?  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_ElecCSTab_VdwLJ_GeomW3P1_sse2_single.c
blob38e95e48e1b31b4d82f6e924521d39afde9843cc
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_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_single
53 * Electrostatics interaction: CubicSplineTable
54 * VdW interaction: LennardJones
55 * Geometry: Water3-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_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 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 nvdwtype = fr->ntype;
125 vdwparam = fr->nbfp;
126 vdwtype = mdatoms->typeA;
127
128 vftab = kernel_data->table_elec->data;
129 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
130
131 /* Setup water-specific parameters */
132 inr = nlist->iinr[0];
133 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
134 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
135 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
136 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
137
138 /* Avoid stupid compiler warnings */
139 jnrA = jnrB = jnrC = jnrD = 0;
140 j_coord_offsetA = 0;
141 j_coord_offsetB = 0;
142 j_coord_offsetC = 0;
143 j_coord_offsetD = 0;
144
145 outeriter = 0;
146 inneriter = 0;
147
148 for(iidx=0;iidx<4*DIM;iidx++)
149 {
150 scratch[iidx] = 0.0;
151 }
152
153 /* Start outer loop over neighborlists */
154 for(iidx=0; iidx<nri; iidx++)
155 {
156 /* Load shift vector for this list */
157 i_shift_offset = DIM*shiftidx[iidx];
158
159 /* Load limits for loop over neighbors */
160 j_index_start = jindex[iidx];
161 j_index_end = jindex[iidx+1];
162
163 /* Get outer coordinate index */
164 inr = iinr[iidx];
165 i_coord_offset = DIM*inr;
166
167 /* Load i particle coords and add shift vector */
168 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
169 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
170
171 fix0 = _mm_setzero_ps();
172 fiy0 = _mm_setzero_ps();
173 fiz0 = _mm_setzero_ps();
174 fix1 = _mm_setzero_ps();
175 fiy1 = _mm_setzero_ps();
176 fiz1 = _mm_setzero_ps();
177 fix2 = _mm_setzero_ps();
178 fiy2 = _mm_setzero_ps();
179 fiz2 = _mm_setzero_ps();
180
181 /* Reset potential sums */
182 velecsum = _mm_setzero_ps();
183 vvdwsum = _mm_setzero_ps();
184
185 /* Start inner kernel loop */
186 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
187 {
188
189 /* Get j neighbor index, and coordinate index */
190 jnrA = jjnr[jidx];
191 jnrB = jjnr[jidx+1];
192 jnrC = jjnr[jidx+2];
193 jnrD = jjnr[jidx+3];
194 j_coord_offsetA = DIM*jnrA;
195 j_coord_offsetB = DIM*jnrB;
196 j_coord_offsetC = DIM*jnrC;
197 j_coord_offsetD = DIM*jnrD;
198
199 /* load j atom coordinates */
200 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
201 x+j_coord_offsetC,x+j_coord_offsetD,
202 &jx0,&jy0,&jz0);
203
204 /* Calculate displacement vector */
205 dx00 = _mm_sub_ps(ix0,jx0);
206 dy00 = _mm_sub_ps(iy0,jy0);
207 dz00 = _mm_sub_ps(iz0,jz0);
208 dx10 = _mm_sub_ps(ix1,jx0);
209 dy10 = _mm_sub_ps(iy1,jy0);
210 dz10 = _mm_sub_ps(iz1,jz0);
211 dx20 = _mm_sub_ps(ix2,jx0);
212 dy20 = _mm_sub_ps(iy2,jy0);
213 dz20 = _mm_sub_ps(iz2,jz0);
214
215 /* Calculate squared distance and things based on it */
216 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
217 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
218 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
219
220 rinv00 = gmx_mm_invsqrt_ps(rsq00);
221 rinv10 = gmx_mm_invsqrt_ps(rsq10);
222 rinv20 = gmx_mm_invsqrt_ps(rsq20);
223
224 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
225
226 /* Load parameters for j particles */
227 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
228 charge+jnrC+0,charge+jnrD+0);
229 vdwjidx0A = 2*vdwtype[jnrA+0];
230 vdwjidx0B = 2*vdwtype[jnrB+0];
231 vdwjidx0C = 2*vdwtype[jnrC+0];
232 vdwjidx0D = 2*vdwtype[jnrD+0];
233
234 fjx0 = _mm_setzero_ps();
235 fjy0 = _mm_setzero_ps();
236 fjz0 = _mm_setzero_ps();
237
238 /**************************
239 * CALCULATE INTERACTIONS *
240 **************************/
241
242 r00 = _mm_mul_ps(rsq00,rinv00);
243
244 /* Compute parameters for interactions between i and j atoms */
245 qq00 = _mm_mul_ps(iq0,jq0);
246 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
247 vdwparam+vdwioffset0+vdwjidx0B,
248 vdwparam+vdwioffset0+vdwjidx0C,
249 vdwparam+vdwioffset0+vdwjidx0D,
250 &c6_00,&c12_00);
251
252 /* Calculate table index by multiplying r with table scale and truncate to integer */
253 rt = _mm_mul_ps(r00,vftabscale);
254 vfitab = _mm_cvttps_epi32(rt);
255 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
256 vfitab = _mm_slli_epi32(vfitab,2);
257
258 /* CUBIC SPLINE TABLE ELECTROSTATICS */
259 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
