?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed simple.h from nb_kernel_sse2_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_sse2_double / nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_sse2_double.c
blob1d56630870e42ff25b9281490488f59f8e4bbff7
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 sse2_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_sse2_double.h"
49 #include "kernelutil_x86_sse2_double.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse2_double
53 * Electrostatics interaction: CubicSplineTable
54 * VdW interaction: CubicSplineTable
55 * Geometry: Water3-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse2_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 vdwioffset1;
84 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 int vdwioffset2;
86 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
87 int vdwjidx0A,vdwjidx0B;
88 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
89 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
90 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
91 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
92 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
93 real *charge;
94 int nvdwtype;
95 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
96 int *vdwtype;
97 real *vdwparam;
98 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
99 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
100 __m128i vfitab;
101 __m128i ifour = _mm_set1_epi32(4);
102 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
103 real *vftab;
104 __m128d dummy_mask,cutoff_mask;
105 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
106 __m128d one = _mm_set1_pd(1.0);
107 __m128d two = _mm_set1_pd(2.0);
108 x = xx[0];
109 f = ff[0];
110
111 nri = nlist->nri;
112 iinr = nlist->iinr;
113 jindex = nlist->jindex;
114 jjnr = nlist->jjnr;
115 shiftidx = nlist->shift;
116 gid = nlist->gid;
117 shiftvec = fr->shift_vec[0];
118 fshift = fr->fshift[0];
119 facel = _mm_set1_pd(fr->epsfac);
120 charge = mdatoms->chargeA;
121 nvdwtype = fr->ntype;
122 vdwparam = fr->nbfp;
123 vdwtype = mdatoms->typeA;
124
125 vftab = kernel_data->table_elec_vdw->data;
126 vftabscale = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
127
128 /* Setup water-specific parameters */
129 inr = nlist->iinr[0];
130 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
131 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
132 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
133 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
134
135 /* Avoid stupid compiler warnings */
136 jnrA = jnrB = 0;
137 j_coord_offsetA = 0;
138 j_coord_offsetB = 0;
139
140 outeriter = 0;
141 inneriter = 0;
142
143 /* Start outer loop over neighborlists */
144 for(iidx=0; iidx<nri; iidx++)
145 {
146 /* Load shift vector for this list */
147 i_shift_offset = DIM*shiftidx[iidx];
148
149 /* Load limits for loop over neighbors */
150 j_index_start = jindex[iidx];
151 j_index_end = jindex[iidx+1];
152
153 /* Get outer coordinate index */
154 inr = iinr[iidx];
155 i_coord_offset = DIM*inr;
156
157 /* Load i particle coords and add shift vector */
158 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
159 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
160
161 fix0 = _mm_setzero_pd();
162 fiy0 = _mm_setzero_pd();
163 fiz0 = _mm_setzero_pd();
164 fix1 = _mm_setzero_pd();
165 fiy1 = _mm_setzero_pd();
166 fiz1 = _mm_setzero_pd();
167 fix2 = _mm_setzero_pd();
168 fiy2 = _mm_setzero_pd();
169 fiz2 = _mm_setzero_pd();
170
171 /* Reset potential sums */
172 velecsum = _mm_setzero_pd();
173 vvdwsum = _mm_setzero_pd();
174
175 /* Start inner kernel loop */
176 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
177 {
178
179 /* Get j neighbor index, and coordinate index */
180 jnrA = jjnr[jidx];
181 jnrB = jjnr[jidx+1];
182 j_coord_offsetA = DIM*jnrA;
183 j_coord_offsetB = DIM*jnrB;
184
185 /* load j atom coordinates */
186 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
187 &jx0,&jy0,&jz0);
188
189 /* Calculate displacement vector */
190 dx00 = _mm_sub_pd(ix0,jx0);
191 dy00 = _mm_sub_pd(iy0,jy0);
192 dz00 = _mm_sub_pd(iz0,jz0);
193 dx10 = _mm_sub_pd(ix1,jx0);
194 dy10 = _mm_sub_pd(iy1,jy0);
195 dz10 = _mm_sub_pd(iz1,jz0);
196 dx20 = _mm_sub_pd(ix2,jx0);
197 dy20 = _mm_sub_pd(iy2,jy0);
198 dz20 = _mm_sub_pd(iz2,jz0);
199
200 /* Calculate squared distance and things based on it */
201 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
202 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
203 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
204
205 rinv00 = gmx_mm_invsqrt_pd(rsq00);
206 rinv10 = gmx_mm_invsqrt_pd(rsq10);
207 rinv20 = gmx_mm_invsqrt_pd(rsq20);
208
209 /* Load parameters for j particles */
210 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
211 vdwjidx0A = 2*vdwtype[jnrA+0];
212 vdwjidx0B = 2*vdwtype[jnrB+0];
213
214 fjx0 = _mm_setzero_pd();
215 fjy0 = _mm_setzero_pd();
216 fjz0 = _mm_setzero_pd();
217
218 /**************************
219 * CALCULATE INTERACTIONS *
220 **************************/
221
222 r00 = _mm_mul_pd(rsq00,rinv00);
223
224 /* Compute parameters for interactions between i and j atoms */
225 qq00 = _mm_mul_pd(iq0,jq0);
226 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
