?  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_ElecRFCut_VdwCSTab_GeomP1P1_sse4_1_single.c
blob09ddbc88c751d019ad504ab07ce089e5399fa8f3
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_ElecRFCut_VdwCSTab_GeomP1P1_VF_sse4_1_single
53 * Electrostatics interaction: ReactionField
54 * VdW interaction: CubicSplineTable
55 * Geometry: Particle-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_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 vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
87 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
88 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
89 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
90 real *charge;
91 int nvdwtype;
92 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
93 int *vdwtype;
94 real *vdwparam;
95 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
96 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
97 __m128i vfitab;
98 __m128i ifour = _mm_set1_epi32(4);
99 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
100 real *vftab;
101 __m128 dummy_mask,cutoff_mask;
102 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
103 __m128 one = _mm_set1_ps(1.0);
104 __m128 two = _mm_set1_ps(2.0);
105 x = xx[0];
106 f = ff[0];
107
108 nri = nlist->nri;
109 iinr = nlist->iinr;
110 jindex = nlist->jindex;
111 jjnr = nlist->jjnr;
112 shiftidx = nlist->shift;
113 gid = nlist->gid;
114 shiftvec = fr->shift_vec[0];
115 fshift = fr->fshift[0];
116 facel = _mm_set1_ps(fr->epsfac);
117 charge = mdatoms->chargeA;
118 krf = _mm_set1_ps(fr->ic->k_rf);
119 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
120 crf = _mm_set1_ps(fr->ic->c_rf);
121 nvdwtype = fr->ntype;
122 vdwparam = fr->nbfp;
123 vdwtype = mdatoms->typeA;
124
125 vftab = kernel_data->table_vdw->data;
126 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
127
128 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
129 rcutoff_scalar = fr->rcoulomb;
130 rcutoff = _mm_set1_ps(rcutoff_scalar);
131 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
132
133 /* Avoid stupid compiler warnings */
134 jnrA = jnrB = jnrC = jnrD = 0;
135 j_coord_offsetA = 0;
136 j_coord_offsetB = 0;
137 j_coord_offsetC = 0;
138 j_coord_offsetD = 0;
139
140 outeriter = 0;
141 inneriter = 0;
142
143 for(iidx=0;iidx<4*DIM;iidx++)
144 {
145 scratch[iidx] = 0.0;
146 }
147
148 /* Start outer loop over neighborlists */
149 for(iidx=0; iidx<nri; iidx++)
150 {
151 /* Load shift vector for this list */
152 i_shift_offset = DIM*shiftidx[iidx];
153
154 /* Load limits for loop over neighbors */
155 j_index_start = jindex[iidx];
156 j_index_end = jindex[iidx+1];
157
158 /* Get outer coordinate index */
159 inr = iinr[iidx];
160 i_coord_offset = DIM*inr;
161
162 /* Load i particle coords and add shift vector */
163 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
164
165 fix0 = _mm_setzero_ps();
166 fiy0 = _mm_setzero_ps();
167 fiz0 = _mm_setzero_ps();
168
169 /* Load parameters for i particles */
170 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
171 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
172
173 /* Reset potential sums */
174 velecsum = _mm_setzero_ps();
175 vvdwsum = _mm_setzero_ps();
176
177 /* Start inner kernel loop */
178 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
179 {
180
181 /* Get j neighbor index, and coordinate index */
182 jnrA = jjnr[jidx];
183 jnrB = jjnr[jidx+1];
184 jnrC = jjnr[jidx+2];
185 jnrD = jjnr[jidx+3];
186 j_coord_offsetA = DIM*jnrA;
187 j_coord_offsetB = DIM*jnrB;
188 j_coord_offsetC = DIM*jnrC;
189 j_coord_offsetD = DIM*jnrD;
190
191 /* load j atom coordinates */
192 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
193 x+j_coord_offsetC,x+j_coord_offsetD,
194 &jx0,&jy0,&jz0);
195
196 /* Calculate displacement vector */
197 dx00 = _mm_sub_ps(ix0,jx0);
198 dy00 = _mm_sub_ps(iy0,jy0);
199 dz00 = _mm_sub_ps(iz0,jz0);
200
