?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Remove nb-parameters from t_forcerec
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_avx_256_single / nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_avx_256_single.c
blob737a40059bc4c46830e6f53a5d55f8720d4f6ab8
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015,2017, 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_256_single 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/gmxlib/nrnb.h"
46
47 #include "kernelutil_x86_avx_256_single.h"
48
49 /*
50 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_avx_256_single
51 * Electrostatics interaction: ReactionField
52 * VdW interaction: CubicSplineTable
53 * Geometry: Water4-Particle
54 * Calculate force/pot: PotentialAndForce
55 */
56 void
57 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_avx_256_single
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
65 {
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
70 */
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
73 int jnrA,jnrB,jnrC,jnrD;
74 int jnrE,jnrF,jnrG,jnrH;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real rcutoff_scalar;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
83 real scratch[4*DIM];
84 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
85 real * vdwioffsetptr0;
86 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
87 real * vdwioffsetptr1;
88 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
89 real * vdwioffsetptr2;
90 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
91 real * vdwioffsetptr3;
92 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
93 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
94 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
95 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
96 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
97 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
98 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
99 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
100 real *charge;
101 int nvdwtype;
102 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
103 int *vdwtype;
104 real *vdwparam;
105 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
106 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
107 __m256i vfitab;
108 __m128i vfitab_lo,vfitab_hi;
109 __m128i ifour = _mm_set1_epi32(4);
110 __m256 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
111 real *vftab;
112 __m256 dummy_mask,cutoff_mask;
113 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
114 __m256 one = _mm256_set1_ps(1.0);
115 __m256 two = _mm256_set1_ps(2.0);
116 x = xx[0];
117 f = ff[0];
118
119 nri = nlist->nri;
120 iinr = nlist->iinr;
121 jindex = nlist->jindex;
122 jjnr = nlist->jjnr;
123 shiftidx = nlist->shift;
124 gid = nlist->gid;
125 shiftvec = fr->shift_vec[0];
126 fshift = fr->fshift[0];
127 facel = _mm256_set1_ps(fr->ic->epsfac);
128 charge = mdatoms->chargeA;
129 krf = _mm256_set1_ps(fr->ic->k_rf);
130 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
131 crf = _mm256_set1_ps(fr->ic->c_rf);
132 nvdwtype = fr->ntype;
133 vdwparam = fr->nbfp;
134 vdwtype = mdatoms->typeA;
135
136 vftab = kernel_data->table_vdw->data;
137 vftabscale = _mm256_set1_ps(kernel_data->table_vdw->scale);
138
139 /* Setup water-specific parameters */
140 inr = nlist->iinr[0];
141 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
142 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
143 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
144 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
145
146 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
147 rcutoff_scalar = fr->ic->rcoulomb;
148 rcutoff = _mm256_set1_ps(rcutoff_scalar);
149 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
150
151 /* Avoid stupid compiler warnings */
152 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
153 j_coord_offsetA = 0;
154 j_coord_offsetB = 0;
155 j_coord_offsetC = 0;
156 j_coord_offsetD = 0;
157 j_coord_offsetE = 0;
158 j_coord_offsetF = 0;
159 j_coord_offsetG = 0;
160 j_coord_offsetH = 0;
161
162 outeriter = 0;
163 inneriter = 0;
164
165 for(iidx=0;iidx<4*DIM;iidx++)
166 {
167 scratch[iidx] = 0.0;
168 }
169
170 /* Start outer loop over neighborlists */
171 for(iidx=0; iidx<nri; iidx++)
172 {
173 /* Load shift vector for this list */
174 i_shift_offset = DIM*shiftidx[iidx];
175
176 /* Load limits for loop over neighbors */
177 j_index_start = jindex[iidx];
178 j_index_end = jindex[iidx+1];
179
180 /* Get outer coordinate index */
181 inr = iinr[iidx];
182 i_coord_offset = DIM*inr;
183
184 /* Load i particle coords and add shift vector */
185 gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
186 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
187
188 fix0 = _mm256_setzero_ps();
189 fiy0 = _mm256_setzero_ps();
190 fiz0 = _mm256_setzero_ps();
191 fix1 = _mm256_setzero_ps();
192 fiy1 = _mm256_setzero_ps();
193 fiz1 = _mm256_setzero_ps();
194 fix2 = _mm256_setzero_ps();
195 fiy2 = _mm256_setzero_ps();
196 fiz2 = _mm256_setzero_ps();
197 fix3 = _mm256_setzero_ps();
198 fiy3 = _mm256_setzero_ps();
199 fiz3 = _mm256_setzero_ps();
200
201 /* Reset potential sums */
202 velecsum = _mm256_setzero_ps();
203 vvdwsum = _mm256_setzero_ps();
204
205 /* Start inner kernel loop */
206 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
207 {
208
209 /* Get j neighbor index, and coordinate index */
210 jnrA = jjnr[jidx];
211 jnrB = jjnr[jidx+1];
212 jnrC = jjnr[jidx+2];
213 jnrD = jjnr[jidx+3];
214 jnrE = jjnr[jidx+4];
215 jnrF = jjnr[jidx+5];
216 jnrG = jjnr[jidx+6];
217 jnrH = jjnr[jidx+7];
218 j_coord_offsetA = DIM*jnrA;
219 j_coord_offsetB = DIM*jnrB;
220 j_coord_offsetC = DIM*jnrC;
221 j_coord_offsetD = DIM*jnrD;
222 j_coord_offsetE = DIM*jnrE;
223 j_coord_offsetF = DIM*jnrF;
224 j_coord_offsetG = DIM*jnrG;
225 j_coord_offsetH = DIM*jnrH;
226
227 /* load j atom coordinates */
228 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
229 x+j_coord_offsetC,x+j_coord_offsetD,
230 x+j_coord_offsetE,x+j_coord_offsetF,
231 x+j_coord_offsetG,x+j_coord_offsetH,
232 &jx0,&jy0,&jz0);
233
234 /* Calculate displacement vector */
235 dx00 = _mm256_sub_ps(ix0,jx0);
236 dy00 = _mm256_sub_ps(iy0,jy0);
237 dz00 = _mm256_sub_ps(iz0,jz0);
238 dx10 = _mm256_sub_ps(ix1,jx0);
239 dy10 = _mm256_sub_ps(iy1,jy0);
240 dz10 = _mm256_sub_ps(iz1,jz0);
241 dx20 = _mm256_sub_ps(ix2,jx0);
242 dy20 = _mm256_sub_ps(iy2,jy0);
243 dz20 = _mm256_sub_ps(iz2,jz0);
244 dx30 = _mm256_sub_ps(ix3,jx0);
245 dy30 = _mm256_sub_ps(iy3,jy0);
246 dz30 = _mm256_sub_ps(iz3,jz0);
247
248 /* Calculate squared distance and things based on it */
249 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
250 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
251 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
252 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
253
254 rinv00 = avx256_invsqrt_f(rsq00);
255 rinv10 = avx256_invsqrt_f(rsq10);