227 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
228
229 /* Calculate table index by multiplying r with table scale and truncate to integer */
230 rt = _mm_mul_pd(r00,vftabscale);
231 vfitab = _mm_cvttpd_epi32(rt);
232 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
233 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
234
235 /* CUBIC SPLINE TABLE ELECTROSTATICS */
236 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
237 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
238 GMX_MM_TRANSPOSE2_PD(Y,F);
239 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
240 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
241 GMX_MM_TRANSPOSE2_PD(G,H);
242 Heps = _mm_mul_pd(vfeps,H);
243 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
244 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
245 velec = _mm_mul_pd(qq00,VV);
246 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
247 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
248
249 /* CUBIC SPLINE TABLE DISPERSION */
250 vfitab = _mm_add_epi32(vfitab,ifour);
251 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
252 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
253 GMX_MM_TRANSPOSE2_PD(Y,F);
254 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
255 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
256 GMX_MM_TRANSPOSE2_PD(G,H);
257 Heps = _mm_mul_pd(vfeps,H);
258 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
259 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
260 vvdw6 = _mm_mul_pd(c6_00,VV);
261 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
262 fvdw6 = _mm_mul_pd(c6_00,FF);
263
264 /* CUBIC SPLINE TABLE REPULSION */
265 vfitab = _mm_add_epi32(vfitab,ifour);
266 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
267 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
268 GMX_MM_TRANSPOSE2_PD(Y,F);
269 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
270 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
271 GMX_MM_TRANSPOSE2_PD(G,H);
272 Heps = _mm_mul_pd(vfeps,H);
273 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
274 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
275 vvdw12 = _mm_mul_pd(c12_00,VV);
276 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
277 fvdw12 = _mm_mul_pd(c12_00,FF);
278 vvdw = _mm_add_pd(vvdw12,vvdw6);
279 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
280
281 /* Update potential sum for this i atom from the interaction with this j atom. */
282 velecsum = _mm_add_pd(velecsum,velec);
283 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
284
285 fscal = _mm_add_pd(felec,fvdw);
286
287 /* Calculate temporary vectorial force */
288 tx = _mm_mul_pd(fscal,dx00);
289 ty = _mm_mul_pd(fscal,dy00);
290 tz = _mm_mul_pd(fscal,dz00);
291
292 /* Update vectorial force */
293 fix0 = _mm_add_pd(fix0,tx);
294 fiy0 = _mm_add_pd(fiy0,ty);
295 fiz0 = _mm_add_pd(fiz0,tz);
296
297 fjx0 = _mm_add_pd(fjx0,tx);
298 fjy0 = _mm_add_pd(fjy0,ty);
299 fjz0 = _mm_add_pd(fjz0,tz);
300
301 /**************************
302 * CALCULATE INTERACTIONS *
303 **************************/
304
305 r10 = _mm_mul_pd(rsq10,rinv10);
306
307 /* Compute parameters for interactions between i and j atoms */
308 qq10 = _mm_mul_pd(iq1,jq0);
309
310 /* Calculate table index by multiplying r with table scale and truncate to integer */
311 rt = _mm_mul_pd(r10,vftabscale);
312 vfitab = _mm_cvttpd_epi32(rt);
313 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
314 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
315
316 /* CUBIC SPLINE TABLE ELECTROSTATICS */
317 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
318 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
319 GMX_MM_TRANSPOSE2_PD(Y,F);
320 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
321 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
322 GMX_MM_TRANSPOSE2_PD(G,H);
323 Heps = _mm_mul_pd(vfeps,H);
324 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
325 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
326 velec = _mm_mul_pd(qq10,VV);
327 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
328 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
329
330 /* Update potential sum for this i atom from the interaction with this j atom. */
331 velecsum = _mm_add_pd(velecsum,velec);
332
333 fscal = felec;
334
335 /* Calculate temporary vectorial force */
336 tx = _mm_mul_pd(fscal,dx10);
337 ty = _mm_mul_pd(fscal,dy10);
338 tz = _mm_mul_pd(fscal,dz10);
339
340 /* Update vectorial force */
341 fix1 = _mm_add_pd(fix1,tx);
342 fiy1 = _mm_add_pd(fiy1,ty);
343 fiz1 = _mm_add_pd(fiz1,tz);
344
345 fjx0 = _mm_add_pd(fjx0,tx);
346 fjy0 = _mm_add_pd(fjy0,ty);
347 fjz0 = _mm_add_pd(fjz0,tz);
348
349 /**************************
350 * CALCULATE INTERACTIONS *
351 **************************/
352
353 r20 = _mm_mul_pd(rsq20,rinv20);
354
355 /* Compute parameters for interactions between i and j atoms */
356 qq20 = _mm_mul_pd(iq2,jq0);
357
358 /* Calculate table index by multiplying r with table scale and truncate to integer */
359 rt = _mm_mul_pd(r20,vftabscale);
360 vfitab = _mm_cvttpd_epi32(rt);
361 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
362 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
363
364 /* CUBIC SPLINE TABLE ELECTROSTATICS */
