File: | gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_sse4_1_single.c |
Location: | line 153, column 5 |
Description: | Value stored to 'j_coord_offsetB' 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_GeomW4P1_VF_sse4_1_single |
54 | * Electrostatics interaction: ReactionField |
55 | * VdW interaction: CubicSplineTable |
56 | * Geometry: Water4-Particle |
57 | * Calculate force/pot: PotentialAndForce |
58 | */ |
59 | void |
60 | nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse4_1_single |
61 | (t_nblist * gmx_restrict nlist, |
62 | rvec * gmx_restrict xx, |
63 | rvec * gmx_restrict ff, |
64 | t_forcerec * gmx_restrict fr, |
65 | t_mdatoms * gmx_restrict mdatoms, |
66 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
67 | t_nrnb * gmx_restrict nrnb) |
68 | { |
69 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
70 | * just 0 for non-waters. |
71 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
72 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
73 | */ |
74 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
75 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
76 | int jnrA,jnrB,jnrC,jnrD; |
77 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
78 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
79 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
80 | real rcutoff_scalar; |
81 | real *shiftvec,*fshift,*x,*f; |
82 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
83 | real scratch[4*DIM3]; |
84 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
85 | int vdwioffset0; |
86 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
87 | int vdwioffset1; |
88 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
89 | int vdwioffset2; |
90 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
91 | int vdwioffset3; |
92 | __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3; |
93 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
94 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
95 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
96 | __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10; |
97 | __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20; |
98 | __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30; |
99 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
100 | real *charge; |
101 | int nvdwtype; |
102 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
103 | int *vdwtype; |
104 | real *vdwparam; |
105 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
106 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
107 | __m128i vfitab; |
108 | __m128i ifour = _mm_set1_epi32(4); |
109 | __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF; |
110 | real *vftab; |
111 | __m128 dummy_mask,cutoff_mask; |
112 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
113 | __m128 one = _mm_set1_ps(1.0); |
114 | __m128 two = _mm_set1_ps(2.0); |
115 | x = xx[0]; |
116 | f = ff[0]; |
117 | |
118 | nri = nlist->nri; |
119 | iinr = nlist->iinr; |
120 | jindex = nlist->jindex; |
121 | jjnr = nlist->jjnr; |
122 | shiftidx = nlist->shift; |
123 | gid = nlist->gid; |
124 | shiftvec = fr->shift_vec[0]; |
125 | fshift = fr->fshift[0]; |
126 | facel = _mm_set1_ps(fr->epsfac); |
127 | charge = mdatoms->chargeA; |
128 | krf = _mm_set1_ps(fr->ic->k_rf); |
129 | krf2 = _mm_set1_ps(fr->ic->k_rf*2.0); |
130 | crf = _mm_set1_ps(fr->ic->c_rf); |
131 | nvdwtype = fr->ntype; |
132 | vdwparam = fr->nbfp; |
133 | vdwtype = mdatoms->typeA; |
134 | |
135 | vftab = kernel_data->table_vdw->data; |
136 | vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale); |
137 | |
138 | /* Setup water-specific parameters */ |
139 | inr = nlist->iinr[0]; |
140 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
141 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
142 | iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3])); |
143 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
144 | |
145 | /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */ |
146 | rcutoff_scalar = fr->rcoulomb; |
147 | rcutoff = _mm_set1_ps(rcutoff_scalar); |
148 | rcutoff2 = _mm_mul_ps(rcutoff,rcutoff); |
149 | |
150 | /* Avoid stupid compiler warnings */ |
151 | jnrA = jnrB = jnrC = jnrD = 0; |
152 | j_coord_offsetA = 0; |
153 | j_coord_offsetB = 0; |
Value stored to 'j_coord_offsetB' is never read | |
154 | j_coord_offsetC = 0; |
155 | j_coord_offsetD = 0; |
156 | |
157 | outeriter = 0; |
158 | inneriter = 0; |
159 | |
160 | for(iidx=0;iidx<4*DIM3;iidx++) |
161 | { |
162 | scratch[iidx] = 0.0; |
163 | } |
164 | |
165 | /* Start outer loop over neighborlists */ |
166 | for(iidx=0; iidx<nri; iidx++) |
167 | { |
168 | /* Load shift vector for this list */ |
169 | i_shift_offset = DIM3*shiftidx[iidx]; |
170 | |
171 | /* Load limits for loop over neighbors */ |
172 | j_index_start = jindex[iidx]; |
173 | j_index_end = jindex[iidx+1]; |
174 | |
175 | /* Get outer coordinate index */ |
176 | inr = iinr[iidx]; |
177 | i_coord_offset = DIM3*inr; |
178 | |
179 | /* Load i particle coords and add shift vector */ |
180 | gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
181 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3); |
182 | |
183 | fix0 = _mm_setzero_ps(); |
184 | fiy0 = _mm_setzero_ps(); |
185 | fiz0 = _mm_setzero_ps(); |
186 | fix1 = _mm_setzero_ps(); |
187 | fiy1 = _mm_setzero_ps(); |
188 | fiz1 = _mm_setzero_ps(); |
189 | fix2 = _mm_setzero_ps(); |
190 | fiy2 = _mm_setzero_ps(); |
191 | fiz2 = _mm_setzero_ps(); |
192 | fix3 = _mm_setzero_ps(); |
193 | fiy3 = _mm_setzero_ps(); |
194 | fiz3 = _mm_setzero_ps(); |
195 | |
196 | /* Reset potential sums */ |
197 | velecsum = _mm_setzero_ps(); |
198 | vvdwsum = _mm_setzero_ps(); |
199 | |
200 | /* Start inner kernel loop */ |
201 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
202 | { |
203 | |
204 | /* Get j neighbor index, and coordinate index */ |
205 | jnrA = jjnr[jidx]; |
206 | jnrB = jjnr[jidx+1]; |
207 | jnrC = jjnr[jidx+2]; |
208 | jnrD = jjnr[jidx+3]; |
209 | j_coord_offsetA = DIM3*jnrA; |
210 | j_coord_offsetB = DIM3*jnrB; |
211 | j_coord_offsetC = DIM3*jnrC; |
