?  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_ElecCSTab_VdwNone_GeomW4P1_sse4_1_single.c
blobdaf1f40fd251a938cbb497a3460dbe9354d2730d
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_ElecCSTab_VdwNone_GeomW4P1_VF_sse4_1_single
53 * Electrostatics interaction: CubicSplineTable
54 * VdW interaction: None
55 * Geometry: Water4-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_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 vdwioffset1;
85 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
86 int vdwioffset2;
87 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 int vdwioffset3;
89 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
90 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
91 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
93 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
94 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
95 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
96 real *charge;
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
119 vftab = kernel_data->table_elec->data;
120 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
121
122 /* Setup water-specific parameters */
123 inr = nlist->iinr[0];
124 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
125 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
126 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
127
128 /* Avoid stupid compiler warnings */
129 jnrA = jnrB = jnrC = jnrD = 0;
130 j_coord_offsetA = 0;
131 j_coord_offsetB = 0;
132 j_coord_offsetC = 0;
133 j_coord_offsetD = 0;
134
135 outeriter = 0;
136 inneriter = 0;
137
138 for(iidx=0;iidx<4*DIM;iidx++)
139 {
140 scratch[iidx] = 0.0;
141 }
142
143 /* Start outer loop over neighborlists */
144 for(iidx=0; iidx<nri; iidx++)
145 {
146 /* Load shift vector for this list */
147 i_shift_offset = DIM*shiftidx[iidx];
148
149 /* Load limits for loop over neighbors */
150 j_index_start = jindex[iidx];
151 j_index_end = jindex[iidx+1];
152
153 /* Get outer coordinate index */
154 inr = iinr[iidx];
155 i_coord_offset = DIM*inr;
156
157 /* Load i particle coords and add shift vector */
158 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
159 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
160
161 fix1 = _mm_setzero_ps();
162 fiy1 = _mm_setzero_ps();
163 fiz1 = _mm_setzero_ps();
164 fix2 = _mm_setzero_ps();
165 fiy2 = _mm_setzero_ps();
166 fiz2 = _mm_setzero_ps();
167 fix3 = _mm_setzero_ps();
168 fiy3 = _mm_setzero_ps();
169 fiz3 = _mm_setzero_ps();
170
171 /* Reset potential sums */
172 velecsum = _mm_setzero_ps();
173
174 /* Start inner kernel loop */
175 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
176 {
177
178 /* Get j neighbor index, and coordinate index */
179 jnrA = jjnr[jidx];
180 jnrB = jjnr[jidx+1];
181 jnrC = jjnr[jidx+2];
182 jnrD = jjnr[jidx+3];
183 j_coord_offsetA = DIM*jnrA;
184 j_coord_offsetB = DIM*jnrB;
185 j_coord_offsetC = DIM*jnrC;
186 j_coord_offsetD = DIM*jnrD;
187
188 /* load j atom coordinates */
189 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
190 x+j_coord_offsetC,x+j_coord_offsetD,
191 &jx0,&jy0,&jz0);
192
193 /* Calculate displacement vector */
194 dx10 = _mm_sub_ps(ix1,jx0);
195 dy10 = _mm_sub_ps(iy1,jy0);
196 dz10 = _mm_sub_ps(iz1,jz0);
197 dx20 = _mm_sub_ps(ix2,jx0);
198 dy20 = _mm_sub_ps(iy2,jy0);
199 dz20 = _mm_sub_ps(iz2,jz0);
200 dx30 = _mm_sub_ps(ix3,jx0);
201 dy30 = _mm_sub_ps(iy3,jy0);
202 dz30 = _mm_sub_ps(iz3,jz0);
203
204 /* Calculate squared distance and things based on it */
205 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
206 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
207 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
208
209 rinv10 = gmx_mm_invsqrt_ps(rsq10);
210 rinv20 = gmx_mm_invsqrt_ps(rsq20);
211 rinv30 = gmx_mm_invsqrt_ps(rsq30);
212
213 /* Load parameters for j particles */
214 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
215 charge+jnrC+0,charge+jnrD+0);
216
217 fjx0 = _mm_setzero_ps();
218 fjy0 = _mm_setzero_ps();
219 fjz0 = _mm_setzero_ps();
220
221 /**************************
222 * CALCULATE INTERACTIONS *
223 **************************/
224
225 r10 = _mm_mul_ps(rsq10,rinv10);
226
227 /* Compute parameters for interactions between i and j atoms */
228 qq10 = _mm_mul_ps(iq1,jq0);
229
230 /* Calculate table index by multiplying r with table scale and truncate to integer */
231 rt = _mm_mul_ps(r10,vftabscale);
232 vfitab = _mm_cvttps_epi32(rt);
233 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
234 vfitab = _mm_slli_epi32(vfitab,2);
235
236 /* CUBIC SPLINE TABLE ELECTROSTATICS */
237 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
238 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
239 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