365 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
366 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
367 GMX_MM_TRANSPOSE2_PD(Y,F);
368 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
369 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
370 GMX_MM_TRANSPOSE2_PD(G,H);
371 Heps = _mm_mul_pd(vfeps,H);
372 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
373 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
374 velec = _mm_mul_pd(qq20,VV);
375 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
376 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
377
378 /* Update potential sum for this i atom from the interaction with this j atom. */
379 velecsum = _mm_add_pd(velecsum,velec);
380
381 fscal = felec;
382
383 /* Calculate temporary vectorial force */
384 tx = _mm_mul_pd(fscal,dx20);
385 ty = _mm_mul_pd(fscal,dy20);
386 tz = _mm_mul_pd(fscal,dz20);
387
388 /* Update vectorial force */
389 fix2 = _mm_add_pd(fix2,tx);
390 fiy2 = _mm_add_pd(fiy2,ty);
391 fiz2 = _mm_add_pd(fiz2,tz);
392
393 fjx0 = _mm_add_pd(fjx0,tx);
394 fjy0 = _mm_add_pd(fjy0,ty);
395 fjz0 = _mm_add_pd(fjz0,tz);
396
397 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
398
399 /* Inner loop uses 162 flops */
400 }
401
402 if(jidx<j_index_end)
403 {
404
405 jnrA = jjnr[jidx];
406 j_coord_offsetA = DIM*jnrA;
407
408 /* load j atom coordinates */
409 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
410 &jx0,&jy0,&jz0);
411
412 /* Calculate displacement vector */
413 dx00 = _mm_sub_pd(ix0,jx0);
414 dy00 = _mm_sub_pd(iy0,jy0);
415 dz00 = _mm_sub_pd(iz0,jz0);
416 dx10 = _mm_sub_pd(ix1,jx0);
417 dy10 = _mm_sub_pd(iy1,jy0);
418 dz10 = _mm_sub_pd(iz1,jz0);
419 dx20 = _mm_sub_pd(ix2,jx0);
420 dy20 = _mm_sub_pd(iy2,jy0);
421 dz20 = _mm_sub_pd(iz2,jz0);
422
423 /* Calculate squared distance and things based on it */
424 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
425 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
426 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
427
428 rinv00 = gmx_mm_invsqrt_pd(rsq00);
429 rinv10 = gmx_mm_invsqrt_pd(rsq10);
430 rinv20 = gmx_mm_invsqrt_pd(rsq20);
431
432 /* Load parameters for j particles */
433 jq0 = _mm_load_sd(charge+jnrA+0);
434 vdwjidx0A = 2*vdwtype[jnrA+0];
435
436 fjx0 = _mm_setzero_pd();
437 fjy0 = _mm_setzero_pd();
438 fjz0 = _mm_setzero_pd();
439
440 /**************************
441 * CALCULATE INTERACTIONS *
442 **************************/
443
444 r00 = _mm_mul_pd(rsq00,rinv00);
445
446 /* Compute parameters for interactions between i and j atoms */
447 qq00 = _mm_mul_pd(iq0,jq0);
448 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
449
450 /* Calculate table index by multiplying r with table scale and truncate to integer */
451 rt = _mm_mul_pd(r00,vftabscale);
452 vfitab = _mm_cvttpd_epi32(rt);
453 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
454 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
455
456 /* CUBIC SPLINE TABLE ELECTROSTATICS */
457 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
458 F = _mm_setzero_pd();
459 GMX_MM_TRANSPOSE2_PD(Y,F);
460 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
461 H = _mm_setzero_pd();
462 GMX_MM_TRANSPOSE2_PD(G,H);
463 Heps = _mm_mul_pd(vfeps,H);
464 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
465 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
466 velec = _mm_mul_pd(qq00,VV);
467 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
468 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
469
470 /* CUBIC SPLINE TABLE DISPERSION */
471 vfitab = _mm_add_epi32(vfitab,ifour);
472 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
473 F = _mm_setzero_pd();
474 GMX_MM_TRANSPOSE2_PD(Y,F);
475 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
476 H = _mm_setzero_pd();
477 GMX_MM_TRANSPOSE2_PD(G,H);
478 Heps = _mm_mul_pd(vfeps,H);
479 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
480 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
481 vvdw6 = _mm_mul_pd(c6_00,VV);
482 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
483 fvdw6 = _mm_mul_pd(c6_00,FF);
484
485 /* CUBIC SPLINE TABLE REPULSION */
486 vfitab = _mm_add_epi32(vfitab,ifour);
487 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
488 F = _mm_setzero_pd();
489 GMX_MM_TRANSPOSE2_PD(Y,F);
490 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
491 H = _mm_setzero_pd();
492 GMX_MM_TRANSPOSE2_PD(G,H);
493 Heps = _mm_mul_pd(vfeps,H);
494 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
495 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
496 vvdw12 = _mm_mul_pd(c12_00,VV);
497 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
498 fvdw12 = _mm_mul_pd(c12_00,FF);
499 vvdw = _mm_add_pd(vvdw12,vvdw6);
500 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
501
502 /* Update potential sum for this i atom from the interaction with this j atom. */
503 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
504 velecsum = _mm_add_pd(velecsum,velec);
505 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
506 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
507
508 fscal = _mm_add_pd(felec,fvdw);
509
510 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
511
512 /* Calculate temporary vectorial force */
