _mm_add_ps(fjy1,ty); |
1771 | fjz1 = _mm_add_ps(fjz1,tz); |
1772 | |
1773 | } |
1774 | |
1775 | /************************** |
1776 | * CALCULATE INTERACTIONS * |
1777 | **************************/ |
1778 | |
1779 | if (gmx_mm_any_lt(rsq22,rcutoff2)) |
1780 | { |
1781 | |
1782 | /* REACTION-FIELD ELECTROSTATICS */ |
1783 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
1784 | |
1785 | cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2); |
1786 | |
1787 | fscal = felec; |
1788 | |
1789 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1790 | |
1791 | /* Calculate temporary vectorial force */ |
1792 | tx = _mm_mul_ps(fscal,dx22); |
1793 | ty = _mm_mul_ps(fscal,dy22); |
1794 | tz = _mm_mul_ps(fscal,dz22); |
1795 | |
1796 | /* Update vectorial force */ |
1797 | fix2 = _mm_add_ps(fix2,tx); |
1798 | fiy2 = _mm_add_ps(fiy2,ty); |
1799 | fiz2 = _mm_add_ps(fiz2,tz); |
1800 | |
1801 | fjx2 = _mm_add_ps(fjx2,tx); |
1802 | fjy2 = _mm_add_ps(fjy2,ty); |
1803 | fjz2 = _mm_add_ps(fjz2,tz); |
1804 | |
1805 | } |
1806 | |
1807 | /************************** |
1808 | * CALCULATE INTERACTIONS * |
1809 | **************************/ |
1810 | |
1811 | if (gmx_mm_any_lt(rsq23,rcutoff2)) |
1812 | { |
1813 | |
1814 | /* REACTION-FIELD ELECTROSTATICS */ |
1815 | felec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_mul_ps(rinv23,rinvsq23),krf2)); |
1816 | |
1817 | cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2); |
1818 | |
1819 | fscal = felec; |
1820 | |
1821 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1822 | |
1823 | /* Calculate temporary vectorial force */ |
1824 | tx = _mm_mul_ps(fscal,dx23); |
1825 | ty = _mm_mul_ps(fscal,dy23); |
1826 | tz = _mm_mul_ps(fscal,dz23); |
1827 | |
1828 | /* Update vectorial force */ |
1829 | fix2 = _mm_add_ps(fix2,tx); |
1830 | fiy2 = _mm_add_ps(fiy2,ty); |
1831 | fiz2 = _mm_add_ps(fiz2,tz); |
1832 | |
1833 | fjx3 = _mm_add_ps(fjx3,tx); |
1834 | fjy3 = _mm_add_ps(fjy3,ty); |
1835 | fjz3 = _mm_add_ps(fjz3,tz); |
1836 | |
1837 | } |
1838 | |
1839 | /************************** |
1840 | * CALCULATE INTERACTIONS * |
1841 | **************************/ |
1842 | |
1843 | if (gmx_mm_any_lt(rsq31,rcutoff2)) |
1844 | { |
1845 | |
1846 | /* REACTION-FIELD ELECTROSTATICS */ |
1847 | felec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_mul_ps(rinv31,rinvsq31),krf2)); |
1848 | |
1849 | cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2); |
1850 | |
1851 | fscal = felec; |
1852 | |
1853 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1854 | |
1855 | /* Calculate temporary vectorial force */ |
1856 | tx = _mm_mul_ps(fscal,dx31); |
1857 | ty = _mm_mul_ps(fscal,dy31); |
1858 | tz = _mm_mul_ps(fscal,dz31); |
1859 | |
1860 | /* Update vectorial force */ |
1861 | fix3 = _mm_add_ps(fix3,tx); |
1862 | fiy3 = _mm_add_ps(fiy3,ty); |
1863 | fiz3 = _mm_add_ps(fiz3,tz); |
1864 | |
1865 | fjx1 = _mm_add_ps(fjx1,tx); |
1866 | fjy1 = _mm_add_ps(fjy1,ty); |
1867 | fjz1 = _mm_add_ps(fjz1,tz); |
1868 | |
1869 | } |
1870 | |
1871 | /************************** |
1872 | * CALCULATE INTERACTIONS * |
1873 | **************************/ |
1874 | |
1875 | if (gmx_mm_any_lt(rsq32,rcutoff2)) |
1876 | { |
