?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed simple.h from nb_kernel_sse2_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_sse2_single / nb_kernel_ElecRF_VdwCSTab_GeomW4P1_sse2_single.c
blob16b558d1b3e6a3846bf568cec3256567262c4ccd
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
8 *
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
13 *
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
23 *
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
31 *
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
34 */
35 /*
36 * Note: this file was generated by the GROMACS sse2_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_sse2_single.h"
49 #include "kernelutil_x86_sse2_single.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_sse2_single
53 * Electrostatics interaction: ReactionField
54 * VdW interaction: CubicSplineTable
55 * Geometry: Water4-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_sse2_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 vdwioffset3;
91 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
92 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
93 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
94 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
95 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
96 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
97 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
98 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
99 real *charge;
100 int nvdwtype;
101 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
102 int *vdwtype;
103 real *vdwparam;
104 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
105 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
106 __m128i vfitab;
107 __m128i ifour = _mm_set1_epi32(4);
108 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
109 real *vftab;
110 __m128 dummy_mask,cutoff_mask;
111 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
112 __m128 one = _mm_set1_ps(1.0);
113 __m128 two = _mm_set1_ps(2.0);
114 x = xx[0];
115 f = ff[0];
116
117 nri = nlist->nri;
118 iinr = nlist->iinr;
119 jindex = nlist->jindex;
120 jjnr = nlist->jjnr;
121 shiftidx = nlist->shift;
122 gid = nlist->gid;
123 shiftvec = fr->shift_vec[0];
124 fshift = fr->fshift[0];
125 facel = _mm_set1_ps(fr->epsfac);
126 charge = mdatoms->chargeA;
127 krf = _mm_set1_ps(fr->ic->k_rf);
128 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
129 crf = _mm_set1_ps(fr->ic->c_rf);
130 nvdwtype = fr->ntype;
131 vdwparam = fr->nbfp;
132 vdwtype = mdatoms->typeA;
133
134 vftab = kernel_data->table_vdw->data;
135 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
136
137 /* Setup water-specific parameters */
138 inr = nlist->iinr[0];
139 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
140 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
141 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
142 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
143
144 /* Avoid stupid compiler warnings */
145 jnrA = jnrB = jnrC = jnrD = 0;
146 j_coord_offsetA = 0;
147 j_coord_offsetB = 0;
148 j_coord_offsetC = 0;
149 j_coord_offsetD = 0;
150
151 outeriter = 0;
152 inneriter = 0;
153
154 for(iidx=0;iidx<4*DIM;iidx++)
155 {
156 scratch[iidx] = 0.0;
157 }
158
159 /* Start outer loop over neighborlists */
160 for(iidx=0; iidx<nri; iidx++)
161 {
162 /* Load shift vector for this list */
163 i_shift_offset = DIM*shiftidx[iidx];
164
165 /* Load limits for loop over neighbors */
166 j_index_start = jindex[iidx];
167 j_index_end = jindex[iidx+1];
168
169 /* Get outer coordinate index */
170 inr = iinr[iidx];
171 i_coord_offset = DIM*inr;
172
173 /* Load i particle coords and add shift vector */
174 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
175 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
176
177 fix0 = _mm_setzero_ps();
178 fiy0 = _mm_setzero_ps();
179 fiz0 = _mm_setzero_ps();
180 fix1 = _mm_setzero_ps();
181 fiy1 = _mm_setzero_ps();
182 fiz1 = _mm_setzero_ps();
183 fix2 = _mm_setzero_ps();
184 fiy2 = _mm_setzero_ps();
185 fiz2 = _mm_setzero_ps();
186 fix3 = _mm_setzero_ps();
187 fiy3 = _mm_setzero_ps();
188 fiz3 = _mm_setzero_ps();
189
190 /* Reset potential sums */
191 velecsum = _mm_setzero_ps();
192 vvdwsum = _mm_setzero_ps();
193
194 /* Start inner kernel loop */
195 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
196 {
197
198 /* Get j neighbor index, and coordinate index */
199 jnrA = jjnr[jidx];
200 jnrB = jjnr[jidx+1];
201 jnrC = jjnr[jidx+2];
202 jnrD = jjnr[jidx+3];
203 j_coord_offsetA = DIM*jnrA;
204 j_coord_offsetB = DIM*jnrB;
205 j_coord_offsetC = DIM*jnrC;
206 j_coord_offsetD = DIM*jnrD;
207
208 /* load j atom coordinates */
209 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
210 x+j_coord_offsetC,x+j_coord_offsetD,
211 &jx0,&jy0,&jz0);
212
213 /* Calculate displacement vector */
214 dx00 = _mm_sub_ps(ix0,jx0);
215 dy00 = _mm_sub_ps(iy0,jy0);
216 dz00 = _mm_sub_ps(iz0,jz0);
217 dx10 = _mm_sub_ps(ix1,jx0);
218 dy10 = _mm_sub_ps(iy1,jy0);
219 dz10 = _mm_sub_ps(iz1,jz0);
220 dx20 = _mm_sub_ps(ix2,jx0);
221 dy20 = _mm_sub_ps(iy2,jy0);
222 dz20 = _mm_sub_ps(iz2,jz0);
223 dx30 = _mm_sub_ps(ix3,jx0);
224 dy30 = _mm_sub_ps(iy3,jy0);
225 dz30 = _mm_sub_ps(iz3,jz0);
