?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed include simple.h from nb_kernel_avx_128_fma_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_avx_128_fma_double / nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_avx_128_fma_double.c
blobbf44726486b143b8cf8db14b04b324eefbd58222
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 avx_128_fma_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_avx_128_fma_double.h"
49 #include "kernelutil_x86_avx_128_fma_double.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
53 * Electrostatics interaction: ReactionField
54 * VdW interaction: CubicSplineTable
55 * Geometry: Water3-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_128_fma_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,twovfeps;
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 krf = _mm_set1_pd(fr->ic->k_rf);
122 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
123 crf = _mm_set1_pd(fr->ic->c_rf);
124 nvdwtype = fr->ntype;
125 vdwparam = fr->nbfp;
126 vdwtype = mdatoms->typeA;
127
128 vftab = kernel_data->table_vdw->data;
129 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
130
131 /* Setup water-specific parameters */
132 inr = nlist->iinr[0];
133 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
134 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
135 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
136 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
137
138 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
139 rcutoff_scalar = fr->rcoulomb;
140 rcutoff = _mm_set1_pd(rcutoff_scalar);
141 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
142
143 /* Avoid stupid compiler warnings */
144 jnrA = jnrB = 0;
145 j_coord_offsetA = 0;
146 j_coord_offsetB = 0;
147
148 outeriter = 0;
149 inneriter = 0;
150
151 /* Start outer loop over neighborlists */
152 for(iidx=0; iidx<nri; iidx++)
153 {
154 /* Load shift vector for this list */
155 i_shift_offset = DIM*shiftidx[iidx];
156
157 /* Load limits for loop over neighbors */
158 j_index_start = jindex[iidx];
159 j_index_end = jindex[iidx+1];
160
161 /* Get outer coordinate index */
162 inr = iinr[iidx];
163 i_coord_offset = DIM*inr;
164
165 /* Load i particle coords and add shift vector */
166 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
167 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
168
169 fix0 = _mm_setzero_pd();
170 fiy0 = _mm_setzero_pd();
171 fiz0 = _mm_setzero_pd();
172 fix1 = _mm_setzero_pd();
173 fiy1 = _mm_setzero_pd();
174 fiz1 = _mm_setzero_pd();
175 fix2 = _mm_setzero_pd();
176 fiy2 = _mm_setzero_pd();
177 fiz2 = _mm_setzero_pd();
178
179 /* Reset potential sums */
180 velecsum = _mm_setzero_pd();
181 vvdwsum = _mm_setzero_pd();
182
183 /* Start inner kernel loop */
184 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
185 {
186
187 /* Get j neighbor index, and coordinate index */
188 jnrA = jjnr[jidx];
189 jnrB = jjnr[jidx+1];
190 j_coord_offsetA = DIM*jnrA;
191 j_coord_offsetB = DIM*jnrB;
192
193 /* load j atom coordinates */
194 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
195 &jx0,&jy0,&jz0);
196
197 /* Calculate displacement vector */
198 dx00 = _mm_sub_pd(ix0,jx0);
199 dy00 = _mm_sub_pd(iy0,jy0);
200 dz00 = _mm_sub_pd(iz0,jz0);
201 dx10 = _mm_sub_pd(ix1,jx0);
202 dy10 = _mm_sub_pd(iy1,jy0);
203 dz10 = _mm_sub_pd(iz1,jz0);
204 dx20 = _mm_sub_pd(ix2,jx0);
205 dy20 = _mm_sub_pd(iy2,jy0);
206 dz20 = _mm_sub_pd(iz2,jz0);
207
208 /* Calculate squared distance and things based on it */
209 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
210 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
211 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
212
213 rinv00 = gmx_mm_invsqrt_pd(rsq00);
214 rinv10 = gmx_mm_invsqrt_pd(rsq10);
215 rinv20 = gmx_mm_invsqrt_pd(rsq20);
216
217 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
218 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
219 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
220
221 /* Load parameters for j particles */
222 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
223 vdwjidx0A = 2*vdwtype[jnrA+0];
224 vdwjidx0B = 2*vdwtype[jnrB+0];
225
226 fjx0 = _mm_setzero_pd();
227 fjy0 = _mm_setzero_pd();
228 fjz0 = _mm_setzero_pd();
229
230 /**************************
231 * CALCULATE INTERACTIONS *
232 **************************/
233
234 if (gmx_mm_any_lt(rsq00,rcutoff2))
235 {
236
237 r00 = _mm_mul_pd(rsq00,rinv00);
238