240 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
241 _MM_TRANSPOSE4_PS(Y,F,G,H);
242 Heps = _mm_mul_ps(vfeps,H);
243 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
244 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
245 velec = _mm_mul_ps(qq10,VV);
246 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
247 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
248
249 /* Update potential sum for this i atom from the interaction with this j atom. */
250 velecsum = _mm_add_ps(velecsum,velec);
251
252 fscal = felec;
253
254 /* Calculate temporary vectorial force */
255 tx = _mm_mul_ps(fscal,dx10);
256 ty = _mm_mul_ps(fscal,dy10);
257 tz = _mm_mul_ps(fscal,dz10);
258
259 /* Update vectorial force */
260 fix1 = _mm_add_ps(fix1,tx);
261 fiy1 = _mm_add_ps(fiy1,ty);
262 fiz1 = _mm_add_ps(fiz1,tz);
263
264 fjx0 = _mm_add_ps(fjx0,tx);
265 fjy0 = _mm_add_ps(fjy0,ty);
266 fjz0 = _mm_add_ps(fjz0,tz);
267
268 /**************************
269 * CALCULATE INTERACTIONS *
270 **************************/
271
272 r20 = _mm_mul_ps(rsq20,rinv20);
273
274 /* Compute parameters for interactions between i and j atoms */
275 qq20 = _mm_mul_ps(iq2,jq0);
276
277 /* Calculate table index by multiplying r with table scale and truncate to integer */
278 rt = _mm_mul_ps(r20,vftabscale);
279 vfitab = _mm_cvttps_epi32(rt);
280 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
281 vfitab = _mm_slli_epi32(vfitab,2);
282
283 /* CUBIC SPLINE TABLE ELECTROSTATICS */
284 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
285 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
286 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
287 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
288 _MM_TRANSPOSE4_PS(Y,F,G,H);
289 Heps = _mm_mul_ps(vfeps,H);
290 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
291 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
292 velec = _mm_mul_ps(qq20,VV);
293 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
294 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
295
296 /* Update potential sum for this i atom from the interaction with this j atom. */
297 velecsum = _mm_add_ps(velecsum,velec);
298
299 fscal = felec;
300
301 /* Calculate temporary vectorial force */
302 tx = _mm_mul_ps(fscal,dx20);
303 ty = _mm_mul_ps(fscal,dy20);
304 tz = _mm_mul_ps(fscal,dz20);
305
306 /* Update vectorial force */
307 fix2 = _mm_add_ps(fix2,tx);
308 fiy2 = _mm_add_ps(fiy2,ty);
309 fiz2 = _mm_add_ps(fiz2,tz);
310
311 fjx0 = _mm_add_ps(fjx0,tx);
312 fjy0 = _mm_add_ps(fjy0,ty);
313 fjz0 = _mm_add_ps(fjz0,tz);
314
315 /**************************
316 * CALCULATE INTERACTIONS *
317 **************************/
318
319 r30 = _mm_mul_ps(rsq30,rinv30);
320
321 /* Compute parameters for interactions between i and j atoms */
322 qq30 = _mm_mul_ps(iq3,jq0);
323
324 /* Calculate table index by multiplying r with table scale and truncate to integer */
325 rt = _mm_mul_ps(r30,vftabscale);
326 vfitab = _mm_cvttps_epi32(rt);
327 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
328 vfitab = _mm_slli_epi32(vfitab,2);
329
330 /* CUBIC SPLINE TABLE ELECTROSTATICS */
331 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
332 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
333 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
334 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
335 _MM_TRANSPOSE4_PS(Y,F,G,H);
336 Heps = _mm_mul_ps(vfeps,H);
337 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
338 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
339 velec = _mm_mul_ps(qq30,VV);
340 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
341 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
342
343 /* Update potential sum for this i atom from the interaction with this j atom. */
344 velecsum = _mm_add_ps(velecsum,velec);
345
346 fscal = felec;
347
348 /* Calculate temporary vectorial force */
349 tx = _mm_mul_ps(fscal,dx30);
350 ty = _mm_mul_ps(fscal,dy30);
351 tz = _mm_mul_ps(fscal,dz30);
352
353 /* Update vectorial force */
354 fix3 = _mm_add_ps(fix3,tx);
355 fiy3 = _mm_add_ps(fiy3,ty);
356 fiz3 = _mm_add_ps(fiz3,tz);
357
358 fjx0 = _mm_add_ps(fjx0,tx);
359 fjy0 = _mm_add_ps(fjy0,ty);
360 fjz0 = _mm_add_ps(fjz0,tz);
361
362 fjptrA = f+j_coord_offsetA;
363 fjptrB = f+j_coord_offsetB;
364 fjptrC = f+j_coord_offsetC;
365 fjptrD = f+j_coord_offsetD;
366
367 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
368
369 /* Inner loop uses 129 flops */
370 }
371
372 if(jidx<j_index_end)
373 {
374
375 /* Get j neighbor index, and coordinate index */
376 jnrlistA = jjnr[jidx];
377 jnrlistB = jjnr[jidx+1];
378 jnrlistC = jjnr[jidx+2];
379 jnrlistD = jjnr[jidx+3];
380 /* Sign of each element will be negative for non-real atoms.