513 tx = _mm_mul_pd(fscal,dx00);
514 ty = _mm_mul_pd(fscal,dy00);
515 tz = _mm_mul_pd(fscal,dz00);
516
517 /* Update vectorial force */
518 fix0 = _mm_add_pd(fix0,tx);
519 fiy0 = _mm_add_pd(fiy0,ty);
520 fiz0 = _mm_add_pd(fiz0,tz);
521
522 fjx0 = _mm_add_pd(fjx0,tx);
523 fjy0 = _mm_add_pd(fjy0,ty);
524 fjz0 = _mm_add_pd(fjz0,tz);
525
526 /**************************
527 * CALCULATE INTERACTIONS *
528 **************************/
529
530 r10 = _mm_mul_pd(rsq10,rinv10);
531
532 /* Compute parameters for interactions between i and j atoms */
533 qq10 = _mm_mul_pd(iq1,jq0);
534
535 /* Calculate table index by multiplying r with table scale and truncate to integer */
536 rt = _mm_mul_pd(r10,vftabscale);
537 vfitab = _mm_cvttpd_epi32(rt);
538 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
539 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
540
541 /* CUBIC SPLINE TABLE ELECTROSTATICS */
542 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
543 F = _mm_setzero_pd();
544 GMX_MM_TRANSPOSE2_PD(Y,F);
545 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
546 H = _mm_setzero_pd();
547 GMX_MM_TRANSPOSE2_PD(G,H);
548 Heps = _mm_mul_pd(vfeps,H);
549 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
550 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
551 velec = _mm_mul_pd(qq10,VV);
552 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
553 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
554
555 /* Update potential sum for this i atom from the interaction with this j atom. */
556 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
557 velecsum = _mm_add_pd(velecsum,velec);
558
559 fscal = felec;
560
561 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
562
563 /* Calculate temporary vectorial force */
564 tx = _mm_mul_pd(fscal,dx10);
565 ty = _mm_mul_pd(fscal,dy10);
566 tz = _mm_mul_pd(fscal,dz10);
567
568 /* Update vectorial force */
569 fix1 = _mm_add_pd(fix1,tx);
570 fiy1 = _mm_add_pd(fiy1,ty);
571 fiz1 = _mm_add_pd(fiz1,tz);
572
573 fjx0 = _mm_add_pd(fjx0,tx);
574 fjy0 = _mm_add_pd(fjy0,ty);
575 fjz0 = _mm_add_pd(fjz0,tz);
576
577 /**************************
578 * CALCULATE INTERACTIONS *
579 **************************/
580
581 r20 = _mm_mul_pd(rsq20,rinv20);
582
583 /* Compute parameters for interactions between i and j atoms */
584 qq20 = _mm_mul_pd(iq2,jq0);
585
586 /* Calculate table index by multiplying r with table scale and truncate to integer */
587 rt = _mm_mul_pd(r20,vftabscale);
588 vfitab = _mm_cvttpd_epi32(rt);
589 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
590 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
591
592 /* CUBIC SPLINE TABLE ELECTROSTATICS */
593 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
594 F = _mm_setzero_pd();
595 GMX_MM_TRANSPOSE2_PD(Y,F);
596 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
597 H = _mm_setzero_pd();
598 GMX_MM_TRANSPOSE2_PD(G,H);
599 Heps = _mm_mul_pd(vfeps,H);
600 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
601 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
602 velec = _mm_mul_pd(qq20,VV);
603 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
604 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
605
606 /* Update potential sum for this i atom from the interaction with this j atom. */
607 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
608 velecsum = _mm_add_pd(velecsum,velec);
609
610 fscal = felec;
611
612 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
613
614 /* Calculate temporary vectorial force */
615 tx = _mm_mul_pd(fscal,dx20);
616 ty = _mm_mul_pd(fscal,dy20);
617 tz = _mm_mul_pd(fscal,dz20);
618
619 /* Update vectorial force */
620 fix2 = _mm_add_pd(fix2,tx);
621 fiy2 = _mm_add_pd(fiy2,ty);
622 fiz2 = _mm_add_pd(fiz2,tz);
623
624 fjx0 = _mm_add_pd(fjx0,tx);
625 fjy0 = _mm_add_pd(fjy0,ty);
626 fjz0 = _mm_add_pd(fjz0,tz);
627
628 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
629
630 /* Inner loop uses 162 flops */
631 }
632
633 /* End of innermost loop */
634
635 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
636 f+i_coord_offset,fshift+i_shift_offset);
637
638 ggid = gid[iidx];
639 /* Update potential energies */
640 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
641 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
642
643 /* Increment number of inner iterations */
644 inneriter += j_index_end - j_index_start;
645
646 /* Outer loop uses 20 flops */
647 }
648
649 /* Increment number of outer iterations */
650 outeriter += nri;
651
652 /* Update outer/inner flops */
653
654 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*162);
655 }
656 /*
657 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse2_double
658 * Electrostatics interaction: CubicSplineTable
659 * VdW interaction: CubicSplineTable
660 * Geometry: Water3-Particle
661 * Calculate force/pot: Force
662 */
663 void
664 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse2_double
665 (t_nblist * gmx_restrict nlist,
666 rvec * gmx_restrict xx,
667 rvec * gmx_restrict ff,
668 t_forcerec * gmx_restrict fr,
669 t_mdatoms * gmx_restrict mdatoms,
670 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
671 t_nrnb * gmx_restrict nrnb)
672 {
673 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
674 * just 0 for non-waters.