226
227 /* Calculate squared distance and things based on it */
228 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
229 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
230 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
231 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
232
233 rinv00 = gmx_mm_invsqrt_ps(rsq00);
234 rinv10 = gmx_mm_invsqrt_ps(rsq10);
235 rinv20 = gmx_mm_invsqrt_ps(rsq20);
236 rinv30 = gmx_mm_invsqrt_ps(rsq30);
237
238 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
239 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
240 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
241
242 /* Load parameters for j particles */
243 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
244 charge+jnrC+0,charge+jnrD+0);
245 vdwjidx0A = 2*vdwtype[jnrA+0];
246 vdwjidx0B = 2*vdwtype[jnrB+0];
247 vdwjidx0C = 2*vdwtype[jnrC+0];
248 vdwjidx0D = 2*vdwtype[jnrD+0];
249
250 fjx0 = _mm_setzero_ps();
251 fjy0 = _mm_setzero_ps();
252 fjz0 = _mm_setzero_ps();
253
254 /**************************
255 * CALCULATE INTERACTIONS *
256 **************************/
257
258 r00 = _mm_mul_ps(rsq00,rinv00);
259
260 /* Compute parameters for interactions between i and j atoms */
261 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
262 vdwparam+vdwioffset0+vdwjidx0B,
263 vdwparam+vdwioffset0+vdwjidx0C,
264 vdwparam+vdwioffset0+vdwjidx0D,
265 &c6_00,&c12_00);
266
267 /* Calculate table index by multiplying r with table scale and truncate to integer */
268 rt = _mm_mul_ps(r00,vftabscale);
269 vfitab = _mm_cvttps_epi32(rt);
270 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
271 vfitab = _mm_slli_epi32(vfitab,3);
272
273 /* CUBIC SPLINE TABLE DISPERSION */
274 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
275 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
276 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
277 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
278 _MM_TRANSPOSE4_PS(Y,F,G,H);
279 Heps = _mm_mul_ps(vfeps,H);
280 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
281 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
282 vvdw6 = _mm_mul_ps(c6_00,VV);
283 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
284 fvdw6 = _mm_mul_ps(c6_00,FF);
285
286 /* CUBIC SPLINE TABLE REPULSION */
287 vfitab = _mm_add_epi32(vfitab,ifour);
288 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
289 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
290 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
291 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
292 _MM_TRANSPOSE4_PS(Y,F,G,H);
293 Heps = _mm_mul_ps(vfeps,H);
294 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
295 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
296 vvdw12 = _mm_mul_ps(c12_00,VV);
297 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
298 fvdw12 = _mm_mul_ps(c12_00,FF);
299 vvdw = _mm_add_ps(vvdw12,vvdw6);
300 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
301
302 /* Update potential sum for this i atom from the interaction with this j atom. */
303 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
304
305 fscal = fvdw;
306
307 /* Calculate temporary vectorial force */
308 tx = _mm_mul_ps(fscal,dx00);
309 ty = _mm_mul_ps(fscal,dy00);
310 tz = _mm_mul_ps(fscal,dz00);
311
312 /* Update vectorial force */
313 fix0 = _mm_add_ps(fix0,tx);
314 fiy0 = _mm_add_ps(fiy0,ty);
315 fiz0 = _mm_add_ps(fiz0,tz);
316
317 fjx0 = _mm_add_ps(fjx0,tx);
318 fjy0 = _mm_add_ps(fjy0,ty);
319 fjz0 = _mm_add_ps(fjz0,tz);
320
321 /**************************
322 * CALCULATE INTERACTIONS *
323 **************************/
324
325 /* Compute parameters for interactions between i and j atoms */
326 qq10 = _mm_mul_ps(iq1,jq0);
327
328 /* REACTION-FIELD ELECTROSTATICS */
329 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
330 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
331
332 /* Update potential sum for this i atom from the interaction with this j atom. */
333 velecsum = _mm_add_ps(velecsum,velec);
334
335 fscal = felec;
336
337 /* Calculate temporary vectorial force */
338 tx = _mm_mul_ps(fscal,dx10);
339 ty = _mm_mul_ps(fscal,dy10);
340 tz = _mm_mul_ps(fscal,dz10);
341
342 /* Update vectorial force */
343 fix1 = _mm_add_ps(fix1,tx);
344 fiy1 = _mm_add_ps(fiy1,ty);
345 fiz1 = _mm_add_ps(fiz1,tz);
346
347 fjx0 = _mm_add_ps(fjx0,tx);
348 fjy0 = _mm_add_ps(fjy0,ty);
349 fjz0 = _mm_add_ps(fjz0,tz);
350
351 /**************************
352 * CALCULATE INTERACTIONS *
353 **************************/
354
355 /* Compute parameters for interactions between i and j atoms */
356 qq20 = _mm_mul_ps(iq2,jq0);
357
358 /* REACTION-FIELD ELECTROSTATICS */
359 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
360 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
361
362 /* Update potential sum for this i atom from the interaction with this j atom. */
363 velecsum = _mm_add_ps(velecsum,velec);
364
365 fscal = felec;
366
367 /* Calculate temporary vectorial force */
368 tx = _mm_mul_ps(fscal,dx20);
369 ty = _mm_mul_ps(fscal,dy20);
370 tz = _mm_mul_ps(fscal,dz20);
371
372 /* Update vectorial force */
373 fix2 = _mm_add_ps(fix2,tx);
374 fiy2 = _mm_add_ps(fiy2,ty);