201 /* Calculate squared distance and things based on it */
202 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
203
204 rinv00 = gmx_mm_invsqrt_ps(rsq00);
205
206 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
207
208 /* Load parameters for j particles */
209 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
210 charge+jnrC+0,charge+jnrD+0);
211 vdwjidx0A = 2*vdwtype[jnrA+0];
212 vdwjidx0B = 2*vdwtype[jnrB+0];
213 vdwjidx0C = 2*vdwtype[jnrC+0];
214 vdwjidx0D = 2*vdwtype[jnrD+0];
215
216 /**************************
217 * CALCULATE INTERACTIONS *
218 **************************/
219
220 if (gmx_mm_any_lt(rsq00,rcutoff2))
221 {
222
223 r00 = _mm_mul_ps(rsq00,rinv00);
224
225 /* Compute parameters for interactions between i and j atoms */
226 qq00 = _mm_mul_ps(iq0,jq0);
227 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
228 vdwparam+vdwioffset0+vdwjidx0B,
229 vdwparam+vdwioffset0+vdwjidx0C,
230 vdwparam+vdwioffset0+vdwjidx0D,
231 &c6_00,&c12_00);
232
233 /* Calculate table index by multiplying r with table scale and truncate to integer */
234 rt = _mm_mul_ps(r00,vftabscale);
235 vfitab = _mm_cvttps_epi32(rt);
236 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
237 vfitab = _mm_slli_epi32(vfitab,3);
238
239 /* REACTION-FIELD ELECTROSTATICS */
240 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
241 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
242
243 /* CUBIC SPLINE TABLE DISPERSION */
244 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
245 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
246 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
247 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
248 _MM_TRANSPOSE4_PS(Y,F,G,H);
249 Heps = _mm_mul_ps(vfeps,H);
250 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
251 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
252 vvdw6 = _mm_mul_ps(c6_00,VV);
253 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
254 fvdw6 = _mm_mul_ps(c6_00,FF);
255
256 /* CUBIC SPLINE TABLE REPULSION */
257 vfitab = _mm_add_epi32(vfitab,ifour);
258 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
259 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
260 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
261 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
262 _MM_TRANSPOSE4_PS(Y,F,G,H);
263 Heps = _mm_mul_ps(vfeps,H);
264 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
265 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
266 vvdw12 = _mm_mul_ps(c12_00,VV);
267 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
268 fvdw12 = _mm_mul_ps(c12_00,FF);
269 vvdw = _mm_add_ps(vvdw12,vvdw6);
270 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
271
272 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
273
274 /* Update potential sum for this i atom from the interaction with this j atom. */
275 velec = _mm_and_ps(velec,cutoff_mask);
276 velecsum = _mm_add_ps(velecsum,velec);
277 vvdw = _mm_and_ps(vvdw,cutoff_mask);
278 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
279
280 fscal = _mm_add_ps(felec,fvdw);
281
282 fscal = _mm_and_ps(fscal,cutoff_mask);
283
284 /* Calculate temporary vectorial force */
285 tx = _mm_mul_ps(fscal,dx00);
286 ty = _mm_mul_ps(fscal,dy00);
287 tz = _mm_mul_ps(fscal,dz00);
288
289 /* Update vectorial force */
290 fix0 = _mm_add_ps(fix0,tx);
291 fiy0 = _mm_add_ps(fiy0,ty);
292 fiz0 = _mm_add_ps(fiz0,tz);
293
294 fjptrA = f+j_coord_offsetA;
295 fjptrB = f+j_coord_offsetB;
296 fjptrC = f+j_coord_offsetC;
297 fjptrD = f+j_coord_offsetD;
298 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
299
300 }
301
302 /* Inner loop uses 72 flops */
303 }
304
305 if(jidx<j_index_end)
306 {
307
308 /* Get j neighbor index, and coordinate index */
309 jnrlistA = jjnr[jidx];
310 jnrlistB = jjnr[jidx+1];
311 jnrlistC = jjnr[jidx+2];
312 jnrlistD = jjnr[jidx+3];
313 /* Sign of each element will be negative for non-real atoms.
