File: | gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_sse4_1_single.c |
Location: | line 824, column 5 |
Description: | Value stored to 'gid' 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_VdwLJSw_GeomW4P1_VF_sse4_1_single |
54 | * Electrostatics interaction: ReactionField |
55 | * VdW interaction: LennardJones |
56 | * Geometry: Water4-Particle |
57 | * Calculate force/pot: PotentialAndForce |
58 | */ |
59 | void |
60 | nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_sse4_1_single |
61 | (t_nblist * gmx_restrict nlist, |
62 | rvec * gmx_restrict xx, |
63 | rvec * gmx_restrict ff, |
64 | t_forcerec * gmx_restrict fr, |
65 | t_mdatoms * gmx_restrict mdatoms, |
66 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
67 | t_nrnb * gmx_restrict nrnb) |
68 | { |
69 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
70 | * just 0 for non-waters. |
71 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
72 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
73 | */ |
74 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
75 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
76 | int jnrA,jnrB,jnrC,jnrD; |
77 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
78 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
79 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
80 | real rcutoff_scalar; |
81 | real *shiftvec,*fshift,*x,*f; |
82 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
83 | real scratch[4*DIM3]; |
84 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
85 | int vdwioffset0; |
86 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
87 | int vdwioffset1; |
88 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
89 | int vdwioffset2; |
90 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
91 | int vdwioffset3; |
92 | __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3; |
93 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
94 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
95 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
96 | __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10; |
97 | __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20; |
98 | __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30; |
99 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
100 | real *charge; |
101 | int nvdwtype; |
102 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
103 | int *vdwtype; |
104 | real *vdwparam; |
105 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
106 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
107 | __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw; |
108 | real rswitch_scalar,d_scalar; |
109 | __m128 dummy_mask,cutoff_mask; |
110 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
111 | __m128 one = _mm_set1_ps(1.0); |
112 | __m128 two = _mm_set1_ps(2.0); |
113 | x = xx[0]; |
114 | f = ff[0]; |
115 | |
116 | nri = nlist->nri; |
117 | iinr = nlist->iinr; |
118 | jindex = nlist->jindex; |
119 | jjnr = nlist->jjnr; |
120 | shiftidx = nlist->shift; |
121 | gid = nlist->gid; |
122 | shiftvec = fr->shift_vec[0]; |
123 | fshift = fr->fshift[0]; |
124 | facel = _mm_set1_ps(fr->epsfac); |
125 | charge = mdatoms->chargeA; |
126 | krf = _mm_set1_ps(fr->ic->k_rf); |
127 | krf2 = _mm_set1_ps(fr->ic->k_rf*2.0); |
128 | crf = _mm_set1_ps(fr->ic->c_rf); |
129 | nvdwtype = fr->ntype; |
130 | vdwparam = fr->nbfp; |
131 | vdwtype = mdatoms->typeA; |
132 | |
133 | /* Setup water-specific parameters */ |
134 | inr = nlist->iinr[0]; |
135 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
136 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
137 | iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3])); |
138 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
139 | |
140 | /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */ |
141 | rcutoff_scalar = fr->rcoulomb; |
142 | rcutoff = _mm_set1_ps(rcutoff_scalar); |
143 | rcutoff2 = _mm_mul_ps(rcutoff,rcutoff); |
144 | |
145 | rswitch_scalar = fr->rvdw_switch; |
146 | rswitch = _mm_set1_ps(rswitch_scalar); |
147 | /* Setup switch parameters */ |
148 | d_scalar = rcutoff_scalar-rswitch_scalar; |
149 | d = _mm_set1_ps(d_scalar); |
150 | swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar)); |
151 | swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar)); |
152 | swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar)); |
153 | swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar)); |
154 | swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar)); |
155 | swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar)); |
156 | |
157 | /* Avoid stupid compiler warnings */ |
158 | jnrA = jnrB = jnrC = jnrD = 0; |
159 | j_coord_offsetA = 0; |
160 | j_coord_offsetB = 0; |
161 | j_coord_offsetC = 0; |
162 | j_coord_offsetD = 0; |
163 | |
164 | outeriter = 0; |
165 | inneriter = 0; |
166 | |
167 | for(iidx=0;iidx<4*DIM3;iidx++) |
168 | { |
169 | scratch[iidx] = 0.0; |
170 | } |
171 | |
172 | /* Start outer loop over neighborlists */ |
173 | for(iidx=0; iidx<nri; iidx++) |
174 | { |
175 | /* Load shift vector for this list */ |
176 | i_shift_offset = DIM3*shiftidx[iidx]; |
177 | |
178 | /* Load limits for loop over neighbors */ |
179 | j_index_start = jindex[iidx]; |
180 | j_index_end = jindex[iidx+1]; |
181 | |
182 | /* Get outer coordinate index */ |
183 | inr = iinr[iidx]; |
184 | i_coord_offset = DIM3*inr; |