239 /* Compute parameters for interactions between i and j atoms */
240 qq00 = _mm_mul_pd(iq0,jq0);
241 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
242 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
243
244 /* Calculate table index by multiplying r with table scale and truncate to integer */
245 rt = _mm_mul_pd(r00,vftabscale);
246 vfitab = _mm_cvttpd_epi32(rt);
247 #ifdef __XOP__
248 vfeps = _mm_frcz_pd(rt);
249 #else
250 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
251 #endif
252 twovfeps = _mm_add_pd(vfeps,vfeps);
253 vfitab = _mm_slli_epi32(vfitab,3);
254
255 /* REACTION-FIELD ELECTROSTATICS */
256 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
257 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
258
259 /* CUBIC SPLINE TABLE DISPERSION */
260 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
261 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
262 GMX_MM_TRANSPOSE2_PD(Y,F);
263 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
264 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
265 GMX_MM_TRANSPOSE2_PD(G,H);
266 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
267 VV = _mm_macc_pd(vfeps,Fp,Y);
268 vvdw6 = _mm_mul_pd(c6_00,VV);
269 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
270 fvdw6 = _mm_mul_pd(c6_00,FF);
271
272 /* CUBIC SPLINE TABLE REPULSION */
273 vfitab = _mm_add_epi32(vfitab,ifour);
274 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
275 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
276 GMX_MM_TRANSPOSE2_PD(Y,F);
277 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
278 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
279 GMX_MM_TRANSPOSE2_PD(G,H);
280 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
281 VV = _mm_macc_pd(vfeps,Fp,Y);
282 vvdw12 = _mm_mul_pd(c12_00,VV);
283 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
284 fvdw12 = _mm_mul_pd(c12_00,FF);
285 vvdw = _mm_add_pd(vvdw12,vvdw6);
286 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
287
288 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
289
290 /* Update potential sum for this i atom from the interaction with this j atom. */
291 velec = _mm_and_pd(velec,cutoff_mask);
292 velecsum = _mm_add_pd(velecsum,velec);
293 vvdw = _mm_and_pd(vvdw,cutoff_mask);
294 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
295
296 fscal = _mm_add_pd(felec,fvdw);
297
298 fscal = _mm_and_pd(fscal,cutoff_mask);
299
300 /* Update vectorial force */
301 fix0 = _mm_macc_pd(dx00,fscal,fix0);
302 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
303 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
304
305 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
306 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
307 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
308
309 }
310
311 /**************************
312 * CALCULATE INTERACTIONS *
313 **************************/
314
315 if (gmx_mm_any_lt(rsq10,rcutoff2))
316 {
317
318 /* Compute parameters for interactions between i and j atoms */
319 qq10 = _mm_mul_pd(iq1,jq0);
320
321 /* REACTION-FIELD ELECTROSTATICS */
322 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
323 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
324
325 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
326
327 /* Update potential sum for this i atom from the interaction with this j atom. */
328 velec = _mm_and_pd(velec,cutoff_mask);
329 velecsum = _mm_add_pd(velecsum,velec);
330
331 fscal = felec;
332
333 fscal = _mm_and_pd(fscal,cutoff_mask);
334
335 /* Update vectorial force */
336 fix1 = _mm_macc_pd(dx10,fscal,fix1);
337 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
338 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
339
340 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
341 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
342 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
343
344 }
345
346 /**************************
347 * CALCULATE INTERACTIONS *
348 **************************/
349
350 if (gmx_mm_any_lt(rsq20,rcutoff2))
351 {
352
353 /* Compute parameters for interactions between i and j atoms */
354 qq20 = _mm_mul_pd(iq2,jq0);
355
356 /* REACTION-FIELD ELECTROSTATICS */
357 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
358 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
359
360 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
361
362 /* Update potential sum for this i atom from the interaction with this j atom. */
