?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Remove nb-parameters from t_forcerec
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_sse2_single / nb_kernel_ElecEw_VdwNone_GeomW4P1_sse2_single.c
blob1a0b5e1500de953f22a8e049576f8c60a919a0af
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
46
47 #include "kernelutil_x86_sse2_single.h"
48
49 /*
50 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_sse2_single
51 * Electrostatics interaction: Ewald
52 * VdW interaction: None
53 * Geometry: Water4-Particle
54 * Calculate force/pot: PotentialAndForce
55 */
56 void
57 nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_sse2_single
58 (t_nblist * gmx_restrict nlist,
59 rvec * gmx_restrict xx,
60 rvec * gmx_restrict ff,
61 struct t_forcerec * gmx_restrict fr,
62 t_mdatoms * gmx_restrict mdatoms,
63 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
64 t_nrnb * gmx_restrict nrnb)
65 {
66 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
67 * just 0 for non-waters.
68 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
69 * jnr indices corresponding to data put in the four positions in the SIMD register.
70 */
71 int i_shift_offset,i_coord_offset,outeriter,inneriter;
72 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
73 int jnrA,jnrB,jnrC,jnrD;
74 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
75 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
77 real rcutoff_scalar;
78 real *shiftvec,*fshift,*x,*f;
79 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
80 real scratch[4*DIM];
81 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
82 int vdwioffset1;
83 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
84 int vdwioffset2;
85 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
86 int vdwioffset3;
87 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
88 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
89 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
91 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
92 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
93 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
94 real *charge;
95 __m128i ewitab;
96 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
97 real *ewtab;
98 __m128 dummy_mask,cutoff_mask;
99 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
100 __m128 one = _mm_set1_ps(1.0);
101 __m128 two = _mm_set1_ps(2.0);
102 x = xx[0];
103 f = ff[0];
104
105 nri = nlist->nri;
106 iinr = nlist->iinr;
107 jindex = nlist->jindex;
108 jjnr = nlist->jjnr;
109 shiftidx = nlist->shift;
110 gid = nlist->gid;
111 shiftvec = fr->shift_vec[0];
112 fshift = fr->fshift[0];
113 facel = _mm_set1_ps(fr->ic->epsfac);
114 charge = mdatoms->chargeA;
115
116 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
117 ewtab = fr->ic->tabq_coul_FDV0;
118 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
119 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
120
121 /* Setup water-specific parameters */
122 inr = nlist->iinr[0];
123 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
124 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
125 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
126
127 /* Avoid stupid compiler warnings */
128 jnrA = jnrB = jnrC = jnrD = 0;
129 j_coord_offsetA = 0;
130 j_coord_offsetB = 0;
131 j_coord_offsetC = 0;
132 j_coord_offsetD = 0;
133
134 outeriter = 0;
135 inneriter = 0;
136
137 for(iidx=0;iidx<4*DIM;iidx++)
138 {
139 scratch[iidx] = 0.0;
140 }
141
142 /* Start outer loop over neighborlists */
143 for(iidx=0; iidx<nri; iidx++)
144 {
145 /* Load shift vector for this list */
146 i_shift_offset = DIM*shiftidx[iidx];
147
148 /* Load limits for loop over neighbors */
149 j_index_start = jindex[iidx];
150 j_index_end = jindex[iidx+1];
151
152 /* Get outer coordinate index */
153 inr = iinr[iidx];
154 i_coord_offset = DIM*inr;
155
156 /* Load i particle coords and add shift vector */
157 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
158 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
159
160 fix1 = _mm_setzero_ps();
161 fiy1 = _mm_setzero_ps();
162 fiz1 = _mm_setzero_ps();
163 fix2 = _mm_setzero_ps();
164 fiy2 = _mm_setzero_ps();
165 fiz2 = _mm_setzero_ps();
166 fix3 = _mm_setzero_ps();
167 fiy3 = _mm_setzero_ps();
168 fiz3 = _mm_setzero_ps();
169
170 /* Reset potential sums */
171 velecsum = _mm_setzero_ps();