375 fiz2 = _mm_add_ps(fiz2,tz);
376
377 fjx0 = _mm_add_ps(fjx0,tx);
378 fjy0 = _mm_add_ps(fjy0,ty);
379 fjz0 = _mm_add_ps(fjz0,tz);
380
381 /**************************
382 * CALCULATE INTERACTIONS *
383 **************************/
384
385 /* Compute parameters for interactions between i and j atoms */
386 qq30 = _mm_mul_ps(iq3,jq0);
387
388 /* REACTION-FIELD ELECTROSTATICS */
389 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
390 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
391
392 /* Update potential sum for this i atom from the interaction with this j atom. */
393 velecsum = _mm_add_ps(velecsum,velec);
394
395 fscal = felec;
396
397 /* Calculate temporary vectorial force */
398 tx = _mm_mul_ps(fscal,dx30);
399 ty = _mm_mul_ps(fscal,dy30);
400 tz = _mm_mul_ps(fscal,dz30);
401
402 /* Update vectorial force */
403 fix3 = _mm_add_ps(fix3,tx);
404 fiy3 = _mm_add_ps(fiy3,ty);
405 fiz3 = _mm_add_ps(fiz3,tz);
406
407 fjx0 = _mm_add_ps(fjx0,tx);
408 fjy0 = _mm_add_ps(fjy0,ty);
409 fjz0 = _mm_add_ps(fjz0,tz);
410
411 fjptrA = f+j_coord_offsetA;
412 fjptrB = f+j_coord_offsetB;
413 fjptrC = f+j_coord_offsetC;
414 fjptrD = f+j_coord_offsetD;
415
416 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
417
418 /* Inner loop uses 152 flops */
419 }
420
421 if(jidx<j_index_end)
422 {
423
424 /* Get j neighbor index, and coordinate index */
425 jnrlistA = jjnr[jidx];
426 jnrlistB = jjnr[jidx+1];
427 jnrlistC = jjnr[jidx+2];
428 jnrlistD = jjnr[jidx+3];
429 /* Sign of each element will be negative for non-real atoms.
430 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
431 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
432 */
433 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
434 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
435 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
436 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
437 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
438 j_coord_offsetA = DIM*jnrA;
439 j_coord_offsetB = DIM*jnrB;
440 j_coord_offsetC = DIM*jnrC;
441 j_coord_offsetD = DIM*jnrD;
442
443 /* load j atom coordinates */
444 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
445 x+j_coord_offsetC,x+j_coord_offsetD,
446 &jx0,&jy0,&jz0);
447
448 /* Calculate displacement vector */
449 dx00 = _mm_sub_ps(ix0,jx0);
450 dy00 = _mm_sub_ps(iy0,jy0);
451 dz00 = _mm_sub_ps(iz0,jz0);
452 dx10 = _mm_sub_ps(ix1,jx0);
453 dy10 = _mm_sub_ps(iy1,jy0);
454 dz10 = _mm_sub_ps(iz1,jz0);
455 dx20 = _mm_sub_ps(ix2,jx0);
456 dy20 = _mm_sub_ps(iy2,jy0);
457 dz20 = _mm_sub_ps(iz2,jz0);
458 dx30 = _mm_sub_ps(ix3,jx0);
459 dy30 = _mm_sub_ps(iy3,jy0);
460 dz30 = _mm_sub_ps(iz3,jz0);
461
462 /* Calculate squared distance and things based on it */
463 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
464 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
465 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
466 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
467
468 rinv00 = gmx_mm_invsqrt_ps(rsq00);
469 rinv10 = gmx_mm_invsqrt_ps(rsq10);
470 rinv20 = gmx_mm_invsqrt_ps(rsq20);
471 rinv30 = gmx_mm_invsqrt_ps(rsq30);
472
473 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
474 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
475 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
476
477 /* Load parameters for j particles */
478 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
479 charge+jnrC+0,charge+jnrD+0);
480 vdwjidx0A = 2*vdwtype[jnrA+0];
481 vdwjidx0B = 2*vdwtype[jnrB+0];
482 vdwjidx0C = 2*vdwtype[jnrC+0];
483 vdwjidx0D = 2*vdwtype[jnrD+0];
484
485 fjx0 = _mm_setzero_ps();
486 fjy0 = _mm_setzero_ps();
487 fjz0 = _mm_setzero_ps();
488
489 /**************************
490 * CALCULATE INTERACTIONS *
491 **************************/
492
493 r00 = _mm_mul_ps(rsq00,rinv00);
494 r00 = _mm_andnot_ps(dummy_mask,r00);
495
496 /* Compute parameters for interactions between i and j atoms */
497 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
498 vdwparam+vdwioffset0+vdwjidx0B,
499 vdwparam+vdwioffset0+vdwjidx0C,
500 vdwparam+vdwioffset0+vdwjidx0D,
501 &c6_00,&c12_00);
502
503 /* Calculate table index by multiplying r with table scale and truncate to integer */
504 rt = _mm_mul_ps(r00,vftabscale);
505 vfitab = _mm_cvttps_epi32(rt);
506 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
507 vfitab = _mm_slli_epi32(vfitab,3);
508
509 /* CUBIC SPLINE TABLE DISPERSION */
510 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
511 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
512 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
513 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
514 _MM_TRANSPOSE4_PS(Y,F,G,H);
515 Heps = _mm_mul_ps(vfeps,H);
516 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
517 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
518 vvdw6 = _mm_mul_ps(c6_00,VV);
519 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
520 fvdw6 = _mm_mul_ps(c6_00,FF);
521
522 /* CUBIC SPLINE TABLE REPULSION */
523 vfitab = _mm_add_epi32(vfitab,ifour);
524 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
525 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
526 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