381 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
382 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
383 */
384 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
385 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
386 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
387 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
388 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
389 j_coord_offsetA = DIM*jnrA;
390 j_coord_offsetB = DIM*jnrB;
391 j_coord_offsetC = DIM*jnrC;
392 j_coord_offsetD = DIM*jnrD;
393
394 /* load j atom coordinates */
395 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
396 x+j_coord_offsetC,x+j_coord_offsetD,
397 &jx0,&jy0,&jz0);
398
399 /* Calculate displacement vector */
400 dx10 = _mm_sub_ps(ix1,jx0);
401 dy10 = _mm_sub_ps(iy1,jy0);
402 dz10 = _mm_sub_ps(iz1,jz0);
403 dx20 = _mm_sub_ps(ix2,jx0);
404 dy20 = _mm_sub_ps(iy2,jy0);
405 dz20 = _mm_sub_ps(iz2,jz0);
406 dx30 = _mm_sub_ps(ix3,jx0);
407 dy30 = _mm_sub_ps(iy3,jy0);
408 dz30 = _mm_sub_ps(iz3,jz0);
409
410 /* Calculate squared distance and things based on it */
411 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
412 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
413 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
414
415 rinv10 = gmx_mm_invsqrt_ps(rsq10);
416 rinv20 = gmx_mm_invsqrt_ps(rsq20);
417 rinv30 = gmx_mm_invsqrt_ps(rsq30);
418
419 /* Load parameters for j particles */
420 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
421 charge+jnrC+0,charge+jnrD+0);
422
423 fjx0 = _mm_setzero_ps();
424 fjy0 = _mm_setzero_ps();
425 fjz0 = _mm_setzero_ps();
426
427 /**************************
428 * CALCULATE INTERACTIONS *
429 **************************/
430
431 r10 = _mm_mul_ps(rsq10,rinv10);
432 r10 = _mm_andnot_ps(dummy_mask,r10);
433
434 /* Compute parameters for interactions between i and j atoms */
435 qq10 = _mm_mul_ps(iq1,jq0);
436
437 /* Calculate table index by multiplying r with table scale and truncate to integer */
438 rt = _mm_mul_ps(r10,vftabscale);
439 vfitab = _mm_cvttps_epi32(rt);
440 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
441 vfitab = _mm_slli_epi32(vfitab,2);
442
443 /* CUBIC SPLINE TABLE ELECTROSTATICS */
444 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
445 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
446 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
447 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
448 _MM_TRANSPOSE4_PS(Y,F,G,H);
449 Heps = _mm_mul_ps(vfeps,H);
450 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
451 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
452 velec = _mm_mul_ps(qq10,VV);
453 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
454 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
455
456 /* Update potential sum for this i atom from the interaction with this j atom. */
457 velec = _mm_andnot_ps(dummy_mask,velec);
458 velecsum = _mm_add_ps(velecsum,velec);
459
460 fscal = felec;
461
462 fscal = _mm_andnot_ps(dummy_mask,fscal);
463
464 /* Calculate temporary vectorial force */
465 tx = _mm_mul_ps(fscal,dx10);
466 ty = _mm_mul_ps(fscal,dy10);
467 tz = _mm_mul_ps(fscal,dz10);
468
469 /* Update vectorial force */
470 fix1 = _mm_add_ps(fix1,tx);
471 fiy1 = _mm_add_ps(fiy1,ty);
472 fiz1 = _mm_add_ps(fiz1,tz);
473
474 fjx0 = _mm_add_ps(fjx0,tx);
475 fjy0 = _mm_add_ps(fjy0,ty);
476 fjz0 = _mm_add_ps(fjz0,tz);
477
478 /**************************
479 * CALCULATE INTERACTIONS *
480 **************************/
481
482 r20 = _mm_mul_ps(rsq20,rinv20);
483 r20 = _mm_andnot_ps(dummy_mask,r20);
484
485 /* Compute parameters for interactions between i and j atoms */
486 qq20 = _mm_mul_ps(iq2,jq0);
487
488 /* Calculate table index by multiplying r with table scale and truncate to integer */
489 rt = _mm_mul_ps(r20,vftabscale);
490 vfitab = _mm_cvttps_epi32(rt);
491 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
492 vfitab = _mm_slli_epi32(vfitab,2);
493
494 /* CUBIC SPLINE TABLE ELECTROSTATICS */
495 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
496 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
497 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
498 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
499 _MM_TRANSPOSE4_PS(Y,F,G,H);
500 Heps = _mm_mul_ps(vfeps,H);
