?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed simple.h from nb_kernel_sse4_1_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_sse4_1_single / nb_kernel_ElecRF_VdwCSTab_GeomW3P1_sse4_1_single.c
blob492bcf5411cb47ea6b3f7ccb365e0abf714406bd
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
8 *
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
13 *
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
23 *
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
31 *
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
34 */
35 /*
36 * Note: this file was generated by the GROMACS sse4_1_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/math/vec.h"
46 #include "gromacs/legacyheaders/nrnb.h"
47
48 #include "gromacs/simd/math_x86_sse4_1_single.h"
49 #include "kernelutil_x86_sse4_1_single.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse4_1_single
53 * Electrostatics interaction: ReactionField
54 * VdW interaction: CubicSplineTable
55 * Geometry: Water3-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse4_1_single
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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
76 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real rcutoff_scalar;
80 real *shiftvec,*fshift,*x,*f;
81 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 real scratch[4*DIM];
83 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 int vdwioffset0;
85 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 int vdwioffset1;
87 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 int vdwioffset2;
89 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
91 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
93 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
94 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
95 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
96 real *charge;
97 int nvdwtype;
98 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
99 int *vdwtype;
100 real *vdwparam;
101 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
102 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
103 __m128i vfitab;
104 __m128i ifour = _mm_set1_epi32(4);
105 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
106 real *vftab;
107 __m128 dummy_mask,cutoff_mask;
108 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
109 __m128 one = _mm_set1_ps(1.0);
110 __m128 two = _mm_set1_ps(2.0);
111 x = xx[0];
112 f = ff[0];
113
114 nri = nlist->nri;
115 iinr = nlist->iinr;
116 jindex = nlist->jindex;
117 jjnr = nlist->jjnr;
118 shiftidx = nlist->shift;
119 gid = nlist->gid;
120 shiftvec = fr->shift_vec[0];
121 fshift = fr->fshift[0];
122 facel = _mm_set1_ps(fr->epsfac);
123 charge = mdatoms->chargeA;
124 krf = _mm_set1_ps(fr->ic->k_rf);
125 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
126 crf = _mm_set1_ps(fr->ic->c_rf);
127 nvdwtype = fr->ntype;
128 vdwparam = fr->nbfp;
129 vdwtype = mdatoms->typeA;
130
131 vftab = kernel_data->table_vdw->data;
132 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
133
134 /* Setup water-specific parameters */
135 inr = nlist->iinr[0];
136 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
137 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
138 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
139 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
140
141 /* Avoid stupid compiler warnings */
142 jnrA = jnrB = jnrC = jnrD = 0;
143 j_coord_offsetA = 0;
144 j_coord_offsetB = 0;
145 j_coord_offsetC = 0;
146 j_coord_offsetD = 0;
147
148 outeriter = 0;
149 inneriter = 0;
150
151 for(iidx=0;iidx<4*DIM;iidx++)
152 {
153 scratch[iidx] = 0.0;
154 }
155
156 /* Start outer loop over neighborlists */
157 for(iidx=0; iidx<nri; iidx++)
158 {
159 /* Load shift vector for this list */
160 i_shift_offset = DIM*shiftidx[iidx];
161
162 /* Load limits for loop over neighbors */
163 j_index_start = jindex[iidx];
164 j_index_end = jindex[iidx+1];
165
166 /* Get outer coordinate index */
167 inr = iinr[iidx];
168 i_coord_offset = DIM*inr;
169
170 /* Load i particle coords and add shift vector */
171 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
172 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
173
174 fix0 = _mm_setzero_ps();
175 fiy0 = _mm_setzero_ps();
176 fiz0 = _mm_setzero_ps();
177 fix1 = _mm_setzero_ps();
178 fiy1 = _mm_setzero_ps();
179 fiz1 = _mm_setzero_ps();
180 fix2 = _mm_setzero_ps();
181 fiy2 = _mm_setzero_ps();
182 fiz2 = _mm_setzero_ps();
183
184 /* Reset potential sums */
185 velecsum = _mm_setzero_ps();
186 vvdwsum = _mm_setzero_ps();
187
188 /* Start inner kernel loop */
189 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
190 {
191
192 /* Get j neighbor index, and coordinate index */
193 jnrA = jjnr[jidx];
194 jnrB = jjnr[jidx+1];
195 jnrC = jjnr[jidx+2];
196 jnrD = jjnr[jidx+3];
197 j_coord_offsetA = DIM*jnrA;
198 j_coord_offsetB = DIM*jnrB;
199 j_coord_offsetC = DIM*jnrC;
200 j_coord_offsetD = DIM*jnrD;
201
202 /* load j atom coordinates */