527 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
528 _MM_TRANSPOSE4_PS(Y,F,G,H);
529 Heps = _mm_mul_ps(vfeps,H);
530 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
531 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
532 vvdw12 = _mm_mul_ps(c12_00,VV);
533 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
534 fvdw12 = _mm_mul_ps(c12_00,FF);
535 vvdw = _mm_add_ps(vvdw12,vvdw6);
536 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
537
538 /* Update potential sum for this i atom from the interaction with this j atom. */
539 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
540 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
541
542 fscal = fvdw;
543
544 fscal = _mm_andnot_ps(dummy_mask,fscal);
545
546 /* Calculate temporary vectorial force */
547 tx = _mm_mul_ps(fscal,dx00);
548 ty = _mm_mul_ps(fscal,dy00);
549 tz = _mm_mul_ps(fscal,dz00);
550
551 /* Update vectorial force */
552 fix0 = _mm_add_ps(fix0,tx);
553 fiy0 = _mm_add_ps(fiy0,ty);
554 fiz0 = _mm_add_ps(fiz0,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 qq10 = _mm_mul_ps(iq1,jq0);
566
567 /* REACTION-FIELD ELECTROSTATICS */
568 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
569 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),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,dx10);
581 ty = _mm_mul_ps(fscal,dy10);
582 tz = _mm_mul_ps(fscal,dz10);
583
584 /* Update vectorial force */
585 fix1 = _mm_add_ps(fix1,tx);
586 fiy1 = _mm_add_ps(fiy1,ty);
587 fiz1 = _mm_add_ps(fiz1,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 /**************************
594 * CALCULATE INTERACTIONS *
595 **************************/
596
597 /* Compute parameters for interactions between i and j atoms */
598 qq20 = _mm_mul_ps(iq2,jq0);
599
600 /* REACTION-FIELD ELECTROSTATICS */
601 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
602 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
603
604 /* Update potential sum for this i atom from the interaction with this j atom. */
605 velec = _mm_andnot_ps(dummy_mask,velec);
606 velecsum = _mm_add_ps(velecsum,velec);
607
608 fscal = felec;
609
610 fscal = _mm_andnot_ps(dummy_mask,fscal);
611
612 /* Calculate temporary vectorial force */
613 tx = _mm_mul_ps(fscal,dx20);
614 ty = _mm_mul_ps(fscal,dy20);
615 tz = _mm_mul_ps(fscal,dz20);
616
617 /* Update vectorial force */
618 fix2 = _mm_add_ps(fix2,tx);
619 fiy2 = _mm_add_ps(fiy2,ty);
620 fiz2 = _mm_add_ps(fiz2,tz);
621
622 fjx0 = _mm_add_ps(fjx0,tx);
623 fjy0 = _mm_add_ps(fjy0,ty);
624 fjz0 = _mm_add_ps(fjz0,tz);
625
626 /**************************
627 * CALCULATE INTERACTIONS *
628 **************************/
629
630 /* Compute parameters for interactions between i and j atoms */
631 qq30 = _mm_mul_ps(iq3,jq0);
632
633 /* REACTION-FIELD ELECTROSTATICS */
634 velec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
635 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
636
637 /* Update potential sum for this i atom from the interaction with this j atom. */
638 velec = _mm_andnot_ps(dummy_mask,velec);
639 velecsum = _mm_add_ps(velecsum,velec);
640
641 fscal = felec;
642
643 fscal = _mm_andnot_ps(dummy_mask,fscal);
644
645 /* Calculate temporary vectorial force */
646 tx = _mm_mul_ps(fscal,dx30);
647 ty = _mm_mul_ps(fscal,dy30);
648 tz = _mm_mul_ps(fscal,dz30);
649
650 /* Update vectorial force */
651 fix3 = _mm_add_ps(fix3,tx);
652 fiy3 = _mm_add_ps(fiy3,ty);
653 fiz3 = _mm_add_ps(fiz3,tz);
654
655 fjx0 = _mm_add_ps(fjx0,tx);
656 fjy0 = _mm_add_ps(fjy0,ty);
657 fjz0 = _mm_add_ps(fjz0,tz);
658
659 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
660 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
661 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
662 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
663
664 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
665
666 /* Inner loop uses 153 flops */
667 }
668
669 /* End of innermost loop */
670
671 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
672 f+i_coord_offset,fshift+i_shift_offset);
673
674 ggid = gid[iidx];
675 /* Update potential energies */
676 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
677 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
678
679 /* Increment number of inner iterations */
680 inneriter += j_index_end - j_index_start;
681
682 /* Outer loop uses 26 flops */
683 }
684
685 /* Increment number of outer iterations */
686 outeriter += nri;
687
688 /* Update outer/inner flops */
689
690 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*153);
691 }
692 /*
693 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse2_single
694 * Electrostatics interaction: ReactionField
695 * VdW interaction: CubicSplineTable
696 * Geometry: Water4-Particle
697 * Calculate force/pot: Force
698 */
699 void
700 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse2_single
701 (t_nblist * gmx_restrict nlist,
702 rvec * gmx_restrict xx,
703 rvec * gmx_restrict ff,
704 t_forcerec * gmx_restrict fr,
705 t_mdatoms * gmx_restrict mdatoms,
706 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
707 t_nrnb * gmx_restrict nrnb)
708 {
709 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
710 * just 0 for non-waters.