212 | j_coord_offsetD = DIM3*jnrD; |
213 | |
214 | /* load j atom coordinates */ |
215 | gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
216 | x+j_coord_offsetC,x+j_coord_offsetD, |
217 | &jx0,&jy0,&jz0); |
218 | |
219 | /* Calculate displacement vector */ |
220 | dx00 = _mm_sub_ps(ix0,jx0); |
221 | dy00 = _mm_sub_ps(iy0,jy0); |
222 | dz00 = _mm_sub_ps(iz0,jz0); |
223 | dx10 = _mm_sub_ps(ix1,jx0); |
224 | dy10 = _mm_sub_ps(iy1,jy0); |
225 | dz10 = _mm_sub_ps(iz1,jz0); |
226 | dx20 = _mm_sub_ps(ix2,jx0); |
227 | dy20 = _mm_sub_ps(iy2,jy0); |
228 | dz20 = _mm_sub_ps(iz2,jz0); |
229 | dx30 = _mm_sub_ps(ix3,jx0); |
230 | dy30 = _mm_sub_ps(iy3,jy0); |
231 | dz30 = _mm_sub_ps(iz3,jz0); |
232 | |
233 | /* Calculate squared distance and things based on it */ |
234 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
235 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
236 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
237 | rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30); |
238 | |
239 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
240 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
241 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
242 | rinv30 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq30); |
243 | |
244 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
245 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
246 | rinvsq30 = _mm_mul_ps(rinv30,rinv30); |
247 | |
248 | /* Load parameters for j particles */ |
249 | jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0, |
250 | charge+jnrC+0,charge+jnrD+0); |
251 | vdwjidx0A = 2*vdwtype[jnrA+0]; |
252 | vdwjidx0B = 2*vdwtype[jnrB+0]; |
253 | vdwjidx0C = 2*vdwtype[jnrC+0]; |
254 | vdwjidx0D = 2*vdwtype[jnrD+0]; |
255 | |
256 | fjx0 = _mm_setzero_ps(); |
257 | fjy0 = _mm_setzero_ps(); |
258 | fjz0 = _mm_setzero_ps(); |
259 | |
260 | /************************** |
261 | * CALCULATE INTERACTIONS * |
262 | **************************/ |
263 | |
264 | r00 = _mm_mul_ps(rsq00,rinv00); |
265 | |
266 | /* Compute parameters for interactions between i and j atoms */ |
267 | gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A, |
268 | vdwparam+vdwioffset0+vdwjidx0B, |
269 | vdwparam+vdwioffset0+vdwjidx0C, |
270 | vdwparam+vdwioffset0+vdwjidx0D, |
271 | &c6_00,&c12_00); |
272 | |
273 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
274 | rt = _mm_mul_ps(r00,vftabscale); |
275 | vfitab = _mm_cvttps_epi32(rt); |
276 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
277 | vfitab = _mm_slli_epi32(vfitab,3); |
278 | |
279 | /* CUBIC SPLINE TABLE DISPERSION */ |
280 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
281 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
282 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
283 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
284 | _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); |
285 | Heps = _mm_mul_ps(vfeps,H); |
286 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
287 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
288 | vvdw6 = _mm_mul_ps(c6_00,VV); |
289 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
290 | fvdw6 = _mm_mul_ps(c6_00,FF); |
291 | |
292 | /* CUBIC SPLINE TABLE REPULSION */ |
293 | vfitab = _mm_add_epi32(vfitab,ifour); |
294 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
295 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
296 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
297 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
298 | _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); |
299 | Heps = _mm_mul_ps(vfeps,H); |
300 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
301 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
302 | vvdw12 = _mm_mul_ps(c12_00,VV); |
303 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
304 | fvdw12 = _mm_mul_ps(c12_00,FF); |
305 | vvdw = _mm_add_ps(vvdw12,vvdw6); |
306 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
307 | |
308 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
309 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
310 | |
311 | fscal = fvdw; |
312 | |
313 | /* Calculate temporary vectorial force */ |
314 | tx = _mm_mul_ps(fscal,dx00); |
315 | ty = _mm_mul_ps(fscal,dy00); |
316 | tz = _mm_mul_ps(fscal,dz00); |
317 | |
318 | /* Update vectorial force */ |
319 | fix0 = _mm_add_ps(fix0,tx); |
320 | fiy0 = _mm_add_ps(fiy0,ty); |
321 | fiz0 = _mm_add_ps(fiz0,tz); |
322 | |
323 | fjx0 = _mm_add_ps(fjx0,tx); |
324 | fjy0 = _mm_add_ps(fjy0,ty); |
325 | fjz0 = _mm_add_ps(fjz0,tz); |
326 | |
327 | /************************** |
328 | * CALCULATE INTERACTIONS * |
329 | **************************/ |
330 | |
331 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
332 | { |
333 | |
334 | /* Compute parameters for interactions between i and j atoms */ |
335 | qq10 = _mm_mul_ps(iq1,jq0); |
336 | |
337 | /* REACTION-FIELD ELECTROSTATICS */ |
338 | velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf)); |
339 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
340 | |
341 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
342 | |
343 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
344 | velec = _mm_and_ps(velec,cutoff_mask); |
345 | velecsum = _mm_add_ps(velecsum,velec); |
346 | |
347 | fscal = felec; |
348 | |
349 | fscal = _mm_and_ps(fscal,cutoff_mask); |
350 | |
351 | /* Calculate temporary vectorial force */ |
352 | tx = _mm_mul_ps(fscal,dx10); |
353 | ty = _mm_mul_ps(fscal,dy10); |
354 | tz = _mm_mul_ps(fscal,dz10); |
355 | |
356 | /* Update vectorial force */ |
357 | fix1 = _mm_add_ps(fix1,tx); |
358 | fiy1 = _mm_add_ps(fiy1,ty); |
359 | fiz1 = _mm_add_ps(fiz1,tz); |
360 | |
361 | fjx0 = _mm_add_ps(fjx0,tx); |
362 | fjy0 = _mm_add_ps(fjy0,ty); |
363 | fjz0 = _mm_add_ps(fjz0,tz); |
364 | |
365 | } |
366 | |
367 | /************************** |
368 | * CALCULATE INTERACTIONS * |
369 | **************************/ |
370 | |
371 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
372 | { |
373 | |
374 | /* Compute parameters for interactions between i and j atoms */ |
375 | qq20 = _mm_mul_ps(iq2,jq0); |