501 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
502 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
503 velec = _mm_mul_ps(qq20,VV);
504 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
505 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
506
507 /* Update potential sum for this i atom from the interaction with this j atom. */
508 velec = _mm_andnot_ps(dummy_mask,velec);
509 velecsum = _mm_add_ps(velecsum,velec);
510
511 fscal = felec;
512
513 fscal = _mm_andnot_ps(dummy_mask,fscal);
514
515 /* Calculate temporary vectorial force */
516 tx = _mm_mul_ps(fscal,dx20);
517 ty = _mm_mul_ps(fscal,dy20);
518 tz = _mm_mul_ps(fscal,dz20);
519
520 /* Update vectorial force */
521 fix2 = _mm_add_ps(fix2,tx);
522 fiy2 = _mm_add_ps(fiy2,ty);
523 fiz2 = _mm_add_ps(fiz2,tz);
524
525 fjx0 = _mm_add_ps(fjx0,tx);
526 fjy0 = _mm_add_ps(fjy0,ty);
527 fjz0 = _mm_add_ps(fjz0,tz);
528
529 /**************************
530 * CALCULATE INTERACTIONS *
531 **************************/
532
533 r30 = _mm_mul_ps(rsq30,rinv30);
534 r30 = _mm_andnot_ps(dummy_mask,r30);
535
536 /* Compute parameters for interactions between i and j atoms */
537 qq30 = _mm_mul_ps(iq3,jq0);
538
539 /* Calculate table index by multiplying r with table scale and truncate to integer */
540 rt = _mm_mul_ps(r30,vftabscale);
541 vfitab = _mm_cvttps_epi32(rt);
542 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
543 vfitab = _mm_slli_epi32(vfitab,2);
544
545 /* CUBIC SPLINE TABLE ELECTROSTATICS */
546 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
547 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
548 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
549 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
550 _MM_TRANSPOSE4_PS(Y,F,G,H);
551 Heps = _mm_mul_ps(vfeps,H);
552 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
553 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
554 velec = _mm_mul_ps(qq30,VV);
555 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
556 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
557
558 /* Update potential sum for this i atom from the interaction with this j atom. */
559 velec = _mm_andnot_ps(dummy_mask,velec);
560 velecsum = _mm_add_ps(velecsum,velec);
561
562 fscal = felec;
563
564 fscal = _mm_andnot_ps(dummy_mask,fscal);
565
566 /* Calculate temporary vectorial force */
567 tx = _mm_mul_ps(fscal,dx30);
568 ty = _mm_mul_ps(fscal,dy30);
569 tz = _mm_mul_ps(fscal,dz30);
570
571 /* Update vectorial force */
572 fix3 = _mm_add_ps(fix3,tx);
573 fiy3 = _mm_add_ps(fiy3,ty);
574 fiz3 = _mm_add_ps(fiz3,tz);
575
576 fjx0 = _mm_add_ps(fjx0,tx);
577 fjy0 = _mm_add_ps(fjy0,ty);
578 fjz0 = _mm_add_ps(fjz0,tz);
579
580 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
581 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
582 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
583 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
584
585 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
586
587 /* Inner loop uses 132 flops */
588 }
589
590 /* End of innermost loop */
591
592 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
593 f+i_coord_offset+DIM,fshift+i_shift_offset);
594
595 ggid = gid[iidx];
596 /* Update potential energies */
597 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
598
599 /* Increment number of inner iterations */
600 inneriter += j_index_end - j_index_start;
601
602 /* Outer loop uses 19 flops */
603 }
604
605 /* Increment number of outer iterations */
606 outeriter += nri;
607
608 /* Update outer/inner flops */
609
610 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*132);
611 }
612 /*
613 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse4_1_single
614 * Electrostatics interaction: CubicSplineTable
615 * VdW interaction: None
616 * Geometry: Water4-Particle
617 * Calculate force/pot: Force
618 */
619 void
620 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_sse4_1_single
621 (t_nblist * gmx_restrict nlist,
622 rvec * gmx_restrict xx,
623 rvec * gmx_restrict ff,
624 t_forcerec * gmx_restrict fr,
625 t_mdatoms * gmx_restrict mdatoms,
626 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
627 t_nrnb * gmx_restrict nrnb)
628 {
629 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
630 * just 0 for non-waters.
