?  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_GeomP1P1_sse2_double.c
blobbd9a5fa0497dcb31e73f481d5fc072426aeca5bd
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_GeomP1P1_VF_sse2_double
53 * Electrostatics interaction: CubicSplineTable
54 * VdW interaction: CubicSplineTable
55 * Geometry: Particle-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_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 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 vfitab;
95 __m128i ifour = _mm_set1_epi32(4);
96 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
97 real *vftab;
98 __m128d dummy_mask,cutoff_mask;
99 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
100 __m128d one = _mm_set1_pd(1.0);
101 __m128d two = _mm_set1_pd(2.0);
102 x = xx[0];
103 f = ff[0];
104
105 nri = nlist->nri;
106 iinr = nlist->iinr;
107 jindex = nlist->jindex;
108 jjnr = nlist->jjnr;
109 shiftidx = nlist->shift;
110 gid = nlist->gid;
111 shiftvec = fr->shift_vec[0];
112 fshift = fr->fshift[0];
113 facel = _mm_set1_pd(fr->epsfac);
114 charge = mdatoms->chargeA;
115 nvdwtype = fr->ntype;
116 vdwparam = fr->nbfp;
117 vdwtype = mdatoms->typeA;
118
119 vftab = kernel_data->table_elec_vdw->data;
120 vftabscale = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
121
122 /* Avoid stupid compiler warnings */
123 jnrA = jnrB = 0;
124 j_coord_offsetA = 0;
125 j_coord_offsetB = 0;
126
127 outeriter = 0;
128 inneriter = 0;
129
130 /* Start outer loop over neighborlists */
131 for(iidx=0; iidx<nri; iidx++)
132 {
133 /* Load shift vector for this list */
134 i_shift_offset = DIM*shiftidx[iidx];
135
136 /* Load limits for loop over neighbors */
137 j_index_start = jindex[iidx];
138 j_index_end = jindex[iidx+1];
139
140 /* Get outer coordinate index */
141 inr = iinr[iidx];
142 i_coord_offset = DIM*inr;
143
144 /* Load i particle coords and add shift vector */
145 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
146
147 fix0 = _mm_setzero_pd();
148 fiy0 = _mm_setzero_pd();
149 fiz0 = _mm_setzero_pd();
150
151 /* Load parameters for i particles */
152 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
153 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
154
155 /* Reset potential sums */
156 velecsum = _mm_setzero_pd();
157 vvdwsum = _mm_setzero_pd();
158
159 /* Start inner kernel loop */
160 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
161 {
162
163 /* Get j neighbor index, and coordinate index */
164 jnrA = jjnr[jidx];
165 jnrB = jjnr[jidx+1];
166 j_coord_offsetA = DIM*jnrA;
167 j_coord_offsetB = DIM*jnrB;
168
169 /* load j atom coordinates */
170 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
171 &jx0,&jy0,&jz0);
172
173 /* Calculate displacement vector */
174 dx00 = _mm_sub_pd(ix0,jx0);
175 dy00 = _mm_sub_pd(iy0,jy0);
176 dz00 = _mm_sub_pd(iz0,jz0);
177
178 /* Calculate squared distance and things based on it */
179 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
180
181 rinv00 = gmx_mm_invsqrt_pd(rsq00);
182
183 /* Load parameters for j particles */
184 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
185 vdwjidx0A = 2*vdwtype[jnrA+0];
186 vdwjidx0B = 2*vdwtype[jnrB+0];
187
188 /**************************
189 * CALCULATE INTERACTIONS *
190 **************************/
191
192 r00 = _mm_mul_pd(rsq00,rinv00);
193
194 /* Compute parameters for interactions between i and j atoms */
195 qq00 = _mm_mul_pd(iq0,jq0);
196 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
197 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
198
199 /* Calculate table index by multiplying r with table scale and truncate to integer */
200 rt = _mm_mul_pd(r00,vftabscale);
201 vfitab = _mm_cvttpd_epi32(rt);
202 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
203 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
204
205 /* CUBIC SPLINE TABLE ELECTROSTATICS */
206 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
207 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
208 GMX_MM_TRANSPOSE2_PD(Y,F);
209 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
210 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
211 GMX_MM_TRANSPOSE2_PD(G,H);
212 Heps = _mm_mul_pd(vfeps,H);
213 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
214 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