711 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
712 * jnr indices corresponding to data put in the four positions in the SIMD register.
713 */
714 int i_shift_offset,i_coord_offset,outeriter,inneriter;
715 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
716 int jnrA,jnrB,jnrC,jnrD;
717 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
718 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
719 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
720 real rcutoff_scalar;
721 real *shiftvec,*fshift,*x,*f;
722 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
723 real scratch[4*DIM];
724 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
725 int vdwioffset0;
726 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
727 int vdwioffset1;
728 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
729 int vdwioffset2;
730 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
731 int vdwioffset3;
732 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
733 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
734 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
735 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
736 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
737 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
738 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
739 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
740 real *charge;
741 int nvdwtype;
742 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
743 int *vdwtype;
744 real *vdwparam;
745 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
746 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
747 __m128i vfitab;
748 __m128i ifour = _mm_set1_epi32(4);
749 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
750 real *vftab;
751 __m128 dummy_mask,cutoff_mask;
752 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
753 __m128 one = _mm_set1_ps(1.0);
754 __m128 two = _mm_set1_ps(2.0);
755 x = xx[0];
756 f = ff[0];
757
758 nri = nlist->nri;
759 iinr = nlist->iinr;
760 jindex = nlist->jindex;
761 jjnr = nlist->jjnr;
762 shiftidx = nlist->shift;
763 gid = nlist->gid;
764 shiftvec = fr->shift_vec[0];
765 fshift = fr->fshift[0];
766 facel = _mm_set1_ps(fr->epsfac);
767 charge = mdatoms->chargeA;
768 krf = _mm_set1_ps(fr->ic->k_rf);
769 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
770 crf = _mm_set1_ps(fr->ic->c_rf);
771 nvdwtype = fr->ntype;
772 vdwparam = fr->nbfp;
773 vdwtype = mdatoms->typeA;
774
775 vftab = kernel_data->table_vdw->data;
776 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
777
778 /* Setup water-specific parameters */
779 inr = nlist->iinr[0];
780 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
781 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
782 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
783 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
784
785 /* Avoid stupid compiler warnings */
786 jnrA = jnrB = jnrC = jnrD = 0;
787 j_coord_offsetA = 0;
788 j_coord_offsetB = 0;
789 j_coord_offsetC = 0;
790 j_coord_offsetD = 0;
791
792 outeriter = 0;
793 inneriter = 0;
794
795 for(iidx=0;iidx<4*DIM;iidx++)
796 {
797 scratch[iidx] = 0.0;
798 }
799
800 /* Start outer loop over neighborlists */
801 for(iidx=0; iidx<nri; iidx++)
802 {
803 /* Load shift vector for this list */
804 i_shift_offset = DIM*shiftidx[iidx];
805
806 /* Load limits for loop over neighbors */
807 j_index_start = jindex[iidx];
808 j_index_end = jindex[iidx+1];
809
810 /* Get outer coordinate index */
811 inr = iinr[iidx];
812 i_coord_offset = DIM*inr;
813
814 /* Load i particle coords and add shift vector */
815 gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
816 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
817
818 fix0 = _mm_setzero_ps();
819 fiy0 = _mm_setzero_ps();
820 fiz0 = _mm_setzero_ps();
821 fix1 = _mm_setzero_ps();
822 fiy1 = _mm_setzero_ps();
823 fiz1 = _mm_setzero_ps();
824 fix2 = _mm_setzero_ps();
825 fiy2 = _mm_setzero_ps();
826 fiz2 = _mm_setzero_ps();
827 fix3 = _mm_setzero_ps();
828 fiy3 = _mm_setzero_ps();
829 fiz3 = _mm_setzero_ps();
830
831 /* Start inner kernel loop */
832 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
833 {
834
835 /* Get j neighbor index, and coordinate index */
836 jnrA = jjnr[jidx];
837 jnrB = jjnr[jidx+1];
838 jnrC = jjnr[jidx+2];
839 jnrD = jjnr[jidx+3];
840 j_coord_offsetA = DIM*jnrA;
841 j_coord_offsetB = DIM*jnrB;
842 j_coord_offsetC = DIM*jnrC;
843 j_coord_offsetD = DIM*jnrD;
844
845 /* load j atom coordinates */
846 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
847 x+j_coord_offsetC,x+j_coord_offsetD,
848 &jx0,&jy0,&jz0);
849
850 /* Calculate displacement vector */