172
173 /* Start inner kernel loop */
174 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
175 {
176
177 /* Get j neighbor index, and coordinate index */
178 jnrA = jjnr[jidx];
179 jnrB = jjnr[jidx+1];
180 jnrC = jjnr[jidx+2];
181 jnrD = jjnr[jidx+3];
182 j_coord_offsetA = DIM*jnrA;
183 j_coord_offsetB = DIM*jnrB;
184 j_coord_offsetC = DIM*jnrC;
185 j_coord_offsetD = DIM*jnrD;
186
187 /* load j atom coordinates */
188 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
189 x+j_coord_offsetC,x+j_coord_offsetD,
190 &jx0,&jy0,&jz0);
191
192 /* Calculate displacement vector */
193 dx10 = _mm_sub_ps(ix1,jx0);
194 dy10 = _mm_sub_ps(iy1,jy0);
195 dz10 = _mm_sub_ps(iz1,jz0);
196 dx20 = _mm_sub_ps(ix2,jx0);
197 dy20 = _mm_sub_ps(iy2,jy0);
198 dz20 = _mm_sub_ps(iz2,jz0);
199 dx30 = _mm_sub_ps(ix3,jx0);
200 dy30 = _mm_sub_ps(iy3,jy0);
201 dz30 = _mm_sub_ps(iz3,jz0);
202
203 /* Calculate squared distance and things based on it */
204 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
205 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
206 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
207
208 rinv10 = sse2_invsqrt_f(rsq10);
209 rinv20 = sse2_invsqrt_f(rsq20);
210 rinv30 = sse2_invsqrt_f(rsq30);
211
212 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
213 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
214 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
215
216 /* Load parameters for j particles */
217 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
218 charge+jnrC+0,charge+jnrD+0);
219
220 fjx0 = _mm_setzero_ps();
221 fjy0 = _mm_setzero_ps();
222 fjz0 = _mm_setzero_ps();
223
224 /**************************
225 * CALCULATE INTERACTIONS *
226 **************************/
227
228 r10 = _mm_mul_ps(rsq10,rinv10);
229
230 /* Compute parameters for interactions between i and j atoms */
231 qq10 = _mm_mul_ps(iq1,jq0);
232
233 /* EWALD ELECTROSTATICS */
234
235 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
236 ewrt = _mm_mul_ps(r10,ewtabscale);
237 ewitab = _mm_cvttps_epi32(ewrt);
238 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
239 ewitab = _mm_slli_epi32(ewitab,2);
240 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
241 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
242 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
243 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
244 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
245 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
246 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
247 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
248 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
249
250 /* Update potential sum for this i atom from the interaction with this j atom. */
251 velecsum = _mm_add_ps(velecsum,velec);
252
253 fscal = felec;
254
255 /* Calculate temporary vectorial force */
256 tx = _mm_mul_ps(fscal,dx10);
257 ty = _mm_mul_ps(fscal,dy10);
258 tz = _mm_mul_ps(fscal,dz10);
259
260 /* Update vectorial force */
261 fix1 = _mm_add_ps(fix1,tx);
262 fiy1 = _mm_add_ps(fiy1,ty);
263 fiz1 = _mm_add_ps(fiz1,tz);
264
265 fjx0 = _mm_add_ps(fjx0,tx);
266 fjy0 = _mm_add_ps(fjy0,ty);
267 fjz0 = _mm_add_ps(fjz0,tz);
268
269 /**************************
270 * CALCULATE INTERACTIONS *
271 **************************/
272
273 r20 = _mm_mul_ps(rsq20,rinv20);
274
275 /* Compute parameters for interactions between i and j atoms */
276 qq20 = _mm_mul_ps(iq2,jq0);
277
278 /* EWALD ELECTROSTATICS */
279
280 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
281 ewrt = _mm_mul_ps(r20,ewtabscale);
282 ewitab = _mm_cvttps_epi32(ewrt);
283 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
284 ewitab = _mm_slli_epi32(ewitab,2);
285 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
286 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
287 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
288 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
289 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
290 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
291 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
292 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
293 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
294
295 /* Update potential sum for this i atom from the interaction with this j atom. */