203 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
204 x+j_coord_offsetC,x+j_coord_offsetD,
205 &jx0,&jy0,&jz0);
206
207 /* Calculate displacement vector */
208 dx00 = _mm_sub_ps(ix0,jx0);
209 dy00 = _mm_sub_ps(iy0,jy0);
210 dz00 = _mm_sub_ps(iz0,jz0);
211 dx10 = _mm_sub_ps(ix1,jx0);
212 dy10 = _mm_sub_ps(iy1,jy0);
213 dz10 = _mm_sub_ps(iz1,jz0);
214 dx20 = _mm_sub_ps(ix2,jx0);
215 dy20 = _mm_sub_ps(iy2,jy0);
216 dz20 = _mm_sub_ps(iz2,jz0);
217
218 /* Calculate squared distance and things based on it */
219 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
220 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
221 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
222
223 rinv00 = gmx_mm_invsqrt_ps(rsq00);
224 rinv10 = gmx_mm_invsqrt_ps(rsq10);
225 rinv20 = gmx_mm_invsqrt_ps(rsq20);
226
227 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
228 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
229 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
230
231 /* Load parameters for j particles */
232 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
233 charge+jnrC+0,charge+jnrD+0);
234 vdwjidx0A = 2*vdwtype[jnrA+0];
235 vdwjidx0B = 2*vdwtype[jnrB+0];
236 vdwjidx0C = 2*vdwtype[jnrC+0];
237 vdwjidx0D = 2*vdwtype[jnrD+0];
238
239 fjx0 = _mm_setzero_ps();
240 fjy0 = _mm_setzero_ps();
241 fjz0 = _mm_setzero_ps();
242
243 /**************************
244 * CALCULATE INTERACTIONS *
245 **************************/
246
247 r00 = _mm_mul_ps(rsq00,rinv00);
248
249 /* Compute parameters for interactions between i and j atoms */
250 qq00 = _mm_mul_ps(iq0,jq0);
251 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
252 vdwparam+vdwioffset0+vdwjidx0B,
253 vdwparam+vdwioffset0+vdwjidx0C,
254 vdwparam+vdwioffset0+vdwjidx0D,
255 &c6_00,&c12_00);
256
257 /* Calculate table index by multiplying r with table scale and truncate to integer */
258 rt = _mm_mul_ps(r00,vftabscale);
259 vfitab = _mm_cvttps_epi32(rt);
260 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
261 vfitab = _mm_slli_epi32(vfitab,3);
262
263 /* REACTION-FIELD ELECTROSTATICS */
264 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
265 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
266
267 /* CUBIC SPLINE TABLE DISPERSION */
268 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
269 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
270 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
271 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
272 _MM_TRANSPOSE4_PS(Y,F,G,H);
273 Heps = _mm_mul_ps(vfeps,H);
274 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
275 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
276 vvdw6 = _mm_mul_ps(c6_00,VV);
277 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
278 fvdw6 = _mm_mul_ps(c6_00,FF);
279
280 /* CUBIC SPLINE TABLE REPULSION */
281 vfitab = _mm_add_epi32(vfitab,ifour);
282 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
283 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
284 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
285 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
286 _MM_TRANSPOSE4_PS(Y,F,G,H);
287 Heps = _mm_mul_ps(vfeps,H);
288 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
289 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
290 vvdw12 = _mm_mul_ps(c12_00,VV);
291 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
292 fvdw12 = _mm_mul_ps(c12_00,FF);
293 vvdw = _mm_add_ps(vvdw12,vvdw6);
294 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
295
296 /* Update potential sum for this i atom from the interaction with this j atom. */
297 velecsum = _mm_add_ps(velecsum,velec);
298 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
299
300 fscal = _mm_add_ps(felec,fvdw);
301
302 /* Calculate temporary vectorial force */
303 tx = _mm_mul_ps(fscal,dx00);
304 ty = _mm_mul_ps(fscal,dy00);
305 tz = _mm_mul_ps(fscal,dz00);
306
307 /* Update vectorial force */
308 fix0 = _mm_add_ps(fix0,tx);
309 fiy0 = _mm_add_ps(fiy0,ty);
310 fiz0 = _mm_add_ps(fiz0,tz);
311
312 fjx0 = _mm_add_ps(fjx0,tx);
313 fjy0 = _mm_add_ps(fjy0,ty);
314 fjz0 = _mm_add_ps(fjz0,tz);
315
316 /**************************
317 * CALCULATE INTERACTIONS *
318 **************************/
319
320 /* Compute parameters for interactions between i and j atoms */
321 qq10 = _mm_mul_ps(iq1,jq0);
322
323 /* REACTION-FIELD ELECTROSTATICS */
324 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
325 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
326
327 /* Update potential sum for this i atom from the interaction with this j atom. */
328 velecsum = _mm_add_ps(velecsum,velec);
329
330 fscal = felec;
331
332 /* Calculate temporary vectorial force */
333 tx = _mm_mul_ps(fscal,dx10);
334 ty = _mm_mul_ps(fscal,dy10);
335 tz = _mm_mul_ps(fscal,dz10);
336
337 /* Update vectorial force */
338 fix1 = _mm_add_ps(fix1,tx);
339 fiy1 = _mm_add_ps(fiy1,ty);
340 fiz1 = _mm_add_ps(fiz1,tz);
341
342 fjx0 = _mm_add_ps(fjx0,tx);
343 fjy0 = _mm_add_ps(fjy0,ty);