260 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
261 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
262 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
263 _MM_TRANSPOSE4_PS(Y,F,G,H);
264 Heps = _mm_mul_ps(vfeps,H);
265 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
266 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
267 velec = _mm_mul_ps(qq00,VV);
268 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
269 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
270
271 /* LENNARD-JONES DISPERSION/REPULSION */
272
273 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
274 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
275 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
276 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
277 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
278
279 /* Update potential sum for this i atom from the interaction with this j atom. */
280 velecsum = _mm_add_ps(velecsum,velec);
281 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
282
283 fscal = _mm_add_ps(felec,fvdw);
284
285 /* Calculate temporary vectorial force */
286 tx = _mm_mul_ps(fscal,dx00);
287 ty = _mm_mul_ps(fscal,dy00);
288 tz = _mm_mul_ps(fscal,dz00);
289
290 /* Update vectorial force */
291 fix0 = _mm_add_ps(fix0,tx);
292 fiy0 = _mm_add_ps(fiy0,ty);
293 fiz0 = _mm_add_ps(fiz0,tz);
294
295 fjx0 = _mm_add_ps(fjx0,tx);
296 fjy0 = _mm_add_ps(fjy0,ty);
297 fjz0 = _mm_add_ps(fjz0,tz);
298
299 /**************************
300 * CALCULATE INTERACTIONS *
301 **************************/
302
303 r10 = _mm_mul_ps(rsq10,rinv10);
304
305 /* Compute parameters for interactions between i and j atoms */
306 qq10 = _mm_mul_ps(iq1,jq0);
307
308 /* Calculate table index by multiplying r with table scale and truncate to integer */
309 rt = _mm_mul_ps(r10,vftabscale);
310 vfitab = _mm_cvttps_epi32(rt);
311 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
312 vfitab = _mm_slli_epi32(vfitab,2);
313
314 /* CUBIC SPLINE TABLE ELECTROSTATICS */
315 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
316 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
317 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
318 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
319 _MM_TRANSPOSE4_PS(Y,F,G,H);
320 Heps = _mm_mul_ps(vfeps,H);
321 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
322 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
323 velec = _mm_mul_ps(qq10,VV);
324 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
325 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
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 r20 = _mm_mul_ps(rsq20,rinv20);
351
352 /* Compute parameters for interactions between i and j atoms */
353 qq20 = _mm_mul_ps(iq2,jq0);
354
355 /* Calculate table index by multiplying r with table scale and truncate to integer */
356 rt = _mm_mul_ps(r20,vftabscale);
357 vfitab = _mm_cvttps_epi32(rt);
358 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
359 vfitab = _mm_slli_epi32(vfitab,2);
360
361 /* CUBIC SPLINE TABLE ELECTROSTATICS */
362 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
363 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
364 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
365 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
366 _MM_TRANSPOSE4_PS(Y,F,G,H);
367 Heps = _mm_mul_ps(vfeps,H);
368 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
369 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
370 velec = _mm_mul_ps(qq20,VV);
371 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
372 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
373
374 /* Update potential sum for this i atom from the interaction with this j atom. */
375 velecsum = _mm_add_ps(velecsum,velec);
376
377 fscal = felec;
378
379 /* Calculate temporary vectorial force */
380 tx = _mm_mul_ps(fscal,dx20);
381 ty = _mm_mul_ps(fscal,dy20);
382 tz = _mm_mul_ps(fscal,dz20);
383
384 /* Update vectorial force */
385 fix2 = _mm_add_ps(fix2,tx);
386 fiy2 = _mm_add_ps(fiy2,ty);
387 fiz2 = _mm_add_ps(fiz2,tz);
388
389 fjx0 = _mm_add_ps(fjx0,tx);
390 fjy0 = _mm_add_ps(fjy0,ty);
391 fjz0 = _mm_add_ps(fjz0,tz);
392
393 fjptrA = f+j_coord_offsetA;
394 fjptrB = f+j_coord_offsetB;
395 fjptrC = f+j_coord_offsetC;
396 fjptrD = f+j_coord_offsetD;
397
398 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
399
400 /* Inner loop uses 142 flops */
401 }
402
403 if(jidx<j_index_end)
404 {
405
406 /* Get j neighbor index, and coordinate index */
407 jnrlistA = jjnr[jidx];
408 jnrlistB = jjnr[jidx+1];
409 jnrlistC = jjnr[jidx+2];
410 jnrlistD = jjnr[jidx+3];
411 /* Sign of each element will be negative for non-real atoms.