256 rinv20 = avx256_invsqrt_f(rsq20);
257 rinv30 = avx256_invsqrt_f(rsq30);
258
259 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
260 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
261 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
262
263 /* Load parameters for j particles */
264 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
265 charge+jnrC+0,charge+jnrD+0,
266 charge+jnrE+0,charge+jnrF+0,
267 charge+jnrG+0,charge+jnrH+0);
268 vdwjidx0A = 2*vdwtype[jnrA+0];
269 vdwjidx0B = 2*vdwtype[jnrB+0];
270 vdwjidx0C = 2*vdwtype[jnrC+0];
271 vdwjidx0D = 2*vdwtype[jnrD+0];
272 vdwjidx0E = 2*vdwtype[jnrE+0];
273 vdwjidx0F = 2*vdwtype[jnrF+0];
274 vdwjidx0G = 2*vdwtype[jnrG+0];
275 vdwjidx0H = 2*vdwtype[jnrH+0];
276
277 fjx0 = _mm256_setzero_ps();
278 fjy0 = _mm256_setzero_ps();
279 fjz0 = _mm256_setzero_ps();
280
281 /**************************
282 * CALCULATE INTERACTIONS *
283 **************************/
284
285 r00 = _mm256_mul_ps(rsq00,rinv00);
286
287 /* Compute parameters for interactions between i and j atoms */
288 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
289 vdwioffsetptr0+vdwjidx0B,
290 vdwioffsetptr0+vdwjidx0C,
291 vdwioffsetptr0+vdwjidx0D,
292 vdwioffsetptr0+vdwjidx0E,
293 vdwioffsetptr0+vdwjidx0F,
294 vdwioffsetptr0+vdwjidx0G,
295 vdwioffsetptr0+vdwjidx0H,
296 &c6_00,&c12_00);
297
298 /* Calculate table index by multiplying r with table scale and truncate to integer */
299 rt = _mm256_mul_ps(r00,vftabscale);
300 vfitab = _mm256_cvttps_epi32(rt);
301 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
302 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
303 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
304 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
305 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
306 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
307
308 /* CUBIC SPLINE TABLE DISPERSION */
309 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
310 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
311 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
312 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
313 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
314 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
315 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
316 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
317 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
318 Heps = _mm256_mul_ps(vfeps,H);
319 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
320 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
321 vvdw6 = _mm256_mul_ps(c6_00,VV);
322 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
323 fvdw6 = _mm256_mul_ps(c6_00,FF);
324
325 /* CUBIC SPLINE TABLE REPULSION */
326 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
327 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
328 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
329 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
330 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
331 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
332 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
333 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
334 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
335 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
336 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
337 Heps = _mm256_mul_ps(vfeps,H);
338 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
339 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
340 vvdw12 = _mm256_mul_ps(c12_00,VV);
341 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
342 fvdw12 = _mm256_mul_ps(c12_00,FF);
343 vvdw = _mm256_add_ps(vvdw12,vvdw6);
344 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
345
346 /* Update potential sum for this i atom from the interaction with this j atom. */
347 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
348
349 fscal = fvdw;
350
351 /* Calculate temporary vectorial force */
352 tx = _mm256_mul_ps(fscal,dx00);
353 ty = _mm256_mul_ps(fscal,dy00);
354 tz = _mm256_mul_ps(fscal,dz00);
355
356 /* Update vectorial force */
357 fix0 = _mm256_add_ps(fix0,tx);
358 fiy0 = _mm256_add_ps(fiy0,ty);
359 fiz0 = _mm256_add_ps(fiz0,tz);
360
361 fjx0 = _mm256_add_ps(fjx0,tx);
362 fjy0 = _mm256_add_ps(fjy0,ty);
363 fjz0 = _mm256_add_ps(fjz0,tz);
364
365 /**************************
366 * CALCULATE INTERACTIONS *
367 **************************/
368
369 if (gmx_mm256_any_lt(rsq10,rcutoff2))
370 {
371
372 /* Compute parameters for interactions between i and j atoms */
373 qq10 = _mm256_mul_ps(iq1,jq0);
374
375 /* REACTION-FIELD ELECTROSTATICS */
376 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
377 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
378
379 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
380
381 /* Update potential sum for this i atom from the interaction with this j atom. */
382 velec = _mm256_and_ps(velec,cutoff_mask);
383 velecsum = _mm256_add_ps(velecsum,velec);
384
385 fscal = felec;
386
387 fscal = _mm256_and_ps(fscal,cutoff_mask);
388
389 /* Calculate temporary vectorial force */
390 tx = _mm256_mul_ps(fscal,dx10);
391 ty = _mm256_mul_ps(fscal,dy10);
392 tz = _mm256_mul_ps(fscal,dz10);
393
394 /* Update vectorial force */
395 fix1 = _mm256_add_ps(fix1,tx);
396 fiy1 = _mm256_add_ps(fiy1,ty);
397 fiz1 = _mm256_add_ps(fiz1,tz);
398
399 fjx0 = _mm256_add_ps(fjx0,tx);
400 fjy0 = _mm256_add_ps(fjy0,ty);
401 fjz0 = _mm256_add_ps(fjz0,tz);
402
403 }
404
405 /**************************
406 * CALCULATE INTERACTIONS *
407 **************************/
408
409 if (gmx_mm256_any_lt(rsq20,rcutoff2))
410 {
411
412 /* Compute parameters for interactions between i and j atoms */
413 qq20 = _mm256_mul_ps(iq2,jq0);
414
415 /* REACTION-FIELD ELECTROSTATICS */
416 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
417 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
418
419 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
420
421 /* Update potential sum for this i atom from the interaction with this j atom. */
422 velec = _mm256_and_ps(velec,cutoff_mask);
423 velecsum = _mm256_add_ps(velecsum,velec);
424
425 fscal = felec;
426
427 fscal = _mm256_and_ps(fscal,cutoff_mask);
428
429 /* Calculate temporary vectorial force */
430 tx = _mm256_mul_ps(fscal,dx20);
431 ty = _mm256_mul_ps(fscal,dy20);
432 tz = _mm256_mul_ps(fscal,dz20);
433
434 /* Update vectorial force */
435 fix2 = _mm256_add_ps(fix2,tx);
436 fiy2 = _mm256_add_ps(fiy2,ty);
437 fiz2 = _mm256_add_ps(fiz2,tz);
438
439 fjx0 = _mm256_add_ps(fjx0,tx);
440 fjy0 = _mm256_add_ps(fjy0,ty);
441 fjz0 = _mm256_add_ps(fjz0,tz);
442
443 }
444
445 /**************************
446 * CALCULATE INTERACTIONS *
447 **************************/
448
449 if (gmx_mm256_any_lt(rsq30,rcutoff2))
450 {
451
452 /* Compute parameters for interactions between i and j atoms */
453 qq30 = _mm256_mul_ps(iq3,jq0);
454