296 velecsum = _mm_add_ps(velecsum,velec);
297
298 fscal = felec;
299
300 /* Calculate temporary vectorial force */
301 tx = _mm_mul_ps(fscal,dx20);
302 ty = _mm_mul_ps(fscal,dy20);
303 tz = _mm_mul_ps(fscal,dz20);
304
305 /* Update vectorial force */
306 fix2 = _mm_add_ps(fix2,tx);
307 fiy2 = _mm_add_ps(fiy2,ty);
308 fiz2 = _mm_add_ps(fiz2,tz);
309
310 fjx0 = _mm_add_ps(fjx0,tx);
311 fjy0 = _mm_add_ps(fjy0,ty);
312 fjz0 = _mm_add_ps(fjz0,tz);
313
314 /**************************
315 * CALCULATE INTERACTIONS *
316 **************************/
317
318 r30 = _mm_mul_ps(rsq30,rinv30);
319
320 /* Compute parameters for interactions between i and j atoms */
321 qq30 = _mm_mul_ps(iq3,jq0);
322
323 /* EWALD ELECTROSTATICS */
324
325 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
326 ewrt = _mm_mul_ps(r30,ewtabscale);
327 ewitab = _mm_cvttps_epi32(ewrt);
328 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
329 ewitab = _mm_slli_epi32(ewitab,2);
330 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
331 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
332 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
333 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
334 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
335 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
336 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
337 velec = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
338 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
339
340 /* Update potential sum for this i atom from the interaction with this j atom. */
341 velecsum = _mm_add_ps(velecsum,velec);
342
343 fscal = felec;
344
345 /* Calculate temporary vectorial force */
346 tx = _mm_mul_ps(fscal,dx30);
347 ty = _mm_mul_ps(fscal,dy30);
348 tz = _mm_mul_ps(fscal,dz30);
349
350 /* Update vectorial force */
351 fix3 = _mm_add_ps(fix3,tx);
352 fiy3 = _mm_add_ps(fiy3,ty);
353 fiz3 = _mm_add_ps(fiz3,tz);
354
355 fjx0 = _mm_add_ps(fjx0,tx);
356 fjy0 = _mm_add_ps(fjy0,ty);
357 fjz0 = _mm_add_ps(fjz0,tz);
358
359 fjptrA = f+j_coord_offsetA;
360 fjptrB = f+j_coord_offsetB;
361 fjptrC = f+j_coord_offsetC;
362 fjptrD = f+j_coord_offsetD;
363
364 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
365
366 /* Inner loop uses 123 flops */
367 }
368
369 if(jidx<j_index_end)
370 {
371
372 /* Get j neighbor index, and coordinate index */
373 jnrlistA = jjnr[jidx];
374 jnrlistB = jjnr[jidx+1];
375 jnrlistC = jjnr[jidx+2];
376 jnrlistD = jjnr[jidx+3];
377 /* Sign of each element will be negative for non-real atoms.
378 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
379 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
380 */
381 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
382 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
383 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
384 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
385 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
386 j_coord_offsetA = DIM*jnrA;
387 j_coord_offsetB = DIM*jnrB;
388 j_coord_offsetC = DIM*jnrC;
389 j_coord_offsetD = DIM*jnrD;
390
391 /* load j atom coordinates */
392 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
393 x+j_coord_offsetC,x+j_coord_offsetD,
394 &jx0,&jy0,&jz0);
395
396 /* Calculate displacement vector */
397 dx10 = _mm_sub_ps(ix1,jx0);
398 dy10 = _mm_sub_ps(iy1,jy0);
399 dz10 = _mm_sub_ps(iz1,jz0);
400 dx20 = _mm_sub_ps(ix2,jx0);
401 dy20 = _mm_sub_ps(iy2,jy0);
402 dz20 = _mm_sub_ps(iz2,jz0);
403 dx30 = _mm_sub_ps(ix3,jx0);
404 dy30 = _mm_sub_ps(iy3,jy0);
405 dz30 = _mm_sub_ps(iz3,jz0);
406
407 /* Calculate squared distance and things based on it */
408 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
409 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
410 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
411
412 rinv10 = sse2_invsqrt_f(rsq10);
413 rinv20 = sse2_invsqrt_f(rsq20);
414 rinv30 = sse2_invsqrt_f(rsq30);
415
416 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
417 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
418 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
419
420 /* Load parameters for j particles */
421 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
422 charge+jnrC+0,charge+jnrD+0);
423
424 fjx0 = _mm_setzero_ps();
425 fjy0 = _mm_setzero_ps();
426 fjz0 = _mm_setzero_ps();