412 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
413 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
414 */
415 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
416 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
417 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
418 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
419 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
420 j_coord_offsetA = DIM*jnrA;
421 j_coord_offsetB = DIM*jnrB;
422 j_coord_offsetC = DIM*jnrC;
423 j_coord_offsetD = DIM*jnrD;
424
425 /* load j atom coordinates */
426 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
427 x+j_coord_offsetC,x+j_coord_offsetD,
428 &jx0,&jy0,&jz0);
429
430 /* Calculate displacement vector */
431 dx00 = _mm_sub_ps(ix0,jx0);
432 dy00 = _mm_sub_ps(iy0,jy0);
433 dz00 = _mm_sub_ps(iz0,jz0);
434 dx10 = _mm_sub_ps(ix1,jx0);
435 dy10 = _mm_sub_ps(iy1,jy0);
436 dz10 = _mm_sub_ps(iz1,jz0);
437 dx20 = _mm_sub_ps(ix2,jx0);
438 dy20 = _mm_sub_ps(iy2,jy0);
439 dz20 = _mm_sub_ps(iz2,jz0);
440
441 /* Calculate squared distance and things based on it */
442 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
443 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
444 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
445
446 rinv00 = gmx_mm_invsqrt_ps(rsq00);
447 rinv10 = gmx_mm_invsqrt_ps(rsq10);
448 rinv20 = gmx_mm_invsqrt_ps(rsq20);
449
450 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
451
452 /* Load parameters for j particles */
453 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
454 charge+jnrC+0,charge+jnrD+0);
455 vdwjidx0A = 2*vdwtype[jnrA+0];
456 vdwjidx0B = 2*vdwtype[jnrB+0];
457 vdwjidx0C = 2*vdwtype[jnrC+0];
458 vdwjidx0D = 2*vdwtype[jnrD+0];
459
460 fjx0 = _mm_setzero_ps();
461 fjy0 = _mm_setzero_ps();
462 fjz0 = _mm_setzero_ps();
463
464 /**************************
465 * CALCULATE INTERACTIONS *
466 **************************/
467
468 r00 = _mm_mul_ps(rsq00,rinv00);
469 r00 = _mm_andnot_ps(dummy_mask,r00);
470
471 /* Compute parameters for interactions between i and j atoms */
472 qq00 = _mm_mul_ps(iq0,jq0);
473 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
474 vdwparam+vdwioffset0+vdwjidx0B,
475 vdwparam+vdwioffset0+vdwjidx0C,
476 vdwparam+vdwioffset0+vdwjidx0D,
477 &c6_00,&c12_00);
478
479 /* Calculate table index by multiplying r with table scale and truncate to integer */
480 rt = _mm_mul_ps(r00,vftabscale);
481 vfitab = _mm_cvttps_epi32(rt);
482 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
483 vfitab = _mm_slli_epi32(vfitab,2);
484
485 /* CUBIC SPLINE TABLE ELECTROSTATICS */
486 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
487 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
488 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
489 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
490 _MM_TRANSPOSE4_PS(Y,F,G,H);
491 Heps = _mm_mul_ps(vfeps,H);
492 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
493 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
494 velec = _mm_mul_ps(qq00,VV);
495 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
496 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
497
498 /* LENNARD-JONES DISPERSION/REPULSION */
499
500 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
501 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
502 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
503 vvdw = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
504 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
505
506 /* Update potential sum for this i atom from the interaction with this j atom. */
507 velec = _mm_andnot_ps(dummy_mask,velec);
508 velecsum = _mm_add_ps(velecsum,velec);
509 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
510 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
511
512 fscal = _mm_add_ps(felec,fvdw);
513
514 fscal = _mm_andnot_ps(dummy_mask,fscal);
515
516 /* Calculate temporary vectorial force */
517 tx = _mm_mul_ps(fscal,dx00);
518 ty = _mm_mul_ps(fscal,dy00);
519 tz = _mm_mul_ps(fscal,dz00);
520
521 /* Update vectorial force */
522 fix0 = _mm_add_ps(fix0,tx);
523 fiy0 = _mm_add_ps(fiy0,ty);
524 fiz0 = _mm_add_ps(fiz0,tz);
525
526 fjx0 = _mm_add_ps(fjx0,tx);
527 fjy0 = _mm_add_ps(fjy0,ty);
528 fjz0 = _mm_add_ps(fjz0,tz);
529
530 /**************************
531 * CALCULATE INTERACTIONS *
532 **************************/
533
534 r10 = _mm_mul_ps(rsq10,rinv10);
535 r10 = _mm_andnot_ps(dummy_mask,r10);
536
537 /* Compute parameters for interactions between i and j atoms */
538 qq10 = _mm_mul_ps(iq1,jq0);
539
540 /* Calculate table index by multiplying r with table scale and truncate to integer */
541 rt = _mm_mul_ps(r10,vftabscale);
542 vfitab = _mm_cvttps_epi32(rt);