631 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
632 * jnr indices corresponding to data put in the four positions in the SIMD register.
633 */
634 int i_shift_offset,i_coord_offset,outeriter,inneriter;
635 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
636 int jnrA,jnrB,jnrC,jnrD;
637 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
638 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
639 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
640 real rcutoff_scalar;
641 real *shiftvec,*fshift,*x,*f;
642 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
643 real scratch[4*DIM];
644 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
645 int vdwioffset1;
646 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
647 int vdwioffset2;
648 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
649 int vdwioffset3;
650 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
651 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
652 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
653 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
654 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
655 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
656 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
657 real *charge;
658 __m128i vfitab;
659 __m128i ifour = _mm_set1_epi32(4);
660 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
661 real *vftab;
662 __m128 dummy_mask,cutoff_mask;
663 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
664 __m128 one = _mm_set1_ps(1.0);
665 __m128 two = _mm_set1_ps(2.0);
666 x = xx[0];
667 f = ff[0];
668
669 nri = nlist->nri;
670 iinr = nlist->iinr;
671 jindex = nlist->jindex;
672 jjnr = nlist->jjnr;
673 shiftidx = nlist->shift;
674 gid = nlist->gid;
675 shiftvec = fr->shift_vec[0];
676 fshift = fr->fshift[0];
677 facel = _mm_set1_ps(fr->epsfac);
678 charge = mdatoms->chargeA;
679
680 vftab = kernel_data->table_elec->data;
681 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
682
683 /* Setup water-specific parameters */
684 inr = nlist->iinr[0];
685 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
686 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
687 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
688
689 /* Avoid stupid compiler warnings */
690 jnrA = jnrB = jnrC = jnrD = 0;
691 j_coord_offsetA = 0;
692 j_coord_offsetB = 0;
693 j_coord_offsetC = 0;
694 j_coord_offsetD = 0;
695
696 outeriter = 0;
697 inneriter = 0;
698
699 for(iidx=0;iidx<4*DIM;iidx++)
700 {
701 scratch[iidx] = 0.0;
702 }
703
704 /* Start outer loop over neighborlists */
705 for(iidx=0; iidx<nri; iidx++)
706 {
707 /* Load shift vector for this list */
708 i_shift_offset = DIM*shiftidx[iidx];
709
710 /* Load limits for loop over neighbors */
711 j_index_start = jindex[iidx];
712 j_index_end = jindex[iidx+1];
713
714 /* Get outer coordinate index */
715 inr = iinr[iidx];
716 i_coord_offset = DIM*inr;
717
718 /* Load i particle coords and add shift vector */
719 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
720 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
721
722 fix1 = _mm_setzero_ps();
723 fiy1 = _mm_setzero_ps();
724 fiz1 = _mm_setzero_ps();
725 fix2 = _mm_setzero_ps();
726 fiy2 = _mm_setzero_ps();
727 fiz2 = _mm_setzero_ps();
728 fix3 = _mm_setzero_ps();
729 fiy3 = _mm_setzero_ps();
730 fiz3 = _mm_setzero_ps();
731
732 /* Start inner kernel loop */
733 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
734 {
735
736 /* Get j neighbor index, and coordinate index */
737 jnrA = jjnr[jidx];
738 jnrB = jjnr[jidx+1];
739 jnrC = jjnr[jidx+2];
740 jnrD = jjnr[jidx+3];
741 j_coord_offsetA = DIM*jnrA;
742 j_coord_offsetB = DIM*jnrB;
743 j_coord_offsetC = DIM*jnrC;
744 j_coord_offsetD = DIM*jnrD;
745
746 /* load j atom coordinates */
747 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
748 x+j_coord_offsetC,x+j_coord_offsetD,
749 &jx0,&jy0,&jz0);
750
751 /* Calculate displacement vector */
752 dx10 = _mm_sub_ps(ix1,jx0);
753 dy10 = _mm_sub_ps(iy1,jy0);
754 dz10 = _mm_sub_ps(iz1,jz0);
755 dx20 = _mm_sub_ps(ix2,jx0);