427
428 /**************************
429 * CALCULATE INTERACTIONS *
430 **************************/
431
432 r10 = _mm_mul_ps(rsq10,rinv10);
433 r10 = _mm_andnot_ps(dummy_mask,r10);
434
435 /* Compute parameters for interactions between i and j atoms */
436 qq10 = _mm_mul_ps(iq1,jq0);
437
438 /* EWALD ELECTROSTATICS */
439
440 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
441 ewrt = _mm_mul_ps(r10,ewtabscale);
442 ewitab = _mm_cvttps_epi32(ewrt);
443 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
444 ewitab = _mm_slli_epi32(ewitab,2);
445 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
446 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
447 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
448 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
449 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
450 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
451 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
452 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
453 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
454
455 /* Update potential sum for this i atom from the interaction with this j atom. */
456 velec = _mm_andnot_ps(dummy_mask,velec);
457 velecsum = _mm_add_ps(velecsum,velec);
458
459 fscal = felec;
460
461 fscal = _mm_andnot_ps(dummy_mask,fscal);
462
463 /* Calculate temporary vectorial force */
464 tx = _mm_mul_ps(fscal,dx10);
465 ty = _mm_mul_ps(fscal,dy10);
466 tz = _mm_mul_ps(fscal,dz10);
467
468 /* Update vectorial force */
469 fix1 = _mm_add_ps(fix1,tx);
470 fiy1 = _mm_add_ps(fiy1,ty);
471 fiz1 = _mm_add_ps(fiz1,tz);
472
473 fjx0 = _mm_add_ps(fjx0,tx);
474 fjy0 = _mm_add_ps(fjy0,ty);
475 fjz0 = _mm_add_ps(fjz0,tz);
476
477 /**************************
478 * CALCULATE INTERACTIONS *
479 **************************/
480
481 r20 = _mm_mul_ps(rsq20,rinv20);
482 r20 = _mm_andnot_ps(dummy_mask,r20);
483
484 /* Compute parameters for interactions between i and j atoms */
485 qq20 = _mm_mul_ps(iq2,jq0);
486
487 /* EWALD ELECTROSTATICS */
488
489 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
490 ewrt = _mm_mul_ps(r20,ewtabscale);
491 ewitab = _mm_cvttps_epi32(ewrt);
492 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
493 ewitab = _mm_slli_epi32(ewitab,2);
494 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
495 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
496 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
497 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
498 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
499 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
500 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
501 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
502 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
503
504 /* Update potential sum for this i atom from the interaction with this j atom. */
505 velec = _mm_andnot_ps(dummy_mask,velec);
506 velecsum = _mm_add_ps(velecsum,velec);
507
508 fscal = felec;
509
510 fscal = _mm_andnot_ps(dummy_mask,fscal);
511
512 /* Calculate temporary vectorial force */
513 tx = _mm_mul_ps(fscal,dx20);
514 ty = _mm_mul_ps(fscal,dy20);
515 tz = _mm_mul_ps(fscal,dz20);
516
517 /* Update vectorial force */
518 fix2 = _mm_add_ps(fix2,tx);
519 fiy2 = _mm_add_ps(fiy2,ty);
520 fiz2 = _mm_add_ps(fiz2,tz);
521
522 fjx0 = _mm_add_ps(fjx0,tx);
523 fjy0 = _mm_add_ps(fjy0,ty);
524 fjz0 = _mm_add_ps(fjz0,tz);
525
526 /**************************
527 * CALCULATE INTERACTIONS *
528 **************************/
529
530 r30 = _mm_mul_ps(rsq30,rinv30);
531 r30 = _mm_andnot_ps(dummy_mask,r30);
532
533 /* Compute parameters for interactions between i and j atoms */
534 qq30 = _mm_mul_ps(iq3,jq0);
535
536 /* EWALD ELECTROSTATICS */
537
538 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
539 ewrt = _mm_mul_ps(r30,ewtabscale);
540 ewitab = _mm_cvttps_epi32(ewrt);
541 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
542 ewitab = _mm_slli_epi32(ewitab,2);
543 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
544 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
545 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
546 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
547 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
548 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
