?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed simple.h from nb_kernel_sse4_1_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_sse4_1_double / nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_sse4_1_double.c
blob38d396c573dcdaa0ec8e9db3a238db9339ac1cfb
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015, 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_double kernel generator.
37 */
38 #include "gmxpre.h"
39
40 #include "config.h"
41
42 #include <math.h>
43
44 #include "../nb_kernel.h"
45 #include "gromacs/math/vec.h"
46 #include "gromacs/legacyheaders/nrnb.h"
47
48 #include "gromacs/simd/math_x86_sse4_1_double.h"
49 #include "kernelutil_x86_sse4_1_double.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse4_1_double
53 * Electrostatics interaction: Ewald
54 * VdW interaction: LennardJones
55 * Geometry: Particle-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse4_1_double
60 (t_nblist * gmx_restrict nlist,
61 rvec * gmx_restrict xx,
62 rvec * gmx_restrict ff,
63 t_forcerec * gmx_restrict fr,
64 t_mdatoms * gmx_restrict mdatoms,
65 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
66 t_nrnb * gmx_restrict nrnb)
67 {
68 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
69 * just 0 for non-waters.
70 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
71 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 */
73 int i_shift_offset,i_coord_offset,outeriter,inneriter;
74 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int jnrA,jnrB;
76 int j_coord_offsetA,j_coord_offsetB;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real rcutoff_scalar;
79 real *shiftvec,*fshift,*x,*f;
80 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 int vdwioffset0;
82 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 int vdwjidx0A,vdwjidx0B;
84 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
85 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
86 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
87 real *charge;
88 int nvdwtype;
89 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
90 int *vdwtype;
91 real *vdwparam;
92 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
93 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
94 __m128i ewitab;
95 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
96 real *ewtab;
97 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
98 real rswitch_scalar,d_scalar;
99 __m128d dummy_mask,cutoff_mask;
100 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
101 __m128d one = _mm_set1_pd(1.0);
102 __m128d two = _mm_set1_pd(2.0);
103 x = xx[0];
104 f = ff[0];
105
106 nri = nlist->nri;
107 iinr = nlist->iinr;
108 jindex = nlist->jindex;
109 jjnr = nlist->jjnr;
110 shiftidx = nlist->shift;
111 gid = nlist->gid;
112 shiftvec = fr->shift_vec[0];
113 fshift = fr->fshift[0];
114 facel = _mm_set1_pd(fr->epsfac);
115 charge = mdatoms->chargeA;
116 nvdwtype = fr->ntype;
117 vdwparam = fr->nbfp;
118 vdwtype = mdatoms->typeA;
119
120 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
121 ewtab = fr->ic->tabq_coul_FDV0;
122 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
123 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
124
125 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
126 rcutoff_scalar = fr->rcoulomb;
127 rcutoff = _mm_set1_pd(rcutoff_scalar);
128 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
129
130 rswitch_scalar = fr->rcoulomb_switch;
131 rswitch = _mm_set1_pd(rswitch_scalar);
132 /* Setup switch parameters */
133 d_scalar = rcutoff_scalar-rswitch_scalar;
134 d = _mm_set1_pd(d_scalar);
135 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
136 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
137 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
138 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
139 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
140 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
141
142 /* Avoid stupid compiler warnings */
143 jnrA = jnrB = 0;
144 j_coord_offsetA = 0;
145 j_coord_offsetB = 0;
146
147 outeriter = 0;
148 inneriter = 0;
149
150 /* Start outer loop over neighborlists */
151 for(iidx=0; iidx<nri; iidx++)
152 {
153 /* Load shift vector for this list */
154 i_shift_offset = DIM*shiftidx[iidx];
155
156 /* Load limits for loop over neighbors */
157 j_index_start = jindex[iidx];
158 j_index_end = jindex[iidx+1];
159
160 /* Get outer coordinate index */
161 inr = iinr[iidx];
162 i_coord_offset = DIM*inr;
163
164 /* Load i particle coords and add shift vector */
165 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
166
167 fix0 = _mm_setzero_pd();
168 fiy0 = _mm_setzero_pd();