376 | |
377 | /* REACTION-FIELD ELECTROSTATICS */ |
378 | velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf)); |
379 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
380 | |
381 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
382 | |
383 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
384 | velec = _mm_and_ps(velec,cutoff_mask); |
385 | velecsum = _mm_add_ps(velecsum,velec); |
386 | |
387 | fscal = felec; |
388 | |
389 | fscal = _mm_and_ps(fscal,cutoff_mask); |
390 | |
391 | /* Calculate temporary vectorial force */ |
392 | tx = _mm_mul_ps(fscal,dx20); |
393 | ty = _mm_mul_ps(fscal,dy20); |
394 | tz = _mm_mul_ps(fscal,dz20); |
395 | |
396 | /* Update vectorial force */ |
397 | fix2 = _mm_add_ps(fix2,tx); |
398 | fiy2 = _mm_add_ps(fiy2,ty); |
399 | fiz2 = _mm_add_ps(fiz2,tz); |
400 | |
401 | fjx0 = _mm_add_ps(fjx0,tx); |
402 | fjy0 = _mm_add_ps(fjy0,ty); |
403 | fjz0 = _mm_add_ps(fjz0,tz); |
404 | |
405 | } |
406 | |
407 | /************************** |
408 | * CALCULATE INTERACTIONS * |
409 | **************************/ |
410 | |
411 | if (gmx_mm_any_lt(rsq30,rcutoff2)) |
412 | { |
413 | |
414 | /* Compute parameters for interactions between i and j atoms */ |
415 | qq30 = _mm_mul_ps(iq3,jq0); |
416 | |
417 | /* REACTION-FIELD ELECTROSTATICS */ |
418 | velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf)); |
419 | felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2)); |
420 | |
421 | cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2); |
422 | |
423 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
424 | velec = _mm_and_ps(velec,cutoff_mask); |
425 | velecsum = _mm_add_ps(velecsum,velec); |
426 | |
427 | fscal = felec; |
428 | |
429 | fscal = _mm_and_ps(fscal,cutoff_mask); |
430 | |
431 | /* Calculate temporary vectorial force */ |
432 | tx = _mm_mul_ps(fscal,dx30); |
433 | ty = _mm_mul_ps(fscal,dy30); |
434 | tz = _mm_mul_ps(fscal,dz30); |
435 | |
436 | /* Update vectorial force */ |
437 | fix3 = _mm_add_ps(fix3,tx); |
438 | fiy3 = _mm_add_ps(fiy3,ty); |
439 | fiz3 = _mm_add_ps(fiz3,tz); |
440 | |
441 | fjx0 = _mm_add_ps(fjx0,tx); |
442 | fjy0 = _mm_add_ps(fjy0,ty); |
443 | fjz0 = _mm_add_ps(fjz0,tz); |
444 | |
445 | } |
446 | |
447 | fjptrA = f+j_coord_offsetA; |
448 | fjptrB = f+j_coord_offsetB; |
449 | fjptrC = f+j_coord_offsetC; |
450 | fjptrD = f+j_coord_offsetD; |
451 | |
452 | gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0); |
453 | |
454 | /* Inner loop uses 164 flops */ |
455 | } |
456 | |
457 | if(jidx<j_index_end) |
458 | { |
459 | |
460 | /* Get j neighbor index, and coordinate index */ |
461 | jnrlistA = jjnr[jidx]; |
462 | jnrlistB = jjnr[jidx+1]; |
463 | jnrlistC = jjnr[jidx+2]; |
464 | jnrlistD = jjnr[jidx+3]; |
465 | /* Sign of each element will be negative for non-real atoms. |
466 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
467 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
468 | */ |
469 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
470 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
471 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
472 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
473 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
474 | j_coord_offsetA = DIM3*jnrA; |
475 | j_coord_offsetB = DIM3*jnrB; |
476 | j_coord_offsetC = DIM3*jnrC; |
477 | j_coord_offsetD = DIM3*jnrD; |
478 | |
479 | /* load j atom coordinates */ |
480 | gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
481 | x+j_coord_offsetC,x+j_coord_offsetD, |
482 | &jx0,&jy0,&jz0); |
483 | |
484 | /* Calculate displacement vector */ |
485 | dx00 = _mm_sub_ps(ix0,jx0); |
486 | dy00 = _mm_sub_ps(iy0,jy0); |
487 | dz00 = _mm_sub_ps(iz0,jz0); |
488 | dx10 = _mm_sub_ps(ix1,jx0); |
489 | dy10 = _mm_sub_ps(iy1,jy0); |
490 | dz10 = _mm_sub_ps(iz1,jz0); |
491 | dx20 = _mm_sub_ps(ix2,jx0); |
492 | dy20 = _mm_sub_ps(iy2,jy0); |
493 | dz20 = _mm_sub_ps(iz2,jz0); |
494 | dx30 = _mm_sub_ps(ix3,jx0); |
495 | dy30 = _mm_sub_ps(iy3,jy0); |
496 | dz30 = _mm_sub_ps(iz3,jz0); |
497 | |
498 | /* Calculate squared distance and things based on it */ |
499 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
500 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
501 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
502 | rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30); |
503 | |
504 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
505 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
506 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
507 | rinv30 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq30); |
508 | |
509 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
510 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
511 | rinvsq30 = _mm_mul_ps(rinv30,rinv30); |
512 | |
513 | /* Load parameters for j particles */ |
514 | jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0, |
515 | charge+jnrC+0,charge+jnrD+0); |
516 | vdwjidx0A = 2*vdwtype[jnrA+0]; |
517 | vdwjidx0B = 2*vdwtype[jnrB+0]; |
518 | vdwjidx0C = 2*vdwtype[jnrC+0]; |
519 | vdwjidx0D = 2*vdwtype[jnrD+0]; |
520 | |
521 | fjx0 = _mm_setzero_ps(); |
522 | fjy0 = _mm_setzero_ps(); |
523 | fjz0 = _mm_setzero_ps(); |
524 | |
525 | /************************** |
526 | * CALCULATE INTERACTIONS * |
527 | **************************/ |
528 | |
529 | r00 = _mm_mul_ps(rsq00,rinv00); |
530 | r00 = _mm_andnot_ps(dummy_mask,r00); |
531 | |
532 | /* Compute parameters for interactions between i and j atoms */ |
533 | gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A, |
534 | vdwparam+vdwioffset0+vdwjidx0B, |
535 | vdwparam+vdwioffset0+vdwjidx0C, |
536 | vdwparam+vdwioffset0+vdwjidx0D, |
537 | &c6_00,&c12_00); |
538 | |
539 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
540 | rt = _mm_mul_ps(r00,vftabscale); |
541 | vfitab = _mm_cvttps_epi32(rt); |
542 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
543 | vfitab = _mm_slli_epi32(vfitab,3); |
544 | |
545 | /* CUBIC SPLINE TABLE DISPERSION */ |
546 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
547 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