543 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
544 vfitab = _mm_slli_epi32(vfitab,2);
545
546 /* CUBIC SPLINE TABLE ELECTROSTATICS */
547 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
548 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
549 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
550 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
551 _MM_TRANSPOSE4_PS(Y,F,G,H);
552 Heps = _mm_mul_ps(vfeps,H);
553 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
554 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
555 velec = _mm_mul_ps(qq10,VV);
556 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
557 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
558
559 /* Update potential sum for this i atom from the interaction with this j atom. */
560 velec = _mm_andnot_ps(dummy_mask,velec);
561 velecsum = _mm_add_ps(velecsum,velec);
562
563 fscal = felec;
564
565 fscal = _mm_andnot_ps(dummy_mask,fscal);
566
567 /* Calculate temporary vectorial force */
568 tx = _mm_mul_ps(fscal,dx10);
569 ty = _mm_mul_ps(fscal,dy10);
570 tz = _mm_mul_ps(fscal,dz10);
571
572 /* Update vectorial force */
573 fix1 = _mm_add_ps(fix1,tx);
574 fiy1 = _mm_add_ps(fiy1,ty);
575 fiz1 = _mm_add_ps(fiz1,tz);
576
577 fjx0 = _mm_add_ps(fjx0,tx);
578 fjy0 = _mm_add_ps(fjy0,ty);
579 fjz0 = _mm_add_ps(fjz0,tz);
580
581 /**************************
582 * CALCULATE INTERACTIONS *
583 **************************/
584
585 r20 = _mm_mul_ps(rsq20,rinv20);
586 r20 = _mm_andnot_ps(dummy_mask,r20);
587
588 /* Compute parameters for interactions between i and j atoms */
589 qq20 = _mm_mul_ps(iq2,jq0);
590
591 /* Calculate table index by multiplying r with table scale and truncate to integer */
592 rt = _mm_mul_ps(r20,vftabscale);
593 vfitab = _mm_cvttps_epi32(rt);
594 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
595 vfitab = _mm_slli_epi32(vfitab,2);
596
597 /* CUBIC SPLINE TABLE ELECTROSTATICS */
598 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
599 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
600 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
601 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
602 _MM_TRANSPOSE4_PS(Y,F,G,H);
603 Heps = _mm_mul_ps(vfeps,H);
604 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
605 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
606 velec = _mm_mul_ps(qq20,VV);
607 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
608 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
609
610 /* Update potential sum for this i atom from the interaction with this j atom. */
611 velec = _mm_andnot_ps(dummy_mask,velec);
612 velecsum = _mm_add_ps(velecsum,velec);
613
614 fscal = felec;
615
616 fscal = _mm_andnot_ps(dummy_mask,fscal);
617
618 /* Calculate temporary vectorial force */
619 tx = _mm_mul_ps(fscal,dx20);
620 ty = _mm_mul_ps(fscal,dy20);
621 tz = _mm_mul_ps(fscal,dz20);
622
623 /* Update vectorial force */
624 fix2 = _mm_add_ps(fix2,tx);
625 fiy2 = _mm_add_ps(fiy2,ty);
626 fiz2 = _mm_add_ps(fiz2,tz);
627
628 fjx0 = _mm_add_ps(fjx0,tx);
629 fjy0 = _mm_add_ps(fjy0,ty);
630 fjz0 = _mm_add_ps(fjz0,tz);
631
632 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
633 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
634 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
635 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
636
637 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
638
639 /* Inner loop uses 145 flops */
640 }
641
642 /* End of innermost loop */
643
644 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
645 f+i_coord_offset,fshift+i_shift_offset);
646
647 ggid = gid[iidx];
648 /* Update potential energies */
649 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
650 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
651
652 /* Increment number of inner iterations */
653 inneriter += j_index_end - j_index_start;
654
655 /* Outer loop uses 20 flops */
656 }
657
658 /* Increment number of outer iterations */
659 outeriter += nri;
660
661 /* Update outer/inner flops */
662
663 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
664 }
665 /*
666 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_single
667 * Electrostatics interaction: CubicSplineTable
668 * VdW interaction: LennardJones
669 * Geometry: Water3-Particle
670 * Calculate force/pot: Force
671 */
672 void
673 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_single
674 (t_nblist * gmx_restrict nlist,
675 rvec * gmx_restrict xx,
676 rvec * gmx_restrict ff,
677 t_forcerec * gmx_restrict fr,
678 t_mdatoms * gmx_restrict mdatoms,
679 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
680 t_nrnb * gmx_restrict nrnb)
681 {
682 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
683 * just 0 for non-waters.