344 fjz0 = _mm_add_ps(fjz0,tz);
345
346 /**************************
347 * CALCULATE INTERACTIONS *
348 **************************/
349
350 /* Compute parameters for interactions between i and j atoms */
351 qq20 = _mm_mul_ps(iq2,jq0);
352
353 /* REACTION-FIELD ELECTROSTATICS */
354 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
355 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
356
357 /* Update potential sum for this i atom from the interaction with this j atom. */
358 velecsum = _mm_add_ps(velecsum,velec);
359
360 fscal = felec;
361
362 /* Calculate temporary vectorial force */
363 tx = _mm_mul_ps(fscal,dx20);
364 ty = _mm_mul_ps(fscal,dy20);
365 tz = _mm_mul_ps(fscal,dz20);
366
367 /* Update vectorial force */
368 fix2 = _mm_add_ps(fix2,tx);
369 fiy2 = _mm_add_ps(fiy2,ty);
370 fiz2 = _mm_add_ps(fiz2,tz);
371
372 fjx0 = _mm_add_ps(fjx0,tx);
373 fjy0 = _mm_add_ps(fjy0,ty);
374 fjz0 = _mm_add_ps(fjz0,tz);
375
376 fjptrA = f+j_coord_offsetA;
377 fjptrB = f+j_coord_offsetB;
378 fjptrC = f+j_coord_offsetC;
379 fjptrD = f+j_coord_offsetD;
380
381 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
382
383 /* Inner loop uses 131 flops */
384 }
385
386 if(jidx<j_index_end)
387 {
388
389 /* Get j neighbor index, and coordinate index */
390 jnrlistA = jjnr[jidx];
391 jnrlistB = jjnr[jidx+1];
392 jnrlistC = jjnr[jidx+2];
393 jnrlistD = jjnr[jidx+3];
394 /* Sign of each element will be negative for non-real atoms.
395 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
396 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
397 */
398 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
399 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
400 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
401 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
402 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
403 j_coord_offsetA = DIM*jnrA;
404 j_coord_offsetB = DIM*jnrB;
405 j_coord_offsetC = DIM*jnrC;
406 j_coord_offsetD = DIM*jnrD;
407
408 /* load j atom coordinates */
409 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
410 x+j_coord_offsetC,x+j_coord_offsetD,
411 &jx0,&jy0,&jz0);
412
413 /* Calculate displacement vector */
414 dx00 = _mm_sub_ps(ix0,jx0);
415 dy00 = _mm_sub_ps(iy0,jy0);
416 dz00 = _mm_sub_ps(iz0,jz0);
417 dx10 = _mm_sub_ps(ix1,jx0);
418 dy10 = _mm_sub_ps(iy1,jy0);
419 dz10 = _mm_sub_ps(iz1,jz0);
420 dx20 = _mm_sub_ps(ix2,jx0);
421 dy20 = _mm_sub_ps(iy2,jy0);
422 dz20 = _mm_sub_ps(iz2,jz0);
423
424 /* Calculate squared distance and things based on it */
425 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
426 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
427 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
428
429 rinv00 = gmx_mm_invsqrt_ps(rsq00);
430 rinv10 = gmx_mm_invsqrt_ps(rsq10);
431 rinv20 = gmx_mm_invsqrt_ps(rsq20);
432
433 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
434 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
435 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
436
437 /* Load parameters for j particles */
438 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
439 charge+jnrC+0,charge+jnrD+0);
440 vdwjidx0A = 2*vdwtype[jnrA+0];
441 vdwjidx0B = 2*vdwtype[jnrB+0];
442 vdwjidx0C = 2*vdwtype[jnrC+0];
443 vdwjidx0D = 2*vdwtype[jnrD+0];
444
445 fjx0 = _mm_setzero_ps();
446 fjy0 = _mm_setzero_ps();
447 fjz0 = _mm_setzero_ps();
448
449 /**************************
450 * CALCULATE INTERACTIONS *
451 **************************/
452
453 r00 = _mm_mul_ps(rsq00,rinv00);
454 r00 = _mm_andnot_ps(dummy_mask,r00);
455
456 /* Compute parameters for interactions between i and j atoms */
457 qq00 = _mm_mul_ps(iq0,jq0);
458 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
459 vdwparam+vdwioffset0+vdwjidx0B,
460 vdwparam+vdwioffset0+vdwjidx0C,
461 vdwparam+vdwioffset0+vdwjidx0D,
462 &c6_00,&c12_00);
463
464 /* Calculate table index by multiplying r with table scale and truncate to integer */
465 rt = _mm_mul_ps(r00,vftabscale);
466 vfitab = _mm_cvttps_epi32(rt);
467 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
468 vfitab = _mm_slli_epi32(vfitab,3);
469
470 /* REACTION-FIELD ELECTROSTATICS */
471 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
472 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
473
474 /* CUBIC SPLINE TABLE DISPERSION */
475 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
476 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
477 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
478 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
479 _MM_TRANSPOSE4_PS(Y,F,G,H);
480 Heps = _mm_mul_ps(vfeps,H);
481 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
482 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
483 vvdw6 = _mm_mul_ps(c6_00,VV);
484 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
485 fvdw6 = _mm_mul_ps(c6_00,FF);
486
487 /* CUBIC SPLINE TABLE REPULSION */
488 vfitab = _mm_add_epi32(vfitab,ifour);
489 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