363 velec = _mm_and_pd(velec,cutoff_mask);
364 velecsum = _mm_add_pd(velecsum,velec);
365
366 fscal = felec;
367
368 fscal = _mm_and_pd(fscal,cutoff_mask);
369
370 /* Update vectorial force */
371 fix2 = _mm_macc_pd(dx20,fscal,fix2);
372 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
373 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
374
375 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
376 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
377 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
378
379 }
380
381 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
382
383 /* Inner loop uses 156 flops */
384 }
385
386 if(jidx<j_index_end)
387 {
388
389 jnrA = jjnr[jidx];
390 j_coord_offsetA = DIM*jnrA;
391
392 /* load j atom coordinates */
393 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
394 &jx0,&jy0,&jz0);
395
396 /* Calculate displacement vector */
397 dx00 = _mm_sub_pd(ix0,jx0);
398 dy00 = _mm_sub_pd(iy0,jy0);
399 dz00 = _mm_sub_pd(iz0,jz0);
400 dx10 = _mm_sub_pd(ix1,jx0);
401 dy10 = _mm_sub_pd(iy1,jy0);
402 dz10 = _mm_sub_pd(iz1,jz0);
403 dx20 = _mm_sub_pd(ix2,jx0);
404 dy20 = _mm_sub_pd(iy2,jy0);
405 dz20 = _mm_sub_pd(iz2,jz0);
406
407 /* Calculate squared distance and things based on it */
408 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
409 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
410 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
411
412 rinv00 = gmx_mm_invsqrt_pd(rsq00);
413 rinv10 = gmx_mm_invsqrt_pd(rsq10);
414 rinv20 = gmx_mm_invsqrt_pd(rsq20);
415
416 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
417 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
418 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
419
420 /* Load parameters for j particles */
421 jq0 = _mm_load_sd(charge+jnrA+0);
422 vdwjidx0A = 2*vdwtype[jnrA+0];
423
424 fjx0 = _mm_setzero_pd();
425 fjy0 = _mm_setzero_pd();
426 fjz0 = _mm_setzero_pd();
427
428 /**************************
429 * CALCULATE INTERACTIONS *
430 **************************/
431
432 if (gmx_mm_any_lt(rsq00,rcutoff2))
433 {
434
435 r00 = _mm_mul_pd(rsq00,rinv00);
436
437 /* Compute parameters for interactions between i and j atoms */
438 qq00 = _mm_mul_pd(iq0,jq0);
439 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
440
441 /* Calculate table index by multiplying r with table scale and truncate to integer */
442 rt = _mm_mul_pd(r00,vftabscale);
443 vfitab = _mm_cvttpd_epi32(rt);
444 #ifdef __XOP__
445 vfeps = _mm_frcz_pd(rt);
446 #else
447 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
448 #endif
449 twovfeps = _mm_add_pd(vfeps,vfeps);
450 vfitab = _mm_slli_epi32(vfitab,3);
451
452 /* REACTION-FIELD ELECTROSTATICS */
453 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
454 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
455
456 /* CUBIC SPLINE TABLE DISPERSION */
457 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
458 F = _mm_setzero_pd();
459 GMX_MM_TRANSPOSE2_PD(Y,F);
460 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
461 H = _mm_setzero_pd();
462 GMX_MM_TRANSPOSE2_PD(G,H);
463 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
464 VV = _mm_macc_pd(vfeps,Fp,Y);
465 vvdw6 = _mm_mul_pd(c6_00,VV);
466 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
467 fvdw6 = _mm_mul_pd(c6_00,FF);
468
469 /* CUBIC SPLINE TABLE REPULSION */
470 vfitab = _mm_add_epi32(vfitab,ifour);
471 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
472 F = _mm_setzero_pd();
473 GMX_MM_TRANSPOSE2_PD(Y,F);
474 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
475 H = _mm_setzero_pd();
476 GMX_MM_TRANSPOSE2_PD(G,H);
477 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
478 VV = _mm_macc_pd(vfeps,Fp,Y);
479 vvdw12 = _mm_mul_pd(c12_00,VV);
480 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
481 fvdw12 = _mm_mul_pd(c12_00,FF);
482 vvdw = _mm_add_pd(vvdw12,vvdw6);
483 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
484
485 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
486
487 /* Update potential sum for this i atom from the interaction with this j atom. */
488 velec = _mm_and_pd(velec,cutoff_mask);
489 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
490 velecsum = _mm_add_pd(velecsum,velec);
491 vvdw = _mm_and_pd(vvdw,cutoff_mask);
492 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
493 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
494
495 fscal = _mm_add_pd(felec,fvdw);
496
497 fscal = _mm_and_pd(fscal,cutoff_mask);