851 dx00 = _mm_sub_ps(ix0,jx0);
852 dy00 = _mm_sub_ps(iy0,jy0);
853 dz00 = _mm_sub_ps(iz0,jz0);
854 dx10 = _mm_sub_ps(ix1,jx0);
855 dy10 = _mm_sub_ps(iy1,jy0);
856 dz10 = _mm_sub_ps(iz1,jz0);
857 dx20 = _mm_sub_ps(ix2,jx0);
858 dy20 = _mm_sub_ps(iy2,jy0);
859 dz20 = _mm_sub_ps(iz2,jz0);
860 dx30 = _mm_sub_ps(ix3,jx0);
861 dy30 = _mm_sub_ps(iy3,jy0);
862 dz30 = _mm_sub_ps(iz3,jz0);
863
864 /* Calculate squared distance and things based on it */
865 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
866 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
867 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
868 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
869
870 rinv00 = gmx_mm_invsqrt_ps(rsq00);
871 rinv10 = gmx_mm_invsqrt_ps(rsq10);
872 rinv20 = gmx_mm_invsqrt_ps(rsq20);
873 rinv30 = gmx_mm_invsqrt_ps(rsq30);
874
875 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
876 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
877 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
878
879 /* Load parameters for j particles */
880 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
881 charge+jnrC+0,charge+jnrD+0);
882 vdwjidx0A = 2*vdwtype[jnrA+0];
883 vdwjidx0B = 2*vdwtype[jnrB+0];
884 vdwjidx0C = 2*vdwtype[jnrC+0];
885 vdwjidx0D = 2*vdwtype[jnrD+0];
886
887 fjx0 = _mm_setzero_ps();
888 fjy0 = _mm_setzero_ps();
889 fjz0 = _mm_setzero_ps();
890
891 /**************************
892 * CALCULATE INTERACTIONS *
893 **************************/
894
895 r00 = _mm_mul_ps(rsq00,rinv00);
896
897 /* Compute parameters for interactions between i and j atoms */
898 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
899 vdwparam+vdwioffset0+vdwjidx0B,
900 vdwparam+vdwioffset0+vdwjidx0C,
901 vdwparam+vdwioffset0+vdwjidx0D,
902 &c6_00,&c12_00);
903
904 /* Calculate table index by multiplying r with table scale and truncate to integer */
905 rt = _mm_mul_ps(r00,vftabscale);
906 vfitab = _mm_cvttps_epi32(rt);
907 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
908 vfitab = _mm_slli_epi32(vfitab,3);
909
910 /* CUBIC SPLINE TABLE DISPERSION */
911 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
912 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
913 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
914 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
915 _MM_TRANSPOSE4_PS(Y,F,G,H);
916 Heps = _mm_mul_ps(vfeps,H);
917 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
918 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
919 fvdw6 = _mm_mul_ps(c6_00,FF);
920
921 /* CUBIC SPLINE TABLE REPULSION */
922 vfitab = _mm_add_epi32(vfitab,ifour);
923 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
924 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
925 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
926 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
927 _MM_TRANSPOSE4_PS(Y,F,G,H);
928 Heps = _mm_mul_ps(vfeps,H);
929 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
930 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
931 fvdw12 = _mm_mul_ps(c12_00,FF);
932 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
933
934 fscal = fvdw;
935
936 /* Calculate temporary vectorial force */
937 tx = _mm_mul_ps(fscal,dx00);
938 ty = _mm_mul_ps(fscal,dy00);
939 tz = _mm_mul_ps(fscal,dz00);
940
941 /* Update vectorial force */
942 fix0 = _mm_add_ps(fix0,tx);
943 fiy0 = _mm_add_ps(fiy0,ty);
944 fiz0 = _mm_add_ps(fiz0,tz);
945
946 fjx0 = _mm_add_ps(fjx0,tx);
947 fjy0 = _mm_add_ps(fjy0,ty);
948 fjz0 = _mm_add_ps(fjz0,tz);
949
950 /**************************
951 * CALCULATE INTERACTIONS *
952 **************************/
953
954 /* Compute parameters for interactions between i and j atoms */
955 qq10 = _mm_mul_ps(iq1,jq0);
956
957 /* REACTION-FIELD ELECTROSTATICS */
958 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
959
960 fscal = felec;
961
962 /* Calculate temporary vectorial force */
963 tx = _mm_mul_ps(fscal,dx10);
964 ty = _mm_mul_ps(fscal,dy10);
965 tz = _mm_mul_ps(fscal,dz10);
966
967 /* Update vectorial force */
968 fix1 = _mm_add_ps(fix1,tx);
969 fiy1 = _mm_add_ps(fiy1,ty);
970 fiz1 = _mm_add_ps(fiz1,tz);
971
972 fjx0 = _mm_add_ps(fjx0,tx);
973 fjy0 = _mm_add_ps(fjy0,ty);
974 fjz0 = _mm_add_ps(fjz0,tz);
975
976 /**************************
977 * CALCULATE INTERACTIONS *
978 **************************/
979
980 /* Compute parameters for interactions between i and j atoms */
981 qq20 = _mm_mul_ps(iq2,jq0);
982
983 /* REACTION-FIELD ELECTROSTATICS */
984 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
985
986 fscal = felec;
987
988 /* Calculate temporary vectorial force */
989 tx = _mm_mul_ps(fscal,dx20);