185 | |
186 | /* Load i particle coords and add shift vector */ |
187 | gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
188 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3); |
189 | |
190 | fix0 = _mm_setzero_ps(); |
191 | fiy0 = _mm_setzero_ps(); |
192 | fiz0 = _mm_setzero_ps(); |
193 | fix1 = _mm_setzero_ps(); |
194 | fiy1 = _mm_setzero_ps(); |
195 | fiz1 = _mm_setzero_ps(); |
196 | fix2 = _mm_setzero_ps(); |
197 | fiy2 = _mm_setzero_ps(); |
198 | fiz2 = _mm_setzero_ps(); |
199 | fix3 = _mm_setzero_ps(); |
200 | fiy3 = _mm_setzero_ps(); |
201 | fiz3 = _mm_setzero_ps(); |
202 | |
203 | /* Reset potential sums */ |
204 | velecsum = _mm_setzero_ps(); |
205 | vvdwsum = _mm_setzero_ps(); |
206 | |
207 | /* Start inner kernel loop */ |
208 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
209 | { |
210 | |
211 | /* Get j neighbor index, and coordinate index */ |
212 | jnrA = jjnr[jidx]; |
213 | jnrB = jjnr[jidx+1]; |
214 | jnrC = jjnr[jidx+2]; |
215 | jnrD = jjnr[jidx+3]; |
216 | j_coord_offsetA = DIM3*jnrA; |
217 | j_coord_offsetB = DIM3*jnrB; |
218 | j_coord_offsetC = DIM3*jnrC; |
219 | j_coord_offsetD = DIM3*jnrD; |
220 | |
221 | /* load j atom coordinates */ |
222 | gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
223 | x+j_coord_offsetC,x+j_coord_offsetD, |
224 | &jx0,&jy0,&jz0); |
225 | |
226 | /* Calculate displacement vector */ |
227 | dx00 = _mm_sub_ps(ix0,jx0); |
228 | dy00 = _mm_sub_ps(iy0,jy0); |
229 | dz00 = _mm_sub_ps(iz0,jz0); |
230 | dx10 = _mm_sub_ps(ix1,jx0); |
231 | dy10 = _mm_sub_ps(iy1,jy0); |
232 | dz10 = _mm_sub_ps(iz1,jz0); |
233 | dx20 = _mm_sub_ps(ix2,jx0); |
234 | dy20 = _mm_sub_ps(iy2,jy0); |
235 | dz20 = _mm_sub_ps(iz2,jz0); |
236 | dx30 = _mm_sub_ps(ix3,jx0); |
237 | dy30 = _mm_sub_ps(iy3,jy0); |
238 | dz30 = _mm_sub_ps(iz3,jz0); |
239 | |
240 | /* Calculate squared distance and things based on it */ |
241 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
242 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
243 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
244 | rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30); |
245 | |
246 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
247 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
248 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
249 | rinv30 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq30); |
250 | |
251 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
252 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
253 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
254 | rinvsq30 = _mm_mul_ps(rinv30,rinv30); |
255 | |
256 | /* Load parameters for j particles */ |
257 | jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0, |
258 | charge+jnrC+0,charge+jnrD+0); |
259 | vdwjidx0A = 2*vdwtype[jnrA+0]; |
260 | vdwjidx0B = 2*vdwtype[jnrB+0]; |
261 | vdwjidx0C = 2*vdwtype[jnrC+0]; |
262 | vdwjidx0D = 2*vdwtype[jnrD+0]; |
263 | |
264 | fjx0 = _mm_setzero_ps(); |
265 | fjy0 = _mm_setzero_ps(); |
266 | fjz0 = _mm_setzero_ps(); |
267 | |
268 | /************************** |
269 | * CALCULATE INTERACTIONS * |
270 | **************************/ |
271 | |
272 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
273 | { |
274 | |
275 | r00 = _mm_mul_ps(rsq00,rinv00); |
276 | |
277 | /* Compute parameters for interactions between i and j atoms */ |
278 | gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A, |
279 | vdwparam+vdwioffset0+vdwjidx0B, |
280 | vdwparam+vdwioffset0+vdwjidx0C, |
281 | vdwparam+vdwioffset0+vdwjidx0D, |
282 | &c6_00,&c12_00); |
283 | |
284 | /* LENNARD-JONES DISPERSION/REPULSION */ |
285 | |
286 | rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00); |
287 | vvdw6 = _mm_mul_ps(c6_00,rinvsix); |
288 | vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix)); |
289 | vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) ); |
290 | fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00); |
291 | |
292 | d = _mm_sub_ps(r00,rswitch); |
293 | d = _mm_max_ps(d,_mm_setzero_ps()); |
294 | d2 = _mm_mul_ps(d,d); |
295 | sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5))))))); |
296 | |
297 | dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4))))); |
298 | |
299 | /* Evaluate switch function */ |
300 | /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */ |
301 | fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) ); |
302 | vvdw = _mm_mul_ps(vvdw,sw); |
303 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
304 | |
305 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
306 | vvdw = _mm_and_ps(vvdw,cutoff_mask); |
307 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
308 | |
309 | fscal = fvdw; |
310 | |
311 | fscal = _mm_and_ps(fscal,cutoff_mask); |
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 | |
329 | /************************** |
330 | * CALCULATE INTERACTIONS * |
331 | **************************/ |
332 | |
333 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
334 | { |
335 | |
336 | /* Compute parameters for interactions between i and j atoms */ |
337 | qq10 = _mm_mul_ps(iq1,jq0); |
338 | |
339 | /* REACTION-FIELD ELECTROSTATICS */ |
340 | velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf)); |
341 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
342 | |
343 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
344 | |
345 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