490 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
491 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
492 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
493 _MM_TRANSPOSE4_PS(Y,F,G,H);
494 Heps = _mm_mul_ps(vfeps,H);
495 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
496 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
497 vvdw12 = _mm_mul_ps(c12_00,VV);
498 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
499 fvdw12 = _mm_mul_ps(c12_00,FF);
500 vvdw = _mm_add_ps(vvdw12,vvdw6);
501 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
502
503 /* Update potential sum for this i atom from the interaction with this j atom. */
504 velec = _mm_andnot_ps(dummy_mask,velec);
505 velecsum = _mm_add_ps(velecsum,velec);
506 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
507 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
508
509 fscal = _mm_add_ps(felec,fvdw);
510
511 fscal = _mm_andnot_ps(dummy_mask,fscal);
512
513 /* Calculate temporary vectorial force */
514 tx = _mm_mul_ps(fscal,dx00);
515 ty = _mm_mul_ps(fscal,dy00);
516 tz = _mm_mul_ps(fscal,dz00);
517
518 /* Update vectorial force */
519 fix0 = _mm_add_ps(fix0,tx);
520 fiy0 = _mm_add_ps(fiy0,ty);
521 fiz0 = _mm_add_ps(fiz0,tz);
522
523 fjx0 = _mm_add_ps(fjx0,tx);
524 fjy0 = _mm_add_ps(fjy0,ty);
525 fjz0 = _mm_add_ps(fjz0,tz);
526
527 /**************************
528 * CALCULATE INTERACTIONS *
529 **************************/
530
531 /* Compute parameters for interactions between i and j atoms */
532 qq10 = _mm_mul_ps(iq1,jq0);
533
534 /* REACTION-FIELD ELECTROSTATICS */
535 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
536 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
537
538 /* Update potential sum for this i atom from the interaction with this j atom. */
539 velec = _mm_andnot_ps(dummy_mask,velec);
540 velecsum = _mm_add_ps(velecsum,velec);
541
542 fscal = felec;
543
544 fscal = _mm_andnot_ps(dummy_mask,fscal);
545
546 /* Calculate temporary vectorial force */
547 tx = _mm_mul_ps(fscal,dx10);
548 ty = _mm_mul_ps(fscal,dy10);
549 tz = _mm_mul_ps(fscal,dz10);
550
551 /* Update vectorial force */
552 fix1 = _mm_add_ps(fix1,tx);
553 fiy1 = _mm_add_ps(fiy1,ty);
554 fiz1 = _mm_add_ps(fiz1,tz);
555
556 fjx0 = _mm_add_ps(fjx0,tx);
557 fjy0 = _mm_add_ps(fjy0,ty);
558 fjz0 = _mm_add_ps(fjz0,tz);
559
560 /**************************
561 * CALCULATE INTERACTIONS *
562 **************************/
563
564 /* Compute parameters for interactions between i and j atoms */
565 qq20 = _mm_mul_ps(iq2,jq0);
566
567 /* REACTION-FIELD ELECTROSTATICS */
568 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
569 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
570
571 /* Update potential sum for this i atom from the interaction with this j atom. */
572 velec = _mm_andnot_ps(dummy_mask,velec);
573 velecsum = _mm_add_ps(velecsum,velec);
574
575 fscal = felec;
576
577 fscal = _mm_andnot_ps(dummy_mask,fscal);
578
579 /* Calculate temporary vectorial force */
580 tx = _mm_mul_ps(fscal,dx20);
581 ty = _mm_mul_ps(fscal,dy20);
582 tz = _mm_mul_ps(fscal,dz20);
583
584 /* Update vectorial force */
585 fix2 = _mm_add_ps(fix2,tx);
586 fiy2 = _mm_add_ps(fiy2,ty);
587 fiz2 = _mm_add_ps(fiz2,tz);
588
589 fjx0 = _mm_add_ps(fjx0,tx);
590 fjy0 = _mm_add_ps(fjy0,ty);
591 fjz0 = _mm_add_ps(fjz0,tz);
592
593 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
594 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
595 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
596 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
597
598 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
599
600 /* Inner loop uses 132 flops */
601 }
602
603 /* End of innermost loop */
604
605 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
606 f+i_coord_offset,fshift+i_shift_offset);
607
608 ggid = gid[iidx];
609 /* Update potential energies */
610 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
611 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
612
613 /* Increment number of inner iterations */
614 inneriter += j_index_end - j_index_start;
615
616 /* Outer loop uses 20 flops */
617 }
618
619 /* Increment number of outer iterations */
620 outeriter += nri;
621
622 /* Update outer/inner flops */
623
624 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*132);
625 }
626 /*
627 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse4_1_single
628 * Electrostatics interaction: ReactionField
629 * VdW interaction: CubicSplineTable
630 * Geometry: Water3-Particle
631 * Calculate force/pot: Force
632 */
633 void
634 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse4_1_single
635 (t_nblist * gmx_restrict nlist,
636 rvec * gmx_restrict xx,
637 rvec * gmx_restrict ff,
638 t_forcerec * gmx_restrict fr,
639 t_mdatoms * gmx_restrict mdatoms,
640 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
641 t_nrnb * gmx_restrict nrnb)
642 {
643 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
644 * just 0 for non-waters.