675 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
676 * jnr indices corresponding to data put in the four positions in the SIMD register.
677 */
678 int i_shift_offset,i_coord_offset,outeriter,inneriter;
679 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
680 int jnrA,jnrB;
681 int j_coord_offsetA,j_coord_offsetB;
682 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
683 real rcutoff_scalar;
684 real *shiftvec,*fshift,*x,*f;
685 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
686 int vdwioffset0;
687 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
688 int vdwioffset1;
689 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
690 int vdwioffset2;
691 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
692 int vdwjidx0A,vdwjidx0B;
693 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
694 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
695 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
696 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
697 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
698 real *charge;
699 int nvdwtype;
700 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
701 int *vdwtype;
702 real *vdwparam;
703 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
704 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
705 __m128i vfitab;
706 __m128i ifour = _mm_set1_epi32(4);
707 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
708 real *vftab;
709 __m128d dummy_mask,cutoff_mask;
710 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
711 __m128d one = _mm_set1_pd(1.0);
712 __m128d two = _mm_set1_pd(2.0);
713 x = xx[0];
714 f = ff[0];
715
716 nri = nlist->nri;
717 iinr = nlist->iinr;
718 jindex = nlist->jindex;
719 jjnr = nlist->jjnr;
720 shiftidx = nlist->shift;
721 gid = nlist->gid;
722 shiftvec = fr->shift_vec[0];
723 fshift = fr->fshift[0];
724 facel = _mm_set1_pd(fr->epsfac);
725 charge = mdatoms->chargeA;
726 nvdwtype = fr->ntype;
727 vdwparam = fr->nbfp;
728 vdwtype = mdatoms->typeA;
729
730 vftab = kernel_data->table_elec_vdw->data;
731 vftabscale = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
732
733 /* Setup water-specific parameters */
734 inr = nlist->iinr[0];
735 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
736 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
737 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
738 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
739
740 /* Avoid stupid compiler warnings */
741 jnrA = jnrB = 0;
742 j_coord_offsetA = 0;
743 j_coord_offsetB = 0;
744
745 outeriter = 0;
746 inneriter = 0;
747
748 /* Start outer loop over neighborlists */
749 for(iidx=0; iidx<nri; iidx++)
750 {
751 /* Load shift vector for this list */
752 i_shift_offset = DIM*shiftidx[iidx];
753
754 /* Load limits for loop over neighbors */
755 j_index_start = jindex[iidx];
756 j_index_end = jindex[iidx+1];
757
758 /* Get outer coordinate index */
759 inr = iinr[iidx];
760 i_coord_offset = DIM*inr;
761
762 /* Load i particle coords and add shift vector */
763 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
764 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
765
766 fix0 = _mm_setzero_pd();
767 fiy0 = _mm_setzero_pd();
768 fiz0 = _mm_setzero_pd();
769 fix1 = _mm_setzero_pd();
770 fiy1 = _mm_setzero_pd();
771 fiz1 = _mm_setzero_pd();
772 fix2 = _mm_setzero_pd();
773 fiy2 = _mm_setzero_pd();
774 fiz2 = _mm_setzero_pd();
775
776 /* Start inner kernel loop */
777 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
778 {
779
780 /* Get j neighbor index, and coordinate index */
781 jnrA = jjnr[jidx];
782 jnrB = jjnr[jidx+1];
783 j_coord_offsetA = DIM*jnrA;
784 j_coord_offsetB = DIM*jnrB;
785
786 /* load j atom coordinates */
787 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
788 &jx0,&jy0,&jz0);
789
790 /* Calculate displacement vector */
791 dx00 = _mm_sub_pd(ix0,jx0);
792 dy00 = _mm_sub_pd(iy0,jy0);
793 dz00 = _mm_sub_pd(iz0,jz0);
794 dx10 = _mm_sub_pd(ix1,jx0);
795 dy10 = _mm_sub_pd(iy1,jy0);
796 dz10 = _mm_sub_pd(iz1,jz0);
797 dx20 = _mm_sub_pd(ix2,jx0);
798 dy20 = _mm_sub_pd(iy2,jy0);
799 dz20 = _mm_sub_pd(iz2,jz0);
800
801 /* Calculate squared distance and things based on it */
802 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
803 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
804 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
805
806 rinv00 = gmx_mm_invsqrt_pd(rsq00);
807 rinv10 = gmx_mm_invsqrt_pd(rsq10);
808 rinv20 = gmx_mm_invsqrt_pd(rsq20);
809
810 /* Load parameters for j particles */