549 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
550 velec = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
551 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
552
553 /* Update potential sum for this i atom from the interaction with this j atom. */
554 velec = _mm_andnot_ps(dummy_mask,velec);
555 velecsum = _mm_add_ps(velecsum,velec);
556
557 fscal = felec;
558
559 fscal = _mm_andnot_ps(dummy_mask,fscal);
560
561 /* Calculate temporary vectorial force */
562 tx = _mm_mul_ps(fscal,dx30);
563 ty = _mm_mul_ps(fscal,dy30);
564 tz = _mm_mul_ps(fscal,dz30);
565
566 /* Update vectorial force */
567 fix3 = _mm_add_ps(fix3,tx);
568 fiy3 = _mm_add_ps(fiy3,ty);
569 fiz3 = _mm_add_ps(fiz3,tz);
570
571 fjx0 = _mm_add_ps(fjx0,tx);
572 fjy0 = _mm_add_ps(fjy0,ty);
573 fjz0 = _mm_add_ps(fjz0,tz);
574
575 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
576 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
577 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
578 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
579
580 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
581
582 /* Inner loop uses 126 flops */
583 }
584
585 /* End of innermost loop */
586
587 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
588 f+i_coord_offset+DIM,fshift+i_shift_offset);
589
590 ggid = gid[iidx];
591 /* Update potential energies */
592 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
593
594 /* Increment number of inner iterations */
595 inneriter += j_index_end - j_index_start;
596
597 /* Outer loop uses 19 flops */
598 }
599
600 /* Increment number of outer iterations */
601 outeriter += nri;
602
603 /* Update outer/inner flops */
604
605 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*126);
606 }
607 /*
608 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW4P1_F_sse2_single
609 * Electrostatics interaction: Ewald
610 * VdW interaction: None
611 * Geometry: Water4-Particle
612 * Calculate force/pot: Force
613 */
614 void
615 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_sse2_single
616 (t_nblist * gmx_restrict nlist,
617 rvec * gmx_restrict xx,
618 rvec * gmx_restrict ff,
619 struct t_forcerec * gmx_restrict fr,
620 t_mdatoms * gmx_restrict mdatoms,
621 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
622 t_nrnb * gmx_restrict nrnb)
623 {
624 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
625 * just 0 for non-waters.
626 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
627 * jnr indices corresponding to data put in the four positions in the SIMD register.
628 */
629 int i_shift_offset,i_coord_offset,outeriter,inneriter;
630 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
631 int jnrA,jnrB,jnrC,jnrD;
632 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
633 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
634 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
635 real rcutoff_scalar;
636 real *shiftvec,*fshift,*x,*f;
637 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
638 real scratch[4*DIM];
639 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
640 int vdwioffset1;
641 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
642 int vdwioffset2;
643 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
644 int vdwioffset3;
645 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
646 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
647 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
648 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
649 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
650 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
651 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
652 real *charge;
653 __m128i ewitab;
654 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
655 real *ewtab;
656 __m128 dummy_mask,cutoff_mask;
657 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
658 __m128 one = _mm_set1_ps(1.0);
659 __m128 two = _mm_set1_ps(2.0);
660 x = xx[0];
661 f = ff[0];
662
663 nri = nlist->nri;
664 iinr = nlist->iinr;
665 jindex = nlist->jindex;
666 jjnr = nlist->jjnr;
667 shiftidx = nlist->shift;
668 gid = nlist->gid;
669 shiftvec = fr->shift_vec[0];
670 fshift = fr->fshift[0];
671 facel = _mm_set1_ps(fr->ic->epsfac);
672 charge = mdatoms->chargeA;
673
674 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
675 ewtab = fr->ic->tabq_coul_F;