1877 | |
1878 | /* REACTION-FIELD ELECTROSTATICS */ |
1879 | felec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_mul_ps(rinv32,rinvsq32),krf2)); |
1880 | |
1881 | cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2); |
1882 | |
1883 | fscal = felec; |
1884 | |
1885 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1886 | |
1887 | /* Calculate temporary vectorial force */ |
1888 | tx = _mm_mul_ps(fscal,dx32); |
1889 | ty = _mm_mul_ps(fscal,dy32); |
1890 | tz = _mm_mul_ps(fscal,dz32); |
1891 | |
1892 | /* Update vectorial force */ |
1893 | fix3 = _mm_add_ps(fix3,tx); |
1894 | fiy3 = _mm_add_ps(fiy3,ty); |
1895 | fiz3 = _mm_add_ps(fiz3,tz); |
1896 | |
1897 | fjx2 = _mm_add_ps(fjx2,tx); |
1898 | fjy2 = _mm_add_ps(fjy2,ty); |
1899 | fjz2 = _mm_add_ps(fjz2,tz); |
1900 | |
1901 | } |
1902 | |
1903 | /************************** |
1904 | * CALCULATE INTERACTIONS * |
1905 | **************************/ |
1906 | |
1907 | if (gmx_mm_any_lt(rsq33,rcutoff2)) |
1908 | { |
1909 | |
1910 | /* REACTION-FIELD ELECTROSTATICS */ |
1911 | felec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_mul_ps(rinv33,rinvsq33),krf2)); |
1912 | |
1913 | cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2); |
1914 | |
1915 | fscal = felec; |
1916 | |
1917 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1918 | |
1919 | /* Calculate temporary vectorial force */ |
1920 | tx = _mm_mul_ps(fscal,dx33); |
1921 | ty = _mm_mul_ps(fscal,dy33); |
1922 | tz = _mm_mul_ps(fscal,dz33); |
1923 | |
1924 | /* Update vectorial force */ |
1925 | fix3 = _mm_add_ps(fix3,tx); |
1926 | fiy3 = _mm_add_ps(fiy3,ty); |
1927 | fiz3 = _mm_add_ps(fiz3,tz); |
1928 | |
1929 | fjx3 = _mm_add_ps(fjx3,tx); |
1930 | fjy3 = _mm_add_ps(fjy3,ty); |
1931 | fjz3 = _mm_add_ps(fjz3,tz); |
1932 | |
1933 | } |
1934 | |
1935 | fjptrA = f+j_coord_offsetA; |
1936 | fjptrB = f+j_coord_offsetB; |
1937 | fjptrC = f+j_coord_offsetC; |
1938 | fjptrD = f+j_coord_offsetD; |
1939 | |
1940 | gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
1941 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1, |
1942 | fjx2,fjy2,fjz2,fjx3,fjy3,fjz3); |
1943 | |
1944 | /* Inner loop uses 324 flops */ |
1945 | } |
1946 | |
1947 | if(jidx<j_index_end) |
1948 | { |
1949 | |
1950 | /* Get j neighbor index, and coordinate index */ |
1951 | jnrlistA = jjnr[jidx]; |
1952 | jnrlistB = jjnr[jidx+1]; |
1953 | jnrlistC = jjnr[jidx+2]; |
1954 | jnrlistD = jjnr[jidx+3]; |
1955 | /* Sign of each element will be negative for non-real atoms. |
1956 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
1957 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
1958 | */ |
1959 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
1960 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
1961 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
1962 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
1963 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
1964 | j_coord_offsetA = DIM3*jnrA; |
1965 | j_coord_offsetB = DIM3*jnrB; |
1966 | j_coord_offsetC = DIM3*jnrC; |
1967 | j_coord_offsetD = DIM3*jnrD; |
1968 | |
1969 | /* load j atom coordinates */ |
1970 | gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