756 dy20 = _mm_sub_ps(iy2,jy0);
757 dz20 = _mm_sub_ps(iz2,jz0);
758 dx30 = _mm_sub_ps(ix3,jx0);
759 dy30 = _mm_sub_ps(iy3,jy0);
760 dz30 = _mm_sub_ps(iz3,jz0);
761
762 /* Calculate squared distance and things based on it */
763 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
764 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
765 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
766
767 rinv10 = gmx_mm_invsqrt_ps(rsq10);
768 rinv20 = gmx_mm_invsqrt_ps(rsq20);
769 rinv30 = gmx_mm_invsqrt_ps(rsq30);
770
771 /* Load parameters for j particles */
772 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
773 charge+jnrC+0,charge+jnrD+0);
774
775 fjx0 = _mm_setzero_ps();
776 fjy0 = _mm_setzero_ps();
777 fjz0 = _mm_setzero_ps();
778
779 /**************************
780 * CALCULATE INTERACTIONS *
781 **************************/
782
783 r10 = _mm_mul_ps(rsq10,rinv10);
784
785 /* Compute parameters for interactions between i and j atoms */
786 qq10 = _mm_mul_ps(iq1,jq0);
787
788 /* Calculate table index by multiplying r with table scale and truncate to integer */
789 rt = _mm_mul_ps(r10,vftabscale);
790 vfitab = _mm_cvttps_epi32(rt);
791 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
792 vfitab = _mm_slli_epi32(vfitab,2);
793
794 /* CUBIC SPLINE TABLE ELECTROSTATICS */
795 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
796 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
797 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
798 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
799 _MM_TRANSPOSE4_PS(Y,F,G,H);
800 Heps = _mm_mul_ps(vfeps,H);
801 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
802 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
803 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
804
805 fscal = felec;
806
807 /* Calculate temporary vectorial force */
808 tx = _mm_mul_ps(fscal,dx10);
809 ty = _mm_mul_ps(fscal,dy10);
810 tz = _mm_mul_ps(fscal,dz10);
811
812 /* Update vectorial force */
813 fix1 = _mm_add_ps(fix1,tx);
814 fiy1 = _mm_add_ps(fiy1,ty);
815 fiz1 = _mm_add_ps(fiz1,tz);
816
817 fjx0 = _mm_add_ps(fjx0,tx);
818 fjy0 = _mm_add_ps(fjy0,ty);
819 fjz0 = _mm_add_ps(fjz0,tz);
820
821 /**************************
822 * CALCULATE INTERACTIONS *
823 **************************/
824
825 r20 = _mm_mul_ps(rsq20,rinv20);
826
827 /* Compute parameters for interactions between i and j atoms */
828 qq20 = _mm_mul_ps(iq2,jq0);
829
830 /* Calculate table index by multiplying r with table scale and truncate to integer */
831 rt = _mm_mul_ps(r20,vftabscale);
832 vfitab = _mm_cvttps_epi32(rt);
833 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
834 vfitab = _mm_slli_epi32(vfitab,2);
835
836 /* CUBIC SPLINE TABLE ELECTROSTATICS */
837 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
838 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
839 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
840 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
841 _MM_TRANSPOSE4_PS(Y,F,G,H);
842 Heps = _mm_mul_ps(vfeps,H);
843 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
844 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
845 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
846
847 fscal = felec;
848
849 /* Calculate temporary vectorial force */
850 tx = _mm_mul_ps(fscal,dx20);
851 ty = _mm_mul_ps(fscal,dy20);
852 tz = _mm_mul_ps(fscal,dz20);
853
854 /* Update vectorial force */
855 fix2 = _mm_add_ps(fix2,tx);
856 fiy2 = _mm_add_ps(fiy2,ty);
857 fiz2 = _mm_add_ps(fiz2,tz);
858
859 fjx0 = _mm_add_ps(fjx0,tx);
860 fjy0 = _mm_add_ps(fjy0,ty);
861 fjz0 = _mm_add_ps(fjz0,tz);
862
863 /**************************
864 * CALCULATE INTERACTIONS *
865 **************************/
866
867 r30 = _mm_mul_ps(rsq30,rinv30);
868
869 /* Compute parameters for interactions between i and j atoms */
870 qq30 = _mm_mul_ps(iq3,jq0);
871
872 /* Calculate table index by multiplying r with table scale and truncate to integer */
873 rt = _mm_mul_ps(r30,vftabscale);
874 vfitab = _mm_cvttps_epi32(rt);
875 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
876 vfitab = _mm_slli_epi32(vfitab,2);