215 velec = _mm_mul_pd(qq00,VV);
216 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
217 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
218
219 /* CUBIC SPLINE TABLE DISPERSION */
220 vfitab = _mm_add_epi32(vfitab,ifour);
221 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
222 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
223 GMX_MM_TRANSPOSE2_PD(Y,F);
224 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
225 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
226 GMX_MM_TRANSPOSE2_PD(G,H);
227 Heps = _mm_mul_pd(vfeps,H);
228 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
229 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
230 vvdw6 = _mm_mul_pd(c6_00,VV);
231 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
232 fvdw6 = _mm_mul_pd(c6_00,FF);
233
234 /* CUBIC SPLINE TABLE REPULSION */
235 vfitab = _mm_add_epi32(vfitab,ifour);
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 vvdw12 = _mm_mul_pd(c12_00,VV);
246 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
247 fvdw12 = _mm_mul_pd(c12_00,FF);
248 vvdw = _mm_add_pd(vvdw12,vvdw6);
249 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
250
251 /* Update potential sum for this i atom from the interaction with this j atom. */
252 velecsum = _mm_add_pd(velecsum,velec);
253 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
254
255 fscal = _mm_add_pd(felec,fvdw);
256
257 /* Calculate temporary vectorial force */
258 tx = _mm_mul_pd(fscal,dx00);
259 ty = _mm_mul_pd(fscal,dy00);
260 tz = _mm_mul_pd(fscal,dz00);
261
262 /* Update vectorial force */
263 fix0 = _mm_add_pd(fix0,tx);
264 fiy0 = _mm_add_pd(fiy0,ty);
265 fiz0 = _mm_add_pd(fiz0,tz);
266
267 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
268
269 /* Inner loop uses 73 flops */
270 }
271
272 if(jidx<j_index_end)
273 {
274
275 jnrA = jjnr[jidx];
276 j_coord_offsetA = DIM*jnrA;
277
278 /* load j atom coordinates */
279 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
280 &jx0,&jy0,&jz0);
281
282 /* Calculate displacement vector */
283 dx00 = _mm_sub_pd(ix0,jx0);
284 dy00 = _mm_sub_pd(iy0,jy0);
285 dz00 = _mm_sub_pd(iz0,jz0);
286
287 /* Calculate squared distance and things based on it */
288 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
289
290 rinv00 = gmx_mm_invsqrt_pd(rsq00);
291
292 /* Load parameters for j particles */
293 jq0 = _mm_load_sd(charge+jnrA+0);
294 vdwjidx0A = 2*vdwtype[jnrA+0];
295
296 /**************************
297 * CALCULATE INTERACTIONS *
298 **************************/
299
300 r00 = _mm_mul_pd(rsq00,rinv00);
301
302 /* Compute parameters for interactions between i and j atoms */
303 qq00 = _mm_mul_pd(iq0,jq0);
304 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
305
306 /* Calculate table index by multiplying r with table scale and truncate to integer */
307 rt = _mm_mul_pd(r00,vftabscale);
308 vfitab = _mm_cvttpd_epi32(rt);
309 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
310 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
311
312 /* CUBIC SPLINE TABLE ELECTROSTATICS */
313 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
314 F = _mm_setzero_pd();
315 GMX_MM_TRANSPOSE2_PD(Y,F);
316 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
317 H = _mm_setzero_pd();
318 GMX_MM_TRANSPOSE2_PD(G,H);
319 Heps = _mm_mul_pd(vfeps,H);
320 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
321 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
322 velec = _mm_mul_pd(qq00,VV);
323 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
324 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
325
326 /* CUBIC SPLINE TABLE DISPERSION */
327 vfitab = _mm_add_epi32(vfitab,ifour);
328 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
329 F = _mm_setzero_pd();
330 GMX_MM_TRANSPOSE2_PD(Y,F);
331 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
332 H = _mm_setzero_pd();
333 GMX_MM_TRANSPOSE2_PD(G,H);
334 Heps = _mm_mul_pd(vfeps,H);
335 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
336 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
337 vvdw6 = _mm_mul_pd(c6_00,VV);
338 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
339 fvdw6 = _mm_mul_pd(c6_00,FF);
340
341 /* CUBIC SPLINE TABLE REPULSION */
342 vfitab = _mm_add_epi32(vfitab,ifour);
343 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