1971 | x+j_coord_offsetC,x+j_coord_offsetD, |
1972 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2, |
1973 | &jy2,&jz2,&jx3,&jy3,&jz3); |
1974 | |
1975 | /* Calculate displacement vector */ |
1976 | dx00 = _mm_sub_ps(ix0,jx0); |
1977 | dy00 = _mm_sub_ps(iy0,jy0); |
1978 | dz00 = _mm_sub_ps(iz0,jz0); |
1979 | dx11 = _mm_sub_ps(ix1,jx1); |
1980 | dy11 = _mm_sub_ps(iy1,jy1); |
1981 | dz11 = _mm_sub_ps(iz1,jz1); |
1982 | dx12 = _mm_sub_ps(ix1,jx2); |
1983 | dy12 = _mm_sub_ps(iy1,jy2); |
1984 | dz12 = _mm_sub_ps(iz1,jz2); |
1985 | dx13 = _mm_sub_ps(ix1,jx3); |
1986 | dy13 = _mm_sub_ps(iy1,jy3); |
1987 | dz13 = _mm_sub_ps(iz1,jz3); |
1988 | dx21 = _mm_sub_ps(ix2,jx1); |
1989 | dy21 = _mm_sub_ps(iy2,jy1); |
1990 | dz21 = _mm_sub_ps(iz2,jz1); |
1991 | dx22 = _mm_sub_ps(ix2,jx2); |
1992 | dy22 = _mm_sub_ps(iy2,jy2); |
1993 | dz22 = _mm_sub_ps(iz2,jz2); |
1994 | dx23 = _mm_sub_ps(ix2,jx3); |
1995 | dy23 = _mm_sub_ps(iy2,jy3); |
1996 | dz23 = _mm_sub_ps(iz2,jz3); |
1997 | dx31 = _mm_sub_ps(ix3,jx1); |
1998 | dy31 = _mm_sub_ps(iy3,jy1); |
1999 | dz31 = _mm_sub_ps(iz3,jz1); |
2000 | dx32 = _mm_sub_ps(ix3,jx2); |
2001 | dy32 = _mm_sub_ps(iy3,jy2); |
2002 | dz32 = _mm_sub_ps(iz3,jz2); |
2003 | dx33 = _mm_sub_ps(ix3,jx3); |
2004 | dy33 = _mm_sub_ps(iy3,jy3); |
2005 | dz33 = _mm_sub_ps(iz3,jz3); |
2006 | |
2007 | /* Calculate squared distance and things based on it */ |
2008 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
2009 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
2010 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
2011 | rsq13 = gmx_mm_calc_rsq_ps(dx13,dy13,dz13); |
2012 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
2013 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
2014 | rsq23 = gmx_mm_calc_rsq_ps(dx23,dy23,dz23); |
2015 | rsq31 = gmx_mm_calc_rsq_ps(dx31,dy31,dz31); |
2016 | rsq32 = gmx_mm_calc_rsq_ps(dx32,dy32,dz32); |
2017 | rsq33 = gmx_mm_calc_rsq_ps(dx33,dy33,dz33); |
2018 | |
2019 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
2020 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
2021 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
2022 | rinv13 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq13); |
2023 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
2024 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
2025 | rinv23 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq23); |
2026 | rinv31 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq31); |
2027 | rinv32 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq32); |
2028 | rinv33 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq33); |
2029 | |
2030 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
2031 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
2032 | rinvsq13 = _mm_mul_ps(rinv13,rinv13); |
2033 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
2034 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
2035 | rinvsq23 = _mm_mul_ps(rinv23,rinv23); |
2036 | rinvsq31 = _mm_mul_ps(rinv31,rinv31); |
2037 | rinvsq32 = _mm_mul_ps(rinv32,rinv32); |
2038 | rinvsq33 = _mm_mul_ps(rinv33,rinv33); |
2039 | |
2040 | fjx0 = _mm_setzero_ps(); |
2041 | fjy0 = _mm_setzero_ps(); |