455 /* REACTION-FIELD ELECTROSTATICS */
456 velec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
457 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
458
459 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
460
461 /* Update potential sum for this i atom from the interaction with this j atom. */
462 velec = _mm256_and_ps(velec,cutoff_mask);
463 velecsum = _mm256_add_ps(velecsum,velec);
464
465 fscal = felec;
466
467 fscal = _mm256_and_ps(fscal,cutoff_mask);
468
469 /* Calculate temporary vectorial force */
470 tx = _mm256_mul_ps(fscal,dx30);
471 ty = _mm256_mul_ps(fscal,dy30);
472 tz = _mm256_mul_ps(fscal,dz30);
473
474 /* Update vectorial force */
475 fix3 = _mm256_add_ps(fix3,tx);
476 fiy3 = _mm256_add_ps(fiy3,ty);
477 fiz3 = _mm256_add_ps(fiz3,tz);
478
479 fjx0 = _mm256_add_ps(fjx0,tx);
480 fjy0 = _mm256_add_ps(fjy0,ty);
481 fjz0 = _mm256_add_ps(fjz0,tz);
482
483 }
484
485 fjptrA = f+j_coord_offsetA;
486 fjptrB = f+j_coord_offsetB;
487 fjptrC = f+j_coord_offsetC;
488 fjptrD = f+j_coord_offsetD;
489 fjptrE = f+j_coord_offsetE;
490 fjptrF = f+j_coord_offsetF;
491 fjptrG = f+j_coord_offsetG;
492 fjptrH = f+j_coord_offsetH;
493
494 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
495
496 /* Inner loop uses 167 flops */
497 }
498
499 if(jidx<j_index_end)
500 {
501
502 /* Get j neighbor index, and coordinate index */
503 jnrlistA = jjnr[jidx];
504 jnrlistB = jjnr[jidx+1];
505 jnrlistC = jjnr[jidx+2];
506 jnrlistD = jjnr[jidx+3];
507 jnrlistE = jjnr[jidx+4];
508 jnrlistF = jjnr[jidx+5];
509 jnrlistG = jjnr[jidx+6];
510 jnrlistH = jjnr[jidx+7];
511 /* Sign of each element will be negative for non-real atoms.
512 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
513 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
514 */
515 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
516 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
517
518 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
519 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
520 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
521 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
522 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
523 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
524 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
525 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
526 j_coord_offsetA = DIM*jnrA;
527 j_coord_offsetB = DIM*jnrB;
528 j_coord_offsetC = DIM*jnrC;
529 j_coord_offsetD = DIM*jnrD;
530 j_coord_offsetE = DIM*jnrE;
531 j_coord_offsetF = DIM*jnrF;
532 j_coord_offsetG = DIM*jnrG;
533 j_coord_offsetH = DIM*jnrH;
534
535 /* load j atom coordinates */
536 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
537 x+j_coord_offsetC,x+j_coord_offsetD,
538 x+j_coord_offsetE,x+j_coord_offsetF,
539 x+j_coord_offsetG,x+j_coord_offsetH,
540 &jx0,&jy0,&jz0);
541
542 /* Calculate displacement vector */
543 dx00 = _mm256_sub_ps(ix0,jx0);
544 dy00 = _mm256_sub_ps(iy0,jy0);
545 dz00 = _mm256_sub_ps(iz0,jz0);
546 dx10 = _mm256_sub_ps(ix1,jx0);
547 dy10 = _mm256_sub_ps(iy1,jy0);
548 dz10 = _mm256_sub_ps(iz1,jz0);
549 dx20 = _mm256_sub_ps(ix2,jx0);
550 dy20 = _mm256_sub_ps(iy2,jy0);
551 dz20 = _mm256_sub_ps(iz2,jz0);
552 dx30 = _mm256_sub_ps(ix3,jx0);
553 dy30 = _mm256_sub_ps(iy3,jy0);
554 dz30 = _mm256_sub_ps(iz3,jz0);
555
556 /* Calculate squared distance and things based on it */
557 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
558 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
559 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
560 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
561
562 rinv00 = avx256_invsqrt_f(rsq00);
563 rinv10 = avx256_invsqrt_f(rsq10);
564 rinv20 = avx256_invsqrt_f(rsq20);
565 rinv30 = avx256_invsqrt_f(rsq30);
566
567 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
568 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
569 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
570
571 /* Load parameters for j particles */
572 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
573 charge+jnrC+0,charge+jnrD+0,
574 charge+jnrE+0,charge+jnrF+0,
575 charge+jnrG+0,charge+jnrH+0);
576 vdwjidx0A = 2*vdwtype[jnrA+0];
577 vdwjidx0B = 2*vdwtype[jnrB+0];
578 vdwjidx0C = 2*vdwtype[jnrC+0];
579 vdwjidx0D = 2*vdwtype[jnrD+0];
580 vdwjidx0E = 2*vdwtype[jnrE+0];
581 vdwjidx0F = 2*vdwtype[jnrF+0];
582 vdwjidx0G = 2*vdwtype[jnrG+0];
583 vdwjidx0H = 2*vdwtype[jnrH+0];
584
585 fjx0 = _mm256_setzero_ps();
586 fjy0 = _mm256_setzero_ps();
587 fjz0 = _mm256_setzero_ps();
588
589 /**************************
590 * CALCULATE INTERACTIONS *
591 **************************/
592
593 r00 = _mm256_mul_ps(rsq00,rinv00);
594 r00 = _mm256_andnot_ps(dummy_mask,r00);
595
596 /* Compute parameters for interactions between i and j atoms */
597 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
598 vdwioffsetptr0+vdwjidx0B,
599 vdwioffsetptr0+vdwjidx0C,
600 vdwioffsetptr0+vdwjidx0D,
601 vdwioffsetptr0+vdwjidx0E,
602 vdwioffsetptr0+vdwjidx0F,
603 vdwioffsetptr0+vdwjidx0G,
604 vdwioffsetptr0+vdwjidx0H,
605 &c6_00,&c12_00);
606
607 /* Calculate table index by multiplying r with table scale and truncate to integer */
608 rt = _mm256_mul_ps(r00,vftabscale);
609 vfitab = _mm256_cvttps_epi32(rt);
610 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
611 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
612 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
613 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
614 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
615 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
616
617 /* CUBIC SPLINE TABLE DISPERSION */
618 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
619 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
620 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
621 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
622 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
623 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
624 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
625 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
626 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
627 Heps = _mm256_mul_ps(vfeps,H);
628 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
629 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
630 vvdw6 = _mm256_mul_ps(c6_00,VV);
631 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
632 fvdw6 = _mm256_mul_ps(c6_00,FF);
633
634 /* CUBIC SPLINE TABLE REPULSION */
635 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
636 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
637 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
638 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
639 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
640 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
641 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
642 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