877
878 /* CUBIC SPLINE TABLE ELECTROSTATICS */
879 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
880 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
881 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
882 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
883 _MM_TRANSPOSE4_PS(Y,F,G,H);
884 Heps = _mm_mul_ps(vfeps,H);
885 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
886 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
887 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
888
889 fscal = felec;
890
891 /* Calculate temporary vectorial force */
892 tx = _mm_mul_ps(fscal,dx30);
893 ty = _mm_mul_ps(fscal,dy30);
894 tz = _mm_mul_ps(fscal,dz30);
895
896 /* Update vectorial force */
897 fix3 = _mm_add_ps(fix3,tx);
898 fiy3 = _mm_add_ps(fiy3,ty);
899 fiz3 = _mm_add_ps(fiz3,tz);
900
901 fjx0 = _mm_add_ps(fjx0,tx);
902 fjy0 = _mm_add_ps(fjy0,ty);
903 fjz0 = _mm_add_ps(fjz0,tz);
904
905 fjptrA = f+j_coord_offsetA;
906 fjptrB = f+j_coord_offsetB;
907 fjptrC = f+j_coord_offsetC;
908 fjptrD = f+j_coord_offsetD;
909
910 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
911
912 /* Inner loop uses 117 flops */
913 }
914
915 if(jidx<j_index_end)
916 {
917
918 /* Get j neighbor index, and coordinate index */
919 jnrlistA = jjnr[jidx];
920 jnrlistB = jjnr[jidx+1];
921 jnrlistC = jjnr[jidx+2];
922 jnrlistD = jjnr[jidx+3];
923 /* Sign of each element will be negative for non-real atoms.
924 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
925 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
926 */
927 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
928 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
929 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
930 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
931 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
932 j_coord_offsetA = DIM*jnrA;
933 j_coord_offsetB = DIM*jnrB;
934 j_coord_offsetC = DIM*jnrC;
935 j_coord_offsetD = DIM*jnrD;
936
937 /* load j atom coordinates */
938 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
939 x+j_coord_offsetC,x+j_coord_offsetD,
940 &jx0,&jy0,&jz0);
941
942 /* Calculate displacement vector */
943 dx10 = _mm_sub_ps(ix1,jx0);
944 dy10 = _mm_sub_ps(iy1,jy0);
945 dz10 = _mm_sub_ps(iz1,jz0);
946 dx20 = _mm_sub_ps(ix2,jx0);
947 dy20 = _mm_sub_ps(iy2,jy0);
948 dz20 = _mm_sub_ps(iz2,jz0);
949 dx30 = _mm_sub_ps(ix3,jx0);
950 dy30 = _mm_sub_ps(iy3,jy0);
951 dz30 = _mm_sub_ps(iz3,jz0);
952
953 /* Calculate squared distance and things based on it */
954 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
955 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
956 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
957
958 rinv10 = gmx_mm_invsqrt_ps(rsq10);
959 rinv20 = gmx_mm_invsqrt_ps(rsq20);
960 rinv30 = gmx_mm_invsqrt_ps(rsq30);
961
962 /* Load parameters for j particles */
963 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
964 charge+jnrC+0,charge+jnrD+0);
965
966 fjx0 = _mm_setzero_ps();
967 fjy0 = _mm_setzero_ps();
968 fjz0 = _mm_setzero_ps();
969
970 /**************************
971 * CALCULATE INTERACTIONS *
972 **************************/
973
974 r10 = _mm_mul_ps(rsq10,rinv10);
975 r10 = _mm_andnot_ps(dummy_mask,r10);
976
977 /* Compute parameters for interactions between i and j atoms */
978 qq10 = _mm_mul_ps(iq1,jq0);
979
980 /* Calculate table index by multiplying r with table scale and truncate to integer */
981 rt = _mm_mul_ps(r10,vftabscale);
982 vfitab = _mm_cvttps_epi32(rt);
983 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
984 vfitab = _mm_slli_epi32(vfitab,2);
985
986 /* CUBIC SPLINE TABLE ELECTROSTATICS */
987 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
988 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
989 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
990 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
991 _MM_TRANSPOSE4_PS(Y,F,G,H);
992 Heps = _mm_mul_ps(vfeps,H);
993 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
994 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