498
499 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
500
501 /* Update vectorial force */
502 fix0 = _mm_macc_pd(dx00,fscal,fix0);
503 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
504 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
505
506 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
507 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
508 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
509
510 }
511
512 /**************************
513 * CALCULATE INTERACTIONS *
514 **************************/
515
516 if (gmx_mm_any_lt(rsq10,rcutoff2))
517 {
518
519 /* Compute parameters for interactions between i and j atoms */
520 qq10 = _mm_mul_pd(iq1,jq0);
521
522 /* REACTION-FIELD ELECTROSTATICS */
523 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
524 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
525
526 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
527
528 /* Update potential sum for this i atom from the interaction with this j atom. */
529 velec = _mm_and_pd(velec,cutoff_mask);
530 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
531 velecsum = _mm_add_pd(velecsum,velec);
532
533 fscal = felec;
534
535 fscal = _mm_and_pd(fscal,cutoff_mask);
536
537 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
538
539 /* Update vectorial force */
540 fix1 = _mm_macc_pd(dx10,fscal,fix1);
541 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
542 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
543
544 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
545 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
546 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
547
548 }
549
550 /**************************
551 * CALCULATE INTERACTIONS *
552 **************************/
553
554 if (gmx_mm_any_lt(rsq20,rcutoff2))
555 {
556
557 /* Compute parameters for interactions between i and j atoms */
558 qq20 = _mm_mul_pd(iq2,jq0);
559
560 /* REACTION-FIELD ELECTROSTATICS */
561 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
562 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
563
564 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
565
566 /* Update potential sum for this i atom from the interaction with this j atom. */
567 velec = _mm_and_pd(velec,cutoff_mask);
568 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
569 velecsum = _mm_add_pd(velecsum,velec);
570
571 fscal = felec;
572
573 fscal = _mm_and_pd(fscal,cutoff_mask);
574
575 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
576
577 /* Update vectorial force */
578 fix2 = _mm_macc_pd(dx20,fscal,fix2);
579 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
580 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
581
582 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
583 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
584 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
585
586 }
587
588 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
589
590 /* Inner loop uses 156 flops */
591 }
592
593 /* End of innermost loop */
594
595 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
596 f+i_coord_offset,fshift+i_shift_offset);
597
598 ggid = gid[iidx];
599 /* Update potential energies */
600 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
601 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
602
603 /* Increment number of inner iterations */
604 inneriter += j_index_end - j_index_start;
605
606 /* Outer loop uses 20 flops */
607 }
608
609 /* Increment number of outer iterations */
610 outeriter += nri;
611
612 /* Update outer/inner flops */
613
614 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*156);
615 }
616 /*
617 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_128_fma_double
618 * Electrostatics interaction: ReactionField
619 * VdW interaction: CubicSplineTable
620 * Geometry: Water3-Particle
621 * Calculate force/pot: Force
622 */
623 void
624 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_128_fma_double
625 (t_nblist * gmx_restrict nlist,
626 rvec * gmx_restrict xx,
627 rvec * gmx_restrict ff,
628 t_forcerec * gmx_restrict fr,
629 t_mdatoms * gmx_restrict mdatoms,
630 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
631 t_nrnb * gmx_restrict nrnb)
632 {
633 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
634 * just 0 for non-waters.
635 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
636 * jnr indices corresponding to data put in the four positions in the SIMD register.