169 fiz0 = _mm_setzero_pd();
170
171 /* Load parameters for i particles */
172 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
173 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
174
175 /* Reset potential sums */
176 velecsum = _mm_setzero_pd();
177 vvdwsum = _mm_setzero_pd();
178
179 /* Start inner kernel loop */
180 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
181 {
182
183 /* Get j neighbor index, and coordinate index */
184 jnrA = jjnr[jidx];
185 jnrB = jjnr[jidx+1];
186 j_coord_offsetA = DIM*jnrA;
187 j_coord_offsetB = DIM*jnrB;
188
189 /* load j atom coordinates */
190 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
191 &jx0,&jy0,&jz0);
192
193 /* Calculate displacement vector */
194 dx00 = _mm_sub_pd(ix0,jx0);
195 dy00 = _mm_sub_pd(iy0,jy0);
196 dz00 = _mm_sub_pd(iz0,jz0);
197
198 /* Calculate squared distance and things based on it */
199 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
200
201 rinv00 = gmx_mm_invsqrt_pd(rsq00);
202
203 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
204
205 /* Load parameters for j particles */
206 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
207 vdwjidx0A = 2*vdwtype[jnrA+0];
208 vdwjidx0B = 2*vdwtype[jnrB+0];
209
210 /**************************
211 * CALCULATE INTERACTIONS *
212 **************************/
213
214 if (gmx_mm_any_lt(rsq00,rcutoff2))
215 {
216
217 r00 = _mm_mul_pd(rsq00,rinv00);
218
219 /* Compute parameters for interactions between i and j atoms */
220 qq00 = _mm_mul_pd(iq0,jq0);
221 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
222 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
223
224 /* EWALD ELECTROSTATICS */
225
226 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
227 ewrt = _mm_mul_pd(r00,ewtabscale);
228 ewitab = _mm_cvttpd_epi32(ewrt);
229 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
230 ewitab = _mm_slli_epi32(ewitab,2);
231 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
232 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
233 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
234 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
235 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
236 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
237 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
238 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
239 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
240 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
241
242 /* LENNARD-JONES DISPERSION/REPULSION */
243
244 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
245 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
246 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
247 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
248 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
249
250 d = _mm_sub_pd(r00,rswitch);
251 d = _mm_max_pd(d,_mm_setzero_pd());
252 d2 = _mm_mul_pd(d,d);
253 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
254
255 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
256
257 /* Evaluate switch function */
258 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
259 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
260 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
261 velec = _mm_mul_pd(velec,sw);
262 vvdw = _mm_mul_pd(vvdw,sw);
263 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
264
265 /* Update potential sum for this i atom from the interaction with this j atom. */
266 velec = _mm_and_pd(velec,cutoff_mask);
267 velecsum = _mm_add_pd(velecsum,velec);
268 vvdw = _mm_and_pd(vvdw,cutoff_mask);
269 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
270
271 fscal = _mm_add_pd(felec,fvdw);
272
273 fscal = _mm_and_pd(fscal,cutoff_mask);
274
275 /* Calculate temporary vectorial force */
276 tx = _mm_mul_pd(fscal,dx00);
277 ty = _mm_mul_pd(fscal,dy00);
278 tz = _mm_mul_pd(fscal,dz00);
279
280 /* Update vectorial force */
281 fix0 = _mm_add_pd(fix0,tx);
282 fiy0 = _mm_add_pd(fiy0,ty);
283 fiz0 = _mm_add_pd(fiz0,tz);
284
285 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
286
287 }
288
289 /* Inner loop uses 83 flops */
290 }
291
292 if(jidx<j_index_end)
293 {
294
295 jnrA = jjnr[jidx];
296 j_coord_offsetA = DIM*jnrA;
297
298 /* load j atom coordinates */
299 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
300 &jx0,&jy0,&jz0);
301
302 /* Calculate displacement vector */
303 dx00 = _mm_sub_pd(ix0,jx0);
304 dy00 = _mm_sub_pd(iy0,jy0);
305 dz00 = _mm_sub_pd(iz0,jz0);
306
307 /* Calculate squared distance and things based on it */
308 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
309