995 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
996
997 fscal = felec;
998
999 fscal = _mm_andnot_ps(dummy_mask,fscal);
1000
1001 /* Calculate temporary vectorial force */
1002 tx = _mm_mul_ps(fscal,dx10);
1003 ty = _mm_mul_ps(fscal,dy10);
1004 tz = _mm_mul_ps(fscal,dz10);
1005
1006 /* Update vectorial force */
1007 fix1 = _mm_add_ps(fix1,tx);
1008 fiy1 = _mm_add_ps(fiy1,ty);
1009 fiz1 = _mm_add_ps(fiz1,tz);
1010
1011 fjx0 = _mm_add_ps(fjx0,tx);
1012 fjy0 = _mm_add_ps(fjy0,ty);
1013 fjz0 = _mm_add_ps(fjz0,tz);
1014
1015 /**************************
1016 * CALCULATE INTERACTIONS *
1017 **************************/
1018
1019 r20 = _mm_mul_ps(rsq20,rinv20);
1020 r20 = _mm_andnot_ps(dummy_mask,r20);
1021
1022 /* Compute parameters for interactions between i and j atoms */
1023 qq20 = _mm_mul_ps(iq2,jq0);
1024
1025 /* Calculate table index by multiplying r with table scale and truncate to integer */
1026 rt = _mm_mul_ps(r20,vftabscale);
1027 vfitab = _mm_cvttps_epi32(rt);
1028 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1029 vfitab = _mm_slli_epi32(vfitab,2);
1030
1031 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1032 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1033 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1034 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1035 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1036 _MM_TRANSPOSE4_PS(Y,F,G,H);
1037 Heps = _mm_mul_ps(vfeps,H);
1038 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1039 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1040 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1041
1042 fscal = felec;
1043
1044 fscal = _mm_andnot_ps(dummy_mask,fscal);
1045
1046 /* Calculate temporary vectorial force */
1047 tx = _mm_mul_ps(fscal,dx20);
1048 ty = _mm_mul_ps(fscal,dy20);
1049 tz = _mm_mul_ps(fscal,dz20);
1050
1051 /* Update vectorial force */
1052 fix2 = _mm_add_ps(fix2,tx);
1053 fiy2 = _mm_add_ps(fiy2,ty);
1054 fiz2 = _mm_add_ps(fiz2,tz);
1055
1056 fjx0 = _mm_add_ps(fjx0,tx);
1057 fjy0 = _mm_add_ps(fjy0,ty);
1058 fjz0 = _mm_add_ps(fjz0,tz);
1059
1060 /**************************
1061 * CALCULATE INTERACTIONS *
1062 **************************/
1063
1064 r30 = _mm_mul_ps(rsq30,rinv30);
1065 r30 = _mm_andnot_ps(dummy_mask,r30);
1066
1067 /* Compute parameters for interactions between i and j atoms */
1068 qq30 = _mm_mul_ps(iq3,jq0);
1069
1070 /* Calculate table index by multiplying r with table scale and truncate to integer */
1071 rt = _mm_mul_ps(r30,vftabscale);
1072 vfitab = _mm_cvttps_epi32(rt);
1073 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
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(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1086
1087 fscal = felec;
1088
1089 fscal = _mm_andnot_ps(dummy_mask,fscal);
1090
1091 /* Calculate temporary vectorial force */
1092 tx = _mm_mul_ps(fscal,dx30);
1093 ty = _mm_mul_ps(fscal,dy30);
1094 tz = _mm_mul_ps(fscal,dz30);
1095
1096 /* Update vectorial force */
1097 fix3 = _mm_add_ps(fix3,tx);
1098 fiy3 = _mm_add_ps(fiy3,ty);
1099 fiz3 = _mm_add_ps(fiz3,tz);
1100
1101 fjx0 = _mm_add_ps(fjx0,tx);
1102 fjy0 = _mm_add_ps(fjy0,ty);
1103 fjz0 = _mm_add_ps(fjz0,tz);
1104
1105 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1106 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1107 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1108 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1109
1110 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1111
1112 /* Inner loop uses 120 flops */
1113 }
1114
1115 /* End of innermost loop */
1116
1117 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1118 f+i_coord_offset+DIM,fshift+i_shift_offset);
1119
1120 /* Increment number of inner iterations */
1121 inneriter += j_index_end - j_index_start;
1122
1123 /* Outer loop uses 18 flops */
1124 }
1125
1126 /* Increment number of outer iterations */
1127 outeriter += nri;
1128
1129 /* Update outer/inner flops */
1130
1131 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);
1132 }
The ultimate molecular dynamics simulation package