684 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
685 * jnr indices corresponding to data put in the four positions in the SIMD register.
686 */
687 int i_shift_offset,i_coord_offset,outeriter,inneriter;
688 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
689 int jnrA,jnrB,jnrC,jnrD;
690 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
691 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
692 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
693 real rcutoff_scalar;
694 real *shiftvec,*fshift,*x,*f;
695 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
696 real scratch[4*DIM];
697 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
698 int vdwioffset0;
699 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
700 int vdwioffset1;
701 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
702 int vdwioffset2;
703 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
704 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
705 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
706 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
707 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
708 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
709 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
710 real *charge;
711 int nvdwtype;
712 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
713 int *vdwtype;
714 real *vdwparam;
715 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
716 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
717 __m128i vfitab;
718 __m128i ifour = _mm_set1_epi32(4);
719 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
720 real *vftab;
721 __m128 dummy_mask,cutoff_mask;
722 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
723 __m128 one = _mm_set1_ps(1.0);
724 __m128 two = _mm_set1_ps(2.0);
725 x = xx[0];
726 f = ff[0];
727
728 nri = nlist->nri;
729 iinr = nlist->iinr;
730 jindex = nlist->jindex;
731 jjnr = nlist->jjnr;
732 shiftidx = nlist->shift;
733 gid = nlist->gid;
734 shiftvec = fr->shift_vec[0];
735 fshift = fr->fshift[0];
736 facel = _mm_set1_ps(fr->epsfac);
737 charge = mdatoms->chargeA;
738 nvdwtype = fr->ntype;
739 vdwparam = fr->nbfp;
740 vdwtype = mdatoms->typeA;
741
742 vftab = kernel_data->table_elec->data;
743 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
744
745 /* Setup water-specific parameters */
746 inr = nlist->iinr[0];
747 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
748 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
749 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
750 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
751
752 /* Avoid stupid compiler warnings */
753 jnrA = jnrB = jnrC = jnrD = 0;
754 j_coord_offsetA = 0;
755 j_coord_offsetB = 0;
756 j_coord_offsetC = 0;
757 j_coord_offsetD = 0;
758
759 outeriter = 0;
760 inneriter = 0;
761
762 for(iidx=0;iidx<4*DIM;iidx++)
763 {
764 scratch[iidx] = 0.0;
765 }
766
767 /* Start outer loop over neighborlists */
768 for(iidx=0; iidx<nri; iidx++)
769 {
770 /* Load shift vector for this list */
771 i_shift_offset = DIM*shiftidx[iidx];
772
773 /* Load limits for loop over neighbors */
774 j_index_start = jindex[iidx];
775 j_index_end = jindex[iidx+1];
776
777 /* Get outer coordinate index */
778 inr = iinr[iidx];
779 i_coord_offset = DIM*inr;
780
781 /* Load i particle coords and add shift vector */
782 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
783 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
784
785 fix0 = _mm_setzero_ps();
786 fiy0 = _mm_setzero_ps();
787 fiz0 = _mm_setzero_ps();
788 fix1 = _mm_setzero_ps();
789 fiy1 = _mm_setzero_ps();
790 fiz1 = _mm_setzero_ps();
791 fix2 = _mm_setzero_ps();
792 fiy2 = _mm_setzero_ps();
793 fiz2 = _mm_setzero_ps();
794
795 /* Start inner kernel loop */
796 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
797 {
798
799 /* Get j neighbor index, and coordinate index */
800 jnrA = jjnr[jidx];
801 jnrB = jjnr[jidx+1];
802 jnrC = jjnr[jidx+2];
803 jnrD = jjnr[jidx+3];
804 j_coord_offsetA = DIM*jnrA;
805 j_coord_offsetB = DIM*jnrB;
806 j_coord_offsetC = DIM*jnrC;
807 j_coord_offsetD = DIM*jnrD;
808
809 /* load j atom coordinates */
810 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
811 x+j_coord_offsetC,x+j_coord_offsetD,
812 &jx0,&jy0,&jz0);
813
814 /* Calculate displacement vector */
815 dx00 = _mm_sub_ps(ix0,jx0);
816 dy00 = _mm_sub_ps(iy0,jy0);
817 dz00 = _mm_sub_ps(iz0,jz0);
818 dx10 = _mm_sub_ps(ix1,jx0);
819 dy10 = _mm_sub_ps(iy1,jy0);
820 dz10 = _mm_sub_ps(iz1,jz0);