344 F = _mm_setzero_pd();
345 GMX_MM_TRANSPOSE2_PD(Y,F);
346 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
347 H = _mm_setzero_pd();
348 GMX_MM_TRANSPOSE2_PD(G,H);
349 Heps = _mm_mul_pd(vfeps,H);
350 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
351 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
352 vvdw12 = _mm_mul_pd(c12_00,VV);
353 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
354 fvdw12 = _mm_mul_pd(c12_00,FF);
355 vvdw = _mm_add_pd(vvdw12,vvdw6);
356 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
357
358 /* Update potential sum for this i atom from the interaction with this j atom. */
359 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
360 velecsum = _mm_add_pd(velecsum,velec);
361 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
362 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
363
364 fscal = _mm_add_pd(felec,fvdw);
365
366 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
367
368 /* Calculate temporary vectorial force */
369 tx = _mm_mul_pd(fscal,dx00);
370 ty = _mm_mul_pd(fscal,dy00);
371 tz = _mm_mul_pd(fscal,dz00);
372
373 /* Update vectorial force */
374 fix0 = _mm_add_pd(fix0,tx);
375 fiy0 = _mm_add_pd(fiy0,ty);
376 fiz0 = _mm_add_pd(fiz0,tz);
377
378 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
379
380 /* Inner loop uses 73 flops */
381 }
382
383 /* End of innermost loop */
384
385 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
386 f+i_coord_offset,fshift+i_shift_offset);
387
388 ggid = gid[iidx];
389 /* Update potential energies */
390 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
391 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
392
393 /* Increment number of inner iterations */
394 inneriter += j_index_end - j_index_start;
395
396 /* Outer loop uses 9 flops */
397 }
398
399 /* Increment number of outer iterations */
400 outeriter += nri;
401
402 /* Update outer/inner flops */
403
404 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
405 }
406 /*
407 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sse2_double
408 * Electrostatics interaction: CubicSplineTable
409 * VdW interaction: CubicSplineTable
410 * Geometry: Particle-Particle
411 * Calculate force/pot: Force
412 */
413 void
414 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sse2_double
415 (t_nblist * gmx_restrict nlist,
416 rvec * gmx_restrict xx,
417 rvec * gmx_restrict ff,
418 t_forcerec * gmx_restrict fr,
419 t_mdatoms * gmx_restrict mdatoms,
420 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
421 t_nrnb * gmx_restrict nrnb)
422 {
423 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
424 * just 0 for non-waters.
425 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
426 * jnr indices corresponding to data put in the four positions in the SIMD register.
427 */
428 int i_shift_offset,i_coord_offset,outeriter,inneriter;
429 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
430 int jnrA,jnrB;
431 int j_coord_offsetA,j_coord_offsetB;
432 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
433 real rcutoff_scalar;
434 real *shiftvec,*fshift,*x,*f;
435 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
436 int vdwioffset0;
437 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
438 int vdwjidx0A,vdwjidx0B;
439 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
440 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
441 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
442 real *charge;
443 int nvdwtype;
444 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
445 int *vdwtype;
446 real *vdwparam;
447 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
448 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
449 __m128i vfitab;
450 __m128i ifour = _mm_set1_epi32(4);
451 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
452 real *vftab;
453 __m128d dummy_mask,cutoff_mask;
454 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
455 __m128d one = _mm_set1_pd(1.0);
456 __m128d two = _mm_set1_pd(2.0);
457 x = xx[0];
458 f = ff[0];
459
460 nri = nlist->nri;
461 iinr = nlist->iinr;
462 jindex = nlist->jindex;
463 jjnr = nlist->jjnr;
464 shiftidx = nlist->shift;
465 gid = nlist->gid;
466 shiftvec = fr->shift_vec[0];
467 fshift = fr->fshift[0];
468 facel = _mm_set1_pd(fr->epsfac);