548 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
549 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
550 | _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); |
551 | Heps = _mm_mul_ps(vfeps,H); |
552 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
553 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
554 | vvdw6 = _mm_mul_ps(c6_00,VV); |
555 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
556 | fvdw6 = _mm_mul_ps(c6_00,FF); |
557 | |
558 | /* CUBIC SPLINE TABLE REPULSION */ |
559 | vfitab = _mm_add_epi32(vfitab,ifour); |
560 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
561 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
562 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
563 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
564 | _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); |
565 | Heps = _mm_mul_ps(vfeps,H); |
566 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
567 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
568 | vvdw12 = _mm_mul_ps(c12_00,VV); |
569 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
570 | fvdw12 = _mm_mul_ps(c12_00,FF); |
571 | vvdw = _mm_add_ps(vvdw12,vvdw6); |
572 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
573 | |
574 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
575 | vvdw = _mm_andnot_ps(dummy_mask,vvdw); |
576 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
577 | |
578 | fscal = fvdw; |
579 | |
580 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
581 | |
582 | /* Calculate temporary vectorial force */ |
583 | tx = _mm_mul_ps(fscal,dx00); |
584 | ty = _mm_mul_ps(fscal,dy00); |
585 | tz = _mm_mul_ps(fscal,dz00); |
586 | |
587 | /* Update vectorial force */ |
588 | fix0 = _mm_add_ps(fix0,tx); |
589 | fiy0 = _mm_add_ps(fiy0,ty); |
590 | fiz0 = _mm_add_ps(fiz0,tz); |
591 | |
592 | fjx0 = _mm_add_ps(fjx0,tx); |
593 | fjy0 = _mm_add_ps(fjy0,ty); |
594 | fjz0 = _mm_add_ps(fjz0,tz); |
595 | |
596 | /************************** |
597 | * CALCULATE INTERACTIONS * |
598 | **************************/ |
599 | |
600 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
601 | { |
602 | |
603 | /* Compute parameters for interactions between i and j atoms */ |
604 | qq10 = _mm_mul_ps(iq1,jq0); |
605 | |
606 | /* REACTION-FIELD ELECTROSTATICS */ |
607 | velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf)); |
608 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
609 | |
610 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
611 | |
612 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
613 | velec = _mm_and_ps(velec,cutoff_mask); |
614 | velec = _mm_andnot_ps(dummy_mask,velec); |
615 | velecsum = _mm_add_ps(velecsum,velec); |
616 | |
617 | fscal = felec; |
618 | |
619 | fscal = _mm_and_ps(fscal,cutoff_mask); |
620 | |
621 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
622 | |
623 | /* Calculate temporary vectorial force */ |
624 | tx = _mm_mul_ps(fscal,dx10); |
625 | ty = _mm_mul_ps(fscal,dy10); |
626 | tz = _mm_mul_ps(fscal,dz10); |
627 | |
628 | /* Update vectorial force */ |
629 | fix1 = _mm_add_ps(fix1,tx); |
630 | fiy1 = _mm_add_ps(fiy1,ty); |
631 | fiz1 = _mm_add_ps(fiz1,tz); |
632 | |
633 | fjx0 = _mm_add_ps(fjx0,tx); |
634 | fjy0 = _mm_add_ps(fjy0,ty); |
635 | fjz0 = _mm_add_ps(fjz0,tz); |
636 | |
637 | } |
638 | |
639 | /************************** |
640 | * CALCULATE INTERACTIONS * |
641 | **************************/ |
642 | |
643 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
644 | { |
645 | |
646 | /* Compute parameters for interactions between i and j atoms */ |
647 | qq20 = _mm_mul_ps(iq2,jq0); |
648 | |
649 | /* REACTION-FIELD ELECTROSTATICS */ |
650 | velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf)); |
651 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
652 | |
653 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
654 | |
655 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
656 | velec = _mm_and_ps(velec,cutoff_mask); |
657 | velec = _mm_andnot_ps(dummy_mask,velec); |
658 | velecsum = _mm_add_ps(velecsum,velec); |
659 | |
660 | fscal = felec; |
661 | |
662 | fscal = _mm_and_ps(fscal,cutoff_mask); |
663 | |
664 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
665 | |
666 | /* Calculate temporary vectorial force */ |
667 | tx = _mm_mul_ps(fscal,dx20); |
668 | ty = _mm_mul_ps(fscal,dy20); |
669 | tz = _mm_mul_ps(fscal,dz20); |
670 | |
671 | /* Update vectorial force */ |
672 | fix2 = _mm_add_ps(fix2,tx); |
673 | fiy2 = _mm_add_ps(fiy2,ty); |
674 | fiz2 = _mm_add_ps(fiz2,tz); |
675 | |
676 | fjx0 = _mm_add_ps(fjx0,tx); |
677 | fjy0 = _mm_add_ps(fjy0,ty); |
678 | fjz0 = _mm_add_ps(fjz0,tz); |
679 | |
680 | } |
681 | |
682 | /************************** |
683 | * CALCULATE INTERACTIONS * |
684 | **************************/ |
685 | |
686 | if (gmx_mm_any_lt(rsq30,rcutoff2)) |
687 | { |
688 | |
689 | /* Compute parameters for interactions between i and j atoms */ |
690 | qq30 = _mm_mul_ps(iq3,jq0); |
691 | |
692 | /* REACTION-FIELD ELECTROSTATICS */ |
693 | velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf)); |
694 | felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2)); |
695 | |
696 | cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2); |
697 | |
698 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
699 | velec = _mm_and_ps(velec,cutoff_mask); |
700 | velec = _mm_andnot_ps(dummy_mask,velec); |
701 | velecsum = _mm_add_ps(velecsum,velec); |
702 | |
703 | fscal = felec; |
704 | |
705 | fscal = _mm_and_ps(fscal,cutoff_mask); |
706 | |
707 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
708 | |
709 | /* Calculate temporary vectorial force */ |
710 | tx = _mm_mul_ps(fscal,dx30); |
711 | ty = _mm_mul_ps(fscal,dy30); |
712 | tz = _mm_mul_ps(fscal,dz30); |
713 | |
714 | /* Update vectorial force */ |
715 | fix3 = _mm_add_ps(fix3,tx); |
716 | fiy3 = _mm_add_ps(fiy3,ty); |
717 | fiz3 = _mm_add_ps(fiz3,tz); |
718 | |
719 | fjx0 = _mm_add_ps(fjx0,tx); |
720 | fjy0 = _mm_add_ps(fjy0,ty); |
721 | fjz0 = _mm_add_ps(fjz0,tz); |
722 | |
723 | } |
724 | |
725 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