314 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
315 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
316 */
317 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
318 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
319 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
320 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
321 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
322 j_coord_offsetA = DIM*jnrA;
323 j_coord_offsetB = DIM*jnrB;
324 j_coord_offsetC = DIM*jnrC;
325 j_coord_offsetD = DIM*jnrD;
326
327 /* load j atom coordinates */
328 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
329 x+j_coord_offsetC,x+j_coord_offsetD,
330 &jx0,&jy0,&jz0);
331
332 /* Calculate displacement vector */
333 dx00 = _mm_sub_ps(ix0,jx0);
334 dy00 = _mm_sub_ps(iy0,jy0);
335 dz00 = _mm_sub_ps(iz0,jz0);
336
337 /* Calculate squared distance and things based on it */
338 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
339
340 rinv00 = gmx_mm_invsqrt_ps(rsq00);
341
342 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
343
344 /* Load parameters for j particles */
345 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
346 charge+jnrC+0,charge+jnrD+0);
347 vdwjidx0A = 2*vdwtype[jnrA+0];
348 vdwjidx0B = 2*vdwtype[jnrB+0];
349 vdwjidx0C = 2*vdwtype[jnrC+0];
350 vdwjidx0D = 2*vdwtype[jnrD+0];
351
352 /**************************
353 * CALCULATE INTERACTIONS *
354 **************************/
355
356 if (gmx_mm_any_lt(rsq00,rcutoff2))
357 {
358
359 r00 = _mm_mul_ps(rsq00,rinv00);
360 r00 = _mm_andnot_ps(dummy_mask,r00);
361
362 /* Compute parameters for interactions between i and j atoms */
363 qq00 = _mm_mul_ps(iq0,jq0);
364 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
365 vdwparam+vdwioffset0+vdwjidx0B,
366 vdwparam+vdwioffset0+vdwjidx0C,
367 vdwparam+vdwioffset0+vdwjidx0D,
368 &c6_00,&c12_00);
369
370 /* Calculate table index by multiplying r with table scale and truncate to integer */
371 rt = _mm_mul_ps(r00,vftabscale);
372 vfitab = _mm_cvttps_epi32(rt);
373 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
374 vfitab = _mm_slli_epi32(vfitab,3);
375
376 /* REACTION-FIELD ELECTROSTATICS */
377 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
378 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
379
380 /* CUBIC SPLINE TABLE DISPERSION */
381 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
382 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
383 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
384 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
385 _MM_TRANSPOSE4_PS(Y,F,G,H);
386 Heps = _mm_mul_ps(vfeps,H);
387 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
388 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
389 vvdw6 = _mm_mul_ps(c6_00,VV);
390 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
391 fvdw6 = _mm_mul_ps(c6_00,FF);
392
393 /* CUBIC SPLINE TABLE REPULSION */
394 vfitab = _mm_add_epi32(vfitab,ifour);
395 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
396 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
397 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
398 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
399 _MM_TRANSPOSE4_PS(Y,F,G,H);
400 Heps = _mm_mul_ps(vfeps,H);
401 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
402 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
403 vvdw12 = _mm_mul_ps(c12_00,VV);
404 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
405 fvdw12 = _mm_mul_ps(c12_00,FF);
406 vvdw = _mm_add_ps(vvdw12,vvdw6);
407 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
408
409 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
410
411 /* Update potential sum for this i atom from the interaction with this j atom. */
412 velec = _mm_and_ps(velec,cutoff_mask);
413 velec = _mm_andnot_ps(dummy_mask,velec);
414 velecsum = _mm_add_ps(velecsum,velec);
415 vvdw = _mm_and_ps(vvdw,cutoff_mask);
416 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
417 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
418
419 fscal = _mm_add_ps(felec,fvdw);
420
421 fscal = _mm_and_ps(fscal,cutoff_mask);
422
423 fscal = _mm_andnot_ps(dummy_mask,fscal);
424
425 /* Calculate temporary vectorial force */
426 tx = _mm_mul_ps(fscal,dx00);
427 ty = _mm_mul_ps(fscal,dy00);
428 tz = _mm_mul_ps(fscal,dz00);
429
430 /* Update vectorial force */
431 fix0 = _mm_add_ps(fix0,tx);
432 fiy0 = _mm_add_ps(fiy0,ty);
433 fiz0 = _mm_add_ps(fiz0,tz);
434
435 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
436 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
437 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
438 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
439 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
440
441 }
442
443 /* Inner loop uses 73 flops */
444 }
445
446 /* End of innermost loop */
447
448 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
449 f+i_coord_offset,fshift+i_shift_offset);
450
451 ggid = gid[iidx];
452 /* Update potential energies */
453 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
454 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
455
456 /* Increment number of inner iterations */
457 inneriter += j_index_end - j_index_start;
458
459 /* Outer loop uses 9 flops */
460 }
461
462 /* Increment number of outer iterations */
463 outeriter += nri;
464
465 /* Update outer/inner flops */
466
467 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
468 }
469 /*
470 * Gromacs nonbonded kernel: nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse4_1_single
471 * Electrostatics interaction: ReactionField
472 * VdW interaction: CubicSplineTable
473 * Geometry: Particle-Particle
474 * Calculate force/pot: Force
475 */
476 void
477 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse4_1_single
478 (t_nblist * gmx_restrict nlist,
479 rvec * gmx_restrict xx,
480 rvec * gmx_restrict ff,
481 t_forcerec * gmx_restrict fr,
482 t_mdatoms * gmx_restrict mdatoms,
483 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
484 t_nrnb * gmx_restrict nrnb)
485 {
486 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
487 * just 0 for non-waters.