676 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
677 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
678
679 /* Setup water-specific parameters */
680 inr = nlist->iinr[0];
681 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
682 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
683 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
684
685 /* Avoid stupid compiler warnings */
686 jnrA = jnrB = jnrC = jnrD = 0;
687 j_coord_offsetA = 0;
688 j_coord_offsetB = 0;
689 j_coord_offsetC = 0;
690 j_coord_offsetD = 0;
691
692 outeriter = 0;
693 inneriter = 0;
694
695 for(iidx=0;iidx<4*DIM;iidx++)
696 {
697 scratch[iidx] = 0.0;
698 }
699
700 /* Start outer loop over neighborlists */
701 for(iidx=0; iidx<nri; iidx++)
702 {
703 /* Load shift vector for this list */
704 i_shift_offset = DIM*shiftidx[iidx];
705
706 /* Load limits for loop over neighbors */
707 j_index_start = jindex[iidx];
708 j_index_end = jindex[iidx+1];
709
710 /* Get outer coordinate index */
711 inr = iinr[iidx];
712 i_coord_offset = DIM*inr;
713
714 /* Load i particle coords and add shift vector */
715 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
716 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
717
718 fix1 = _mm_setzero_ps();
719 fiy1 = _mm_setzero_ps();
720 fiz1 = _mm_setzero_ps();
721 fix2 = _mm_setzero_ps();
722 fiy2 = _mm_setzero_ps();
723 fiz2 = _mm_setzero_ps();
724 fix3 = _mm_setzero_ps();
725 fiy3 = _mm_setzero_ps();
726 fiz3 = _mm_setzero_ps();
727
728 /* Start inner kernel loop */
729 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
730 {
731
732 /* Get j neighbor index, and coordinate index */
733 jnrA = jjnr[jidx];
734 jnrB = jjnr[jidx+1];
735 jnrC = jjnr[jidx+2];
736 jnrD = jjnr[jidx+3];
737 j_coord_offsetA = DIM*jnrA;
738 j_coord_offsetB = DIM*jnrB;
739 j_coord_offsetC = DIM*jnrC;
740 j_coord_offsetD = DIM*jnrD;
741
742 /* load j atom coordinates */
743 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
744 x+j_coord_offsetC,x+j_coord_offsetD,
745 &jx0,&jy0,&jz0);
746
747 /* Calculate displacement vector */
748 dx10 = _mm_sub_ps(ix1,jx0);
749 dy10 = _mm_sub_ps(iy1,jy0);
750 dz10 = _mm_sub_ps(iz1,jz0);
751 dx20 = _mm_sub_ps(ix2,jx0);
752 dy20 = _mm_sub_ps(iy2,jy0);
753 dz20 = _mm_sub_ps(iz2,jz0);
754 dx30 = _mm_sub_ps(ix3,jx0);
755 dy30 = _mm_sub_ps(iy3,jy0);
756 dz30 = _mm_sub_ps(iz3,jz0);
757
758 /* Calculate squared distance and things based on it */
759 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
760 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
761 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
762
763 rinv10 = sse2_invsqrt_f(rsq10);
764 rinv20 = sse2_invsqrt_f(rsq20);
765 rinv30 = sse2_invsqrt_f(rsq30);
766
767 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
768 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
769 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
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 /* EWALD ELECTROSTATICS */
789
790 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
791 ewrt = _mm_mul_ps(r10,ewtabscale);
792 ewitab = _mm_cvttps_epi32(ewrt);
793 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
794 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
795 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
796 &ewtabF,&ewtabFn);
797 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
798 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
799
800 fscal = felec;
801
802 /* Calculate temporary vectorial force */
803 tx = _mm_mul_ps(fscal,dx10);
804 ty = _mm_mul_ps(fscal,dy10);
805 tz = _mm_mul_ps(fscal,dz10);
806
807 /* Update vectorial force */
808 fix1 = _mm_add_ps(fix1,tx);
809 fiy1 = _mm_add_ps(fiy1,ty);
810 fiz1 = _mm_add_ps(fiz1,tz);
811
812 fjx0 = _mm_add_ps(fjx0,tx);
813 fjy0 = _mm_add_ps(fjy0,ty);
814 fjz0 = _mm_add_ps(fjz0,tz);
815
816 /**************************
817 * CALCULATE INTERACTIONS *
818 **************************/
819
820 r20 = _mm_mul_ps(rsq20,rinv20);
821
822 /* Compute parameters for interactions between i and j atoms */
823 qq20 = _mm_mul_ps(iq2,jq0);
824
825 /* EWALD ELECTROSTATICS */
826
827 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
828 ewrt = _mm_mul_ps(r20,ewtabscale);
829 ewitab = _mm_cvttps_epi32(ewrt);
830 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
831 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