310 rinv00 = gmx_mm_invsqrt_pd(rsq00);
311
312 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
313
314 /* Load parameters for j particles */
315 jq0 = _mm_load_sd(charge+jnrA+0);
316 vdwjidx0A = 2*vdwtype[jnrA+0];
317
318 /**************************
319 * CALCULATE INTERACTIONS *
320 **************************/
321
322 if (gmx_mm_any_lt(rsq00,rcutoff2))
323 {
324
325 r00 = _mm_mul_pd(rsq00,rinv00);
326
327 /* Compute parameters for interactions between i and j atoms */
328 qq00 = _mm_mul_pd(iq0,jq0);
329 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
330
331 /* EWALD ELECTROSTATICS */
332
333 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
334 ewrt = _mm_mul_pd(r00,ewtabscale);
335 ewitab = _mm_cvttpd_epi32(ewrt);
336 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
337 ewitab = _mm_slli_epi32(ewitab,2);
338 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
339 ewtabD = _mm_setzero_pd();
340 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
341 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
342 ewtabFn = _mm_setzero_pd();
343 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
344 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
345 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
346 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
347 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
348
349 /* LENNARD-JONES DISPERSION/REPULSION */
350
351 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
352 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
353 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
354 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
355 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
356
357 d = _mm_sub_pd(r00,rswitch);
358 d = _mm_max_pd(d,_mm_setzero_pd());
359 d2 = _mm_mul_pd(d,d);
360 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
361
362 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
363
364 /* Evaluate switch function */
365 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
366 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
367 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
368 velec = _mm_mul_pd(velec,sw);
369 vvdw = _mm_mul_pd(vvdw,sw);
370 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
371
372 /* Update potential sum for this i atom from the interaction with this j atom. */
373 velec = _mm_and_pd(velec,cutoff_mask);
374 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
375 velecsum = _mm_add_pd(velecsum,velec);
376 vvdw = _mm_and_pd(vvdw,cutoff_mask);
377 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
378 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
379
380 fscal = _mm_add_pd(felec,fvdw);
381
382 fscal = _mm_and_pd(fscal,cutoff_mask);
383
384 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
385
386 /* Calculate temporary vectorial force */
387 tx = _mm_mul_pd(fscal,dx00);
388 ty = _mm_mul_pd(fscal,dy00);
389 tz = _mm_mul_pd(fscal,dz00);
390
391 /* Update vectorial force */
392 fix0 = _mm_add_pd(fix0,tx);
393 fiy0 = _mm_add_pd(fiy0,ty);
394 fiz0 = _mm_add_pd(fiz0,tz);
395
396 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
397
398 }
399
400 /* Inner loop uses 83 flops */
401 }
402
403 /* End of innermost loop */
404
405 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
406 f+i_coord_offset,fshift+i_shift_offset);
407
408 ggid = gid[iidx];
409 /* Update potential energies */
410 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
411 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
412
413 /* Increment number of inner iterations */
414 inneriter += j_index_end - j_index_start;
415
416 /* Outer loop uses 9 flops */
417 }
418
419 /* Increment number of outer iterations */
420 outeriter += nri;
421
422 /* Update outer/inner flops */
423
424 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*83);
425 }
426 /*
427 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse4_1_double
428 * Electrostatics interaction: Ewald
429 * VdW interaction: LennardJones
430 * Geometry: Particle-Particle
431 * Calculate force/pot: Force
432 */
433 void
434 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse4_1_double
435 (t_nblist * gmx_restrict nlist,
436 rvec * gmx_restrict xx,
437 rvec * gmx_restrict ff,
438 t_forcerec * gmx_restrict fr,
439 t_mdatoms * gmx_restrict mdatoms,
440 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
441 t_nrnb * gmx_restrict nrnb)
442 {
443 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
444 * just 0 for non-waters.
445 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
446 * jnr indices corresponding to data put in the four positions in the SIMD register.