346 | velec = _mm_and_ps(velec,cutoff_mask); |
347 | velecsum = _mm_add_ps(velecsum,velec); |
348 | |
349 | fscal = felec; |
350 | |
351 | fscal = _mm_and_ps(fscal,cutoff_mask); |
352 | |
353 | /* Calculate temporary vectorial force */ |
354 | tx = _mm_mul_ps(fscal,dx10); |
355 | ty = _mm_mul_ps(fscal,dy10); |
356 | tz = _mm_mul_ps(fscal,dz10); |
357 | |
358 | /* Update vectorial force */ |
359 | fix1 = _mm_add_ps(fix1,tx); |
360 | fiy1 = _mm_add_ps(fiy1,ty); |
361 | fiz1 = _mm_add_ps(fiz1,tz); |
362 | |
363 | fjx0 = _mm_add_ps(fjx0,tx); |
364 | fjy0 = _mm_add_ps(fjy0,ty); |
365 | fjz0 = _mm_add_ps(fjz0,tz); |
366 | |
367 | } |
368 | |
369 | /************************** |
370 | * CALCULATE INTERACTIONS * |
371 | **************************/ |
372 | |
373 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
374 | { |
375 | |
376 | /* Compute parameters for interactions between i and j atoms */ |
377 | qq20 = _mm_mul_ps(iq2,jq0); |
378 | |
379 | /* REACTION-FIELD ELECTROSTATICS */ |
380 | velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf)); |
381 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
382 | |
383 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
384 | |
385 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
386 | velec = _mm_and_ps(velec,cutoff_mask); |
387 | velecsum = _mm_add_ps(velecsum,velec); |
388 | |
389 | fscal = felec; |
390 | |
391 | fscal = _mm_and_ps(fscal,cutoff_mask); |
392 | |
393 | /* Calculate temporary vectorial force */ |
394 | tx = _mm_mul_ps(fscal,dx20); |
395 | ty = _mm_mul_ps(fscal,dy20); |
396 | tz = _mm_mul_ps(fscal,dz20); |
397 | |
398 | /* Update vectorial force */ |
399 | fix2 = _mm_add_ps(fix2,tx); |
400 | fiy2 = _mm_add_ps(fiy2,ty); |
401 | fiz2 = _mm_add_ps(fiz2,tz); |
402 | |
403 | fjx0 = _mm_add_ps(fjx0,tx); |
404 | fjy0 = _mm_add_ps(fjy0,ty); |
405 | fjz0 = _mm_add_ps(fjz0,tz); |
406 | |
407 | } |
408 | |
409 | /************************** |
410 | * CALCULATE INTERACTIONS * |
411 | **************************/ |
412 | |
413 | if (gmx_mm_any_lt(rsq30,rcutoff2)) |
414 | { |
415 | |
416 | /* Compute parameters for interactions between i and j atoms */ |
417 | qq30 = _mm_mul_ps(iq3,jq0); |
418 | |
419 | /* REACTION-FIELD ELECTROSTATICS */ |
420 | velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf)); |
421 | felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2)); |
422 | |
423 | cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2); |
424 | |
425 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
426 | velec = _mm_and_ps(velec,cutoff_mask); |
427 | velecsum = _mm_add_ps(velecsum,velec); |
428 | |
429 | fscal = felec; |
430 | |
431 | fscal = _mm_and_ps(fscal,cutoff_mask); |
432 | |
433 | /* Calculate temporary vectorial force */ |
434 | tx = _mm_mul_ps(fscal,dx30); |
435 | ty = _mm_mul_ps(fscal,dy30); |
436 | tz = _mm_mul_ps(fscal,dz30); |
437 | |
438 | /* Update vectorial force */ |
439 | fix3 = _mm_add_ps(fix3,tx); |
440 | fiy3 = _mm_add_ps(fiy3,ty); |
441 | fiz3 = _mm_add_ps(fiz3,tz); |
442 | |
443 | fjx0 = _mm_add_ps(fjx0,tx); |
444 | fjy0 = _mm_add_ps(fjy0,ty); |
445 | fjz0 = _mm_add_ps(fjz0,tz); |
446 | |
447 | } |
448 | |
449 | fjptrA = f+j_coord_offsetA; |
450 | fjptrB = f+j_coord_offsetB; |
451 | fjptrC = f+j_coord_offsetC; |
452 | fjptrD = f+j_coord_offsetD; |
453 | |
454 | gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0); |
455 | |
456 | /* Inner loop uses 167 flops */ |
457 | } |
458 | |
459 | if(jidx<j_index_end) |
460 | { |
461 | |
462 | /* Get j neighbor index, and coordinate index */ |
463 | jnrlistA = jjnr[jidx]; |
464 | jnrlistB = jjnr[jidx+1]; |
465 | jnrlistC = jjnr[jidx+2]; |
466 | jnrlistD = jjnr[jidx+3]; |
467 | /* Sign of each element will be negative for non-real atoms. |
468 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
469 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
470 | */ |
471 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
472 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
473 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
474 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
475 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
476 | j_coord_offsetA = DIM3*jnrA; |
477 | j_coord_offsetB = DIM3*jnrB; |
478 | j_coord_offsetC = DIM3*jnrC; |
479 | j_coord_offsetD = DIM3*jnrD; |
480 | |
481 | /* load j atom coordinates */ |
482 | gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
483 | x+j_coord_offsetC,x+j_coord_offsetD, |
484 | &jx0,&jy0,&jz0); |
485 | |
486 | /* Calculate displacement vector */ |
487 | dx00 = _mm_sub_ps(ix0,jx0); |
488 | dy00 = _mm_sub_ps(iy0,jy0); |
489 | dz00 = _mm_sub_ps(iz0,jz0); |
490 | dx10 = _mm_sub_ps(ix1,jx0); |
491 | dy10 = _mm_sub_ps(iy1,jy0); |
492 | dz10 = _mm_sub_ps(iz1,jz0); |
493 | dx20 = _mm_sub_ps(ix2,jx0); |
494 | dy20 = _mm_sub_ps(iy2,jy0); |
495 | dz20 = _mm_sub_ps(iz2,jz0); |
496 | dx30 = _mm_sub_ps(ix3,jx0); |
497 | dy30 = _mm_sub_ps(iy3,jy0); |
498 | dz30 = _mm_sub_ps(iz3,jz0); |
499 | |
500 | /* Calculate squared distance and things based on it */ |
501 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
502 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
503 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
504 | rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30); |
505 | |
506 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