469 charge = mdatoms->chargeA;
470 nvdwtype = fr->ntype;
471 vdwparam = fr->nbfp;
472 vdwtype = mdatoms->typeA;
473
474 vftab = kernel_data->table_elec_vdw->data;
475 vftabscale = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
476
477 /* Avoid stupid compiler warnings */
478 jnrA = jnrB = 0;
479 j_coord_offsetA = 0;
480 j_coord_offsetB = 0;
481
482 outeriter = 0;
483 inneriter = 0;
484
485 /* Start outer loop over neighborlists */
486 for(iidx=0; iidx<nri; iidx++)
487 {
488 /* Load shift vector for this list */
489 i_shift_offset = DIM*shiftidx[iidx];
490
491 /* Load limits for loop over neighbors */
492 j_index_start = jindex[iidx];
493 j_index_end = jindex[iidx+1];
494
495 /* Get outer coordinate index */
496 inr = iinr[iidx];
497 i_coord_offset = DIM*inr;
498
499 /* Load i particle coords and add shift vector */
500 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
501
502 fix0 = _mm_setzero_pd();
503 fiy0 = _mm_setzero_pd();
504 fiz0 = _mm_setzero_pd();
505
506 /* Load parameters for i particles */
507 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
508 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
509
510 /* Start inner kernel loop */
511 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
512 {
513
514 /* Get j neighbor index, and coordinate index */
515 jnrA = jjnr[jidx];
516 jnrB = jjnr[jidx+1];
517 j_coord_offsetA = DIM*jnrA;
518 j_coord_offsetB = DIM*jnrB;
519
520 /* load j atom coordinates */
521 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
522 &jx0,&jy0,&jz0);
523
524 /* Calculate displacement vector */
525 dx00 = _mm_sub_pd(ix0,jx0);
526 dy00 = _mm_sub_pd(iy0,jy0);
527 dz00 = _mm_sub_pd(iz0,jz0);
528
529 /* Calculate squared distance and things based on it */
530 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
531
532 rinv00 = gmx_mm_invsqrt_pd(rsq00);
533
534 /* Load parameters for j particles */
535 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
536 vdwjidx0A = 2*vdwtype[jnrA+0];
537 vdwjidx0B = 2*vdwtype[jnrB+0];
538
539 /**************************
540 * CALCULATE INTERACTIONS *
541 **************************/
542
543 r00 = _mm_mul_pd(rsq00,rinv00);
544
545 /* Compute parameters for interactions between i and j atoms */
546 qq00 = _mm_mul_pd(iq0,jq0);
547 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
548 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
549
550 /* Calculate table index by multiplying r with table scale and truncate to integer */
551 rt = _mm_mul_pd(r00,vftabscale);
552 vfitab = _mm_cvttpd_epi32(rt);
553 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
554 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
555
556 /* CUBIC SPLINE TABLE ELECTROSTATICS */
557 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
558 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
559 GMX_MM_TRANSPOSE2_PD(Y,F);
560 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
561 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
562 GMX_MM_TRANSPOSE2_PD(G,H);
563 Heps = _mm_mul_pd(vfeps,H);
564 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
565 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
566 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
567
568 /* CUBIC SPLINE TABLE DISPERSION */
569 vfitab = _mm_add_epi32(vfitab,ifour);
570 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
571 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
572 GMX_MM_TRANSPOSE2_PD(Y,F);
573 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
574 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
575 GMX_MM_TRANSPOSE2_PD(G,H);
576 Heps = _mm_mul_pd(vfeps,H);
577 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
578 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
579 fvdw6 = _mm_mul_pd(c6_00,FF);
580
581 /* CUBIC SPLINE TABLE REPULSION */
582 vfitab = _mm_add_epi32(vfitab,ifour);
583 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
584 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
585 GMX_MM_TRANSPOSE2_PD(Y,F);
586 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
587 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
588 GMX_MM_TRANSPOSE2_PD(G,H);
589 Heps = _mm_mul_pd(vfeps,H);
590 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
591 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