447 */
448 int i_shift_offset,i_coord_offset,outeriter,inneriter;
449 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
450 int jnrA,jnrB;
451 int j_coord_offsetA,j_coord_offsetB;
452 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
453 real rcutoff_scalar;
454 real *shiftvec,*fshift,*x,*f;
455 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
456 int vdwioffset0;
457 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
458 int vdwjidx0A,vdwjidx0B;
459 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
460 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
461 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
462 real *charge;
463 int nvdwtype;
464 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
465 int *vdwtype;
466 real *vdwparam;
467 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
468 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
469 __m128i ewitab;
470 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
471 real *ewtab;
472 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
473 real rswitch_scalar,d_scalar;
474 __m128d dummy_mask,cutoff_mask;
475 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
476 __m128d one = _mm_set1_pd(1.0);
477 __m128d two = _mm_set1_pd(2.0);
478 x = xx[0];
479 f = ff[0];
480
481 nri = nlist->nri;
482 iinr = nlist->iinr;
483 jindex = nlist->jindex;
484 jjnr = nlist->jjnr;
485 shiftidx = nlist->shift;
486 gid = nlist->gid;
487 shiftvec = fr->shift_vec[0];
488 fshift = fr->fshift[0];
489 facel = _mm_set1_pd(fr->epsfac);
490 charge = mdatoms->chargeA;
491 nvdwtype = fr->ntype;
492 vdwparam = fr->nbfp;
493 vdwtype = mdatoms->typeA;
494
495 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
496 ewtab = fr->ic->tabq_coul_FDV0;
497 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
498 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
499
500 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
501 rcutoff_scalar = fr->rcoulomb;
502 rcutoff = _mm_set1_pd(rcutoff_scalar);
503 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
504
505 rswitch_scalar = fr->rcoulomb_switch;
506 rswitch = _mm_set1_pd(rswitch_scalar);
507 /* Setup switch parameters */
508 d_scalar = rcutoff_scalar-rswitch_scalar;
509 d = _mm_set1_pd(d_scalar);
510 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
511 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
512 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
513 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
514 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
515 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
516
517 /* Avoid stupid compiler warnings */
518 jnrA = jnrB = 0;
519 j_coord_offsetA = 0;
520 j_coord_offsetB = 0;
521
522 outeriter = 0;
523 inneriter = 0;
524
525 /* Start outer loop over neighborlists */
526 for(iidx=0; iidx<nri; iidx++)
527 {
528 /* Load shift vector for this list */
529 i_shift_offset = DIM*shiftidx[iidx];
530
531 /* Load limits for loop over neighbors */
532 j_index_start = jindex[iidx];
533 j_index_end = jindex[iidx+1];
534
535 /* Get outer coordinate index */
536 inr = iinr[iidx];
537 i_coord_offset = DIM*inr;
538
539 /* Load i particle coords and add shift vector */
540 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
541
542 fix0 = _mm_setzero_pd();
543 fiy0 = _mm_setzero_pd();
544 fiz0 = _mm_setzero_pd();
545
546 /* Load parameters for i particles */
547 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
548 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
549
550 /* Start inner kernel loop */
551 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
552 {
553
554 /* Get j neighbor index, and coordinate index */
555 jnrA = jjnr[jidx];
556 jnrB = jjnr[jidx+1];
557 j_coord_offsetA = DIM*jnrA;
558 j_coord_offsetB = DIM*jnrB;
559
560 /* load j atom coordinates */
561 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
562 &jx0,&jy0,&jz0);
563
564 /* Calculate displacement vector */
565 dx00 = _mm_sub_pd(ix0,jx0);
566 dy00 = _mm_sub_pd(iy0,jy0);
567 dz00 = _mm_sub_pd(iz0,jz0);
568
569 /* Calculate squared distance and things based on it */
570 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
571
572 rinv00 = gmx_mm_invsqrt_pd(rsq00);
573
574 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
575
576 /* Load parameters for j particles */
577 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
578 vdwjidx0A = 2*vdwtype[jnrA+0];
579 vdwjidx0B = 2*vdwtype[jnrB+0];
580
581 /**************************
582 * CALCULATE INTERACTIONS *
583 **************************/
584
585 if (gmx_mm_any_lt(rsq00,rcutoff2))
586 {
587
588 r00 = _mm_mul_pd(rsq00,rinv00);
589
590 /* Compute parameters for interactions between i and j atoms */
591 qq00 = _mm_mul_pd(iq0,jq0);
592 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
593 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
594
595 /* EWALD ELECTROSTATICS */
596
597 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
598 ewrt = _mm_mul_pd(r00,ewtabscale);
599 ewitab = _mm_cvttpd_epi32(ewrt);
600 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
601 ewitab = _mm_slli_epi32(ewitab,2);
602 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
603 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
604 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
605 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
606 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