637 */
638 int i_shift_offset,i_coord_offset,outeriter,inneriter;
639 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
640 int jnrA,jnrB;
641 int j_coord_offsetA,j_coord_offsetB;
642 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
643 real rcutoff_scalar;
644 real *shiftvec,*fshift,*x,*f;
645 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
646 int vdwioffset0;
647 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
648 int vdwioffset1;
649 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
650 int vdwioffset2;
651 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
652 int vdwjidx0A,vdwjidx0B;
653 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
654 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
655 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
656 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
657 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
658 real *charge;
659 int nvdwtype;
660 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
661 int *vdwtype;
662 real *vdwparam;
663 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
664 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
665 __m128i vfitab;
666 __m128i ifour = _mm_set1_epi32(4);
667 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
668 real *vftab;
669 __m128d dummy_mask,cutoff_mask;
670 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
671 __m128d one = _mm_set1_pd(1.0);
672 __m128d two = _mm_set1_pd(2.0);
673 x = xx[0];
674 f = ff[0];
675
676 nri = nlist->nri;
677 iinr = nlist->iinr;
678 jindex = nlist->jindex;
679 jjnr = nlist->jjnr;
680 shiftidx = nlist->shift;
681 gid = nlist->gid;
682 shiftvec = fr->shift_vec[0];
683 fshift = fr->fshift[0];
684 facel = _mm_set1_pd(fr->epsfac);
685 charge = mdatoms->chargeA;
686 krf = _mm_set1_pd(fr->ic->k_rf);
687 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
688 crf = _mm_set1_pd(fr->ic->c_rf);
689 nvdwtype = fr->ntype;
690 vdwparam = fr->nbfp;
691 vdwtype = mdatoms->typeA;
692
693 vftab = kernel_data->table_vdw->data;
694 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
695
696 /* Setup water-specific parameters */
697 inr = nlist->iinr[0];
698 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
699 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
700 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
701 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
702
703 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
704 rcutoff_scalar = fr->rcoulomb;
705 rcutoff = _mm_set1_pd(rcutoff_scalar);
706 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
707
708 /* Avoid stupid compiler warnings */
709 jnrA = jnrB = 0;
710 j_coord_offsetA = 0;
711 j_coord_offsetB = 0;
712
713 outeriter = 0;
714 inneriter = 0;
715
716 /* Start outer loop over neighborlists */
717 for(iidx=0; iidx<nri; iidx++)
718 {
719 /* Load shift vector for this list */
720 i_shift_offset = DIM*shiftidx[iidx];
721
722 /* Load limits for loop over neighbors */
723 j_index_start = jindex[iidx];
724 j_index_end = jindex[iidx+1];
725
726 /* Get outer coordinate index */
727 inr = iinr[iidx];
728 i_coord_offset = DIM*inr;
729
730 /* Load i particle coords and add shift vector */
731 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
732 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
733
734 fix0 = _mm_setzero_pd();
735 fiy0 = _mm_setzero_pd();
736 fiz0 = _mm_setzero_pd();
737 fix1 = _mm_setzero_pd();
738 fiy1 = _mm_setzero_pd();
739 fiz1 = _mm_setzero_pd();
740 fix2 = _mm_setzero_pd();
741 fiy2 = _mm_setzero_pd();
742 fiz2 = _mm_setzero_pd();
743
744 /* Start inner kernel loop */
745 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
746 {
747
748 /* Get j neighbor index, and coordinate index */
749 jnrA = jjnr[jidx];
750 jnrB = jjnr[jidx+1];
751 j_coord_offsetA = DIM*jnrA;
752 j_coord_offsetB = DIM*jnrB;
753
754 /* load j atom coordinates */
755 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
756 &jx0,&jy0,&jz0);
757
758 /* Calculate displacement vector */
759 dx00 = _mm_sub_pd(ix0,jx0);
760 dy00 = _mm_sub_pd(iy0,jy0);
761 dz00 = _mm_sub_pd(iz0,jz0);
762 dx10 = _mm_sub_pd(ix1,jx0);
763 dy10 = _mm_sub_pd(iy1,jy0);
764 dz10 = _mm_sub_pd(iz1,jz0);
765 dx20 = _mm_sub_pd(ix2,jx0);
766 dy20 = _mm_sub_pd(iy2,jy0);
767 dz20 = _mm_sub_pd(iz2,jz0);
768
769 /* Calculate squared distance and things based on it */
770 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
771 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
772 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
773
774 rinv00 = gmx_mm_invsqrt_pd(rsq00);
775 rinv10 = gmx_mm_invsqrt_pd(rsq10);
776 rinv20 = gmx_mm_invsqrt_pd(rsq20);
777
778 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
779 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
780 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
781
782 /* Load parameters for j particles */
783 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
784 vdwjidx0A = 2*vdwtype[jnrA+0];
785 vdwjidx0B = 2*vdwtype[jnrB+0];
786
787 fjx0 = _mm_setzero_pd();
788 fjy0 = _mm_setzero_pd();
789 fjz0 = _mm_setzero_pd();
790
791 /**************************
792 * CALCULATE INTERACTIONS *
793 **************************/
794
795 if (gmx_mm_any_lt(rsq00,rcutoff2))
796 {