726 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
727 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
728 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
729 | |
730 | gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0); |
731 | |
732 | /* Inner loop uses 165 flops */ |
733 | } |
734 | |
735 | /* End of innermost loop */ |
736 | |
737 | gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3, |
738 | f+i_coord_offset,fshift+i_shift_offset); |
739 | |
740 | ggid = gid[iidx]; |
741 | /* Update potential energies */ |
742 | gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid); |
743 | gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid); |
744 | |
745 | /* Increment number of inner iterations */ |
746 | inneriter += j_index_end - j_index_start; |
747 | |
748 | /* Outer loop uses 26 flops */ |
749 | } |
750 | |
751 | /* Increment number of outer iterations */ |
752 | outeriter += nri; |
753 | |
754 | /* Update outer/inner flops */ |
755 | |
756 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*165)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W4_VF] += outeriter*26 + inneriter *165; |
757 | } |
758 | /* |
759 | * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_single |
760 | * Electrostatics interaction: ReactionField |
761 | * VdW interaction: CubicSplineTable |
762 | * Geometry: Water4-Particle |
763 | * Calculate force/pot: Force |
764 | */ |
765 | void |
766 | nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_single |
767 | (t_nblist * gmx_restrict nlist, |
768 | rvec * gmx_restrict xx, |
769 | rvec * gmx_restrict ff, |
770 | t_forcerec * gmx_restrict fr, |
771 | t_mdatoms * gmx_restrict mdatoms, |
772 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
773 | t_nrnb * gmx_restrict nrnb) |
774 | { |
775 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
776 | * just 0 for non-waters. |
777 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
778 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
779 | */ |
780 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
781 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
782 | int jnrA,jnrB,jnrC,jnrD; |
783 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
784 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
785 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
786 | real rcutoff_scalar; |
787 | real *shiftvec,*fshift,*x,*f; |
788 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
789 | real scratch[4*DIM3]; |
790 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
791 | int vdwioffset0; |
792 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
793 | int vdwioffset1; |
794 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
795 | int vdwioffset2; |
796 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
797 | int vdwioffset3; |
798 | __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3; |
799 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
800 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
801 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
802 | __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10; |
803 | __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20; |
804 | __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30; |
805 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
806 | real *charge; |
807 | int nvdwtype; |
808 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
809 | int *vdwtype; |
810 | real *vdwparam; |
811 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
812 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
813 | __m128i vfitab; |
814 | __m128i ifour = _mm_set1_epi32(4); |
815 | __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF; |
816 | real *vftab; |
817 | __m128 dummy_mask,cutoff_mask; |
818 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
819 | __m128 one = _mm_set1_ps(1.0); |
820 | __m128 two = _mm_set1_ps(2.0); |
821 | x = xx[0]; |
822 | f = ff[0]; |
823 | |
824 | nri = nlist->nri; |
825 | iinr = nlist->iinr; |
826 | jindex = nlist->jindex; |
827 | jjnr = nlist->jjnr; |
828 | shiftidx = nlist->shift; |
829 | gid = nlist->gid; |
830 | shiftvec = fr->shift_vec[0]; |
831 | fshift = fr->fshift[0]; |
832 | facel = _mm_set1_ps(fr->epsfac); |
833 | charge = mdatoms->chargeA; |
834 | krf = _mm_set1_ps(fr->ic->k_rf); |
835 | krf2 = _mm_set1_ps(fr->ic->k_rf*2.0); |
836 | crf = _mm_set1_ps(fr->ic->c_rf); |
837 | nvdwtype = fr->ntype; |
838 | vdwparam = fr->nbfp; |
839 | vdwtype = mdatoms->typeA; |
840 | |
841 | vftab = kernel_data->table_vdw->data; |
842 | vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale); |
843 | |
844 | /* Setup water-specific parameters */ |
845 | inr = nlist->iinr[0]; |
846 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
847 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
848 | iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3])); |
849 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
850 | |
851 | /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */ |
852 | rcutoff_scalar = fr->rcoulomb; |
853 | rcutoff = _mm_set1_ps(rcutoff_scalar); |
854 | rcutoff2 = _mm_mul_ps(rcutoff,rcutoff); |
855 | |
856 | /* Avoid stupid compiler warnings */ |
857 | jnrA = jnrB = jnrC = jnrD = 0; |
858 | j_coord_offsetA = 0; |
859 | j_coord_offsetB = 0; |
860 | j_coord_offsetC = 0; |
861 | j_coord_offsetD = 0; |
862 | |
863 | outeriter = 0; |
864 | inneriter = 0; |
865 | |
866 | for(iidx=0;iidx<4*DIM3;iidx++) |
867 | { |
868 | scratch[iidx] = 0.0; |
869 | } |
870 | |
871 | /* Start outer loop over neighborlists */ |
872 | for(iidx=0; iidx<nri; iidx++) |
873 | { |
874 | /* Load shift vector for this list */ |
875 | i_shift_offset = DIM3*shiftidx[iidx]; |
876 | |
877 | /* Load limits for loop over neighbors */ |
878 | j_index_start = jindex[iidx]; |
879 | j_index_end = jindex[iidx+1]; |
880 | |
881 | /* Get outer coordinate index */ |
882 | inr = iinr[iidx]; |
883 | i_coord_offset = DIM3*inr; |
884 | |
885 | /* Load i particle coords and add shift vector */ |
886 | gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