488 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
489 * jnr indices corresponding to data put in the four positions in the SIMD register.
490 */
491 int i_shift_offset,i_coord_offset,outeriter,inneriter;
492 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
493 int jnrA,jnrB,jnrC,jnrD;
494 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
495 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
496 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
497 real rcutoff_scalar;
498 real *shiftvec,*fshift,*x,*f;
499 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
500 real scratch[4*DIM];
501 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
502 int vdwioffset0;
503 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
504 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
505 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
506 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
507 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
508 real *charge;
509 int nvdwtype;
510 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
511 int *vdwtype;
512 real *vdwparam;
513 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
514 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
515 __m128i vfitab;
516 __m128i ifour = _mm_set1_epi32(4);
517 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
518 real *vftab;
519 __m128 dummy_mask,cutoff_mask;
520 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
521 __m128 one = _mm_set1_ps(1.0);
522 __m128 two = _mm_set1_ps(2.0);
523 x = xx[0];
524 f = ff[0];
525
526 nri = nlist->nri;
527 iinr = nlist->iinr;
528 jindex = nlist->jindex;
529 jjnr = nlist->jjnr;
530 shiftidx = nlist->shift;
531 gid = nlist->gid;
532 shiftvec = fr->shift_vec[0];
533 fshift = fr->fshift[0];
534 facel = _mm_set1_ps(fr->epsfac);
535 charge = mdatoms->chargeA;
536 krf = _mm_set1_ps(fr->ic->k_rf);
537 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
538 crf = _mm_set1_ps(fr->ic->c_rf);
539 nvdwtype = fr->ntype;
540 vdwparam = fr->nbfp;
541 vdwtype = mdatoms->typeA;
542
543 vftab = kernel_data->table_vdw->data;
544 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
545
546 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
547 rcutoff_scalar = fr->rcoulomb;
548 rcutoff = _mm_set1_ps(rcutoff_scalar);
549 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
550
551 /* Avoid stupid compiler warnings */
552 jnrA = jnrB = jnrC = jnrD = 0;
553 j_coord_offsetA = 0;
554 j_coord_offsetB = 0;
555 j_coord_offsetC = 0;
556 j_coord_offsetD = 0;
557
558 outeriter = 0;
559 inneriter = 0;
560
561 for(iidx=0;iidx<4*DIM;iidx++)
562 {
563 scratch[iidx] = 0.0;
564 }
565
566 /* Start outer loop over neighborlists */
567 for(iidx=0; iidx<nri; iidx++)
568 {
569 /* Load shift vector for this list */
570 i_shift_offset = DIM*shiftidx[iidx];
571
572 /* Load limits for loop over neighbors */
573 j_index_start = jindex[iidx];
574 j_index_end = jindex[iidx+1];
575
576 /* Get outer coordinate index */
577 inr = iinr[iidx];
578 i_coord_offset = DIM*inr;
579
580 /* Load i particle coords and add shift vector */
581 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
582
583 fix0 = _mm_setzero_ps();
584 fiy0 = _mm_setzero_ps();
585 fiz0 = _mm_setzero_ps();
586
587 /* Load parameters for i particles */
588 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
589 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
590
591 /* Start inner kernel loop */
592 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
593 {
594
595 /* Get j neighbor index, and coordinate index */
596 jnrA = jjnr[jidx];
597 jnrB = jjnr[jidx+1];
598 jnrC = jjnr[jidx+2];
599 jnrD = jjnr[jidx+3];
600 j_coord_offsetA = DIM*jnrA;
601 j_coord_offsetB = DIM*jnrB;