645 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
646 * jnr indices corresponding to data put in the four positions in the SIMD register.
647 */
648 int i_shift_offset,i_coord_offset,outeriter,inneriter;
649 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
650 int jnrA,jnrB,jnrC,jnrD;
651 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
652 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
653 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
654 real rcutoff_scalar;
655 real *shiftvec,*fshift,*x,*f;
656 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
657 real scratch[4*DIM];
658 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
659 int vdwioffset0;
660 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
661 int vdwioffset1;
662 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
663 int vdwioffset2;
664 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
665 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
666 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
667 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
668 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
669 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
670 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
671 real *charge;
672 int nvdwtype;
673 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
674 int *vdwtype;
675 real *vdwparam;
676 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
677 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
678 __m128i vfitab;
679 __m128i ifour = _mm_set1_epi32(4);
680 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
681 real *vftab;
682 __m128 dummy_mask,cutoff_mask;
683 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
684 __m128 one = _mm_set1_ps(1.0);
685 __m128 two = _mm_set1_ps(2.0);
686 x = xx[0];
687 f = ff[0];
688
689 nri = nlist->nri;
690 iinr = nlist->iinr;
691 jindex = nlist->jindex;
692 jjnr = nlist->jjnr;
693 shiftidx = nlist->shift;
694 gid = nlist->gid;
695 shiftvec = fr->shift_vec[0];
696 fshift = fr->fshift[0];
697 facel = _mm_set1_ps(fr->epsfac);
698 charge = mdatoms->chargeA;
699 krf = _mm_set1_ps(fr->ic->k_rf);
700 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
701 crf = _mm_set1_ps(fr->ic->c_rf);
702 nvdwtype = fr->ntype;
703 vdwparam = fr->nbfp;
704 vdwtype = mdatoms->typeA;
705
706 vftab = kernel_data->table_vdw->data;
707 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
708
709 /* Setup water-specific parameters */
710 inr = nlist->iinr[0];
711 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
712 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
713 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
714 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
715
716 /* Avoid stupid compiler warnings */
717 jnrA = jnrB = jnrC = jnrD = 0;
718 j_coord_offsetA = 0;
719 j_coord_offsetB = 0;
720 j_coord_offsetC = 0;
721 j_coord_offsetD = 0;
722
723 outeriter = 0;
724 inneriter = 0;
725
726 for(iidx=0;iidx<4*DIM;iidx++)
727 {
728 scratch[iidx] = 0.0;
729 }
730
731 /* Start outer loop over neighborlists */
732 for(iidx=0; iidx<nri; iidx++)
733 {
734 /* Load shift vector for this list */
735 i_shift_offset = DIM*shiftidx[iidx];
736
737 /* Load limits for loop over neighbors */
738 j_index_start = jindex[iidx];
739 j_index_end = jindex[iidx+1];
740
741 /* Get outer coordinate index */
742 inr = iinr[iidx];
743 i_coord_offset = DIM*inr;
744
745 /* Load i particle coords and add shift vector */
746 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
747 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
748
749 fix0 = _mm_setzero_ps();
750 fiy0 = _mm_setzero_ps();
751 fiz0 = _mm_setzero_ps();
752 fix1 = _mm_setzero_ps();
753 fiy1 = _mm_setzero_ps();
754 fiz1 = _mm_setzero_ps();
755 fix2 = _mm_setzero_ps();
756 fiy2 = _mm_setzero_ps();
757 fiz2 = _mm_setzero_ps();
758
759 /* Start inner kernel loop */
760 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
761 {
762
763 /* Get j neighbor index, and coordinate index */
764 jnrA = jjnr[jidx];
765 jnrB = jjnr[jidx+1];
766 jnrC = jjnr[jidx+2];
767 jnrD = jjnr[jidx+3];
768 j_coord_offsetA = DIM*jnrA;
769 j_coord_offsetB = DIM*jnrB;