887 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3); |
888 | |
889 | fix0 = _mm_setzero_ps(); |
890 | fiy0 = _mm_setzero_ps(); |
891 | fiz0 = _mm_setzero_ps(); |
892 | fix1 = _mm_setzero_ps(); |
893 | fiy1 = _mm_setzero_ps(); |
894 | fiz1 = _mm_setzero_ps(); |
895 | fix2 = _mm_setzero_ps(); |
896 | fiy2 = _mm_setzero_ps(); |
897 | fiz2 = _mm_setzero_ps(); |
898 | fix3 = _mm_setzero_ps(); |
899 | fiy3 = _mm_setzero_ps(); |
900 | fiz3 = _mm_setzero_ps(); |
901 | |
902 | /* Start inner kernel loop */ |
903 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
904 | { |
905 | |
906 | /* Get j neighbor index, and coordinate index */ |
907 | jnrA = jjnr[jidx]; |
908 | jnrB = jjnr[jidx+1]; |
909 | jnrC = jjnr[jidx+2]; |
910 | jnrD = jjnr[jidx+3]; |
911 | j_coord_offsetA = DIM3*jnrA; |
912 | j_coord_offsetB = DIM3*jnrB; |
913 | j_coord_offsetC = DIM3*jnrC; |
914 | j_coord_offsetD = DIM3*jnrD; |
915 | |
916 | /* load j atom coordinates */ |
917 | gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
918 | x+j_coord_offsetC,x+j_coord_offsetD, |
919 | &jx0,&jy0,&jz0); |
920 | |
921 | /* Calculate displacement vector */ |
922 | dx00 = _mm_sub_ps(ix0,jx0); |
923 | dy00 = _mm_sub_ps(iy0,jy0); |
924 | dz00 = _mm_sub_ps(iz0,jz0); |
925 | dx10 = _mm_sub_ps(ix1,jx0); |
926 | dy10 = _mm_sub_ps(iy1,jy0); |
927 | dz10 = _mm_sub_ps(iz1,jz0); |
928 | dx20 = _mm_sub_ps(ix2,jx0); |
929 | dy20 = _mm_sub_ps(iy2,jy0); |
930 | dz20 = _mm_sub_ps(iz2,jz0); |
931 | dx30 = _mm_sub_ps(ix3,jx0); |
932 | dy30 = _mm_sub_ps(iy3,jy0); |
933 | dz30 = _mm_sub_ps(iz3,jz0); |
934 | |
935 | /* Calculate squared distance and things based on it */ |
936 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
937 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
938 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
939 | rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30); |
940 | |
941 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
942 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
943 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
944 | rinv30 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq30); |
945 | |
946 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
947 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
948 | rinvsq30 = _mm_mul_ps(rinv30,rinv30); |
949 | |
950 | /* Load parameters for j particles */ |
951 | jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0, |
952 | charge+jnrC+0,charge+jnrD+0); |
953 | vdwjidx0A = 2*vdwtype[jnrA+0]; |
954 | vdwjidx0B = 2*vdwtype[jnrB+0]; |
955 | vdwjidx0C = 2*vdwtype[jnrC+0]; |
956 | vdwjidx0D = 2*vdwtype[jnrD+0]; |
957 | |
958 | fjx0 = _mm_setzero_ps(); |
959 | fjy0 = _mm_setzero_ps(); |
960 | fjz0 = _mm_setzero_ps(); |
961 | |
962 | /************************** |
963 | * CALCULATE INTERACTIONS * |
964 | **************************/ |
965 | |
966 | r00 = _mm_mul_ps(rsq00,rinv00); |
967 | |
968 | /* Compute parameters for interactions between i and j atoms */ |
969 | gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A, |
970 | vdwparam+vdwioffset0+vdwjidx0B, |
971 | vdwparam+vdwioffset0+vdwjidx0C, |
972 | vdwparam+vdwioffset0+vdwjidx0D, |
973 | &c6_00,&c12_00); |
974 | |
975 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
976 | rt = _mm_mul_ps(r00,vftabscale); |
977 | vfitab = _mm_cvttps_epi32(rt); |
978 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
979 | vfitab = _mm_slli_epi32(vfitab,3); |
980 | |
981 | /* CUBIC SPLINE TABLE DISPERSION */ |
982 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
983 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
984 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
985 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
986 | _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); |
987 | Heps = _mm_mul_ps(vfeps,H); |
988 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
989 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
990 | fvdw6 = _mm_mul_ps(c6_00,FF); |
991 | |
992 | /* CUBIC SPLINE TABLE REPULSION */ |
993 | vfitab = _mm_add_epi32(vfitab,ifour); |
994 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
995 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
996 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
997 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
998 | _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); |
999 | Heps = _mm_mul_ps(vfeps,H); |
1000 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1001 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1002 | fvdw12 = _mm_mul_ps(c12_00,FF); |
1003 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
1004 | |
1005 | fscal = fvdw; |
1006 | |
1007 | /* Calculate temporary vectorial force */ |
1008 | tx = _mm_mul_ps(fscal,dx00); |
1009 | ty = _mm_mul_ps(fscal,dy00); |
1010 | tz = _mm_mul_ps(fscal,dz00); |
1011 | |
1012 | /* Update vectorial force */ |
1013 | fix0 = _mm_add_ps(fix0,tx); |
1014 | fiy0 = _mm_add_ps(fiy0,ty); |
1015 | fiz0 = _mm_add_ps(fiz0,tz); |
1016 | |
1017 | fjx0 = _mm_add_ps(fjx0,tx); |
1018 | fjy0 = _mm_add_ps(fjy0,ty); |
1019 | fjz0 = _mm_add_ps(fjz0,tz); |
1020 | |
1021 | /************************** |
1022 | * CALCULATE INTERACTIONS * |
1023 | **************************/ |
1024 | |
1025 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
1026 | { |
1027 | |
1028 | /* Compute parameters for interactions between i and j atoms */ |
1029 | qq10 = _mm_mul_ps(iq1,jq0); |
1030 | |
1031 | /* REACTION-FIELD ELECTROSTATICS */ |
1032 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
1033 | |
1034 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
1035 | |
1036 | fscal = felec; |
1037 | |
1038 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1039 | |
1040 | /* Calculate temporary vectorial force */ |
1041 | tx = _mm_mul_ps(fscal,dx10); |
1042 | ty = _mm_mul_ps(fscal,dy10); |
1043 | tz = _mm_mul_ps(fscal,dz10); |
1044 | |
1045 | /* Update vectorial force */ |
1046 | fix1 = _mm_add_ps(fix1,tx); |
1047 | fiy1 = _mm_add_ps(fiy1,ty); |
1048 | fiz1 = _mm_add_ps(fiz1,tz); |
1049 | |
1050 | fjx0 = _mm_add_ps(fjx0,tx); |
1051 | fjy0 = _mm_add_ps(fjy0,ty); |