821 dx20 = _mm_sub_ps(ix2,jx0);
822 dy20 = _mm_sub_ps(iy2,jy0);
823 dz20 = _mm_sub_ps(iz2,jz0);
824
825 /* Calculate squared distance and things based on it */
826 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
827 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
828 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
829
830 rinv00 = gmx_mm_invsqrt_ps(rsq00);
831 rinv10 = gmx_mm_invsqrt_ps(rsq10);
832 rinv20 = gmx_mm_invsqrt_ps(rsq20);
833
834 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
835
836 /* Load parameters for j particles */
837 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
838 charge+jnrC+0,charge+jnrD+0);
839 vdwjidx0A = 2*vdwtype[jnrA+0];
840 vdwjidx0B = 2*vdwtype[jnrB+0];
841 vdwjidx0C = 2*vdwtype[jnrC+0];
842 vdwjidx0D = 2*vdwtype[jnrD+0];
843
844 fjx0 = _mm_setzero_ps();
845 fjy0 = _mm_setzero_ps();
846 fjz0 = _mm_setzero_ps();
847
848 /**************************
849 * CALCULATE INTERACTIONS *
850 **************************/
851
852 r00 = _mm_mul_ps(rsq00,rinv00);
853
854 /* Compute parameters for interactions between i and j atoms */
855 qq00 = _mm_mul_ps(iq0,jq0);
856 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
857 vdwparam+vdwioffset0+vdwjidx0B,
858 vdwparam+vdwioffset0+vdwjidx0C,
859 vdwparam+vdwioffset0+vdwjidx0D,
860 &c6_00,&c12_00);
861
862 /* Calculate table index by multiplying r with table scale and truncate to integer */
863 rt = _mm_mul_ps(r00,vftabscale);
864 vfitab = _mm_cvttps_epi32(rt);
865 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
866 vfitab = _mm_slli_epi32(vfitab,2);
867
868 /* CUBIC SPLINE TABLE ELECTROSTATICS */
869 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
870 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
871 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
872 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
873 _MM_TRANSPOSE4_PS(Y,F,G,H);
874 Heps = _mm_mul_ps(vfeps,H);
875 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
876 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
877 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
878
879 /* LENNARD-JONES DISPERSION/REPULSION */
880
881 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
882 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
883
884 fscal = _mm_add_ps(felec,fvdw);
885
886 /* Calculate temporary vectorial force */
887 tx = _mm_mul_ps(fscal,dx00);
888 ty = _mm_mul_ps(fscal,dy00);
889 tz = _mm_mul_ps(fscal,dz00);
890
891 /* Update vectorial force */
892 fix0 = _mm_add_ps(fix0,tx);
893 fiy0 = _mm_add_ps(fiy0,ty);
894 fiz0 = _mm_add_ps(fiz0,tz);
895
896 fjx0 = _mm_add_ps(fjx0,tx);
897 fjy0 = _mm_add_ps(fjy0,ty);
898 fjz0 = _mm_add_ps(fjz0,tz);
899
900 /**************************
901 * CALCULATE INTERACTIONS *
902 **************************/
903
904 r10 = _mm_mul_ps(rsq10,rinv10);
905
906 /* Compute parameters for interactions between i and j atoms */
907 qq10 = _mm_mul_ps(iq1,jq0);
908
909 /* Calculate table index by multiplying r with table scale and truncate to integer */
910 rt = _mm_mul_ps(r10,vftabscale);
911 vfitab = _mm_cvttps_epi32(rt);
912 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
913 vfitab = _mm_slli_epi32(vfitab,2);
914
915 /* CUBIC SPLINE TABLE ELECTROSTATICS */
916 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
917 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
918 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
919 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
920 _MM_TRANSPOSE4_PS(Y,F,G,H);
921 Heps = _mm_mul_ps(vfeps,H);
922 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
923 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
924 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
925
926 fscal = felec;
927
928 /* Calculate temporary vectorial force */
929 tx = _mm_mul_ps(fscal,dx10);
930 ty = _mm_mul_ps(fscal,dy10);
931 tz = _mm_mul_ps(fscal,dz10);
932
933 /* Update vectorial force */
934 fix1 = _mm_add_ps(fix1,tx);
935 fiy1 = _mm_add_ps(fiy1,ty);
936 fiz1 = _mm_add_ps(fiz1,tz);
937
938 fjx0 = _mm_add_ps(fjx0,tx);
939 fjy0 = _mm_add_ps(fjy0,ty);
940 fjz0 = _mm_add_ps(fjz0,tz);
941
942 /**************************
943 * CALCULATE INTERACTIONS *
944 **************************/
945
946 r20 = _mm_mul_ps(rsq20,rinv20);
947
948 /* Compute parameters for interactions between i and j atoms */
949 qq20 = _mm_mul_ps(iq2,jq0);
950
951 /* Calculate table index by multiplying r with table scale and truncate to integer */
952 rt = _mm_mul_ps(r20,vftabscale);
953 vfitab = _mm_cvttps_epi32(rt);