643 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
644 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
645 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
646 Heps = _mm256_mul_ps(vfeps,H);
647 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
648 VV = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
649 vvdw12 = _mm256_mul_ps(c12_00,VV);
650 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
651 fvdw12 = _mm256_mul_ps(c12_00,FF);
652 vvdw = _mm256_add_ps(vvdw12,vvdw6);
653 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
654
655 /* Update potential sum for this i atom from the interaction with this j atom. */
656 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
657 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
658
659 fscal = fvdw;
660
661 fscal = _mm256_andnot_ps(dummy_mask,fscal);
662
663 /* Calculate temporary vectorial force */
664 tx = _mm256_mul_ps(fscal,dx00);
665 ty = _mm256_mul_ps(fscal,dy00);
666 tz = _mm256_mul_ps(fscal,dz00);
667
668 /* Update vectorial force */
669 fix0 = _mm256_add_ps(fix0,tx);
670 fiy0 = _mm256_add_ps(fiy0,ty);
671 fiz0 = _mm256_add_ps(fiz0,tz);
672
673 fjx0 = _mm256_add_ps(fjx0,tx);
674 fjy0 = _mm256_add_ps(fjy0,ty);
675 fjz0 = _mm256_add_ps(fjz0,tz);
676
677 /**************************
678 * CALCULATE INTERACTIONS *
679 **************************/
680
681 if (gmx_mm256_any_lt(rsq10,rcutoff2))
682 {
683
684 /* Compute parameters for interactions between i and j atoms */
685 qq10 = _mm256_mul_ps(iq1,jq0);
686
687 /* REACTION-FIELD ELECTROSTATICS */
688 velec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
689 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
690
691 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
692
693 /* Update potential sum for this i atom from the interaction with this j atom. */
694 velec = _mm256_and_ps(velec,cutoff_mask);
695 velec = _mm256_andnot_ps(dummy_mask,velec);
696 velecsum = _mm256_add_ps(velecsum,velec);
697
698 fscal = felec;
699
700 fscal = _mm256_and_ps(fscal,cutoff_mask);
701
702 fscal = _mm256_andnot_ps(dummy_mask,fscal);
703
704 /* Calculate temporary vectorial force */
705 tx = _mm256_mul_ps(fscal,dx10);
706 ty = _mm256_mul_ps(fscal,dy10);
707 tz = _mm256_mul_ps(fscal,dz10);
708
709 /* Update vectorial force */
710 fix1 = _mm256_add_ps(fix1,tx);
711 fiy1 = _mm256_add_ps(fiy1,ty);
712 fiz1 = _mm256_add_ps(fiz1,tz);
713
714 fjx0 = _mm256_add_ps(fjx0,tx);
715 fjy0 = _mm256_add_ps(fjy0,ty);
716 fjz0 = _mm256_add_ps(fjz0,tz);
717
718 }
719
720 /**************************
721 * CALCULATE INTERACTIONS *
722 **************************/
723
724 if (gmx_mm256_any_lt(rsq20,rcutoff2))
725 {
726
727 /* Compute parameters for interactions between i and j atoms */
728 qq20 = _mm256_mul_ps(iq2,jq0);
729
730 /* REACTION-FIELD ELECTROSTATICS */
731 velec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
732 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
733
734 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
735
736 /* Update potential sum for this i atom from the interaction with this j atom. */
737 velec = _mm256_and_ps(velec,cutoff_mask);
738 velec = _mm256_andnot_ps(dummy_mask,velec);
739 velecsum = _mm256_add_ps(velecsum,velec);
740
741 fscal = felec;
742
743 fscal = _mm256_and_ps(fscal,cutoff_mask);
744
745 fscal = _mm256_andnot_ps(dummy_mask,fscal);
746
747 /* Calculate temporary vectorial force */
748 tx = _mm256_mul_ps(fscal,dx20);
749 ty = _mm256_mul_ps(fscal,dy20);
750 tz = _mm256_mul_ps(fscal,dz20);
751
752 /* Update vectorial force */
753 fix2 = _mm256_add_ps(fix2,tx);
754 fiy2 = _mm256_add_ps(fiy2,ty);
755 fiz2 = _mm256_add_ps(fiz2,tz);
756
757 fjx0 = _mm256_add_ps(fjx0,tx);
758 fjy0 = _mm256_add_ps(fjy0,ty);
759 fjz0 = _mm256_add_ps(fjz0,tz);
760
761 }
762
763 /**************************
764 * CALCULATE INTERACTIONS *
765 **************************/
766
767 if (gmx_mm256_any_lt(rsq30,rcutoff2))
768 {
769
770 /* Compute parameters for interactions between i and j atoms */
771 qq30 = _mm256_mul_ps(iq3,jq0);
772
773 /* REACTION-FIELD ELECTROSTATICS */
774 velec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
775 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
776
777 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
778
779 /* Update potential sum for this i atom from the interaction with this j atom. */
780 velec = _mm256_and_ps(velec,cutoff_mask);
781 velec = _mm256_andnot_ps(dummy_mask,velec);
782 velecsum = _mm256_add_ps(velecsum,velec);
783
784 fscal = felec;
785
786 fscal = _mm256_and_ps(fscal,cutoff_mask);
787
788 fscal = _mm256_andnot_ps(dummy_mask,fscal);
789
790 /* Calculate temporary vectorial force */
791 tx = _mm256_mul_ps(fscal,dx30);
792 ty = _mm256_mul_ps(fscal,dy30);
793 tz = _mm256_mul_ps(fscal,dz30);
794
795 /* Update vectorial force */
796 fix3 = _mm256_add_ps(fix3,tx);
797 fiy3 = _mm256_add_ps(fiy3,ty);
798 fiz3 = _mm256_add_ps(fiz3,tz);
799
800 fjx0 = _mm256_add_ps(fjx0,tx);
801 fjy0 = _mm256_add_ps(fjy0,ty);
802 fjz0 = _mm256_add_ps(fjz0,tz);
803
804 }
805
806 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
807 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
808 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
809 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
810 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
811 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
812 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
813 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
814
815 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
816
817 /* Inner loop uses 168 flops */
818 }
819
820 /* End of innermost loop */
821
822 gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
823 f+i_coord_offset,fshift+i_shift_offset);
824
825 ggid = gid[iidx];
826 /* Update potential energies */
827 gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
828 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
829
830 /* Increment number of inner iterations */
831 inneriter += j_index_end - j_index_start;
832
833 /* Outer loop uses 26 flops */
834 }
835
836 /* Increment number of outer iterations */
837 outeriter += nri;
838
839 /* Update outer/inner flops */
840
841 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*168);
842 }
843 /*
844 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_256_single
845 * Electrostatics interaction: ReactionField
846 * VdW interaction: CubicSplineTable
847 * Geometry: Water4-Particle
848 * Calculate force/pot: Force
849 */
850 void
851 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_256_single
852 (t_nblist * gmx_restrict nlist,
853 rvec * gmx_restrict xx,
854 rvec * gmx_restrict ff,
855 struct t_forcerec * gmx_restrict fr,
856 t_mdatoms * gmx_restrict mdatoms,
857 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
858 t_nrnb * gmx_restrict nrnb)
859 {
860 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
861 * just 0 for non-waters.