797
798 r00 = _mm_mul_pd(rsq00,rinv00);
799
800 /* Compute parameters for interactions between i and j atoms */
801 qq00 = _mm_mul_pd(iq0,jq0);
802 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
803 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
804
805 /* Calculate table index by multiplying r with table scale and truncate to integer */
806 rt = _mm_mul_pd(r00,vftabscale);
807 vfitab = _mm_cvttpd_epi32(rt);
808 #ifdef __XOP__
809 vfeps = _mm_frcz_pd(rt);
810 #else
811 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
812 #endif
813 twovfeps = _mm_add_pd(vfeps,vfeps);
814 vfitab = _mm_slli_epi32(vfitab,3);
815
816 /* REACTION-FIELD ELECTROSTATICS */
817 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
818
819 /* CUBIC SPLINE TABLE DISPERSION */
820 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
821 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
822 GMX_MM_TRANSPOSE2_PD(Y,F);
823 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
824 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
825 GMX_MM_TRANSPOSE2_PD(G,H);
826 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
827 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
828 fvdw6 = _mm_mul_pd(c6_00,FF);
829
830 /* CUBIC SPLINE TABLE REPULSION */
831 vfitab = _mm_add_epi32(vfitab,ifour);
832 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
833 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
834 GMX_MM_TRANSPOSE2_PD(Y,F);
835 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
836 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
837 GMX_MM_TRANSPOSE2_PD(G,H);
838 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
839 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
840 fvdw12 = _mm_mul_pd(c12_00,FF);
841 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
842
843 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
844
845 fscal = _mm_add_pd(felec,fvdw);
846
847 fscal = _mm_and_pd(fscal,cutoff_mask);
848
849 /* Update vectorial force */
850 fix0 = _mm_macc_pd(dx00,fscal,fix0);
851 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
852 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
853
854 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
855 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
856 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
857
858 }
859
860 /**************************
861 * CALCULATE INTERACTIONS *
862 **************************/
863
864 if (gmx_mm_any_lt(rsq10,rcutoff2))
865 {
866
867 /* Compute parameters for interactions between i and j atoms */
868 qq10 = _mm_mul_pd(iq1,jq0);
869
870 /* REACTION-FIELD ELECTROSTATICS */
871 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
872
873 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
874
875 fscal = felec;
876
877 fscal = _mm_and_pd(fscal,cutoff_mask);
878
879 /* Update vectorial force */
880 fix1 = _mm_macc_pd(dx10,fscal,fix1);
881 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
882 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
883
884 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
885 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
886 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
887
888 }
889
890 /**************************
891 * CALCULATE INTERACTIONS *
892 **************************/
893
894 if (gmx_mm_any_lt(rsq20,rcutoff2))
895 {
896
897 /* Compute parameters for interactions between i and j atoms */
898 qq20 = _mm_mul_pd(iq2,jq0);
899
900 /* REACTION-FIELD ELECTROSTATICS */
901 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
902
903 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
904
905 fscal = felec;
906
907 fscal = _mm_and_pd(fscal,cutoff_mask);
908
909 /* Update vectorial force */
910 fix2 = _mm_macc_pd(dx20,fscal,fix2);
911 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
912 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
913
914 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
915 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
916 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
917
918 }
919
920 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
921
922 /* Inner loop uses 129 flops */
923 }
924
925 if(jidx<j_index_end)
926 {
927
928 jnrA = jjnr[jidx];
929 j_coord_offsetA = DIM*jnrA;
930
931 /* load j atom coordinates */
932 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
933 &jx0,&jy0,&jz0);
934
935 /* Calculate displacement vector */
936 dx00 = _mm_sub_pd(ix0,jx0);
937 dy00 = _mm_sub_pd(iy0,jy0);
938 dz00 = _mm_sub_pd(iz0,jz0);
939 dx10 = _mm_sub_pd(ix1,jx0);
940 dy10 = _mm_sub_pd(iy1,jy0);
941 dz10 = _mm_sub_pd(iz1,jz0);
942 dx20 = _mm_sub_pd(ix2,jx0);
943 dy20 = _mm_sub_pd(iy2,jy0);
944 dz20 = _mm_sub_pd(iz2,jz0);
945
946 /* Calculate squared distance and things based on it */
947 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
948 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
949 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
950
951 rinv00 = gmx_mm_invsqrt_pd(rsq00);
952 rinv10 = gmx_mm_invsqrt_pd(rsq10);
953 rinv20 = gmx_mm_invsqrt_pd(rsq20);
954
955 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
956 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
957 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