2042 | fjz0 = _mm_setzero_ps(); |
2043 | fjx1 = _mm_setzero_ps(); |
2044 | fjy1 = _mm_setzero_ps(); |
2045 | fjz1 = _mm_setzero_ps(); |
2046 | fjx2 = _mm_setzero_ps(); |
2047 | fjy2 = _mm_setzero_ps(); |
2048 | fjz2 = _mm_setzero_ps(); |
2049 | fjx3 = _mm_setzero_ps(); |
2050 | fjy3 = _mm_setzero_ps(); |
2051 | fjz3 = _mm_setzero_ps(); |
2052 | |
2053 | /************************** |
2054 | * CALCULATE INTERACTIONS * |
2055 | **************************/ |
2056 | |
2057 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
2058 | { |
2059 | |
2060 | r00 = _mm_mul_ps(rsq00,rinv00); |
2061 | r00 = _mm_andnot_ps(dummy_mask,r00); |
2062 | |
2063 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
2064 | rt = _mm_mul_ps(r00,vftabscale); |
2065 | vfitab = _mm_cvttps_epi32(rt); |
2066 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
2067 | vfitab = _mm_slli_epi32(vfitab,3); |
2068 | |
2069 | /* CUBIC SPLINE TABLE DISPERSION */ |
2070 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
2071 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
2072 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
2073 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
2074 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
2075 | Heps = _mm_mul_ps(vfeps,H); |
2076 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
2077 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
2078 | fvdw6 = _mm_mul_ps(c6_00,FF); |
2079 | |
2080 | /* CUBIC SPLINE TABLE REPULSION */ |
2081 | vfitab = _mm_add_epi32(vfitab,ifour); |
2082 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
2083 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
2084 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
2085 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
2086 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
2087 | Heps = _mm_mul_ps(vfeps,H); |
2088 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
2089 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
2090 | fvdw12 = _mm_mul_ps(c12_00,FF); |
2091 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
2092 | |
2093 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
2094 | |
2095 | fscal = fvdw; |
2096 | |
2097 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2098 | |
2099 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2100 | |
2101 | /* Calculate temporary vectorial force */ |
2102 | tx = _mm_mul_ps(fscal,dx00); |
2103 | ty = _mm_mul_ps(fscal,dy00); |
2104 | tz = _mm_mul_ps(fscal,dz00); |
2105 | |
2106 | /* Update vectorial force */ |
2107 | fix0 = _mm_add_ps(fix0,tx); |
2108 | fiy0 = _mm_add_ps(fiy0,ty); |
2109 | fiz0 = _mm_add_ps(fiz0,tz); |
2110 | |
2111 | fjx0 = _mm_add_ps(fjx0,tx); |
2112 | fjy0 = _mm_add_ps(fjy0,ty); |
2113 | fjz0 = _mm_add_ps(fjz0,tz); |
2114 | |
2115 | } |
2116 | |
2117 | /************************** |
2118 | * CALCULATE INTERACTIONS * |
2119 | **************************/ |
2120 | |
2121 | if (gmx_mm_any_lt(rsq11,rcutoff2)) |
2122 | { |
2123 | |
2124 | /* REACTION-FIELD ELECTROSTATICS */ |
2125 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
2126 | |
2127 | cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2); |
2128 | |