507 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
508 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
509 | rinv30 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq30); |
510 | |
511 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
512 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
513 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
514 | rinvsq30 = _mm_mul_ps(rinv30,rinv30); |
515 | |
516 | /* Load parameters for j particles */ |
517 | jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0, |
518 | charge+jnrC+0,charge+jnrD+0); |
519 | vdwjidx0A = 2*vdwtype[jnrA+0]; |
520 | vdwjidx0B = 2*vdwtype[jnrB+0]; |
521 | vdwjidx0C = 2*vdwtype[jnrC+0]; |
522 | vdwjidx0D = 2*vdwtype[jnrD+0]; |
523 | |
524 | fjx0 = _mm_setzero_ps(); |
525 | fjy0 = _mm_setzero_ps(); |
526 | fjz0 = _mm_setzero_ps(); |
527 | |
528 | /************************** |
529 | * CALCULATE INTERACTIONS * |
530 | **************************/ |
531 | |
532 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
533 | { |
534 | |
535 | r00 = _mm_mul_ps(rsq00,rinv00); |
536 | r00 = _mm_andnot_ps(dummy_mask,r00); |
537 | |
538 | /* Compute parameters for interactions between i and j atoms */ |
539 | gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A, |
540 | vdwparam+vdwioffset0+vdwjidx0B, |
541 | vdwparam+vdwioffset0+vdwjidx0C, |
542 | vdwparam+vdwioffset0+vdwjidx0D, |
543 | &c6_00,&c12_00); |
544 | |
545 | /* LENNARD-JONES DISPERSION/REPULSION */ |
546 | |
547 | rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00); |
548 | vvdw6 = _mm_mul_ps(c6_00,rinvsix); |
549 | vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix)); |
550 | vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) ); |
551 | fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00); |
552 | |
553 | d = _mm_sub_ps(r00,rswitch); |
554 | d = _mm_max_ps(d,_mm_setzero_ps()); |
555 | d2 = _mm_mul_ps(d,d); |
556 | sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5))))))); |
557 | |
558 | dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4))))); |
559 | |
560 | /* Evaluate switch function */ |
561 | /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */ |
562 | fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) ); |
563 | vvdw = _mm_mul_ps(vvdw,sw); |
564 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
565 | |
566 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
567 | vvdw = _mm_and_ps(vvdw,cutoff_mask); |
568 | vvdw = _mm_andnot_ps(dummy_mask,vvdw); |
569 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
570 | |
571 | fscal = fvdw; |
572 | |
573 | fscal = _mm_and_ps(fscal,cutoff_mask); |
574 | |
575 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
576 | |
577 | /* Calculate temporary vectorial force */ |
578 | tx = _mm_mul_ps(fscal,dx00); |
579 | ty = _mm_mul_ps(fscal,dy00); |
580 | tz = _mm_mul_ps(fscal,dz00); |
581 | |
582 | /* Update vectorial force */ |
583 | fix0 = _mm_add_ps(fix0,tx); |
584 | fiy0 = _mm_add_ps(fiy0,ty); |
585 | fiz0 = _mm_add_ps(fiz0,tz); |
586 | |
587 | fjx0 = _mm_add_ps(fjx0,tx); |
588 | fjy0 = _mm_add_ps(fjy0,ty); |
589 | fjz0 = _mm_add_ps(fjz0,tz); |
590 | |
591 | } |
592 | |
593 | /************************** |
594 | * CALCULATE INTERACTIONS * |
595 | **************************/ |
596 | |
597 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
598 | { |
599 | |
600 | /* Compute parameters for interactions between i and j atoms */ |
601 | qq10 = _mm_mul_ps(iq1,jq0); |
602 | |
603 | /* REACTION-FIELD ELECTROSTATICS */ |
604 | velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf)); |
605 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
606 | |
607 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
608 | |
609 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
610 | velec = _mm_and_ps(velec,cutoff_mask); |
611 | velec = _mm_andnot_ps(dummy_mask,velec); |
612 | velecsum = _mm_add_ps(velecsum,velec); |
613 | |
614 | fscal = felec; |
615 | |
616 | fscal = _mm_and_ps(fscal,cutoff_mask); |
617 | |
618 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
619 | |
620 | /* Calculate temporary vectorial force */ |
621 | tx = _mm_mul_ps(fscal,dx10); |
622 | ty = _mm_mul_ps(fscal,dy10); |
623 | tz = _mm_mul_ps(fscal,dz10); |
624 | |
625 | /* Update vectorial force */ |
626 | fix1 = _mm_add_ps(fix1,tx); |
627 | fiy1 = _mm_add_ps(fiy1,ty); |
628 | fiz1 = _mm_add_ps(fiz1,tz); |
629 | |
630 | fjx0 = _mm_add_ps(fjx0,tx); |
631 | fjy0 = _mm_add_ps(fjy0,ty); |
632 | fjz0 = _mm_add_ps(fjz0,tz); |
633 | |
634 | } |
635 | |
636 | /************************** |
637 | * CALCULATE INTERACTIONS * |
638 | **************************/ |
639 | |
640 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
641 | { |
642 | |
643 | /* Compute parameters for interactions between i and j atoms */ |
644 | qq20 = _mm_mul_ps(iq2,jq0); |
645 | |
646 | /* REACTION-FIELD ELECTROSTATICS */ |
647 | velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf)); |
648 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
649 | |
650 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
651 | |
652 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
653 | velec = _mm_and_ps(velec,cutoff_mask); |
654 | velec = _mm_andnot_ps(dummy_mask,velec); |
655 | velecsum = _mm_add_ps(velecsum,velec); |
656 | |
657 | fscal = felec; |
658 | |
659 | fscal = _mm_and_ps(fscal,cutoff_mask); |
660 | |