832 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
833 &ewtabF,&ewtabFn);
834 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
835 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
836
837 fscal = felec;
838
839 /* Calculate temporary vectorial force */
840 tx = _mm_mul_ps(fscal,dx20);
841 ty = _mm_mul_ps(fscal,dy20);
842 tz = _mm_mul_ps(fscal,dz20);
843
844 /* Update vectorial force */
845 fix2 = _mm_add_ps(fix2,tx);
846 fiy2 = _mm_add_ps(fiy2,ty);
847 fiz2 = _mm_add_ps(fiz2,tz);
848
849 fjx0 = _mm_add_ps(fjx0,tx);
850 fjy0 = _mm_add_ps(fjy0,ty);
851 fjz0 = _mm_add_ps(fjz0,tz);
852
853 /**************************
854 * CALCULATE INTERACTIONS *
855 **************************/
856
857 r30 = _mm_mul_ps(rsq30,rinv30);
858
859 /* Compute parameters for interactions between i and j atoms */
860 qq30 = _mm_mul_ps(iq3,jq0);
861
862 /* EWALD ELECTROSTATICS */
863
864 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
865 ewrt = _mm_mul_ps(r30,ewtabscale);
866 ewitab = _mm_cvttps_epi32(ewrt);
867 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
868 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
869 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
870 &ewtabF,&ewtabFn);
871 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
872 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
873
874 fscal = felec;
875
876 /* Calculate temporary vectorial force */
877 tx = _mm_mul_ps(fscal,dx30);
878 ty = _mm_mul_ps(fscal,dy30);
879 tz = _mm_mul_ps(fscal,dz30);
880
881 /* Update vectorial force */
882 fix3 = _mm_add_ps(fix3,tx);
883 fiy3 = _mm_add_ps(fiy3,ty);
884 fiz3 = _mm_add_ps(fiz3,tz);
885
886 fjx0 = _mm_add_ps(fjx0,tx);
887 fjy0 = _mm_add_ps(fjy0,ty);
888 fjz0 = _mm_add_ps(fjz0,tz);
889
890 fjptrA = f+j_coord_offsetA;
891 fjptrB = f+j_coord_offsetB;
892 fjptrC = f+j_coord_offsetC;
893 fjptrD = f+j_coord_offsetD;
894
895 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
896
897 /* Inner loop uses 108 flops */
898 }
899
900 if(jidx<j_index_end)
901 {
902
903 /* Get j neighbor index, and coordinate index */
904 jnrlistA = jjnr[jidx];
905 jnrlistB = jjnr[jidx+1];
906 jnrlistC = jjnr[jidx+2];
907 jnrlistD = jjnr[jidx+3];
908 /* Sign of each element will be negative for non-real atoms.
909 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
910 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
911 */
912 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
913 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
914 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
915 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
916 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
917 j_coord_offsetA = DIM*jnrA;
918 j_coord_offsetB = DIM*jnrB;
919 j_coord_offsetC = DIM*jnrC;
920 j_coord_offsetD = DIM*jnrD;
921
922 /* load j atom coordinates */
923 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
924 x+j_coord_offsetC,x+j_coord_offsetD,
925 &jx0,&jy0,&jz0);
926
927 /* Calculate displacement vector */
928 dx10 = _mm_sub_ps(ix1,jx0);
929 dy10 = _mm_sub_ps(iy1,jy0);
930 dz10 = _mm_sub_ps(iz1,jz0);
931 dx20 = _mm_sub_ps(ix2,jx0);
932 dy20 = _mm_sub_ps(iy2,jy0);
933 dz20 = _mm_sub_ps(iz2,jz0);
934 dx30 = _mm_sub_ps(ix3,jx0);
935 dy30 = _mm_sub_ps(iy3,jy0);
936 dz30 = _mm_sub_ps(iz3,jz0);
937
938 /* Calculate squared distance and things based on it */
939 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
940 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
941 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
942
943 rinv10 = sse2_invsqrt_f(rsq10);
944 rinv20 = sse2_invsqrt_f(rsq20);
945 rinv30 = sse2_invsqrt_f(rsq30);
946
947 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
948 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
949 rinvsq30 = _mm_mul_ps(rinv30,rinv30);
950
951 /* Load parameters for j particles */
952 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
953 charge+jnrC+0,charge+jnrD+0);
954
955 fjx0 = _mm_setzero_ps();
956 fjy0 = _mm_setzero_ps();
957 fjz0 = _mm_setzero_ps();
958
959 /**************************
960 * CALCULATE INTERACTIONS *
961 **************************/
962
963 r10 = _mm_mul_ps(rsq10,rinv10);
964 r10 = _mm_andnot_ps(dummy_mask,r10);
965
966 /* Compute parameters for interactions between i and j atoms */
967 qq10 = _mm_mul_ps(iq1,jq0);
968