602 j_coord_offsetC = DIM*jnrC;
603 j_coord_offsetD = DIM*jnrD;
604
605 /* load j atom coordinates */
606 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
607 x+j_coord_offsetC,x+j_coord_offsetD,
608 &jx0,&jy0,&jz0);
609
610 /* Calculate displacement vector */
611 dx00 = _mm_sub_ps(ix0,jx0);
612 dy00 = _mm_sub_ps(iy0,jy0);
613 dz00 = _mm_sub_ps(iz0,jz0);
614
615 /* Calculate squared distance and things based on it */
616 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
617
618 rinv00 = gmx_mm_invsqrt_ps(rsq00);
619
620 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
621
622 /* Load parameters for j particles */
623 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
624 charge+jnrC+0,charge+jnrD+0);
625 vdwjidx0A = 2*vdwtype[jnrA+0];
626 vdwjidx0B = 2*vdwtype[jnrB+0];
627 vdwjidx0C = 2*vdwtype[jnrC+0];
628 vdwjidx0D = 2*vdwtype[jnrD+0];
629
630 /**************************
631 * CALCULATE INTERACTIONS *
632 **************************/
633
634 if (gmx_mm_any_lt(rsq00,rcutoff2))
635 {
636
637 r00 = _mm_mul_ps(rsq00,rinv00);
638
639 /* Compute parameters for interactions between i and j atoms */
640 qq00 = _mm_mul_ps(iq0,jq0);
641 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
642 vdwparam+vdwioffset0+vdwjidx0B,
643 vdwparam+vdwioffset0+vdwjidx0C,
644 vdwparam+vdwioffset0+vdwjidx0D,
645 &c6_00,&c12_00);
646
647 /* Calculate table index by multiplying r with table scale and truncate to integer */
648 rt = _mm_mul_ps(r00,vftabscale);
649 vfitab = _mm_cvttps_epi32(rt);
650 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
651 vfitab = _mm_slli_epi32(vfitab,3);
652
653 /* REACTION-FIELD ELECTROSTATICS */
654 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
655
656 /* CUBIC SPLINE TABLE DISPERSION */
657 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
658 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
659 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
660 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
661 _MM_TRANSPOSE4_PS(Y,F,G,H);
662 Heps = _mm_mul_ps(vfeps,H);
663 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
664 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
665 fvdw6 = _mm_mul_ps(c6_00,FF);
666
667 /* CUBIC SPLINE TABLE REPULSION */
668 vfitab = _mm_add_epi32(vfitab,ifour);
669 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
670 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
671 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
672 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
673 _MM_TRANSPOSE4_PS(Y,F,G,H);
674 Heps = _mm_mul_ps(vfeps,H);
675 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
676 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
677 fvdw12 = _mm_mul_ps(c12_00,FF);
678 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
679
680 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
681
682 fscal = _mm_add_ps(felec,fvdw);
683
684 fscal = _mm_and_ps(fscal,cutoff_mask);
685
686 /* Calculate temporary vectorial force */
687 tx = _mm_mul_ps(fscal,dx00);
688 ty = _mm_mul_ps(fscal,dy00);
689 tz = _mm_mul_ps(fscal,dz00);
690
691 /* Update vectorial force */
692 fix0 = _mm_add_ps(fix0,tx);
693 fiy0 = _mm_add_ps(fiy0,ty);
694 fiz0 = _mm_add_ps(fiz0,tz);
695
696 fjptrA = f+j_coord_offsetA;
697 fjptrB = f+j_coord_offsetB;
698 fjptrC = f+j_coord_offsetC;
699 fjptrD = f+j_coord_offsetD;
700 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
701
702 }
703
704 /* Inner loop uses 57 flops */
705 }
706
707 if(jidx<j_index_end)
708 {
709
710 /* Get j neighbor index, and coordinate index */
711 jnrlistA = jjnr[jidx];
712 jnrlistB = jjnr[jidx+1];
713 jnrlistC = jjnr[jidx+2];
714 jnrlistD = jjnr[jidx+3];
715 /* Sign of each element will be negative for non-real atoms.