661 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
662 | |
663 | /* Calculate temporary vectorial force */ |
664 | tx = _mm_mul_ps(fscal,dx20); |
665 | ty = _mm_mul_ps(fscal,dy20); |
666 | tz = _mm_mul_ps(fscal,dz20); |
667 | |
668 | /* Update vectorial force */ |
669 | fix2 = _mm_add_ps(fix2,tx); |
670 | fiy2 = _mm_add_ps(fiy2,ty); |
671 | fiz2 = _mm_add_ps(fiz2,tz); |
672 | |
673 | fjx0 = _mm_add_ps(fjx0,tx); |
674 | fjy0 = _mm_add_ps(fjy0,ty); |
675 | fjz0 = _mm_add_ps(fjz0,tz); |
676 | |
677 | } |
678 | |
679 | /************************** |
680 | * CALCULATE INTERACTIONS * |
681 | **************************/ |
682 | |
683 | if (gmx_mm_any_lt(rsq30,rcutoff2)) |
684 | { |
685 | |
686 | /* Compute parameters for interactions between i and j atoms */ |
687 | qq30 = _mm_mul_ps(iq3,jq0); |
688 | |
689 | /* REACTION-FIELD ELECTROSTATICS */ |
690 | velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf)); |
691 | felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2)); |
692 | |
693 | cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2); |
694 | |
695 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
696 | velec = _mm_and_ps(velec,cutoff_mask); |
697 | velec = _mm_andnot_ps(dummy_mask,velec); |
698 | velecsum = _mm_add_ps(velecsum,velec); |
699 | |
700 | fscal = felec; |
701 | |
702 | fscal = _mm_and_ps(fscal,cutoff_mask); |
703 | |
704 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
705 | |
706 | /* Calculate temporary vectorial force */ |
707 | tx = _mm_mul_ps(fscal,dx30); |
708 | ty = _mm_mul_ps(fscal,dy30); |
709 | tz = _mm_mul_ps(fscal,dz30); |
710 | |
711 | /* Update vectorial force */ |
712 | fix3 = _mm_add_ps(fix3,tx); |
713 | fiy3 = _mm_add_ps(fiy3,ty); |
714 | fiz3 = _mm_add_ps(fiz3,tz); |
715 | |
716 | fjx0 = _mm_add_ps(fjx0,tx); |
717 | fjy0 = _mm_add_ps(fjy0,ty); |
718 | fjz0 = _mm_add_ps(fjz0,tz); |
719 | |
720 | } |
721 | |
722 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
723 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
724 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
725 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
726 | |
727 | gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0); |
728 | |
729 | /* Inner loop uses 168 flops */ |
730 | } |
731 | |
732 | /* End of innermost loop */ |
733 | |
734 | gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3, |
735 | f+i_coord_offset,fshift+i_shift_offset); |
736 | |
737 | ggid = gid[iidx]; |
738 | /* Update potential energies */ |
739 | gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid); |
740 | gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid); |
741 | |
742 | /* Increment number of inner iterations */ |
743 | inneriter += j_index_end - j_index_start; |
744 | |
745 | /* Outer loop uses 26 flops */ |
746 | } |
747 | |
748 | /* Increment number of outer iterations */ |
749 | outeriter += nri; |
750 | |
751 | /* Update outer/inner flops */ |
752 | |
753 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*168)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W4_VF] += outeriter*26 + inneriter *168; |
754 | } |
755 | /* |
756 | * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse4_1_single |
757 | * Electrostatics interaction: ReactionField |
758 | * VdW interaction: LennardJones |
759 | * Geometry: Water4-Particle |
760 | * Calculate force/pot: Force |
761 | */ |
762 | void |
763 | nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse4_1_single |
764 | (t_nblist * gmx_restrict nlist, |
765 | rvec * gmx_restrict xx, |
766 | rvec * gmx_restrict ff, |
767 | t_forcerec * gmx_restrict fr, |
768 | t_mdatoms * gmx_restrict mdatoms, |
769 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
770 | t_nrnb * gmx_restrict nrnb) |
771 | { |
772 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
773 | * just 0 for non-waters. |
774 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
775 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
776 | */ |
777 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
778 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
779 | int jnrA,jnrB,jnrC,jnrD; |
780 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
781 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
782 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
783 | real rcutoff_scalar; |
784 | real *shiftvec,*fshift,*x,*f; |
785 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
786 | real scratch[4*DIM3]; |
787 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
788 | int vdwioffset0; |
789 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
790 | int vdwioffset1; |
791 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
792 | int vdwioffset2; |
793 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
794 | int vdwioffset3; |
795 | __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3; |
796 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
797 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
798 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
799 | __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10; |
800 | __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20; |
801 | __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30; |
802 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
803 | real *charge; |
804 | int nvdwtype; |
805 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