954 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
955 vfitab = _mm_slli_epi32(vfitab,2);
956
957 /* CUBIC SPLINE TABLE ELECTROSTATICS */
958 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
959 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
960 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
961 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
962 _MM_TRANSPOSE4_PS(Y,F,G,H);
963 Heps = _mm_mul_ps(vfeps,H);
964 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
965 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
966 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
967
968 fscal = felec;
969
970 /* Calculate temporary vectorial force */
971 tx = _mm_mul_ps(fscal,dx20);
972 ty = _mm_mul_ps(fscal,dy20);
973 tz = _mm_mul_ps(fscal,dz20);
974
975 /* Update vectorial force */
976 fix2 = _mm_add_ps(fix2,tx);
977 fiy2 = _mm_add_ps(fiy2,ty);
978 fiz2 = _mm_add_ps(fiz2,tz);
979
980 fjx0 = _mm_add_ps(fjx0,tx);
981 fjy0 = _mm_add_ps(fjy0,ty);
982 fjz0 = _mm_add_ps(fjz0,tz);
983
984 fjptrA = f+j_coord_offsetA;
985 fjptrB = f+j_coord_offsetB;
986 fjptrC = f+j_coord_offsetC;
987 fjptrD = f+j_coord_offsetD;
988
989 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
990
991 /* Inner loop uses 125 flops */
992 }
993
994 if(jidx<j_index_end)
995 {
996
997 /* Get j neighbor index, and coordinate index */
998 jnrlistA = jjnr[jidx];
999 jnrlistB = jjnr[jidx+1];
1000 jnrlistC = jjnr[jidx+2];
1001 jnrlistD = jjnr[jidx+3];
1002 /* Sign of each element will be negative for non-real atoms.
1003 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1004 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1005 */
1006 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1007 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1008 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1009 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1010 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1011 j_coord_offsetA = DIM*jnrA;
1012 j_coord_offsetB = DIM*jnrB;
1013 j_coord_offsetC = DIM*jnrC;
1014 j_coord_offsetD = DIM*jnrD;
1015
1016 /* load j atom coordinates */
1017 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1018 x+j_coord_offsetC,x+j_coord_offsetD,
1019 &jx0,&jy0,&jz0);
1020
1021 /* Calculate displacement vector */
1022 dx00 = _mm_sub_ps(ix0,jx0);
1023 dy00 = _mm_sub_ps(iy0,jy0);
1024 dz00 = _mm_sub_ps(iz0,jz0);
1025 dx10 = _mm_sub_ps(ix1,jx0);
1026 dy10 = _mm_sub_ps(iy1,jy0);
1027 dz10 = _mm_sub_ps(iz1,jz0);
1028 dx20 = _mm_sub_ps(ix2,jx0);
1029 dy20 = _mm_sub_ps(iy2,jy0);
1030 dz20 = _mm_sub_ps(iz2,jz0);
1031
1032 /* Calculate squared distance and things based on it */
1033 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1034 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1035 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1036
1037 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1038 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1039 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1040
1041 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1042
1043 /* Load parameters for j particles */
1044 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1045 charge+jnrC+0,charge+jnrD+0);
1046 vdwjidx0A = 2*vdwtype[jnrA+0];
1047 vdwjidx0B = 2*vdwtype[jnrB+0];
1048 vdwjidx0C = 2*vdwtype[jnrC+0];
1049 vdwjidx0D = 2*vdwtype[jnrD+0];
1050
1051 fjx0 = _mm_setzero_ps();
1052 fjy0 = _mm_setzero_ps();
1053 fjz0 = _mm_setzero_ps();
1054
1055 /**************************
1056 * CALCULATE INTERACTIONS *
1057 **************************/
1058
1059 r00 = _mm_mul_ps(rsq00,rinv00);
1060 r00 = _mm_andnot_ps(dummy_mask,r00);
1061
1062 /* Compute parameters for interactions between i and j atoms */
1063 qq00 = _mm_mul_ps(iq0,jq0);
1064 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1065 vdwparam+vdwioffset0+vdwjidx0B,
1066 vdwparam+vdwioffset0+vdwjidx0C,
1067 vdwparam+vdwioffset0+vdwjidx0D,
1068 &c6_00,&c12_00);
1069
1070 /* Calculate table index by multiplying r with table scale and truncate to integer */
1071 rt = _mm_mul_ps(r00,vftabscale);
1072 vfitab = _mm_cvttps_epi32(rt);
1073 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1074 vfitab = _mm_slli_epi32(vfitab,2);
1075
1076 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1077 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1078 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1079 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1080 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1081 _MM_TRANSPOSE4_PS(Y,F,G,H);