716 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
717 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
718 */
719 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
720 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
721 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
722 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
723 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
724 j_coord_offsetA = DIM*jnrA;
725 j_coord_offsetB = DIM*jnrB;
726 j_coord_offsetC = DIM*jnrC;
727 j_coord_offsetD = DIM*jnrD;
728
729 /* load j atom coordinates */
730 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
731 x+j_coord_offsetC,x+j_coord_offsetD,
732 &jx0,&jy0,&jz0);
733
734 /* Calculate displacement vector */
735 dx00 = _mm_sub_ps(ix0,jx0);
736 dy00 = _mm_sub_ps(iy0,jy0);
737 dz00 = _mm_sub_ps(iz0,jz0);
738
739 /* Calculate squared distance and things based on it */
740 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
741
742 rinv00 = gmx_mm_invsqrt_ps(rsq00);
743
744 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
745
746 /* Load parameters for j particles */
747 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
748 charge+jnrC+0,charge+jnrD+0);
749 vdwjidx0A = 2*vdwtype[jnrA+0];
750 vdwjidx0B = 2*vdwtype[jnrB+0];
751 vdwjidx0C = 2*vdwtype[jnrC+0];
752 vdwjidx0D = 2*vdwtype[jnrD+0];
753
754 /**************************
755 * CALCULATE INTERACTIONS *
756 **************************/
757
758 if (gmx_mm_any_lt(rsq00,rcutoff2))
759 {
760
761 r00 = _mm_mul_ps(rsq00,rinv00);
762 r00 = _mm_andnot_ps(dummy_mask,r00);
763
764 /* Compute parameters for interactions between i and j atoms */
765 qq00 = _mm_mul_ps(iq0,jq0);
766 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
767 vdwparam+vdwioffset0+vdwjidx0B,
768 vdwparam+vdwioffset0+vdwjidx0C,
769 vdwparam+vdwioffset0+vdwjidx0D,
770 &c6_00,&c12_00);
771
772 /* Calculate table index by multiplying r with table scale and truncate to integer */
773 rt = _mm_mul_ps(r00,vftabscale);
774 vfitab = _mm_cvttps_epi32(rt);
775 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
776 vfitab = _mm_slli_epi32(vfitab,3);
777
778 /* REACTION-FIELD ELECTROSTATICS */
779 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
780
781 /* CUBIC SPLINE TABLE DISPERSION */
782 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
783 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
784 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
785 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
786 _MM_TRANSPOSE4_PS(Y,F,G,H);
787 Heps = _mm_mul_ps(vfeps,H);
788 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
789 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
790 fvdw6 = _mm_mul_ps(c6_00,FF);
791
792 /* CUBIC SPLINE TABLE REPULSION */
793 vfitab = _mm_add_epi32(vfitab,ifour);
794 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
795 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
796 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
797 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
798 _MM_TRANSPOSE4_PS(Y,F,G,H);
799 Heps = _mm_mul_ps(vfeps,H);
800 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
801 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
802 fvdw12 = _mm_mul_ps(c12_00,FF);
803 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
804
805 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
806
807 fscal = _mm_add_ps(felec,fvdw);
808
809 fscal = _mm_and_ps(fscal,cutoff_mask);
810
811 fscal = _mm_andnot_ps(dummy_mask,fscal);
812
813 /* Calculate temporary vectorial force */
814 tx = _mm_mul_ps(fscal,dx00);
815 ty = _mm_mul_ps(fscal,dy00);
816 tz = _mm_mul_ps(fscal,dz00);
817
818 /* Update vectorial force */
819 fix0 = _mm_add_ps(fix0,tx);
820 fiy0 = _mm_add_ps(fiy0,ty);
821 fiz0 = _mm_add_ps(fiz0,tz);
822
823 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
824 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
825 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
826 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
827 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
828
829 }
830
831 /* Inner loop uses 58 flops */
832 }
833
834 /* End of innermost loop */
835
836 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
837 f+i_coord_offset,fshift+i_shift_offset);
838
839 /* Increment number of inner iterations */
840 inneriter += j_index_end - j_index_start;
841
842 /* Outer loop uses 7 flops */
843 }
844
845 /* Increment number of outer iterations */
846 outeriter += nri;
847
848 /* Update outer/inner flops */
849
850 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*58);
851 }
The ultimate molecular dynamics simulation package