806 | int *vdwtype; |
807 | real *vdwparam; |
808 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
809 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
810 | __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw; |
811 | real rswitch_scalar,d_scalar; |
812 | __m128 dummy_mask,cutoff_mask; |
813 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
814 | __m128 one = _mm_set1_ps(1.0); |
815 | __m128 two = _mm_set1_ps(2.0); |
816 | x = xx[0]; |
817 | f = ff[0]; |
818 | |
819 | nri = nlist->nri; |
820 | iinr = nlist->iinr; |
821 | jindex = nlist->jindex; |
822 | jjnr = nlist->jjnr; |
823 | shiftidx = nlist->shift; |
824 | gid = nlist->gid; |
Value stored to 'gid' is never read | |
825 | shiftvec = fr->shift_vec[0]; |
826 | fshift = fr->fshift[0]; |
827 | facel = _mm_set1_ps(fr->epsfac); |
828 | charge = mdatoms->chargeA; |
829 | krf = _mm_set1_ps(fr->ic->k_rf); |
830 | krf2 = _mm_set1_ps(fr->ic->k_rf*2.0); |
831 | crf = _mm_set1_ps(fr->ic->c_rf); |
832 | nvdwtype = fr->ntype; |
833 | vdwparam = fr->nbfp; |
834 | vdwtype = mdatoms->typeA; |
835 | |
836 | /* Setup water-specific parameters */ |
837 | inr = nlist->iinr[0]; |
838 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
839 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
840 | iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3])); |
841 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
842 | |
843 | /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */ |
844 | rcutoff_scalar = fr->rcoulomb; |
845 | rcutoff = _mm_set1_ps(rcutoff_scalar); |
846 | rcutoff2 = _mm_mul_ps(rcutoff,rcutoff); |
847 | |
848 | rswitch_scalar = fr->rvdw_switch; |
849 | rswitch = _mm_set1_ps(rswitch_scalar); |
850 | /* Setup switch parameters */ |
851 | d_scalar = rcutoff_scalar-rswitch_scalar; |
852 | d = _mm_set1_ps(d_scalar); |
853 | swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar)); |
854 | swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar)); |
855 | swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar)); |
856 | swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar)); |
857 | swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar)); |
858 | swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar)); |
859 | |
860 | /* Avoid stupid compiler warnings */ |
861 | jnrA = jnrB = jnrC = jnrD = 0; |
862 | j_coord_offsetA = 0; |
863 | j_coord_offsetB = 0; |
864 | j_coord_offsetC = 0; |
865 | j_coord_offsetD = 0; |
866 | |
867 | outeriter = 0; |
868 | inneriter = 0; |
869 | |
870 | for(iidx=0;iidx<4*DIM3;iidx++) |
871 | { |
872 | scratch[iidx] = 0.0; |
873 | } |
874 | |
875 | /* Start outer loop over neighborlists */ |
876 | for(iidx=0; iidx<nri; iidx++) |
877 | { |
878 | /* Load shift vector for this list */ |
879 | i_shift_offset = DIM3*shiftidx[iidx]; |
880 | |
881 | /* Load limits for loop over neighbors */ |
882 | j_index_start = jindex[iidx]; |
883 | j_index_end = jindex[iidx+1]; |
884 | |
885 | /* Get outer coordinate index */ |
886 | inr = iinr[iidx]; |
887 | i_coord_offset = DIM3*inr; |
888 | |
889 | /* Load i particle coords and add shift vector */ |
890 | gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
891 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3); |
892 | |
893 | fix0 = _mm_setzero_ps(); |
894 | fiy0 = _mm_setzero_ps(); |
895 | fiz0 = _mm_setzero_ps(); |
896 | fix1 = _mm_setzero_ps(); |
897 | fiy1 = _mm_setzero_ps(); |
898 | fiz1 = _mm_setzero_ps(); |
899 | fix2 = _mm_setzero_ps(); |
900 | fiy2 = _mm_setzero_ps(); |
901 | fiz2 = _mm_setzero_ps(); |
902 | fix3 = _mm_setzero_ps(); |
903 | fiy3 = _mm_setzero_ps(); |
904 | fiz3 = _mm_setzero_ps(); |
905 | |
906 | /* Start inner kernel loop */ |
907 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
908 | { |
909 | |
910 | /* Get j neighbor index, and coordinate index */ |
911 | jnrA = jjnr[jidx]; |
912 | jnrB = jjnr[jidx+1]; |
913 | jnrC = jjnr[jidx+2]; |
914 | jnrD = jjnr[jidx+3]; |
915 | j_coord_offsetA = DIM3*jnrA; |
916 | j_coord_offsetB = DIM3*jnrB; |
917 | j_coord_offsetC = DIM3*jnrC; |
918 | j_coord_offsetD = DIM3*jnrD; |
919 | |
920 | /* load j atom coordinates */ |
921 | gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
922 | x+j_coord_offsetC,x+j_coord_offsetD, |
923 | &jx0,&jy0,&jz0); |
924 | |
925 | /* Calculate displacement vector */ |
926 | dx00 = _mm_sub_ps(ix0,jx0); |
927 | dy00 = _mm_sub_ps(iy0,jy0); |
928 | dz00 = _mm_sub_ps(iz0,jz0); |
929 | dx10 = _mm_sub_ps(ix1,jx0); |
930 | dy10 = _mm_sub_ps(iy1,jy0); |
931 | dz10 = _mm_sub_ps(iz1,jz0); |
932 | dx20 = _mm_sub_ps(ix2,jx0); |
933 | dy20 = _mm_sub_ps(iy2,jy0); |
934 | dz20 = _mm_sub_ps(iz2,jz0); |
935 | dx30 = _mm_sub_ps(ix3,jx0); |
936 | dy30 = _mm_sub_ps(iy3,jy0); |
937 | dz30 = _mm_sub_ps(iz3,jz0); |
938 | |
939 | /* Calculate squared distance and things based on it */ |
940 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
941 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
942 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
943 | rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30); |
944 | |
945 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
946 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
947 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
948 | rinv30 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq30); |
949 | |
950 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