811 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
812 vdwjidx0A = 2*vdwtype[jnrA+0];
813 vdwjidx0B = 2*vdwtype[jnrB+0];
814
815 fjx0 = _mm_setzero_pd();
816 fjy0 = _mm_setzero_pd();
817 fjz0 = _mm_setzero_pd();
818
819 /**************************
820 * CALCULATE INTERACTIONS *
821 **************************/
822
823 r00 = _mm_mul_pd(rsq00,rinv00);
824
825 /* Compute parameters for interactions between i and j atoms */
826 qq00 = _mm_mul_pd(iq0,jq0);
827 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
828 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
829
830 /* Calculate table index by multiplying r with table scale and truncate to integer */
831 rt = _mm_mul_pd(r00,vftabscale);
832 vfitab = _mm_cvttpd_epi32(rt);
833 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
834 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
835
836 /* CUBIC SPLINE TABLE ELECTROSTATICS */
837 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
838 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
839 GMX_MM_TRANSPOSE2_PD(Y,F);
840 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
841 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
842 GMX_MM_TRANSPOSE2_PD(G,H);
843 Heps = _mm_mul_pd(vfeps,H);
844 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
845 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
846 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
847
848 /* CUBIC SPLINE TABLE DISPERSION */
849 vfitab = _mm_add_epi32(vfitab,ifour);
850 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
851 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
852 GMX_MM_TRANSPOSE2_PD(Y,F);
853 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
854 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
855 GMX_MM_TRANSPOSE2_PD(G,H);
856 Heps = _mm_mul_pd(vfeps,H);
857 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
858 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
859 fvdw6 = _mm_mul_pd(c6_00,FF);
860
861 /* CUBIC SPLINE TABLE REPULSION */
862 vfitab = _mm_add_epi32(vfitab,ifour);
863 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
864 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
865 GMX_MM_TRANSPOSE2_PD(Y,F);
866 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
867 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
868 GMX_MM_TRANSPOSE2_PD(G,H);
869 Heps = _mm_mul_pd(vfeps,H);
870 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
871 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
872 fvdw12 = _mm_mul_pd(c12_00,FF);
873 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
874
875 fscal = _mm_add_pd(felec,fvdw);
876
877 /* Calculate temporary vectorial force */
878 tx = _mm_mul_pd(fscal,dx00);
879 ty = _mm_mul_pd(fscal,dy00);
880 tz = _mm_mul_pd(fscal,dz00);
881
882 /* Update vectorial force */
883 fix0 = _mm_add_pd(fix0,tx);
884 fiy0 = _mm_add_pd(fiy0,ty);
885 fiz0 = _mm_add_pd(fiz0,tz);
886
887 fjx0 = _mm_add_pd(fjx0,tx);
888 fjy0 = _mm_add_pd(fjy0,ty);
889 fjz0 = _mm_add_pd(fjz0,tz);
890
891 /**************************
892 * CALCULATE INTERACTIONS *
893 **************************/
894
895 r10 = _mm_mul_pd(rsq10,rinv10);
896
897 /* Compute parameters for interactions between i and j atoms */
898 qq10 = _mm_mul_pd(iq1,jq0);
899
900 /* Calculate table index by multiplying r with table scale and truncate to integer */
901 rt = _mm_mul_pd(r10,vftabscale);
902 vfitab = _mm_cvttpd_epi32(rt);
903 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
904 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
905
906 /* CUBIC SPLINE TABLE ELECTROSTATICS */
907 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
908 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
909 GMX_MM_TRANSPOSE2_PD(Y,F);
910 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
911 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
912 GMX_MM_TRANSPOSE2_PD(G,H);
913 Heps = _mm_mul_pd(vfeps,H);
914 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
915 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
916 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
917
918 fscal = felec;
919
920 /* Calculate temporary vectorial force */
921 tx = _mm_mul_pd(fscal,dx10);
922 ty = _mm_mul_pd(fscal,dy10);
923 tz = _mm_mul_pd(fscal,dz10);
924
925 /* Update vectorial force */
926 fix1 = _mm_add_pd(fix1,tx);
927 fiy1 = _mm_add_pd(fiy1,ty);
928 fiz1 = _mm_add_pd(fiz1,tz);
929
930 fjx0 = _mm_add_pd(fjx0,tx);
931 fjy0 = _mm_add_pd(fjy0,ty);
932 fjz0 = _mm_add_pd(fjz0,tz);
933
934 /**************************
935 * CALCULATE INTERACTIONS *