862 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
863 * jnr indices corresponding to data put in the four positions in the SIMD register.
864 */
865 int i_shift_offset,i_coord_offset,outeriter,inneriter;
866 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
867 int jnrA,jnrB,jnrC,jnrD;
868 int jnrE,jnrF,jnrG,jnrH;
869 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
870 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
871 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
872 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
873 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
874 real rcutoff_scalar;
875 real *shiftvec,*fshift,*x,*f;
876 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
877 real scratch[4*DIM];
878 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
879 real * vdwioffsetptr0;
880 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
881 real * vdwioffsetptr1;
882 __m256 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
883 real * vdwioffsetptr2;
884 __m256 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
885 real * vdwioffsetptr3;
886 __m256 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
887 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
888 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
889 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
890 __m256 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
891 __m256 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
892 __m256 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
893 __m256 velec,felec,velecsum,facel,crf,krf,krf2;
894 real *charge;
895 int nvdwtype;
896 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
897 int *vdwtype;
898 real *vdwparam;
899 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
900 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
901 __m256i vfitab;
902 __m128i vfitab_lo,vfitab_hi;
903 __m128i ifour = _mm_set1_epi32(4);
904 __m256 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
905 real *vftab;
906 __m256 dummy_mask,cutoff_mask;
907 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
908 __m256 one = _mm256_set1_ps(1.0);
909 __m256 two = _mm256_set1_ps(2.0);
910 x = xx[0];
911 f = ff[0];
912
913 nri = nlist->nri;
914 iinr = nlist->iinr;
915 jindex = nlist->jindex;
916 jjnr = nlist->jjnr;
917 shiftidx = nlist->shift;
918 gid = nlist->gid;
919 shiftvec = fr->shift_vec[0];
920 fshift = fr->fshift[0];
921 facel = _mm256_set1_ps(fr->ic->epsfac);
922 charge = mdatoms->chargeA;
923 krf = _mm256_set1_ps(fr->ic->k_rf);
924 krf2 = _mm256_set1_ps(fr->ic->k_rf*2.0);
925 crf = _mm256_set1_ps(fr->ic->c_rf);
926 nvdwtype = fr->ntype;
927 vdwparam = fr->nbfp;
928 vdwtype = mdatoms->typeA;
929
930 vftab = kernel_data->table_vdw->data;
931 vftabscale = _mm256_set1_ps(kernel_data->table_vdw->scale);
932
933 /* Setup water-specific parameters */
934 inr = nlist->iinr[0];
935 iq1 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
936 iq2 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
937 iq3 = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
938 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
939
940 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
941 rcutoff_scalar = fr->ic->rcoulomb;
942 rcutoff = _mm256_set1_ps(rcutoff_scalar);
943 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
944
945 /* Avoid stupid compiler warnings */
946 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
947 j_coord_offsetA = 0;
948 j_coord_offsetB = 0;
949 j_coord_offsetC = 0;
950 j_coord_offsetD = 0;
951 j_coord_offsetE = 0;
952 j_coord_offsetF = 0;
953 j_coord_offsetG = 0;
954 j_coord_offsetH = 0;
955
956 outeriter = 0;
957 inneriter = 0;
958
959 for(iidx=0;iidx<4*DIM;iidx++)
960 {
961 scratch[iidx] = 0.0;
962 }
963
964 /* Start outer loop over neighborlists */
965 for(iidx=0; iidx<nri; iidx++)
966 {
967 /* Load shift vector for this list */
968 i_shift_offset = DIM*shiftidx[iidx];
969
970 /* Load limits for loop over neighbors */
971 j_index_start = jindex[iidx];
972 j_index_end = jindex[iidx+1];
973
974 /* Get outer coordinate index */
975 inr = iinr[iidx];
976 i_coord_offset = DIM*inr;
977
978 /* Load i particle coords and add shift vector */
979 gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
980 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
981
982 fix0 = _mm256_setzero_ps();
983 fiy0 = _mm256_setzero_ps();
984 fiz0 = _mm256_setzero_ps();
985 fix1 = _mm256_setzero_ps();
986 fiy1 = _mm256_setzero_ps();
987 fiz1 = _mm256_setzero_ps();
988 fix2 = _mm256_setzero_ps();
989 fiy2 = _mm256_setzero_ps();
990 fiz2 = _mm256_setzero_ps();
991 fix3 = _mm256_setzero_ps();
992 fiy3 = _mm256_setzero_ps();
993 fiz3 = _mm256_setzero_ps();
994
995 /* Start inner kernel loop */
996 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
997 {
998
999 /* Get j neighbor index, and coordinate index */
1000 jnrA = jjnr[jidx];
1001 jnrB = jjnr[jidx+1];
1002 jnrC = jjnr[jidx+2];
1003 jnrD = jjnr[jidx+3];
1004 jnrE = jjnr[jidx+4];
1005 jnrF = jjnr[jidx+5];
1006 jnrG = jjnr[jidx+6];
1007 jnrH = jjnr[jidx+7];
1008 j_coord_offsetA = DIM*jnrA;
1009 j_coord_offsetB = DIM*jnrB;
1010 j_coord_offsetC = DIM*jnrC;
1011 j_coord_offsetD = DIM*jnrD;
1012 j_coord_offsetE = DIM*jnrE;
1013 j_coord_offsetF = DIM*jnrF;
1014 j_coord_offsetG = DIM*jnrG;
1015 j_coord_offsetH = DIM*jnrH;
1016
1017 /* load j atom coordinates */
1018 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1019 x+j_coord_offsetC,x+j_coord_offsetD,
1020 x+j_coord_offsetE,x+j_coord_offsetF,
1021 x+j_coord_offsetG,x+j_coord_offsetH,
1022 &jx0,&jy0,&jz0);
1023
1024 /* Calculate displacement vector */
1025 dx00 = _mm256_sub_ps(ix0,jx0);
1026 dy00 = _mm256_sub_ps(iy0,jy0);
1027 dz00 = _mm256_sub_ps(iz0,jz0);
1028 dx10 = _mm256_sub_ps(ix1,jx0);
1029 dy10 = _mm256_sub_ps(iy1,jy0);
1030 dz10 = _mm256_sub_ps(iz1,jz0);
1031 dx20 = _mm256_sub_ps(ix2,jx0);
1032 dy20 = _mm256_sub_ps(iy2,jy0);
1033 dz20 = _mm256_sub_ps(iz2,jz0);
1034 dx30 = _mm256_sub_ps(ix3,jx0);
1035 dy30 = _mm256_sub_ps(iy3,jy0);
1036 dz30 = _mm256_sub_ps(iz3,jz0);
1037
1038 /* Calculate squared distance and things based on it */