951 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
952 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
953 | rinvsq30 = _mm_mul_ps(rinv30,rinv30); |
954 | |
955 | /* Load parameters for j particles */ |
956 | jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0, |
957 | charge+jnrC+0,charge+jnrD+0); |
958 | vdwjidx0A = 2*vdwtype[jnrA+0]; |
959 | vdwjidx0B = 2*vdwtype[jnrB+0]; |
960 | vdwjidx0C = 2*vdwtype[jnrC+0]; |
961 | vdwjidx0D = 2*vdwtype[jnrD+0]; |
962 | |
963 | fjx0 = _mm_setzero_ps(); |
964 | fjy0 = _mm_setzero_ps(); |
965 | fjz0 = _mm_setzero_ps(); |
966 | |
967 | /************************** |
968 | * CALCULATE INTERACTIONS * |
969 | **************************/ |
970 | |
971 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
972 | { |
973 | |
974 | r00 = _mm_mul_ps(rsq00,rinv00); |
975 | |
976 | /* Compute parameters for interactions between i and j atoms */ |
977 | gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A, |
978 | vdwparam+vdwioffset0+vdwjidx0B, |
979 | vdwparam+vdwioffset0+vdwjidx0C, |
980 | vdwparam+vdwioffset0+vdwjidx0D, |
981 | &c6_00,&c12_00); |
982 | |
983 | /* LENNARD-JONES DISPERSION/REPULSION */ |
984 | |
985 | rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00); |
986 | vvdw6 = _mm_mul_ps(c6_00,rinvsix); |
987 | vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix)); |
988 | vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) ); |
989 | fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00); |
990 | |
991 | d = _mm_sub_ps(r00,rswitch); |
992 | d = _mm_max_ps(d,_mm_setzero_ps()); |
993 | d2 = _mm_mul_ps(d,d); |
994 | sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5))))))); |
995 | |
996 | dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4))))); |
997 | |
998 | /* Evaluate switch function */ |
999 | /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */ |
1000 | fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) ); |
1001 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
1002 | |
1003 | fscal = fvdw; |
1004 | |
1005 | fscal = _mm_and_ps(fscal,cutoff_mask); |
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 | |
1023 | /************************** |
1024 | * CALCULATE INTERACTIONS * |
1025 | **************************/ |
1026 | |
1027 | if (gmx_mm_any_lt(rsq10,rcutoff2)) |
1028 | { |
1029 | |
1030 | /* Compute parameters for interactions between i and j atoms */ |
1031 | qq10 = _mm_mul_ps(iq1,jq0); |
1032 | |
1033 | /* REACTION-FIELD ELECTROSTATICS */ |
1034 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
1035 | |
1036 | cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2); |
1037 | |
1038 | fscal = felec; |
1039 | |
1040 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1041 | |
1042 | /* Calculate temporary vectorial force */ |
1043 | tx = _mm_mul_ps(fscal,dx10); |
1044 | ty = _mm_mul_ps(fscal,dy10); |
1045 | tz = _mm_mul_ps(fscal,dz10); |
1046 | |
1047 | /* Update vectorial force */ |
1048 | fix1 = _mm_add_ps(fix1,tx); |
1049 | fiy1 = _mm_add_ps(fiy1,ty); |
1050 | fiz1 = _mm_add_ps(fiz1,tz); |
1051 | |
1052 | fjx0 = _mm_add_ps(fjx0,tx); |
1053 | fjy0 = _mm_add_ps(fjy0,ty); |
1054 | fjz0 = _mm_add_ps(fjz0,tz); |
1055 | |
1056 | } |
1057 | |
1058 | /************************** |
1059 | * CALCULATE INTERACTIONS * |
1060 | **************************/ |
1061 | |
1062 | if (gmx_mm_any_lt(rsq20,rcutoff2)) |
1063 | { |
1064 | |
1065 | /* Compute parameters for interactions between i and j atoms */ |
1066 | qq20 = _mm_mul_ps(iq2,jq0); |
1067 | |
1068 | /* REACTION-FIELD ELECTROSTATICS */ |
1069 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
1070 | |
1071 | cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2); |
1072 | |
1073 | fscal = felec; |
1074 | |
1075 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1076 | |
1077 | /* Calculate temporary vectorial force */ |
1078 | tx = _mm_mul_ps(fscal,dx20); |
1079 | ty = _mm_mul_ps(fscal,dy20); |
1080 | tz = _mm_mul_ps(fscal,dz20); |
1081 | |
1082 | /* Update vectorial force */ |
1083 | fix2 = _mm_add_ps(fix2,tx); |
1084 | fiy2 = _mm_add_ps(fiy2,ty); |
1085 | fiz2 = _mm_add_ps(fiz2,tz); |
1086 | |
1087 | fjx0 = _mm_add_ps(fjx0,tx); |
1088 | fjy0 = _mm_add_ps(fjy0,ty); |
1089 | fjz0 = _mm_add_ps(fjz0,tz); |
1090 | |
1091 | } |
1092 | |
1093 | /************************** |
1094 | * CALCULATE INTERACTIONS * |
1095 | **************************/ |
1096 | |
1097 | if (gmx_mm_any_lt(rsq30,rcutoff2)) |
1098 | { |
1099 | |
1100 | /* Compute parameters for interactions between i and j atoms */ |
1101 | qq30 = _mm_mul_ps(iq3,jq0); |
1102 | |
1103 | /* REACTION-FIELD ELECTROSTATICS */ |
1104 | felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2)); |
1105 | |
1106 | cutoff_mask = _mm_cmplt_ps(rsq30,rcutoff2); |
1107 | |
1108 | fscal = felec; |
1109 | |
1110 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1111 | |
1112 | /* Calculate temporary vectorial force */ |
1113 | tx = _mm_mul_ps(fscal,dx30); |
1114 | ty = _mm_mul_ps(fscal,dy30); |
1115 | tz = _mm_mul_ps(fscal,dz30); |
1116 | |
1117 | /* Update vectorial force */ |
1118 | fix3 = _mm_add_ps(fix3,tx); |
1119 | fiy3 = _mm_add_ps(fiy3,ty); |
1120 | fiz3 = _mm_add_ps(fiz3,tz); |
1121 | |
1122 | fjx0 = _mm_add_ps(fjx0,tx); |
1123 | fjy0 = _mm_add_ps(fjy0,ty); |
1124 | fjz0 = _mm_add_ps(fjz0,tz); |
1125 | |
1126 | } |
1127 | |
1128 | fjptrA = f+j_coord_offsetA; |
1129 | fjptrB = f+j_coord_offsetB; |