592 fvdw12 = _mm_mul_pd(c12_00,FF);
593 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
594
595 fscal = _mm_add_pd(felec,fvdw);
596
597 /* Calculate temporary vectorial force */
598 tx = _mm_mul_pd(fscal,dx00);
599 ty = _mm_mul_pd(fscal,dy00);
600 tz = _mm_mul_pd(fscal,dz00);
601
602 /* Update vectorial force */
603 fix0 = _mm_add_pd(fix0,tx);
604 fiy0 = _mm_add_pd(fiy0,ty);
605 fiz0 = _mm_add_pd(fiz0,tz);
606
607 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
608
609 /* Inner loop uses 61 flops */
610 }
611
612 if(jidx<j_index_end)
613 {
614
615 jnrA = jjnr[jidx];
616 j_coord_offsetA = DIM*jnrA;
617
618 /* load j atom coordinates */
619 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
620 &jx0,&jy0,&jz0);
621
622 /* Calculate displacement vector */
623 dx00 = _mm_sub_pd(ix0,jx0);
624 dy00 = _mm_sub_pd(iy0,jy0);
625 dz00 = _mm_sub_pd(iz0,jz0);
626
627 /* Calculate squared distance and things based on it */
628 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
629
630 rinv00 = gmx_mm_invsqrt_pd(rsq00);
631
632 /* Load parameters for j particles */
633 jq0 = _mm_load_sd(charge+jnrA+0);
634 vdwjidx0A = 2*vdwtype[jnrA+0];
635
636 /**************************
637 * CALCULATE INTERACTIONS *
638 **************************/
639
640 r00 = _mm_mul_pd(rsq00,rinv00);
641
642 /* Compute parameters for interactions between i and j atoms */
643 qq00 = _mm_mul_pd(iq0,jq0);
644 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
645
646 /* Calculate table index by multiplying r with table scale and truncate to integer */
647 rt = _mm_mul_pd(r00,vftabscale);
648 vfitab = _mm_cvttpd_epi32(rt);
649 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
650 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
651
652 /* CUBIC SPLINE TABLE ELECTROSTATICS */
653 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
654 F = _mm_setzero_pd();
655 GMX_MM_TRANSPOSE2_PD(Y,F);
656 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
657 H = _mm_setzero_pd();
658 GMX_MM_TRANSPOSE2_PD(G,H);
659 Heps = _mm_mul_pd(vfeps,H);
660 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
661 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
662 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
663
664 /* CUBIC SPLINE TABLE DISPERSION */
665 vfitab = _mm_add_epi32(vfitab,ifour);
666 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
667 F = _mm_setzero_pd();
668 GMX_MM_TRANSPOSE2_PD(Y,F);
669 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
670 H = _mm_setzero_pd();
671 GMX_MM_TRANSPOSE2_PD(G,H);
672 Heps = _mm_mul_pd(vfeps,H);
673 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
674 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
675 fvdw6 = _mm_mul_pd(c6_00,FF);
676
677 /* CUBIC SPLINE TABLE REPULSION */
678 vfitab = _mm_add_epi32(vfitab,ifour);
679 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
680 F = _mm_setzero_pd();
681 GMX_MM_TRANSPOSE2_PD(Y,F);
682 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
683 H = _mm_setzero_pd();
684 GMX_MM_TRANSPOSE2_PD(G,H);
685 Heps = _mm_mul_pd(vfeps,H);
686 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
687 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
688 fvdw12 = _mm_mul_pd(c12_00,FF);
689 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
690
691 fscal = _mm_add_pd(felec,fvdw);
692
693 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
694
695 /* Calculate temporary vectorial force */
696 tx = _mm_mul_pd(fscal,dx00);
697 ty = _mm_mul_pd(fscal,dy00);
698 tz = _mm_mul_pd(fscal,dz00);
699
700 /* Update vectorial force */
701 fix0 = _mm_add_pd(fix0,tx);
702 fiy0 = _mm_add_pd(fiy0,ty);
703 fiz0 = _mm_add_pd(fiz0,tz);
704
705 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
706
707 /* Inner loop uses 61 flops */
708 }
709
710 /* End of innermost loop */
711
712 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
713 f+i_coord_offset,fshift+i_shift_offset);
714
715 /* Increment number of inner iterations */
716 inneriter += j_index_end - j_index_start;
717
718 /* Outer loop uses 7 flops */
719 }
720
721 /* Increment number of outer iterations */
722 outeriter += nri;
723
724 /* Update outer/inner flops */
725
726 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*61);
727 }
The ultimate molecular dynamics simulation package