1082 Heps = _mm_mul_ps(vfeps,H);
1083 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1084 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1085 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1086
1087 /* LENNARD-JONES DISPERSION/REPULSION */
1088
1089 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1090 fvdw = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1091
1092 fscal = _mm_add_ps(felec,fvdw);
1093
1094 fscal = _mm_andnot_ps(dummy_mask,fscal);
1095
1096 /* Calculate temporary vectorial force */
1097 tx = _mm_mul_ps(fscal,dx00);
1098 ty = _mm_mul_ps(fscal,dy00);
1099 tz = _mm_mul_ps(fscal,dz00);
1100
1101 /* Update vectorial force */
1102 fix0 = _mm_add_ps(fix0,tx);
1103 fiy0 = _mm_add_ps(fiy0,ty);
1104 fiz0 = _mm_add_ps(fiz0,tz);
1105
1106 fjx0 = _mm_add_ps(fjx0,tx);
1107 fjy0 = _mm_add_ps(fjy0,ty);
1108 fjz0 = _mm_add_ps(fjz0,tz);
1109
1110 /**************************
1111 * CALCULATE INTERACTIONS *
1112 **************************/
1113
1114 r10 = _mm_mul_ps(rsq10,rinv10);
1115 r10 = _mm_andnot_ps(dummy_mask,r10);
1116
1117 /* Compute parameters for interactions between i and j atoms */
1118 qq10 = _mm_mul_ps(iq1,jq0);
1119
1120 /* Calculate table index by multiplying r with table scale and truncate to integer */
1121 rt = _mm_mul_ps(r10,vftabscale);
1122 vfitab = _mm_cvttps_epi32(rt);
1123 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1124 vfitab = _mm_slli_epi32(vfitab,2);
1125
1126 /* CUBIC SPLINE TABLE ELECTROSTATICS */
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 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1136
1137 fscal = felec;
1138
1139 fscal = _mm_andnot_ps(dummy_mask,fscal);
1140
1141 /* Calculate temporary vectorial force */
1142 tx = _mm_mul_ps(fscal,dx10);
1143 ty = _mm_mul_ps(fscal,dy10);
1144 tz = _mm_mul_ps(fscal,dz10);
1145
1146 /* Update vectorial force */
1147 fix1 = _mm_add_ps(fix1,tx);
1148 fiy1 = _mm_add_ps(fiy1,ty);
1149 fiz1 = _mm_add_ps(fiz1,tz);
1150
1151 fjx0 = _mm_add_ps(fjx0,tx);
1152 fjy0 = _mm_add_ps(fjy0,ty);
1153 fjz0 = _mm_add_ps(fjz0,tz);
1154
1155 /**************************
1156 * CALCULATE INTERACTIONS *
1157 **************************/
1158
1159 r20 = _mm_mul_ps(rsq20,rinv20);
1160 r20 = _mm_andnot_ps(dummy_mask,r20);
1161
1162 /* Compute parameters for interactions between i and j atoms */
1163 qq20 = _mm_mul_ps(iq2,jq0);
1164
1165 /* Calculate table index by multiplying r with table scale and truncate to integer */
1166 rt = _mm_mul_ps(r20,vftabscale);
1167 vfitab = _mm_cvttps_epi32(rt);
1168 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1169 vfitab = _mm_slli_epi32(vfitab,2);
1170
1171 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1172 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1173 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1174 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1175 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1176 _MM_TRANSPOSE4_PS(Y,F,G,H);
1177 Heps = _mm_mul_ps(vfeps,H);
1178 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1179 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1180 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1181
1182 fscal = felec;
1183
1184 fscal = _mm_andnot_ps(dummy_mask,fscal);
1185
1186 /* Calculate temporary vectorial force */
1187 tx = _mm_mul_ps(fscal,dx20);
1188 ty = _mm_mul_ps(fscal,dy20);
1189 tz = _mm_mul_ps(fscal,dz20);
1190
1191 /* Update vectorial force */
1192 fix2 = _mm_add_ps(fix2,tx);
1193 fiy2 = _mm_add_ps(fiy2,ty);
1194 fiz2 = _mm_add_ps(fiz2,tz);
1195
1196 fjx0 = _mm_add_ps(fjx0,tx);
1197 fjy0 = _mm_add_ps(fjy0,ty);
1198 fjz0 = _mm_add_ps(fjz0,tz);
1199
1200 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1201 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1202 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1203 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1204
1205 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1206
1207 /* Inner loop uses 128 flops */
1208 }
1209
1210 /* End of innermost loop */
1211
1212 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1213 f+i_coord_offset,fshift+i_shift_offset);
1214
1215 /* Increment number of inner iterations */
1216 inneriter += j_index_end - j_index_start;
1217
1218 /* Outer loop uses 18 flops */
1219 }
1220
1221 /* Increment number of outer iterations */
1222 outeriter += nri;
1223
1224 /* Update outer/inner flops */
1225
1226 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*128);
1227 }
The ultimate molecular dynamics simulation package