1130 | fjptrC = f+j_coord_offsetC; |
1131 | fjptrD = f+j_coord_offsetD; |
1132 | |
1133 | gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0); |
1134 | |
1135 | /* Inner loop uses 146 flops */ |
1136 | } |
1137 | |
1138 | if(jidx<j_index_end) |
1139 | { |
1140 | |
1141 | /* Get j neighbor index, and coordinate index */ |
1142 | jnrlistA = jjnr[jidx]; |
1143 | jnrlistB = jjnr[jidx+1]; |
1144 | jnrlistC = jjnr[jidx+2]; |
1145 | jnrlistD = jjnr[jidx+3]; |
1146 | /* Sign of each element will be negative for non-real atoms. |
1147 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
1148 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
1149 | */ |
1150 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
1151 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
1152 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
1153 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
1154 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
1155 | j_coord_offsetA = DIM3*jnrA; |
1156 | j_coord_offsetB = DIM3*jnrB; |
1157 | j_coord_offsetC = DIM3*jnrC; |
1158 | j_coord_offsetD = DIM3*jnrD; |
1159 | |
1160 | /* load j atom coordinates */ |
1161 | gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
1162 | x+j_coord_offsetC,x+j_coord_offsetD, |
1163 | &jx0,&jy0,&jz0); |
1164 | |
1165 | /* Calculate displacement vector */ |
1166 | dx00 = _mm_sub_ps(ix0,jx0); |
1167 | dy00 = _mm_sub_ps(iy0,jy0); |
1168 | dz00 = _mm_sub_ps(iz0,jz0); |
1169 | dx10 = _mm_sub_ps(ix1,jx0); |
1170 | dy10 = _mm_sub_ps(iy1,jy0); |
1171 | dz10 = _mm_sub_ps(iz1,jz0); |
1172 | dx20 = _mm_sub_ps(ix2,jx0); |
1173 | dy20 = _mm_sub_ps(iy2,jy0); |
1174 | dz20 = _mm_sub_ps(iz2,jz0); |
1175 | dx30 = _mm_sub_ps(ix3,jx0); |
1176 | dy30 = _mm_sub_ps(iy3,jy0); |
1177 | dz30 = _mm_sub_ps(iz3,jz0); |
1178 | |
1179 | /* Calculate squared distance and things based on it */ |
1180 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
1181 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
1182 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
1183 | rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30); |
1184 | |
1185 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
1186 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
1187 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
1188 | rinv30 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq30); |
1189 | |
1190 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
1191 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
1192 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
1193 | rinvsq30 = _mm_mul_ps(rinv30,rinv30); |
1194 | |
1195 | /* Load parameters for j particles */ |
1196 | jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0, |
1197 | charge+jnrC+0,charge+jnrD+0); |
1198 | vdwjidx0A = 2*vdwtype[jnrA+0]; |
1199 | vdwjidx0B = 2*vdwtype[jnrB+0]; |
1200 | vdwjidx0C = 2*vdwtype[jnrC+0]; |
1201 | vdwjidx0D = 2*vdwtype[jnrD+0]; |
1202 | |
1203 | fjx0 = _mm_setzero_ps(); |
1204 | fjy0 = _mm_setzero_ps(); |
1205 | fjz0 = _mm_setzero_ps(); |
1206 | |
1207 | /************************** |
1208 | * CALCULATE INTERACTIONS * |
1209 | **************************/ |
1210 | |
1211 | if (gmx_mm_any_lt(rsq00,rcutoff2)) |
1212 | { |
1213 | |
1214 | r00 = _mm_mul_ps(rsq00,rinv00); |
1215 | r00 = _mm_andnot_ps(dummy_mask,r00); |
1216 | |
1217 | /* Compute parameters for interactions between i and j atoms */ |
1218 | gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A, |
1219 | vdwparam+vdwioffset0+vdwjidx0B, |
1220 | vdwparam+vdwioffset0+vdwjidx0C, |
1221 | vdwparam+vdwioffset0+vdwjidx0D, |
1222 | &c6_00,&c12_00); |
1223 | |
1224 | /* LENNARD-JONES DISPERSION/REPULSION */ |
1225 | |
1226 | rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00); |
1227 | vvdw6 = _mm_mul_ps(c6_00,rinvsix); |
1228 | vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix)); |
1229 | vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) ); |
1230 | fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00); |
1231 | |
1232 | d = _mm_sub_ps(r00,rswitch); |
1233 | d = _mm_max_ps(d,_mm_setzero_ps()); |
1234 | d2 = _mm_mul_ps(d,d); |
1235 | sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5))))))); |
1236 | |
1237 | dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4))))); |
1238 | |
1239 | /* Evaluate switch function */ |
1240 | /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */ |
1241 | fvdw = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) ); |
1242 | cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2); |
1243 | |
1244 | fscal = fvdw; |
1245 | |
1246 | fscal = _mm_and_ps(fscal,cutoff_mask); |
1247 | |
1248 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1249 | |
1250 | /* Calculate temporary vectorial force */ |
1251 | tx = _mm_mul_ps(fscal,dx00); |
1252 | ty = _mm_mul_ps(fscal,dy00); |
1253 | tz = _mm_mul_ps(fscal,dz00); |
1254 | |
1255 | /* Update vectorial force */ |
1256 | fix0 = _mm_add_ps(fix0,tx); |
1257 | fiy0 = _mm_add_ps(fiy0,ty); |
1258 | fiz0 = _mm_add_ps(fiz0,tz); |
1259 | |
1260 | fjx0 = _mm_add_ps(fjx0,tx); |
1261 | fjy0 = _mm_add_ps(fjy0,ty); |
1262 | fjz0 = _mm_add_ps(fjz0,tz); |
1263 | |
1264 | } |
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 147 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*147)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W4_F] += outeriter*24 + inneriter *147; |
1404 | } |