2129 | fscal = felec; |
2130 | |
2131 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2132 | |
2133 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2134 | |
2135 | /* Calculate temporary vectorial force */ |
2136 | tx = _mm_mul_ps(fscal,dx11); |
2137 | ty = _mm_mul_ps(fscal,dy11); |
2138 | tz = _mm_mul_ps(fscal,dz11); |
2139 | |
2140 | /* Update vectorial force */ |
2141 | fix1 = _mm_add_ps(fix1,tx); |
2142 | fiy1 = _mm_add_ps(fiy1,ty); |
2143 | fiz1 = _mm_add_ps(fiz1,tz); |
2144 | |
2145 | fjx1 = _mm_add_ps(fjx1,tx); |
2146 | fjy1 = _mm_add_ps(fjy1,ty); |
2147 | fjz1 = _mm_add_ps(fjz1,tz); |
2148 | |
2149 | } |
2150 | |
2151 | /************************** |
2152 | * CALCULATE INTERACTIONS * |
2153 | **************************/ |
2154 | |
2155 | if (gmx_mm_any_lt(rsq12,rcutoff2)) |
2156 | { |
2157 | |
2158 | /* REACTION-FIELD ELECTROSTATICS */ |
2159 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
2160 | |
2161 | cutoff_mask = _mm_cmplt_ps(rsq12,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,dx12); |
2171 | ty = _mm_mul_ps(fscal,dy12); |
2172 | tz = _mm_mul_ps(fscal,dz12); |
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 | fjx2 = _mm_add_ps(fjx2,tx); |
2180 | fjy2 = _mm_add_ps(fjy2,ty); |
2181 | fjz2 = _mm_add_ps(fjz2,tz); |
2182 | |
2183 | } |
2184 | |
2185 | /************************** |
2186 | * CALCULATE INTERACTIONS * |
2187 | **************************/ |
2188 | |
2189 | if (gmx_mm_any_lt(rsq13,rcutoff2)) |
2190 | { |
2191 | |
2192 | /* REACTION-FIELD ELECTROSTATICS */ |
2193 | felec = _mm_mul_ps(qq13,_mm_sub_ps(_mm_mul_ps(rinv13,rinvsq13),krf2)); |
2194 | |
2195 | cutoff_mask = _mm_cmplt_ps(rsq13,rcutoff2); |
2196 | |
2197 | fscal = felec; |
2198 | |
2199 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2200 | |
2201 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2202 | |
2203 | /* Calculate temporary vectorial force */ |
2204 | tx = _mm_mul_ps(fscal,dx13); |
2205 | ty = _mm_mul_ps(fscal,dy13); |
2206 | tz = _mm_mul_ps(fscal,dz13); |
2207 | |
2208 | /* Update vectorial force */ |
2209 | fix1 = _mm_add_ps(fix1,tx); |
2210 | fiy1 = _mm_add_ps(fiy1,ty); |
2211 | fiz1 = _mm_add_ps(fiz1,tz); |
2212 | |
2213 | fjx3 = _mm_add_ps(fjx3,tx); |
2214 | fjy3 = _mm_add_ps(fjy3,ty); |
2215 | fjz3 = _mm_add_ps(fjz3,tz); |
2216 | |
2217 | } |
2218 | |
2219 | /************************** |
2220 | * CALCULATE INTERACTIONS * |
2221 | **************************/ |
2222 | |
2223 | if (gmx_mm_any_lt(rsq21,rcutoff2)) |
2224 | { |
2225 | |
2226 | /* REACTION-FIELD ELECTROSTATICS */ |
2227 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
2228 | |
2229 | cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2); |
2230 | |
2231 | fscal = felec; |
2232 | |
2233 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2234 | |
2235 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2236 | |
2237 | /* Calculate temporary vectorial force */ |
2238 | tx = _mm_mul_ps(fscal,dx21); |
2239 | ty = _mm_mul_ps(fscal,dy21); |
2240 | tz = _mm_mul_ps(fscal,dz21); |
2241 | |
2242 | /* Update vectorial force */ |
2243 | fix2 = _mm_add_ps(fix2,tx); |
2244 | fiy2 = _mm_add_ps(fiy2,ty); |
2245 | fiz2 = _mm_add_ps(fiz2,tz); |