990 ty = _mm_mul_ps(fscal,dy20);
991 tz = _mm_mul_ps(fscal,dz20);
992
993 /* Update vectorial force */
994 fix2 = _mm_add_ps(fix2,tx);
995 fiy2 = _mm_add_ps(fiy2,ty);
996 fiz2 = _mm_add_ps(fiz2,tz);
997
998 fjx0 = _mm_add_ps(fjx0,tx);
999 fjy0 = _mm_add_ps(fjy0,ty);
1000 fjz0 = _mm_add_ps(fjz0,tz);
1001
1002 /**************************
1003 * CALCULATE INTERACTIONS *
1004 **************************/
1005
1006 /* Compute parameters for interactions between i and j atoms */
1007 qq30 = _mm_mul_ps(iq3,jq0);
1008
1009 /* REACTION-FIELD ELECTROSTATICS */
1010 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1011
1012 fscal = felec;
1013
1014 /* Calculate temporary vectorial force */
1015 tx = _mm_mul_ps(fscal,dx30);
1016 ty = _mm_mul_ps(fscal,dy30);
1017 tz = _mm_mul_ps(fscal,dz30);
1018
1019 /* Update vectorial force */
1020 fix3 = _mm_add_ps(fix3,tx);
1021 fiy3 = _mm_add_ps(fiy3,ty);
1022 fiz3 = _mm_add_ps(fiz3,tz);
1023
1024 fjx0 = _mm_add_ps(fjx0,tx);
1025 fjy0 = _mm_add_ps(fjy0,ty);
1026 fjz0 = _mm_add_ps(fjz0,tz);
1027
1028 fjptrA = f+j_coord_offsetA;
1029 fjptrB = f+j_coord_offsetB;
1030 fjptrC = f+j_coord_offsetC;
1031 fjptrD = f+j_coord_offsetD;
1032
1033 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1034
1035 /* Inner loop uses 129 flops */
1036 }
1037
1038 if(jidx<j_index_end)
1039 {
1040
1041 /* Get j neighbor index, and coordinate index */
1042 jnrlistA = jjnr[jidx];
1043 jnrlistB = jjnr[jidx+1];
1044 jnrlistC = jjnr[jidx+2];
1045 jnrlistD = jjnr[jidx+3];
1046 /* Sign of each element will be negative for non-real atoms.
1047 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1048 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1049 */
1050 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1051 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1052 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1053 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1054 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1055 j_coord_offsetA = DIM*jnrA;
1056 j_coord_offsetB = DIM*jnrB;
1057 j_coord_offsetC = DIM*jnrC;
1058 j_coord_offsetD = DIM*jnrD;
1059
1060 /* load j atom coordinates */
1061 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1062 x+j_coord_offsetC,x+j_coord_offsetD,
1063 &jx0,&jy0,&jz0);
1064
1065 /* Calculate displacement vector */
1066 dx00 = _mm_sub_ps(ix0,jx0);
1067 dy00 = _mm_sub_ps(iy0,jy0);
1068 dz00 = _mm_sub_ps(iz0,jz0);
1069 dx10 = _mm_sub_ps(ix1,jx0);
1070 dy10 = _mm_sub_ps(iy1,jy0);
1071 dz10 = _mm_sub_ps(iz1,jz0);
1072 dx20 = _mm_sub_ps(ix2,jx0);
1073 dy20 = _mm_sub_ps(iy2,jy0);
1074 dz20 = _mm_sub_ps(iz2,jz0);
1075 dx30 = _mm_sub_ps(ix3,jx0);
1076 dy30 = _mm_sub_ps(iy3,jy0);
1077 dz30 = _mm_sub_ps(iz3,jz0);
1078
1079 /* Calculate squared distance and things based on it */
1080 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1081 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1082 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1083 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1084
1085 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1086 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1087 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1088 rinv30 = gmx_mm_invsqrt_ps(rsq30);
1089
1090 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1091 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1092 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
1093
1094 /* Load parameters for j particles */
1095 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1096 charge+jnrC+0,charge+jnrD+0);
1097 vdwjidx0A = 2*vdwtype[jnrA+0];
1098 vdwjidx0B = 2*vdwtype[jnrB+0];
1099 vdwjidx0C = 2*vdwtype[jnrC+0];
1100 vdwjidx0D = 2*vdwtype[jnrD+0];
1101
1102 fjx0 = _mm_setzero_ps();
1103 fjy0 = _mm_setzero_ps();
1104 fjz0 = _mm_setzero_ps();
1105
1106 /**************************
1107 * CALCULATE INTERACTIONS *
1108 **************************/
1109
1110 r00 = _mm_mul_ps(rsq00,rinv00);
1111 r00 = _mm_andnot_ps(dummy_mask,r00);
1112
1113 /* Compute parameters for interactions between i and j atoms */
1114 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1115 vdwparam+vdwioffset0+vdwjidx0B,
1116 vdwparam+vdwioffset0+vdwjidx0C,
1117 vdwparam+vdwioffset0+vdwjidx0D,
1118 &c6_00,&c12_00);
1119
1120 /* Calculate table index by multiplying r with table scale and truncate to integer */
1121 rt = _mm_mul_ps(r00,vftabscale);
1122 vfitab = _mm_cvttps_epi32(rt);
1123 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1124 vfitab = _mm_slli_epi32(vfitab,3);
1125
1126 /* CUBIC SPLINE TABLE DISPERSION */