770 j_coord_offsetC = DIM*jnrC;
771 j_coord_offsetD = DIM*jnrD;
772
773 /* load j atom coordinates */
774 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
775 x+j_coord_offsetC,x+j_coord_offsetD,
776 &jx0,&jy0,&jz0);
777
778 /* Calculate displacement vector */
779 dx00 = _mm_sub_ps(ix0,jx0);
780 dy00 = _mm_sub_ps(iy0,jy0);
781 dz00 = _mm_sub_ps(iz0,jz0);
782 dx10 = _mm_sub_ps(ix1,jx0);
783 dy10 = _mm_sub_ps(iy1,jy0);
784 dz10 = _mm_sub_ps(iz1,jz0);
785 dx20 = _mm_sub_ps(ix2,jx0);
786 dy20 = _mm_sub_ps(iy2,jy0);
787 dz20 = _mm_sub_ps(iz2,jz0);
788
789 /* Calculate squared distance and things based on it */
790 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
791 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
792 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
793
794 rinv00 = gmx_mm_invsqrt_ps(rsq00);
795 rinv10 = gmx_mm_invsqrt_ps(rsq10);
796 rinv20 = gmx_mm_invsqrt_ps(rsq20);
797
798 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
799 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
800 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
801
802 /* Load parameters for j particles */
803 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
804 charge+jnrC+0,charge+jnrD+0);
805 vdwjidx0A = 2*vdwtype[jnrA+0];
806 vdwjidx0B = 2*vdwtype[jnrB+0];
807 vdwjidx0C = 2*vdwtype[jnrC+0];
808 vdwjidx0D = 2*vdwtype[jnrD+0];
809
810 fjx0 = _mm_setzero_ps();
811 fjy0 = _mm_setzero_ps();
812 fjz0 = _mm_setzero_ps();
813
814 /**************************
815 * CALCULATE INTERACTIONS *
816 **************************/
817
818 r00 = _mm_mul_ps(rsq00,rinv00);
819
820 /* Compute parameters for interactions between i and j atoms */
821 qq00 = _mm_mul_ps(iq0,jq0);
822 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
823 vdwparam+vdwioffset0+vdwjidx0B,
824 vdwparam+vdwioffset0+vdwjidx0C,
825 vdwparam+vdwioffset0+vdwjidx0D,
826 &c6_00,&c12_00);
827
828 /* Calculate table index by multiplying r with table scale and truncate to integer */
829 rt = _mm_mul_ps(r00,vftabscale);
830 vfitab = _mm_cvttps_epi32(rt);
831 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
832 vfitab = _mm_slli_epi32(vfitab,3);
833
834 /* REACTION-FIELD ELECTROSTATICS */
835 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
836
837 /* CUBIC SPLINE TABLE DISPERSION */
838 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
839 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
840 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
841 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
842 _MM_TRANSPOSE4_PS(Y,F,G,H);
843 Heps = _mm_mul_ps(vfeps,H);
844 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
845 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
846 fvdw6 = _mm_mul_ps(c6_00,FF);
847
848 /* CUBIC SPLINE TABLE REPULSION */
849 vfitab = _mm_add_epi32(vfitab,ifour);
850 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
851 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
852 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
853 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
854 _MM_TRANSPOSE4_PS(Y,F,G,H);
855 Heps = _mm_mul_ps(vfeps,H);
856 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
857 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
858 fvdw12 = _mm_mul_ps(c12_00,FF);
859 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
860
861 fscal = _mm_add_ps(felec,fvdw);
862
863 /* Calculate temporary vectorial force */
864 tx = _mm_mul_ps(fscal,dx00);
865 ty = _mm_mul_ps(fscal,dy00);
866 tz = _mm_mul_ps(fscal,dz00);
867
868 /* Update vectorial force */
869 fix0 = _mm_add_ps(fix0,tx);
870 fiy0 = _mm_add_ps(fiy0,ty);
871 fiz0 = _mm_add_ps(fiz0,tz);
872
873 fjx0 = _mm_add_ps(fjx0,tx);
874 fjy0 = _mm_add_ps(fjy0,ty);
875 fjz0 = _mm_add_ps(fjz0,tz);
876
877 /**************************
878 * CALCULATE INTERACTIONS *
879 **************************/
880
881 /* Compute parameters for interactions between i and j atoms */
882 qq10 = _mm_mul_ps(iq1,jq0);
883
884 /* REACTION-FIELD ELECTROSTATICS */
885 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
886
887 fscal = felec;
888
889 /* Calculate temporary vectorial force */
890 tx = _mm_mul_ps(fscal,dx10);