2246 | |
2247 | fjx1 = _mm_add_ps(fjx1,tx); |
2248 | fjy1 = _mm_add_ps(fjy1,ty); |
2249 | fjz1 = _mm_add_ps(fjz1,tz); |
2250 | |
2251 | } |
2252 | |
2253 | /************************** |
2254 | * CALCULATE INTERACTIONS * |
2255 | **************************/ |
2256 | |
2257 | if (gmx_mm_any_lt(rsq22,rcutoff2)) |
2258 | { |
2259 | |
2260 | /* REACTION-FIELD ELECTROSTATICS */ |
2261 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
2262 | |
2263 | cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2); |
2264 | |
2265 | fscal = felec; |
2266 | |
2267 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2268 | |
2269 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2270 | |
2271 | /* Calculate temporary vectorial force */ |
2272 | tx = _mm_mul_ps(fscal,dx22); |
2273 | ty = _mm_mul_ps(fscal,dy22); |
2274 | tz = _mm_mul_ps(fscal,dz22); |
2275 | |
2276 | /* Update vectorial force */ |
2277 | fix2 = _mm_add_ps(fix2,tx); |
2278 | fiy2 = _mm_add_ps(fiy2,ty); |
2279 | fiz2 = _mm_add_ps(fiz2,tz); |
2280 | |
2281 | fjx2 = _mm_add_ps(fjx2,tx); |
2282 | fjy2 = _mm_add_ps(fjy2,ty); |
2283 | fjz2 = _mm_add_ps(fjz2,tz); |
2284 | |
2285 | } |
2286 | |
2287 | /************************** |
2288 | * CALCULATE INTERACTIONS * |
2289 | **************************/ |
2290 | |
2291 | if (gmx_mm_any_lt(rsq23,rcutoff2)) |
2292 | { |
2293 | |
2294 | /* REACTION-FIELD ELECTROSTATICS */ |
2295 | felec = _mm_mul_ps(qq23,_mm_sub_ps(_mm_mul_ps(rinv23,rinvsq23),krf2)); |
2296 | |
2297 | cutoff_mask = _mm_cmplt_ps(rsq23,rcutoff2); |
2298 | |
2299 | fscal = felec; |
2300 | |
2301 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2302 | |
2303 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2304 | |
2305 | /* Calculate temporary vectorial force */ |
2306 | tx = _mm_mul_ps(fscal,dx23); |
2307 | ty = _mm_mul_ps(fscal,dy23); |
2308 | tz = _mm_mul_ps(fscal,dz23); |
2309 | |
2310 | /* Update vectorial force */ |
2311 | fix2 = _mm_add_ps(fix2,tx); |
2312 | fiy2 = _mm_add_ps(fiy2,ty); |
2313 | fiz2 = _mm_add_ps(fiz2,tz); |
2314 | |
2315 | fjx3 = _mm_add_ps(fjx3,tx); |
2316 | fjy3 = _mm_add_ps(fjy3,ty); |
2317 | fjz3 = _mm_add_ps(fjz3,tz); |
2318 | |
2319 | } |
2320 | |
2321 | /************************** |
2322 | * CALCULATE INTERACTIONS * |
2323 | **************************/ |
2324 | |
2325 | if (gmx_mm_any_lt(rsq31,rcutoff2)) |
2326 | { |
2327 | |
2328 | /* REACTION-FIELD ELECTROSTATICS */ |
2329 | felec = _mm_mul_ps(qq31,_mm_sub_ps(_mm_mul_ps(rinv31,rinvsq31),krf2)); |
2330 | |
2331 | cutoff_mask = _mm_cmplt_ps(rsq31,rcutoff2); |
2332 | |
2333 | fscal = felec; |
2334 | |
2335 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2336 | |
2337 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2338 | |
2339 | /* Calculate temporary vectorial force */ |
2340 | tx = _mm_mul_ps(fscal,dx31); |
2341 | ty = _mm_mul_ps(fscal,dy31); |
2342 | tz = _mm_mul_ps(fscal,dz31); |
2343 | |
2344 | /* Update vectorial force */ |
2345 | fix3 = _mm_add_ps(fix3,tx); |
2346 | fiy3 = _mm_add_ps(fiy3,ty); |
2347 | fiz3 = _mm_add_ps(fiz3,tz); |
2348 | |
2349 | fjx1 = _mm_add_ps(fjx1,tx); |
2350 | fjy1 = _mm_add_ps(fjy1,ty); |
2351 | fjz1 = _mm_add_ps(fjz1,tz); |
2352 | |