1127 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1128 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1129 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1130 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1131 _MM_TRANSPOSE4_PS(Y,F,G,H);
1132 Heps = _mm_mul_ps(vfeps,H);
1133 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1134 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1135 fvdw6 = _mm_mul_ps(c6_00,FF);
1136
1137 /* CUBIC SPLINE TABLE REPULSION */
1138 vfitab = _mm_add_epi32(vfitab,ifour);
1139 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1140 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1141 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1142 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1143 _MM_TRANSPOSE4_PS(Y,F,G,H);
1144 Heps = _mm_mul_ps(vfeps,H);
1145 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1146 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1147 fvdw12 = _mm_mul_ps(c12_00,FF);
1148 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1149
1150 fscal = fvdw;
1151
1152 fscal = _mm_andnot_ps(dummy_mask,fscal);
1153
1154 /* Calculate temporary vectorial force */
1155 tx = _mm_mul_ps(fscal,dx00);
1156 ty = _mm_mul_ps(fscal,dy00);
1157 tz = _mm_mul_ps(fscal,dz00);
1158
1159 /* Update vectorial force */
1160 fix0 = _mm_add_ps(fix0,tx);
1161 fiy0 = _mm_add_ps(fiy0,ty);
1162 fiz0 = _mm_add_ps(fiz0,tz);
1163
1164 fjx0 = _mm_add_ps(fjx0,tx);
1165 fjy0 = _mm_add_ps(fjy0,ty);
1166 fjz0 = _mm_add_ps(fjz0,tz);
1167
1168 /**************************
1169 * CALCULATE INTERACTIONS *
1170 **************************/
1171
1172 /* Compute parameters for interactions between i and j atoms */
1173 qq10 = _mm_mul_ps(iq1,jq0);
1174
1175 /* REACTION-FIELD ELECTROSTATICS */
1176 felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1177
1178 fscal = felec;
1179
1180 fscal = _mm_andnot_ps(dummy_mask,fscal);
1181
1182 /* Calculate temporary vectorial force */
1183 tx = _mm_mul_ps(fscal,dx10);
1184 ty = _mm_mul_ps(fscal,dy10);
1185 tz = _mm_mul_ps(fscal,dz10);
1186
1187 /* Update vectorial force */
1188 fix1 = _mm_add_ps(fix1,tx);
1189 fiy1 = _mm_add_ps(fiy1,ty);
1190 fiz1 = _mm_add_ps(fiz1,tz);
1191
1192 fjx0 = _mm_add_ps(fjx0,tx);
1193 fjy0 = _mm_add_ps(fjy0,ty);
1194 fjz0 = _mm_add_ps(fjz0,tz);
1195
1196 /**************************
1197 * CALCULATE INTERACTIONS *
1198 **************************/
1199
1200 /* Compute parameters for interactions between i and j atoms */
1201 qq20 = _mm_mul_ps(iq2,jq0);
1202
1203 /* REACTION-FIELD ELECTROSTATICS */
1204 felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1205
1206 fscal = felec;
1207
1208 fscal = _mm_andnot_ps(dummy_mask,fscal);
1209
1210 /* Calculate temporary vectorial force */
1211 tx = _mm_mul_ps(fscal,dx20);
1212 ty = _mm_mul_ps(fscal,dy20);
1213 tz = _mm_mul_ps(fscal,dz20);
1214
1215 /* Update vectorial force */
1216 fix2 = _mm_add_ps(fix2,tx);
1217 fiy2 = _mm_add_ps(fiy2,ty);
1218 fiz2 = _mm_add_ps(fiz2,tz);
1219
1220 fjx0 = _mm_add_ps(fjx0,tx);
1221 fjy0 = _mm_add_ps(fjy0,ty);
1222 fjz0 = _mm_add_ps(fjz0,tz);
1223
1224 /**************************
1225 * CALCULATE INTERACTIONS *
1226 **************************/
1227
1228 /* Compute parameters for interactions between i and j atoms */
1229 qq30 = _mm_mul_ps(iq3,jq0);
1230
1231 /* REACTION-FIELD ELECTROSTATICS */
1232 felec = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1233
1234 fscal = felec;
1235
1236 fscal = _mm_andnot_ps(dummy_mask,fscal);
1237
1238 /* Calculate temporary vectorial force */
1239 tx = _mm_mul_ps(fscal,dx30);
1240 ty = _mm_mul_ps(fscal,dy30);
1241 tz = _mm_mul_ps(fscal,dz30);
1242
1243 /* Update vectorial force */
1244 fix3 = _mm_add_ps(fix3,tx);
1245 fiy3 = _mm_add_ps(fiy3,ty);
1246 fiz3 = _mm_add_ps(fiz3,tz);
1247
1248 fjx0 = _mm_add_ps(fjx0,tx);
1249 fjy0 = _mm_add_ps(fjy0,ty);
1250 fjz0 = _mm_add_ps(fjz0,tz);
1251
1252 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1253 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1254 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1255 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1256
1257 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1258
1259 /* Inner loop uses 130 flops */
1260 }
1261
1262 /* End of innermost loop */
1263
1264 gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1265 f+i_coord_offset,fshift+i_shift_offset);
1266
1267 /* Increment number of inner iterations */
1268 inneriter += j_index_end - j_index_start;
1269
1270 /* Outer loop uses 24 flops */
1271 }
1272
1273 /* Increment number of outer iterations */
1274 outeriter += nri;
1275
1276 /* Update outer/inner flops */
1277
1278 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1279 }
The ultimate molecular dynamics simulation package