1052 | fjz0 = _mm_add_ps(fjz0,tz); |
1053 | |
1054 | } |
1055 | |
1056 | /************************** |
1057 | * CALCULATE INTERACTIONS * |
1058 | **************************/ |
1059 | |
1060 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
1061 | { |
1062 | |
1063 | /* Compute parameters for interactions between i and j atoms */ |
1064 | qq20 = _mm_mul_ps(iq2,jq0); |
1065 | |
1066 | /* REACTION-FIELD ELECTROSTATICS */ |
1067 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
1068 | |
1069 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
1070 | |
1071 | fscal = felec; |
1072 | |
1073 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1074 | |
1075 | /* Calculate temporary vectorial force */ |
1076 | tx = _mm_mul_ps(fscal,dx20); |
1077 | ty = _mm_mul_ps(fscal,dy20); |
1078 | tz = _mm_mul_ps(fscal,dz20); |
1079 | |
1080 | /* Update vectorial force */ |
1081 | fix2 = _mm_add_ps(fix2,tx); |
1082 | fiy2 = _mm_add_ps(fiy2,ty); |
1083 | fiz2 = _mm_add_ps(fiz2,tz); |
1084 | |
1085 | fjx0 = _mm_add_ps(fjx0,tx); |
1086 | fjy0 = _mm_add_ps(fjy0,ty); |
1087 | fjz0 = _mm_add_ps(fjz0,tz); |
1088 | |
1089 | } |
1090 | |
1091 | /************************** |
1092 | * CALCULATE INTERACTIONS * |
1093 | **************************/ |
1094 | |
1095 | if (gmx_mm_any_lt(rsq30,rcutoff2)) |
1096 | { |
1097 | |
1098 | /* Compute parameters for interactions between i and j atoms */ |
1099 | qq30 = _mm_mul_ps(iq3,jq0); |
1100 | |
1101 | /* REACTION-FIELD ELECTROSTATICS */ |
1102 | felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2)); |
1103 | |
1104 | cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2); |
1105 | |
1106 | fscal = felec; |
1107 | |
1108 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1109 | |
1110 | /* Calculate temporary vectorial force */ |
1111 | tx = _mm_mul_ps(fscal,dx30); |
1112 | ty = _mm_mul_ps(fscal,dy30); |
1113 | tz = _mm_mul_ps(fscal,dz30); |
1114 | |
1115 | /* Update vectorial force */ |
1116 | fix3 = _mm_add_ps(fix3,tx); |
1117 | fiy3 = _mm_add_ps(fiy3,ty); |
1118 | fiz3 = _mm_add_ps(fiz3,tz); |
1119 | |
1120 | fjx0 = _mm_add_ps(fjx0,tx); |
1121 | fjy0 = _mm_add_ps(fjy0,ty); |
1122 | fjz0 = _mm_add_ps(fjz0,tz); |
1123 | |
1124 | } |
1125 | |
1126 | fjptrA = f+j_coord_offsetA; |
1127 | fjptrB = f+j_coord_offsetB; |
1128 | fjptrC = f+j_coord_offsetC; |
1129 | fjptrD = f+j_coord_offsetD; |
1130 | |
1131 | gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0); |
1132 | |
1133 | /* Inner loop uses 138 flops */ |
1134 | } |
1135 | |
1136 | if(jidx<j_index_end) |
1137 | { |
1138 | |
1139 | /* Get j neighbor index, and coordinate index */ |
1140 | jnrlistA = jjnr[jidx]; |
1141 | jnrlistB = jjnr[jidx+1]; |
1142 | jnrlistC = jjnr[jidx+2]; |
1143 | jnrlistD = jjnr[jidx+3]; |
1144 | /* Sign of each element will be negative for non-real atoms. |
1145 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
1146 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
1147 | */ |
1148 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
1149 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
1150 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
1151 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
1152 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
1153 | j_coord_offsetA = DIM3*jnrA; |
1154 | j_coord_offsetB = DIM3*jnrB; |
1155 | j_coord_offsetC = DIM3*jnrC; |
1156 | j_coord_offsetD = DIM3*jnrD; |
1157 | |
1158 | /* load j atom coordinates */ |
1159 | gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
1160 | x+j_coord_offsetC,x+j_coord_offsetD, |
1161 | &jx0,&jy0,&jz0); |
1162 | |
1163 | /* Calculate displacement vector */ |
1164 | dx00 = _mm_sub_ps(ix0,jx0); |
1165 | dy00 = _mm_sub_ps(iy0,jy0); |
1166 | dz00 = _mm_sub_ps(iz0,jz0); |
1167 | dx10 = _mm_sub_ps(ix1,jx0); |
1168 | dy10 = _mm_sub_ps(iy1,jy0); |
1169 | dz10 = _mm_sub_ps(iz1,jz0); |
1170 | dx20 = _mm_sub_ps(ix2,jx0); |
1171 | dy20 = _mm_sub_ps(iy2,jy0); |
1172 | dz20 = _mm_sub_ps(iz2,jz0); |
1173 | dx30 = _mm_sub_ps(ix3,jx0); |
1174 | dy30 = _mm_sub_ps(iy3,jy0); |
1175 | dz30 = _mm_sub_ps(iz3,jz0); |
1176 | |
1177 | /* Calculate squared distance and things based on it */ |
1178 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
1179 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
1180 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
1181 | rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30); |
1182 | |
1183 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
1184 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
1185 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
1186 | rinv30 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq30); |
1187 | |
1188 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
1189 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
1190 | rinvsq30 = _mm_mul_ps(rinv30,rinv30); |
1191 | |
1192 | /* Load parameters for j particles */ |
1193 | jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0, |
1194 | charge+jnrC+0,charge+jnrD+0); |
1195 | vdwjidx0A = 2*vdwtype[jnrA+0]; |
1196 | vdwjidx0B = 2*vdwtype[jnrB+0]; |
1197 | vdwjidx0C = 2*vdwtype[jnrC+0]; |
1198 | vdwjidx0D = 2*vdwtype[jnrD+0]; |
1199 | |
1200 | fjx0 = _mm_setzero_ps(); |
1201 | fjy0 = _mm_setzero_ps(); |
1202 | fjz0 = _mm_setzero_ps(); |
1203 | |
1204 | /************************** |
1205 | * CALCULATE INTERACTIONS * |
1206 | **************************/ |
1207 | |
1208 | r00 = _mm_mul_ps(rsq00,rinv00); |
1209 | r00 = _mm_andnot_ps(dummy_mask,r00); |
1210 | |
1211 | /* Compute parameters for interactions between i and j atoms */ |
1212 | gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A, |
1213 | vdwparam+vdwioffset0+vdwjidx0B, |
1214 | vdwparam+vdwioffset0+vdwjidx0C, |
1215 | vdwparam+vdwioffset0+vdwjidx0D, |
1216 | &c6_00,&c12_00); |
1217 | |
1218 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
1219 | rt = _mm_mul_ps(r00,vftabscale); |
1220 | vfitab = _mm_cvttps_epi32(rt); |
1221 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
1222 | vfitab = _mm_slli_epi32(vfitab,3); |
1223 | |
1224 | /* CUBIC SPLINE TABLE DISPERSION */ |
1225 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