1039 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1040 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1041 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1042 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1043
1044 rinv00 = avx256_invsqrt_f(rsq00);
1045 rinv10 = avx256_invsqrt_f(rsq10);
1046 rinv20 = avx256_invsqrt_f(rsq20);
1047 rinv30 = avx256_invsqrt_f(rsq30);
1048
1049 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1050 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1051 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
1052
1053 /* Load parameters for j particles */
1054 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1055 charge+jnrC+0,charge+jnrD+0,
1056 charge+jnrE+0,charge+jnrF+0,
1057 charge+jnrG+0,charge+jnrH+0);
1058 vdwjidx0A = 2*vdwtype[jnrA+0];
1059 vdwjidx0B = 2*vdwtype[jnrB+0];
1060 vdwjidx0C = 2*vdwtype[jnrC+0];
1061 vdwjidx0D = 2*vdwtype[jnrD+0];
1062 vdwjidx0E = 2*vdwtype[jnrE+0];
1063 vdwjidx0F = 2*vdwtype[jnrF+0];
1064 vdwjidx0G = 2*vdwtype[jnrG+0];
1065 vdwjidx0H = 2*vdwtype[jnrH+0];
1066
1067 fjx0 = _mm256_setzero_ps();
1068 fjy0 = _mm256_setzero_ps();
1069 fjz0 = _mm256_setzero_ps();
1070
1071 /**************************
1072 * CALCULATE INTERACTIONS *
1073 **************************/
1074
1075 r00 = _mm256_mul_ps(rsq00,rinv00);
1076
1077 /* Compute parameters for interactions between i and j atoms */
1078 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1079 vdwioffsetptr0+vdwjidx0B,
1080 vdwioffsetptr0+vdwjidx0C,
1081 vdwioffsetptr0+vdwjidx0D,
1082 vdwioffsetptr0+vdwjidx0E,
1083 vdwioffsetptr0+vdwjidx0F,
1084 vdwioffsetptr0+vdwjidx0G,
1085 vdwioffsetptr0+vdwjidx0H,
1086 &c6_00,&c12_00);
1087
1088 /* Calculate table index by multiplying r with table scale and truncate to integer */
1089 rt = _mm256_mul_ps(r00,vftabscale);
1090 vfitab = _mm256_cvttps_epi32(rt);
1091 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1092 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1093 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
1094 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
1095 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
1096 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
1097
1098 /* CUBIC SPLINE TABLE DISPERSION */
1099 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1100 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1101 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1102 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1103 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1104 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1105 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1106 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1107 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1108 Heps = _mm256_mul_ps(vfeps,H);
1109 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1110 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1111 fvdw6 = _mm256_mul_ps(c6_00,FF);
1112
1113 /* CUBIC SPLINE TABLE REPULSION */
1114 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
1115 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
1116 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1117 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1118 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1119 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1120 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1121 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1122 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1123 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1124 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1125 Heps = _mm256_mul_ps(vfeps,H);
1126 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1127 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1128 fvdw12 = _mm256_mul_ps(c12_00,FF);
1129 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1130
1131 fscal = fvdw;
1132
1133 /* Calculate temporary vectorial force */
1134 tx = _mm256_mul_ps(fscal,dx00);
1135 ty = _mm256_mul_ps(fscal,dy00);
1136 tz = _mm256_mul_ps(fscal,dz00);
1137
1138 /* Update vectorial force */
1139 fix0 = _mm256_add_ps(fix0,tx);
1140 fiy0 = _mm256_add_ps(fiy0,ty);
1141 fiz0 = _mm256_add_ps(fiz0,tz);
1142
1143 fjx0 = _mm256_add_ps(fjx0,tx);
1144 fjy0 = _mm256_add_ps(fjy0,ty);
1145 fjz0 = _mm256_add_ps(fjz0,tz);
1146
1147 /**************************
1148 * CALCULATE INTERACTIONS *
1149 **************************/
1150
1151 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1152 {
1153
1154 /* Compute parameters for interactions between i and j atoms */
1155 qq10 = _mm256_mul_ps(iq1,jq0);
1156
1157 /* REACTION-FIELD ELECTROSTATICS */
1158 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1159
1160 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1161
1162 fscal = felec;
1163
1164 fscal = _mm256_and_ps(fscal,cutoff_mask);
1165
1166 /* Calculate temporary vectorial force */
1167 tx = _mm256_mul_ps(fscal,dx10);
1168 ty = _mm256_mul_ps(fscal,dy10);
1169 tz = _mm256_mul_ps(fscal,dz10);
1170
1171 /* Update vectorial force */
1172 fix1 = _mm256_add_ps(fix1,tx);
1173 fiy1 = _mm256_add_ps(fiy1,ty);
1174 fiz1 = _mm256_add_ps(fiz1,tz);
1175
1176 fjx0 = _mm256_add_ps(fjx0,tx);
1177 fjy0 = _mm256_add_ps(fjy0,ty);
1178 fjz0 = _mm256_add_ps(fjz0,tz);
1179
1180 }
1181
1182 /**************************
1183 * CALCULATE INTERACTIONS *
1184 **************************/
1185
1186 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1187 {
1188
1189 /* Compute parameters for interactions between i and j atoms */
1190 qq20 = _mm256_mul_ps(iq2,jq0);
1191
1192 /* REACTION-FIELD ELECTROSTATICS */
1193 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1194
1195 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1196
1197 fscal = felec;
1198
1199 fscal = _mm256_and_ps(fscal,cutoff_mask);
1200
1201 /* Calculate temporary vectorial force */
1202 tx = _mm256_mul_ps(fscal,dx20);
1203 ty = _mm256_mul_ps(fscal,dy20);
1204 tz = _mm256_mul_ps(fscal,dz20);
1205
1206 /* Update vectorial force */
1207 fix2 = _mm256_add_ps(fix2,tx);
1208 fiy2 = _mm256_add_ps(fiy2,ty);
1209 fiz2 = _mm256_add_ps(fiz2,tz);