891 ty = _mm_mul_ps(fscal,dy10);
892 tz = _mm_mul_ps(fscal,dz10);
893
894 /* Update vectorial force */
895 fix1 = _mm_add_ps(fix1,tx);
896 fiy1 = _mm_add_ps(fiy1,ty);
897 fiz1 = _mm_add_ps(fiz1,tz);
898
899 fjx0 = _mm_add_ps(fjx0,tx);
900 fjy0 = _mm_add_ps(fjy0,ty);
901 fjz0 = _mm_add_ps(fjz0,tz);
902
903 /**************************
904 * CALCULATE INTERACTIONS *
905 **************************/
906
907 /* Compute parameters for interactions between i and j atoms */
908 qq20 = _mm_mul_ps(iq2,jq0);
909
910 /* REACTION-FIELD ELECTROSTATICS */
911 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
912
913 fscal = felec;
914
915 /* Calculate temporary vectorial force */
916 tx = _mm_mul_ps(fscal,dx20);
917 ty = _mm_mul_ps(fscal,dy20);
918 tz = _mm_mul_ps(fscal,dz20);
919
920 /* Update vectorial force */
921 fix2 = _mm_add_ps(fix2,tx);
922 fiy2 = _mm_add_ps(fiy2,ty);
923 fiz2 = _mm_add_ps(fiz2,tz);
924
925 fjx0 = _mm_add_ps(fjx0,tx);
926 fjy0 = _mm_add_ps(fjy0,ty);
927 fjz0 = _mm_add_ps(fjz0,tz);
928
929 fjptrA = f+j_coord_offsetA;
930 fjptrB = f+j_coord_offsetB;
931 fjptrC = f+j_coord_offsetC;
932 fjptrD = f+j_coord_offsetD;
933
934 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
935
936 /* Inner loop uses 108 flops */
937 }
938
939 if(jidx<j_index_end)
940 {
941
942 /* Get j neighbor index, and coordinate index */
943 jnrlistA = jjnr[jidx];
944 jnrlistB = jjnr[jidx+1];
945 jnrlistC = jjnr[jidx+2];
946 jnrlistD = jjnr[jidx+3];
947 /* Sign of each element will be negative for non-real atoms.
948 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
949 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
950 */
951 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
952 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
953 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
954 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
955 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
956 j_coord_offsetA = DIM*jnrA;
957 j_coord_offsetB = DIM*jnrB;
958 j_coord_offsetC = DIM*jnrC;
959 j_coord_offsetD = DIM*jnrD;
960
961 /* load j atom coordinates */
962 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
963 x+j_coord_offsetC,x+j_coord_offsetD,
964 &jx0,&jy0,&jz0);
965
966 /* Calculate displacement vector */
967 dx00 = _mm_sub_ps(ix0,jx0);
968 dy00 = _mm_sub_ps(iy0,jy0);
969 dz00 = _mm_sub_ps(iz0,jz0);
970 dx10 = _mm_sub_ps(ix1,jx0);
971 dy10 = _mm_sub_ps(iy1,jy0);
972 dz10 = _mm_sub_ps(iz1,jz0);
973 dx20 = _mm_sub_ps(ix2,jx0);
974 dy20 = _mm_sub_ps(iy2,jy0);
975 dz20 = _mm_sub_ps(iz2,jz0);
976
977 /* Calculate squared distance and things based on it */
978 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
979 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
980 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
981
982 rinv00 = gmx_mm_invsqrt_ps(rsq00);
983 rinv10 = gmx_mm_invsqrt_ps(rsq10);
984 rinv20 = gmx_mm_invsqrt_ps(rsq20);
985
986 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
987 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
988 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
989
990 /* Load parameters for j particles */
991 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
992 charge+jnrC+0,charge+jnrD+0);
993 vdwjidx0A = 2*vdwtype[jnrA+0];
994 vdwjidx0B = 2*vdwtype[jnrB+0];
995 vdwjidx0C = 2*vdwtype[jnrC+0];
996 vdwjidx0D = 2*vdwtype[jnrD+0];
997
998 fjx0 = _mm_setzero_ps();
999 fjy0 = _mm_setzero_ps();
1000 fjz0 = _mm_setzero_ps();
1001
1002 /**************************
1003 * CALCULATE INTERACTIONS *
1004 **************************/
1005
1006 r00 = _mm_mul_ps(rsq00,rinv00);
1007 r00 = _mm_andnot_ps(dummy_mask,r00);
1008
1009 /* Compute parameters for interactions between i and j atoms */
1010 qq00 = _mm_mul_ps(iq0,jq0);
1011 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1012 vdwparam+vdwioffset0+vdwjidx0B,
1013 vdwparam+vdwioffset0+vdwjidx0C,
1014 vdwparam+vdwioffset0+vdwjidx0D,
1015 &c6_00,&c12_00);
1016
1017 /* Calculate table index by multiplying r with table scale and truncate to integer */
1018 rt = _mm_mul_ps(r00,vftabscale);
1019 vfitab = _mm_cvttps_epi32(rt);
1020 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1021 vfitab = _mm_slli_epi32(vfitab,3);