936 **************************/
937
938 r20 = _mm_mul_pd(rsq20,rinv20);
939
940 /* Compute parameters for interactions between i and j atoms */
941 qq20 = _mm_mul_pd(iq2,jq0);
942
943 /* Calculate table index by multiplying r with table scale and truncate to integer */
944 rt = _mm_mul_pd(r20,vftabscale);
945 vfitab = _mm_cvttpd_epi32(rt);
946 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
947 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
948
949 /* CUBIC SPLINE TABLE ELECTROSTATICS */
950 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
951 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
952 GMX_MM_TRANSPOSE2_PD(Y,F);
953 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
954 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
955 GMX_MM_TRANSPOSE2_PD(G,H);
956 Heps = _mm_mul_pd(vfeps,H);
957 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
958 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
959 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
960
961 fscal = felec;
962
963 /* Calculate temporary vectorial force */
964 tx = _mm_mul_pd(fscal,dx20);
965 ty = _mm_mul_pd(fscal,dy20);
966 tz = _mm_mul_pd(fscal,dz20);
967
968 /* Update vectorial force */
969 fix2 = _mm_add_pd(fix2,tx);
970 fiy2 = _mm_add_pd(fiy2,ty);
971 fiz2 = _mm_add_pd(fiz2,tz);
972
973 fjx0 = _mm_add_pd(fjx0,tx);
974 fjy0 = _mm_add_pd(fjy0,ty);
975 fjz0 = _mm_add_pd(fjz0,tz);
976
977 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
978
979 /* Inner loop uses 142 flops */
980 }
981
982 if(jidx<j_index_end)
983 {
984
985 jnrA = jjnr[jidx];
986 j_coord_offsetA = DIM*jnrA;
987
988 /* load j atom coordinates */
989 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
990 &jx0,&jy0,&jz0);
991
992 /* Calculate displacement vector */
993 dx00 = _mm_sub_pd(ix0,jx0);
994 dy00 = _mm_sub_pd(iy0,jy0);
995 dz00 = _mm_sub_pd(iz0,jz0);
996 dx10 = _mm_sub_pd(ix1,jx0);
997 dy10 = _mm_sub_pd(iy1,jy0);
998 dz10 = _mm_sub_pd(iz1,jz0);
999 dx20 = _mm_sub_pd(ix2,jx0);
1000 dy20 = _mm_sub_pd(iy2,jy0);
1001 dz20 = _mm_sub_pd(iz2,jz0);
1002
1003 /* Calculate squared distance and things based on it */
1004 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1005 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1006 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1007
1008 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1009 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1010 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1011
1012 /* Load parameters for j particles */
1013 jq0 = _mm_load_sd(charge+jnrA+0);
1014 vdwjidx0A = 2*vdwtype[jnrA+0];
1015
1016 fjx0 = _mm_setzero_pd();
1017 fjy0 = _mm_setzero_pd();
1018 fjz0 = _mm_setzero_pd();
1019
1020 /**************************
1021 * CALCULATE INTERACTIONS *
1022 **************************/
1023
1024 r00 = _mm_mul_pd(rsq00,rinv00);
1025
1026 /* Compute parameters for interactions between i and j atoms */
1027 qq00 = _mm_mul_pd(iq0,jq0);
1028 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1029
1030 /* Calculate table index by multiplying r with table scale and truncate to integer */
1031 rt = _mm_mul_pd(r00,vftabscale);
1032 vfitab = _mm_cvttpd_epi32(rt);
1033 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1034 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1035
1036 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1037 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1038 F = _mm_setzero_pd();
1039 GMX_MM_TRANSPOSE2_PD(Y,F);
1040 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1041 H = _mm_setzero_pd();
1042 GMX_MM_TRANSPOSE2_PD(G,H);
1043 Heps = _mm_mul_pd(vfeps,H);
1044 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1045 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1046 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1047
1048 /* CUBIC SPLINE TABLE DISPERSION */
1049 vfitab = _mm_add_epi32(vfitab,ifour);
1050 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1051 F = _mm_setzero_pd();
1052 GMX_MM_TRANSPOSE2_PD(Y,F);
1053 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1054 H = _mm_setzero_pd();
1055 GMX_MM_TRANSPOSE2_PD(G,H);
1056 Heps = _mm_mul_pd(vfeps,H);
1057 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1058 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1059 fvdw6 = _mm_mul_pd(c6_00,FF);
1060
1061 /* CUBIC SPLINE TABLE REPULSION */
1062 vfitab = _mm_add_epi32(vfitab,ifour);
1063 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1064 F = _mm_setzero_pd();
1065 GMX_MM_TRANSPOSE2_PD(Y,F);