1210
1211 fjx0 = _mm256_add_ps(fjx0,tx);
1212 fjy0 = _mm256_add_ps(fjy0,ty);
1213 fjz0 = _mm256_add_ps(fjz0,tz);
1214
1215 }
1216
1217 /**************************
1218 * CALCULATE INTERACTIONS *
1219 **************************/
1220
1221 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1222 {
1223
1224 /* Compute parameters for interactions between i and j atoms */
1225 qq30 = _mm256_mul_ps(iq3,jq0);
1226
1227 /* REACTION-FIELD ELECTROSTATICS */
1228 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1229
1230 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1231
1232 fscal = felec;
1233
1234 fscal = _mm256_and_ps(fscal,cutoff_mask);
1235
1236 /* Calculate temporary vectorial force */
1237 tx = _mm256_mul_ps(fscal,dx30);
1238 ty = _mm256_mul_ps(fscal,dy30);
1239 tz = _mm256_mul_ps(fscal,dz30);
1240
1241 /* Update vectorial force */
1242 fix3 = _mm256_add_ps(fix3,tx);
1243 fiy3 = _mm256_add_ps(fiy3,ty);
1244 fiz3 = _mm256_add_ps(fiz3,tz);
1245
1246 fjx0 = _mm256_add_ps(fjx0,tx);
1247 fjy0 = _mm256_add_ps(fjy0,ty);
1248 fjz0 = _mm256_add_ps(fjz0,tz);
1249
1250 }
1251
1252 fjptrA = f+j_coord_offsetA;
1253 fjptrB = f+j_coord_offsetB;
1254 fjptrC = f+j_coord_offsetC;
1255 fjptrD = f+j_coord_offsetD;
1256 fjptrE = f+j_coord_offsetE;
1257 fjptrF = f+j_coord_offsetF;
1258 fjptrG = f+j_coord_offsetG;
1259 fjptrH = f+j_coord_offsetH;
1260
1261 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1262
1263 /* Inner loop uses 141 flops */
1264 }
1265
1266 if(jidx<j_index_end)
1267 {
1268
1269 /* Get j neighbor index, and coordinate index */
1270 jnrlistA = jjnr[jidx];
1271 jnrlistB = jjnr[jidx+1];
1272 jnrlistC = jjnr[jidx+2];
1273 jnrlistD = jjnr[jidx+3];
1274 jnrlistE = jjnr[jidx+4];
1275 jnrlistF = jjnr[jidx+5];
1276 jnrlistG = jjnr[jidx+6];
1277 jnrlistH = jjnr[jidx+7];
1278 /* Sign of each element will be negative for non-real atoms.
1279 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1280 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1281 */
1282 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1283 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1284
1285 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1286 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1287 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1288 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1289 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
1290 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
1291 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
1292 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
1293 j_coord_offsetA = DIM*jnrA;
1294 j_coord_offsetB = DIM*jnrB;
1295 j_coord_offsetC = DIM*jnrC;
1296 j_coord_offsetD = DIM*jnrD;
1297 j_coord_offsetE = DIM*jnrE;
1298 j_coord_offsetF = DIM*jnrF;
1299 j_coord_offsetG = DIM*jnrG;
1300 j_coord_offsetH = DIM*jnrH;
1301
1302 /* load j atom coordinates */
1303 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1304 x+j_coord_offsetC,x+j_coord_offsetD,
1305 x+j_coord_offsetE,x+j_coord_offsetF,
1306 x+j_coord_offsetG,x+j_coord_offsetH,
1307 &jx0,&jy0,&jz0);
1308
1309 /* Calculate displacement vector */
1310 dx00 = _mm256_sub_ps(ix0,jx0);
1311 dy00 = _mm256_sub_ps(iy0,jy0);
1312 dz00 = _mm256_sub_ps(iz0,jz0);
1313 dx10 = _mm256_sub_ps(ix1,jx0);
1314 dy10 = _mm256_sub_ps(iy1,jy0);
1315 dz10 = _mm256_sub_ps(iz1,jz0);
1316 dx20 = _mm256_sub_ps(ix2,jx0);
1317 dy20 = _mm256_sub_ps(iy2,jy0);
1318 dz20 = _mm256_sub_ps(iz2,jz0);
1319 dx30 = _mm256_sub_ps(ix3,jx0);
1320 dy30 = _mm256_sub_ps(iy3,jy0);
1321 dz30 = _mm256_sub_ps(iz3,jz0);
1322
1323 /* Calculate squared distance and things based on it */
1324 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1325 rsq10 = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1326 rsq20 = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1327 rsq30 = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1328
1329 rinv00 = avx256_invsqrt_f(rsq00);
1330 rinv10 = avx256_invsqrt_f(rsq10);
1331 rinv20 = avx256_invsqrt_f(rsq20);
1332 rinv30 = avx256_invsqrt_f(rsq30);
1333
1334 rinvsq10 = _mm256_mul_ps(rinv10,rinv10);
1335 rinvsq20 = _mm256_mul_ps(rinv20,rinv20);
1336 rinvsq30 = _mm256_mul_ps(rinv30,rinv30);
1337
1338 /* Load parameters for j particles */
1339 jq0 = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1340 charge+jnrC+0,charge+jnrD+0,
1341 charge+jnrE+0,charge+jnrF+0,
1342 charge+jnrG+0,charge+jnrH+0);
1343 vdwjidx0A = 2*vdwtype[jnrA+0];
1344 vdwjidx0B = 2*vdwtype[jnrB+0];
1345 vdwjidx0C = 2*vdwtype[jnrC+0];
1346 vdwjidx0D = 2*vdwtype[jnrD+0];
1347 vdwjidx0E = 2*vdwtype[jnrE+0];
1348 vdwjidx0F = 2*vdwtype[jnrF+0];
1349 vdwjidx0G = 2*vdwtype[jnrG+0];
1350 vdwjidx0H = 2*vdwtype[jnrH+0];
1351
1352 fjx0 = _mm256_setzero_ps();
1353 fjy0 = _mm256_setzero_ps();
1354 fjz0 = _mm256_setzero_ps();
1355
1356 /**************************
1357 * CALCULATE INTERACTIONS *
1358 **************************/
1359
1360 r00 = _mm256_mul_ps(rsq00,rinv00);
1361 r00 = _mm256_andnot_ps(dummy_mask,r00);
1362
1363 /* Compute parameters for interactions between i and j atoms */
1364 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1365 vdwioffsetptr0+vdwjidx0B,
1366 vdwioffsetptr0+vdwjidx0C,
1367 vdwioffsetptr0+vdwjidx0D,
1368 vdwioffsetptr0+vdwjidx0E,
1369 vdwioffsetptr0+vdwjidx0F,
1370 vdwioffsetptr0+vdwjidx0G,
1371 vdwioffsetptr0+vdwjidx0H,
1372 &c6_00,&c12_00);
1373
1374 /* Calculate table index by multiplying r with table scale and truncate to integer */
1375 rt = _mm256_mul_ps(r00,vftabscale);
1376 vfitab = _mm256_cvttps_epi32(rt);
1377 vfeps = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1378 /* AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1379 vfitab_lo = _mm256_extractf128_si256(vfitab,0x0);
1380 vfitab_hi = _mm256_extractf128_si256(vfitab,0x1);
1381 vfitab_lo = _mm_slli_epi32(vfitab_lo,3);
1382 vfitab_hi = _mm_slli_epi32(vfitab_hi,3);
1383
1384 /* CUBIC SPLINE TABLE DISPERSION */