1022
1023 /* REACTION-FIELD ELECTROSTATICS */
1024 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1025
1026 /* CUBIC SPLINE TABLE DISPERSION */
1027 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1028 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1029 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1030 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1031 _MM_TRANSPOSE4_PS(Y,F,G,H);
1032 Heps = _mm_mul_ps(vfeps,H);
1033 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1034 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1035 fvdw6 = _mm_mul_ps(c6_00,FF);
1036
1037 /* CUBIC SPLINE TABLE REPULSION */
1038 vfitab = _mm_add_epi32(vfitab,ifour);
1039 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1040 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1041 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1042 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1043 _MM_TRANSPOSE4_PS(Y,F,G,H);
1044 Heps = _mm_mul_ps(vfeps,H);
1045 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1046 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1047 fvdw12 = _mm_mul_ps(c12_00,FF);
1048 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1049
1050 fscal = _mm_add_ps(felec,fvdw);
1051
1052 fscal = _mm_andnot_ps(dummy_mask,fscal);
1053
1054 /* Calculate temporary vectorial force */
1055 tx = _mm_mul_ps(fscal,dx00);
1056 ty = _mm_mul_ps(fscal,dy00);
1057 tz = _mm_mul_ps(fscal,dz00);
1058
1059 /* Update vectorial force */
1060 fix0 = _mm_add_ps(fix0,tx);
1061 fiy0 = _mm_add_ps(fiy0,ty);
1062 fiz0 = _mm_add_ps(fiz0,tz);
1063
1064 fjx0 = _mm_add_ps(fjx0,tx);
1065 fjy0 = _mm_add_ps(fjy0,ty);
1066 fjz0 = _mm_add_ps(fjz0,tz);
1067
1068 /**************************
1069 * CALCULATE INTERACTIONS *
1070 **************************/
1071
1072 /* Compute parameters for interactions between i and j atoms */
1073 qq10 = _mm_mul_ps(iq1,jq0);
1074
1075 /* REACTION-FIELD ELECTROSTATICS */
1076 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1077
1078 fscal = felec;
1079
1080 fscal = _mm_andnot_ps(dummy_mask,fscal);
1081
1082 /* Calculate temporary vectorial force */
1083 tx = _mm_mul_ps(fscal,dx10);
1084 ty = _mm_mul_ps(fscal,dy10);
1085 tz = _mm_mul_ps(fscal,dz10);
1086
1087 /* Update vectorial force */
1088 fix1 = _mm_add_ps(fix1,tx);
1089 fiy1 = _mm_add_ps(fiy1,ty);
1090 fiz1 = _mm_add_ps(fiz1,tz);
1091
1092 fjx0 = _mm_add_ps(fjx0,tx);
1093 fjy0 = _mm_add_ps(fjy0,ty);
1094 fjz0 = _mm_add_ps(fjz0,tz);
1095
1096 /**************************
1097 * CALCULATE INTERACTIONS *
1098 **************************/
1099
1100 /* Compute parameters for interactions between i and j atoms */
1101 qq20 = _mm_mul_ps(iq2,jq0);
1102
1103 /* REACTION-FIELD ELECTROSTATICS */
1104 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1105
1106 fscal = felec;
1107
1108 fscal = _mm_andnot_ps(dummy_mask,fscal);
1109
1110 /* Calculate temporary vectorial force */
1111 tx = _mm_mul_ps(fscal,dx20);
1112 ty = _mm_mul_ps(fscal,dy20);
1113 tz = _mm_mul_ps(fscal,dz20);
1114
1115 /* Update vectorial force */
1116 fix2 = _mm_add_ps(fix2,tx);
1117 fiy2 = _mm_add_ps(fiy2,ty);
1118 fiz2 = _mm_add_ps(fiz2,tz);
1119
1120 fjx0 = _mm_add_ps(fjx0,tx);
1121 fjy0 = _mm_add_ps(fjy0,ty);
1122 fjz0 = _mm_add_ps(fjz0,tz);
1123
1124 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1125 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1126 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1127 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1128
1129 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1130
1131 /* Inner loop uses 109 flops */
1132 }
1133
1134 /* End of innermost loop */
1135
1136 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1137 f+i_coord_offset,fshift+i_shift_offset);
1138
1139 /* Increment number of inner iterations */
1140 inneriter += j_index_end - j_index_start;
1141
1142 /* Outer loop uses 18 flops */
1143 }
1144
1145 /* Increment number of outer iterations */
1146 outeriter += nri;
1147
1148 /* Update outer/inner flops */
1149
1150 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1151 }
The ultimate molecular dynamics simulation package