1066 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1067 H = _mm_setzero_pd();
1068 GMX_MM_TRANSPOSE2_PD(G,H);
1069 Heps = _mm_mul_pd(vfeps,H);
1070 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1071 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1072 fvdw12 = _mm_mul_pd(c12_00,FF);
1073 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1074
1075 fscal = _mm_add_pd(felec,fvdw);
1076
1077 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1078
1079 /* Calculate temporary vectorial force */
1080 tx = _mm_mul_pd(fscal,dx00);
1081 ty = _mm_mul_pd(fscal,dy00);
1082 tz = _mm_mul_pd(fscal,dz00);
1083
1084 /* Update vectorial force */
1085 fix0 = _mm_add_pd(fix0,tx);
1086 fiy0 = _mm_add_pd(fiy0,ty);
1087 fiz0 = _mm_add_pd(fiz0,tz);
1088
1089 fjx0 = _mm_add_pd(fjx0,tx);
1090 fjy0 = _mm_add_pd(fjy0,ty);
1091 fjz0 = _mm_add_pd(fjz0,tz);
1092
1093 /**************************
1094 * CALCULATE INTERACTIONS *
1095 **************************/
1096
1097 r10 = _mm_mul_pd(rsq10,rinv10);
1098
1099 /* Compute parameters for interactions between i and j atoms */
1100 qq10 = _mm_mul_pd(iq1,jq0);
1101
1102 /* Calculate table index by multiplying r with table scale and truncate to integer */
1103 rt = _mm_mul_pd(r10,vftabscale);
1104 vfitab = _mm_cvttpd_epi32(rt);
1105 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1106 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1107
1108 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1109 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1110 F = _mm_setzero_pd();
1111 GMX_MM_TRANSPOSE2_PD(Y,F);
1112 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1113 H = _mm_setzero_pd();
1114 GMX_MM_TRANSPOSE2_PD(G,H);
1115 Heps = _mm_mul_pd(vfeps,H);
1116 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1117 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1118 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1119
1120 fscal = felec;
1121
1122 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1123
1124 /* Calculate temporary vectorial force */
1125 tx = _mm_mul_pd(fscal,dx10);
1126 ty = _mm_mul_pd(fscal,dy10);
1127 tz = _mm_mul_pd(fscal,dz10);
1128
1129 /* Update vectorial force */
1130 fix1 = _mm_add_pd(fix1,tx);
1131 fiy1 = _mm_add_pd(fiy1,ty);
1132 fiz1 = _mm_add_pd(fiz1,tz);
1133
1134 fjx0 = _mm_add_pd(fjx0,tx);
1135 fjy0 = _mm_add_pd(fjy0,ty);
1136 fjz0 = _mm_add_pd(fjz0,tz);
1137
1138 /**************************
1139 * CALCULATE INTERACTIONS *
1140 **************************/
1141
1142 r20 = _mm_mul_pd(rsq20,rinv20);
1143
1144 /* Compute parameters for interactions between i and j atoms */
1145 qq20 = _mm_mul_pd(iq2,jq0);
1146
1147 /* Calculate table index by multiplying r with table scale and truncate to integer */
1148 rt = _mm_mul_pd(r20,vftabscale);
1149 vfitab = _mm_cvttpd_epi32(rt);
1150 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1151 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1152
1153 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1154 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1155 F = _mm_setzero_pd();
1156 GMX_MM_TRANSPOSE2_PD(Y,F);
1157 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1158 H = _mm_setzero_pd();
1159 GMX_MM_TRANSPOSE2_PD(G,H);
1160 Heps = _mm_mul_pd(vfeps,H);
1161 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1162 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1163 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1164
1165 fscal = felec;
1166
1167 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1168
1169 /* Calculate temporary vectorial force */
1170 tx = _mm_mul_pd(fscal,dx20);
1171 ty = _mm_mul_pd(fscal,dy20);
1172 tz = _mm_mul_pd(fscal,dz20);
1173
1174 /* Update vectorial force */
1175 fix2 = _mm_add_pd(fix2,tx);
1176 fiy2 = _mm_add_pd(fiy2,ty);
1177 fiz2 = _mm_add_pd(fiz2,tz);
1178
1179 fjx0 = _mm_add_pd(fjx0,tx);
1180 fjy0 = _mm_add_pd(fjy0,ty);
1181 fjz0 = _mm_add_pd(fjz0,tz);
1182
1183 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1184
1185 /* Inner loop uses 142 flops */
1186 }
1187
1188 /* End of innermost loop */
1189
1190 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1191 f+i_coord_offset,fshift+i_shift_offset);
1192
1193 /* Increment number of inner iterations */
1194 inneriter += j_index_end - j_index_start;
1195
1196 /* Outer loop uses 18 flops */
1197 }
1198
1199 /* Increment number of outer iterations */
1200 outeriter += nri;
1201
1202 /* Update outer/inner flops */
1203
1204 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*142);
1205 }
The ultimate molecular dynamics simulation package