607 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
608 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
609 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
610 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
611 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
612
613 /* LENNARD-JONES DISPERSION/REPULSION */
614
615 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
616 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
617 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
618 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
619 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
620
621 d = _mm_sub_pd(r00,rswitch);
622 d = _mm_max_pd(d,_mm_setzero_pd());
623 d2 = _mm_mul_pd(d,d);
624 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
625
626 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
627
628 /* Evaluate switch function */
629 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
630 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
631 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
632 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
633
634 fscal = _mm_add_pd(felec,fvdw);
635
636 fscal = _mm_and_pd(fscal,cutoff_mask);
637
638 /* Calculate temporary vectorial force */
639 tx = _mm_mul_pd(fscal,dx00);
640 ty = _mm_mul_pd(fscal,dy00);
641 tz = _mm_mul_pd(fscal,dz00);
642
643 /* Update vectorial force */
644 fix0 = _mm_add_pd(fix0,tx);
645 fiy0 = _mm_add_pd(fiy0,ty);
646 fiz0 = _mm_add_pd(fiz0,tz);
647
648 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
649
650 }
651
652 /* Inner loop uses 77 flops */
653 }
654
655 if(jidx<j_index_end)
656 {
657
658 jnrA = jjnr[jidx];
659 j_coord_offsetA = DIM*jnrA;
660
661 /* load j atom coordinates */
662 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
663 &jx0,&jy0,&jz0);
664
665 /* Calculate displacement vector */
666 dx00 = _mm_sub_pd(ix0,jx0);
667 dy00 = _mm_sub_pd(iy0,jy0);
668 dz00 = _mm_sub_pd(iz0,jz0);
669
670 /* Calculate squared distance and things based on it */
671 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
672
673 rinv00 = gmx_mm_invsqrt_pd(rsq00);
674
675 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
676
677 /* Load parameters for j particles */
678 jq0 = _mm_load_sd(charge+jnrA+0);
679 vdwjidx0A = 2*vdwtype[jnrA+0];
680
681 /**************************
682 * CALCULATE INTERACTIONS *
683 **************************/
684
685 if (gmx_mm_any_lt(rsq00,rcutoff2))
686 {
687
688 r00 = _mm_mul_pd(rsq00,rinv00);
689
690 /* Compute parameters for interactions between i and j atoms */
691 qq00 = _mm_mul_pd(iq0,jq0);
692 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
693
694 /* EWALD ELECTROSTATICS */
695
696 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
697 ewrt = _mm_mul_pd(r00,ewtabscale);
698 ewitab = _mm_cvttpd_epi32(ewrt);
699 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
700 ewitab = _mm_slli_epi32(ewitab,2);
701 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
702 ewtabD = _mm_setzero_pd();
703 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
704 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
705 ewtabFn = _mm_setzero_pd();
706 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
707 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
708 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
709 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
710 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
711
712 /* LENNARD-JONES DISPERSION/REPULSION */
713
714 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
715 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
716 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
717 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
718 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
719
720 d = _mm_sub_pd(r00,rswitch);
721 d = _mm_max_pd(d,_mm_setzero_pd());
722 d2 = _mm_mul_pd(d,d);
723 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
724
725 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
726
727 /* Evaluate switch function */
728 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
729 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
730 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
731 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
732
733 fscal = _mm_add_pd(felec,fvdw);
734
735 fscal = _mm_and_pd(fscal,cutoff_mask);
736
737 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
738
739 /* Calculate temporary vectorial force */
740 tx = _mm_mul_pd(fscal,dx00);
741 ty = _mm_mul_pd(fscal,dy00);
742 tz = _mm_mul_pd(fscal,dz00);
743
744 /* Update vectorial force */
745 fix0 = _mm_add_pd(fix0,tx);
746 fiy0 = _mm_add_pd(fiy0,ty);
747 fiz0 = _mm_add_pd(fiz0,tz);
748
749 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
750
751 }
752
753 /* Inner loop uses 77 flops */
754 }
755
756 /* End of innermost loop */
757
758 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
759 f+i_coord_offset,fshift+i_shift_offset);
760
761 /* Increment number of inner iterations */
762 inneriter += j_index_end - j_index_start;
763
764 /* Outer loop uses 7 flops */
765 }
766
767 /* Increment number of outer iterations */
768 outeriter += nri;
769
770 /* Update outer/inner flops */
771
772 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*77);
773 }
The ultimate molecular dynamics simulation package