1385 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1386 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1387 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1388 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1389 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1390 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1391 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1392 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1393 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1394 Heps = _mm256_mul_ps(vfeps,H);
1395 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1396 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1397 fvdw6 = _mm256_mul_ps(c6_00,FF);
1398
1399 /* CUBIC SPLINE TABLE REPULSION */
1400 vfitab_lo = _mm_add_epi32(vfitab_lo,ifour);
1401 vfitab_hi = _mm_add_epi32(vfitab_hi,ifour);
1402 Y = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1403 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1404 F = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1405 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1406 G = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1407 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1408 H = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1409 _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1410 GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1411 Heps = _mm256_mul_ps(vfeps,H);
1412 Fp = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1413 FF = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1414 fvdw12 = _mm256_mul_ps(c12_00,FF);
1415 fvdw = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1416
1417 fscal = fvdw;
1418
1419 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1420
1421 /* Calculate temporary vectorial force */
1422 tx = _mm256_mul_ps(fscal,dx00);
1423 ty = _mm256_mul_ps(fscal,dy00);
1424 tz = _mm256_mul_ps(fscal,dz00);
1425
1426 /* Update vectorial force */
1427 fix0 = _mm256_add_ps(fix0,tx);
1428 fiy0 = _mm256_add_ps(fiy0,ty);
1429 fiz0 = _mm256_add_ps(fiz0,tz);
1430
1431 fjx0 = _mm256_add_ps(fjx0,tx);
1432 fjy0 = _mm256_add_ps(fjy0,ty);
1433 fjz0 = _mm256_add_ps(fjz0,tz);
1434
1435 /**************************
1436 * CALCULATE INTERACTIONS *
1437 **************************/
1438
1439 if (gmx_mm256_any_lt(rsq10,rcutoff2))
1440 {
1441
1442 /* Compute parameters for interactions between i and j atoms */
1443 qq10 = _mm256_mul_ps(iq1,jq0);
1444
1445 /* REACTION-FIELD ELECTROSTATICS */
1446 felec = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1447
1448 cutoff_mask = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1449
1450 fscal = felec;
1451
1452 fscal = _mm256_and_ps(fscal,cutoff_mask);
1453
1454 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1455
1456 /* Calculate temporary vectorial force */
1457 tx = _mm256_mul_ps(fscal,dx10);
1458 ty = _mm256_mul_ps(fscal,dy10);
1459 tz = _mm256_mul_ps(fscal,dz10);
1460
1461 /* Update vectorial force */
1462 fix1 = _mm256_add_ps(fix1,tx);
1463 fiy1 = _mm256_add_ps(fiy1,ty);
1464 fiz1 = _mm256_add_ps(fiz1,tz);
1465
1466 fjx0 = _mm256_add_ps(fjx0,tx);
1467 fjy0 = _mm256_add_ps(fjy0,ty);
1468 fjz0 = _mm256_add_ps(fjz0,tz);
1469
1470 }
1471
1472 /**************************
1473 * CALCULATE INTERACTIONS *
1474 **************************/
1475
1476 if (gmx_mm256_any_lt(rsq20,rcutoff2))
1477 {
1478
1479 /* Compute parameters for interactions between i and j atoms */
1480 qq20 = _mm256_mul_ps(iq2,jq0);
1481
1482 /* REACTION-FIELD ELECTROSTATICS */
1483 felec = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1484
1485 cutoff_mask = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1486
1487 fscal = felec;
1488
1489 fscal = _mm256_and_ps(fscal,cutoff_mask);
1490
1491 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1492
1493 /* Calculate temporary vectorial force */
1494 tx = _mm256_mul_ps(fscal,dx20);
1495 ty = _mm256_mul_ps(fscal,dy20);
1496 tz = _mm256_mul_ps(fscal,dz20);
1497
1498 /* Update vectorial force */
1499 fix2 = _mm256_add_ps(fix2,tx);
1500 fiy2 = _mm256_add_ps(fiy2,ty);
1501 fiz2 = _mm256_add_ps(fiz2,tz);
1502
1503 fjx0 = _mm256_add_ps(fjx0,tx);
1504 fjy0 = _mm256_add_ps(fjy0,ty);
1505 fjz0 = _mm256_add_ps(fjz0,tz);
1506
1507 }
1508
1509 /**************************
1510 * CALCULATE INTERACTIONS *
1511 **************************/
1512
1513 if (gmx_mm256_any_lt(rsq30,rcutoff2))
1514 {
1515
1516 /* Compute parameters for interactions between i and j atoms */
1517 qq30 = _mm256_mul_ps(iq3,jq0);
1518
1519 /* REACTION-FIELD ELECTROSTATICS */
1520 felec = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1521
1522 cutoff_mask = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1523
1524 fscal = felec;
1525
1526 fscal = _mm256_and_ps(fscal,cutoff_mask);
1527
1528 fscal = _mm256_andnot_ps(dummy_mask,fscal);
1529
1530 /* Calculate temporary vectorial force */
1531 tx = _mm256_mul_ps(fscal,dx30);
1532 ty = _mm256_mul_ps(fscal,dy30);
1533 tz = _mm256_mul_ps(fscal,dz30);
1534
1535 /* Update vectorial force */
1536 fix3 = _mm256_add_ps(fix3,tx);
1537 fiy3 = _mm256_add_ps(fiy3,ty);
1538 fiz3 = _mm256_add_ps(fiz3,tz);
1539
1540 fjx0 = _mm256_add_ps(fjx0,tx);
1541 fjy0 = _mm256_add_ps(fjy0,ty);
1542 fjz0 = _mm256_add_ps(fjz0,tz);
1543
1544 }
1545
1546 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1547 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1548 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1549 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1550 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1551 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1552 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1553 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1554
1555 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1556
1557 /* Inner loop uses 142 flops */
1558 }
1559
1560 /* End of innermost loop */
1561
1562 gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1563 f+i_coord_offset,fshift+i_shift_offset);
1564
1565 /* Increment number of inner iterations */
1566 inneriter += j_index_end - j_index_start;
1567
1568 /* Outer loop uses 24 flops */
1569 }
1570
1571 /* Increment number of outer iterations */
1572 outeriter += nri;
1573
1574 /* Update outer/inner flops */
1575
1576 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*142);
1577 }
The ultimate molecular dynamics simulation package