2353 | } |
2354 | |
2355 | /************************** |
2356 | * CALCULATE INTERACTIONS * |
2357 | **************************/ |
2358 | |
2359 | if (gmx_mm_any_lt(rsq32,rcutoff2)) |
2360 | { |
2361 | |
2362 | /* REACTION-FIELD ELECTROSTATICS */ |
2363 | felec = _mm_mul_ps(qq32,_mm_sub_ps(_mm_mul_ps(rinv32,rinvsq32),krf2)); |
2364 | |
2365 | cutoff_mask = _mm_cmplt_ps(rsq32,rcutoff2); |
2366 | |
2367 | fscal = felec; |
2368 | |
2369 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2370 | |
2371 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2372 | |
2373 | /* Calculate temporary vectorial force */ |
2374 | tx = _mm_mul_ps(fscal,dx32); |
2375 | ty = _mm_mul_ps(fscal,dy32); |
2376 | tz = _mm_mul_ps(fscal,dz32); |
2377 | |
2378 | /* Update vectorial force */ |
2379 | fix3 = _mm_add_ps(fix3,tx); |
2380 | fiy3 = _mm_add_ps(fiy3,ty); |
2381 | fiz3 = _mm_add_ps(fiz3,tz); |
2382 | |
2383 | fjx2 = _mm_add_ps(fjx2,tx); |
2384 | fjy2 = _mm_add_ps(fjy2,ty); |
2385 | fjz2 = _mm_add_ps(fjz2,tz); |
2386 | |
2387 | } |
2388 | |
2389 | /************************** |
2390 | * CALCULATE INTERACTIONS * |
2391 | **************************/ |
2392 | |
2393 | if (gmx_mm_any_lt(rsq33,rcutoff2)) |
2394 | { |
2395 | |
2396 | /* REACTION-FIELD ELECTROSTATICS */ |
2397 | felec = _mm_mul_ps(qq33,_mm_sub_ps(_mm_mul_ps(rinv33,rinvsq33),krf2)); |
2398 | |
2399 | cutoff_mask = _mm_cmplt_ps(rsq33,rcutoff2); |
2400 | |
2401 | fscal = felec; |
2402 | |
2403 | fscal = _mm_and_ps(fscal,cutoff_mask); |
2404 | |
2405 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
2406 | |
2407 | /* Calculate temporary vectorial force */ |
2408 | tx = _mm_mul_ps(fscal,dx33); |
2409 | ty = _mm_mul_ps(fscal,dy33); |
2410 | tz = _mm_mul_ps(fscal,dz33); |
2411 | |
2412 | /* Update vectorial force */ |
2413 | fix3 = _mm_add_ps(fix3,tx); |
2414 | fiy3 = _mm_add_ps(fiy3,ty); |
2415 | fiz3 = _mm_add_ps(fiz3,tz); |
2416 | |
2417 | fjx3 = _mm_add_ps(fjx3,tx); |
2418 | fjy3 = _mm_add_ps(fjy3,ty); |
2419 | fjz3 = _mm_add_ps(fjz3,tz); |
2420 | |
2421 | } |
2422 | |
2423 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
2424 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
2425 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
2426 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
2427 | |
2428 | gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
2429 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1, |
2430 | fjx2,fjy2,fjz2,fjx3,fjy3,fjz3); |
2431 | |
2432 | /* Inner loop uses 325 flops */ |
2433 | } |
2434 | |
2435 | /* End of innermost loop */ |
2436 | |
2437 | gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3, |
2438 | f+i_coord_offset,fshift+i_shift_offset); |
2439 | |
2440 | /* Increment number of inner iterations */ |
2441 | inneriter += j_index_end - j_index_start; |
2442 | |
2443 | /* Outer loop uses 24 flops */ |
2444 | } |
2445 | |
2446 | /* Increment number of outer iterations */ |
2447 | outeriter += nri; |
2448 | |
2449 | /* Update outer/inner flops */ |
2450 | |
2451 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*325)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W4W4_F] += outeriter*24 + inneriter *325; |
2452 | } |