1226 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
1227 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
1228 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
1229 | _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); |
1230 | Heps = _mm_mul_ps(vfeps,H); |
1231 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1232 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1233 | fvdw6 = _mm_mul_ps(c6_00,FF); |
1234 | |
1235 | /* CUBIC SPLINE TABLE REPULSION */ |
1236 | vfitab = _mm_add_epi32(vfitab,ifour); |
1237 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
1238 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
1239 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
1240 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
1241 | _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); |
1242 | Heps = _mm_mul_ps(vfeps,H); |
1243 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1244 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1245 | fvdw12 = _mm_mul_ps(c12_00,FF); |
1246 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
1247 | |
1248 | fscal = fvdw; |
1249 | |
1250 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1251 | |
1252 | /* Calculate temporary vectorial force */ |
1253 | tx = _mm_mul_ps(fscal,dx00); |
1254 | ty = _mm_mul_ps(fscal,dy00); |
1255 | tz = _mm_mul_ps(fscal,dz00); |
1256 | |
1257 | /* Update vectorial force */ |
1258 | fix0 = _mm_add_ps(fix0,tx); |
1259 | fiy0 = _mm_add_ps(fiy0,ty); |
1260 | fiz0 = _mm_add_ps(fiz0,tz); |
1261 | |
1262 | fjx0 = _mm_add_ps(fjx0,tx); |
1263 | fjy0 = _mm_add_ps(fjy0,ty); |
1264 | fjz0 = _mm_add_ps(fjz0,tz); |
1265 | |
1266 | /************************** |
1267 | * CALCULATE INTERACTIONS * |
1268 | **************************/ |
1269 | |
1270 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
1271 | { |
1272 | |
1273 | /* Compute parameters for interactions between i and j atoms */ |
1274 | qq10 = _mm_mul_ps(iq1,jq0); |
1275 | |
1276 | /* REACTION-FIELD ELECTROSTATICS */ |
1277 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
1278 | |
1279 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
1280 | |
1281 | fscal = felec; |
1282 | |
1283 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1284 | |
1285 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1286 | |
1287 | /* Calculate temporary vectorial force */ |
1288 | tx = _mm_mul_ps(fscal,dx10); |
1289 | ty = _mm_mul_ps(fscal,dy10); |
1290 | tz = _mm_mul_ps(fscal,dz10); |
1291 | |
1292 | /* Update vectorial force */ |
1293 | fix1 = _mm_add_ps(fix1,tx); |
1294 | fiy1 = _mm_add_ps(fiy1,ty); |
1295 | fiz1 = _mm_add_ps(fiz1,tz); |
1296 | |
1297 | fjx0 = _mm_add_ps(fjx0,tx); |
1298 | fjy0 = _mm_add_ps(fjy0,ty); |
1299 | fjz0 = _mm_add_ps(fjz0,tz); |
1300 | |
1301 | } |
1302 | |
1303 | /************************** |
1304 | * CALCULATE INTERACTIONS * |
1305 | **************************/ |
1306 | |
1307 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
1308 | { |
1309 | |
1310 | /* Compute parameters for interactions between i and j atoms */ |
1311 | qq20 = _mm_mul_ps(iq2,jq0); |
1312 | |
1313 | /* REACTION-FIELD ELECTROSTATICS */ |
1314 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
1315 | |
1316 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
1317 | |
1318 | fscal = felec; |
1319 | |
1320 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1321 | |
1322 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1323 | |
1324 | /* Calculate temporary vectorial force */ |
1325 | tx = _mm_mul_ps(fscal,dx20); |
1326 | ty = _mm_mul_ps(fscal,dy20); |
1327 | tz = _mm_mul_ps(fscal,dz20); |
1328 | |
1329 | /* Update vectorial force */ |
1330 | fix2 = _mm_add_ps(fix2,tx); |
1331 | fiy2 = _mm_add_ps(fiy2,ty); |
1332 | fiz2 = _mm_add_ps(fiz2,tz); |
1333 | |
1334 | fjx0 = _mm_add_ps(fjx0,tx); |
1335 | fjy0 = _mm_add_ps(fjy0,ty); |
1336 | fjz0 = _mm_add_ps(fjz0,tz); |
1337 | |
1338 | } |
1339 | |
1340 | /************************** |
1341 | * CALCULATE INTERACTIONS * |
1342 | **************************/ |
1343 | |
1344 | if (gmx_mm_any_lt(rsq30,rcutoff2)) |
1345 | { |
1346 | |
1347 | /* Compute parameters for interactions between i and j atoms */ |
1348 | qq30 = _mm_mul_ps(iq3,jq0); |
1349 | |
1350 | /* REACTION-FIELD ELECTROSTATICS */ |
1351 | felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2)); |
1352 | |
1353 | cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2); |
1354 | |
1355 | fscal = felec; |
1356 | |
1357 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1358 | |
1359 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1360 | |
1361 | /* Calculate temporary vectorial force */ |
1362 | tx = _mm_mul_ps(fscal,dx30); |
1363 | ty = _mm_mul_ps(fscal,dy30); |
1364 | tz = _mm_mul_ps(fscal,dz30); |
1365 | |
1366 | /* Update vectorial force */ |
1367 | fix3 = _mm_add_ps(fix3,tx); |
1368 | fiy3 = _mm_add_ps(fiy3,ty); |
1369 | fiz3 = _mm_add_ps(fiz3,tz); |
1370 | |
1371 | fjx0 = _mm_add_ps(fjx0,tx); |
1372 | fjy0 = _mm_add_ps(fjy0,ty); |
1373 | fjz0 = _mm_add_ps(fjz0,tz); |
1374 | |
1375 | } |
1376 | |
1377 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
1378 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
1379 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
1380 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
1381 | |
1382 | gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0); |
1383 | |
1384 | /* Inner loop uses 139 flops */ |
1385 | } |
1386 | |
1387 | /* End of innermost loop */ |
1388 | |
1389 | gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3, |
1390 | f+i_coord_offset,fshift+i_shift_offset); |
1391 | |
1392 | /* Increment number of inner iterations */ |
1393 | inneriter += j_index_end - j_index_start; |
1394 | |
1395 | /* Outer loop uses 24 flops */ |
1396 | } |
1397 | |
1398 | /* Increment number of outer iterations */ |
1399 | outeriter += nri; |
1400 | |
1401 | /* Update outer/inner flops */ |
1402 | |
1403 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*139)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W4_F] += outeriter*24 + inneriter *139; |
1404 | } |