969 /* EWALD ELECTROSTATICS */
970
971 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
972 ewrt = _mm_mul_ps(r10,ewtabscale);
973 ewitab = _mm_cvttps_epi32(ewrt);
974 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
975 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
976 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
977 &ewtabF,&ewtabFn);
978 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
979 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
980
981 fscal = felec;
982
983 fscal = _mm_andnot_ps(dummy_mask,fscal);
984
985 /* Calculate temporary vectorial force */
986 tx = _mm_mul_ps(fscal,dx10);
987 ty = _mm_mul_ps(fscal,dy10);
988 tz = _mm_mul_ps(fscal,dz10);
989
990 /* Update vectorial force */
991 fix1 = _mm_add_ps(fix1,tx);
992 fiy1 = _mm_add_ps(fiy1,ty);
993 fiz1 = _mm_add_ps(fiz1,tz);
994
995 fjx0 = _mm_add_ps(fjx0,tx);
996 fjy0 = _mm_add_ps(fjy0,ty);
997 fjz0 = _mm_add_ps(fjz0,tz);
998
999 /**************************
1000 * CALCULATE INTERACTIONS *
1001 **************************/
1002
1003 r20 = _mm_mul_ps(rsq20,rinv20);
1004 r20 = _mm_andnot_ps(dummy_mask,r20);
1005
1006 /* Compute parameters for interactions between i and j atoms */
1007 qq20 = _mm_mul_ps(iq2,jq0);
1008
1009 /* EWALD ELECTROSTATICS */
1010
1011 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1012 ewrt = _mm_mul_ps(r20,ewtabscale);
1013 ewitab = _mm_cvttps_epi32(ewrt);
1014 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1015 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1016 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1017 &ewtabF,&ewtabFn);
1018 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1019 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1020
1021 fscal = felec;
1022
1023 fscal = _mm_andnot_ps(dummy_mask,fscal);
1024
1025 /* Calculate temporary vectorial force */
1026 tx = _mm_mul_ps(fscal,dx20);
1027 ty = _mm_mul_ps(fscal,dy20);
1028 tz = _mm_mul_ps(fscal,dz20);
1029
1030 /* Update vectorial force */
1031 fix2 = _mm_add_ps(fix2,tx);
1032 fiy2 = _mm_add_ps(fiy2,ty);
1033 fiz2 = _mm_add_ps(fiz2,tz);
1034
1035 fjx0 = _mm_add_ps(fjx0,tx);
1036 fjy0 = _mm_add_ps(fjy0,ty);
1037 fjz0 = _mm_add_ps(fjz0,tz);
1038
1039 /**************************
1040 * CALCULATE INTERACTIONS *
1041 **************************/
1042
1043 r30 = _mm_mul_ps(rsq30,rinv30);
1044 r30 = _mm_andnot_ps(dummy_mask,r30);
1045
1046 /* Compute parameters for interactions between i and j atoms */
1047 qq30 = _mm_mul_ps(iq3,jq0);
1048
1049 /* EWALD ELECTROSTATICS */
1050
1051 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1052 ewrt = _mm_mul_ps(r30,ewtabscale);
1053 ewitab = _mm_cvttps_epi32(ewrt);
1054 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1055 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1056 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1057 &ewtabF,&ewtabFn);
1058 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1059 felec = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1060
1061 fscal = felec;
1062
1063 fscal = _mm_andnot_ps(dummy_mask,fscal);
1064
1065 /* Calculate temporary vectorial force */
1066 tx = _mm_mul_ps(fscal,dx30);
1067 ty = _mm_mul_ps(fscal,dy30);
1068 tz = _mm_mul_ps(fscal,dz30);
1069
1070 /* Update vectorial force */
1071 fix3 = _mm_add_ps(fix3,tx);
1072 fiy3 = _mm_add_ps(fiy3,ty);
1073 fiz3 = _mm_add_ps(fiz3,tz);
1074
1075 fjx0 = _mm_add_ps(fjx0,tx);
1076 fjy0 = _mm_add_ps(fjy0,ty);
1077 fjz0 = _mm_add_ps(fjz0,tz);
1078
1079 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1080 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1081 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1082 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1083
1084 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1085
1086 /* Inner loop uses 111 flops */
1087 }
1088
1089 /* End of innermost loop */
1090
1091 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1092 f+i_coord_offset+DIM,fshift+i_shift_offset);
1093
1094 /* Increment number of inner iterations */
1095 inneriter += j_index_end - j_index_start;
1096
1097 /* Outer loop uses 18 flops */
1098 }
1099
1100 /* Increment number of outer iterations */
1101 outeriter += nri;
1102
1103 /* Update outer/inner flops */
1104
1105 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*111);
1106 }
The ultimate molecular dynamics simulation package