958
959 /* Load parameters for j particles */
960 jq0 = _mm_load_sd(charge+jnrA+0);
961 vdwjidx0A = 2*vdwtype[jnrA+0];
962
963 fjx0 = _mm_setzero_pd();
964 fjy0 = _mm_setzero_pd();
965 fjz0 = _mm_setzero_pd();
966
967 /**************************
968 * CALCULATE INTERACTIONS *
969 **************************/
970
971 if (gmx_mm_any_lt(rsq00,rcutoff2))
972 {
973
974 r00 = _mm_mul_pd(rsq00,rinv00);
975
976 /* Compute parameters for interactions between i and j atoms */
977 qq00 = _mm_mul_pd(iq0,jq0);
978 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
979
980 /* Calculate table index by multiplying r with table scale and truncate to integer */
981 rt = _mm_mul_pd(r00,vftabscale);
982 vfitab = _mm_cvttpd_epi32(rt);
983 #ifdef __XOP__
984 vfeps = _mm_frcz_pd(rt);
985 #else
986 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
987 #endif
988 twovfeps = _mm_add_pd(vfeps,vfeps);
989 vfitab = _mm_slli_epi32(vfitab,3);
990
991 /* REACTION-FIELD ELECTROSTATICS */
992 felec = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
993
994 /* CUBIC SPLINE TABLE DISPERSION */
995 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
996 F = _mm_setzero_pd();
997 GMX_MM_TRANSPOSE2_PD(Y,F);
998 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
999 H = _mm_setzero_pd();
1000 GMX_MM_TRANSPOSE2_PD(G,H);
1001 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1002 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1003 fvdw6 = _mm_mul_pd(c6_00,FF);
1004
1005 /* CUBIC SPLINE TABLE REPULSION */
1006 vfitab = _mm_add_epi32(vfitab,ifour);
1007 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1008 F = _mm_setzero_pd();
1009 GMX_MM_TRANSPOSE2_PD(Y,F);
1010 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1011 H = _mm_setzero_pd();
1012 GMX_MM_TRANSPOSE2_PD(G,H);
1013 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1014 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1015 fvdw12 = _mm_mul_pd(c12_00,FF);
1016 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1017
1018 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1019
1020 fscal = _mm_add_pd(felec,fvdw);
1021
1022 fscal = _mm_and_pd(fscal,cutoff_mask);
1023
1024 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1025
1026 /* Update vectorial force */
1027 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1028 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1029 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1030
1031 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1032 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1033 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1034
1035 }
1036
1037 /**************************
1038 * CALCULATE INTERACTIONS *
1039 **************************/
1040
1041 if (gmx_mm_any_lt(rsq10,rcutoff2))
1042 {
1043
1044 /* Compute parameters for interactions between i and j atoms */
1045 qq10 = _mm_mul_pd(iq1,jq0);
1046
1047 /* REACTION-FIELD ELECTROSTATICS */
1048 felec = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
1049
1050 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1051
1052 fscal = felec;
1053
1054 fscal = _mm_and_pd(fscal,cutoff_mask);
1055
1056 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1057
1058 /* Update vectorial force */
1059 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1060 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1061 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1062
1063 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1064 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1065 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1066
1067 }
1068
1069 /**************************
1070 * CALCULATE INTERACTIONS *
1071 **************************/
1072
1073 if (gmx_mm_any_lt(rsq20,rcutoff2))
1074 {
1075
1076 /* Compute parameters for interactions between i and j atoms */
1077 qq20 = _mm_mul_pd(iq2,jq0);
1078
1079 /* REACTION-FIELD ELECTROSTATICS */
1080 felec = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1081
1082 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1083
1084 fscal = felec;
1085
1086 fscal = _mm_and_pd(fscal,cutoff_mask);
1087
1088 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1089
1090 /* Update vectorial force */
1091 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1092 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1093 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1094
1095 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1096 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1097 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1098
1099 }
1100
1101 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1102
1103 /* Inner loop uses 129 flops */
1104 }
1105
1106 /* End of innermost loop */
1107
1108 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1109 f+i_coord_offset,fshift+i_shift_offset);
1110
1111 /* Increment number of inner iterations */
1112 inneriter += j_index_end - j_index_start;
1113
1114 /* Outer loop uses 18 flops */
1115 }
1116
1117 /* Increment number of outer iterations */
1118 outeriter += nri;
1119
1120 /* Update outer/inner flops */
1121
1122 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*129);
1123 }
The ultimate molecular dynamics simulation package