?  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_ElecEwSh_VdwNone_GeomW3P1_sse2_single.c
blobefa1d788bfe076230112bb219551017e5e852142
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_ElecEwSh_VdwNone_GeomW3P1_VF_sse2_single
53 * Electrostatics interaction: Ewald
54 * VdW interaction: None
55 * Geometry: Water3-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecEwSh_VdwNone_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 __m128i ewitab;
98 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
99 real *ewtab;
100 __m128 dummy_mask,cutoff_mask;
101 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
102 __m128 one = _mm_set1_ps(1.0);
103 __m128 two = _mm_set1_ps(2.0);
104 x = xx[0];
105 f = ff[0];
106
107 nri = nlist->nri;
108 iinr = nlist->iinr;
109 jindex = nlist->jindex;
110 jjnr = nlist->jjnr;
111 shiftidx = nlist->shift;
112 gid = nlist->gid;
113 shiftvec = fr->shift_vec[0];
114 fshift = fr->fshift[0];
115 facel = _mm_set1_ps(fr->epsfac);
116 charge = mdatoms->chargeA;
117
118 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
119 ewtab = fr->ic->tabq_coul_FDV0;
120 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
121 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
122
123 /* Setup water-specific parameters */
124 inr = nlist->iinr[0];
125 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
126 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
127 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
128
129 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
130 rcutoff_scalar = fr->rcoulomb;
131 rcutoff = _mm_set1_ps(rcutoff_scalar);
132 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
133
134 /* Avoid stupid compiler warnings */
135 jnrA = jnrB = jnrC = jnrD = 0;
136 j_coord_offsetA = 0;
137 j_coord_offsetB = 0;
138 j_coord_offsetC = 0;
139 j_coord_offsetD = 0;
140
141 outeriter = 0;
142 inneriter = 0;
143
144 for(iidx=0;iidx<4*DIM;iidx++)
145 {
146 scratch[iidx] = 0.0;
147 }
148
149 /* Start outer loop over neighborlists */
150 for(iidx=0; iidx<nri; iidx++)
151 {
152 /* Load shift vector for this list */
153 i_shift_offset = DIM*shiftidx[iidx];
154
155 /* Load limits for loop over neighbors */
156 j_index_start = jindex[iidx];
157 j_index_end = jindex[iidx+1];
158
159 /* Get outer coordinate index */
160 inr = iinr[iidx];
161 i_coord_offset = DIM*inr;
162
163 /* Load i particle coords and add shift vector */
164 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
165 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
166
167 fix0 = _mm_setzero_ps();
168 fiy0 = _mm_setzero_ps();
169 fiz0 = _mm_setzero_ps();
170 fix1 = _mm_setzero_ps();
171 fiy1 = _mm_setzero_ps();
172 fiz1 = _mm_setzero_ps();
173 fix2 = _mm_setzero_ps();
174 fiy2 = _mm_setzero_ps();
175 fiz2 = _mm_setzero_ps();
176
177 /* Reset potential sums */
178 velecsum = _mm_setzero_ps();
179
180 /* Start inner kernel loop */
181 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
182 {
183
184 /* Get j neighbor index, and coordinate index */
185 jnrA = jjnr[jidx];
186 jnrB = jjnr[jidx+1];
187 jnrC = jjnr[jidx+2];
188 jnrD = jjnr[jidx+3];
189 j_coord_offsetA = DIM*jnrA;
190 j_coord_offsetB = DIM*jnrB;
191 j_coord_offsetC = DIM*jnrC;
192 j_coord_offsetD = DIM*jnrD;
193
194 /* load j atom coordinates */
195 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
196 x+j_coord_offsetC,x+j_coord_offsetD,
197 &jx0,&jy0,&jz0);
198
199 /* Calculate displacement vector */
200 dx00 = _mm_sub_ps(ix0,jx0);
201 dy00 = _mm_sub_ps(iy0,jy0);
202 dz00 = _mm_sub_ps(iz0,jz0);
203 dx10 = _mm_sub_ps(ix1,jx0);
204 dy10 = _mm_sub_ps(iy1,jy0);
205 dz10 = _mm_sub_ps(iz1,jz0);
206 dx20 = _mm_sub_ps(ix2,jx0);
207 dy20 = _mm_sub_ps(iy2,jy0);
208 dz20 = _mm_sub_ps(iz2,jz0);
209
210 /* Calculate squared distance and things based on it */
211 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
212 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
213 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
214
215 rinv00 = gmx_mm_invsqrt_ps(rsq00);
216 rinv10 = gmx_mm_invsqrt_ps(rsq10);
217 rinv20 = gmx_mm_invsqrt_ps(rsq20);
218
219 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
220 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
221 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
222
223 /* Load parameters for j particles */
224 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
225 charge+jnrC+0,charge+jnrD+0);
226
227 fjx0 = _mm_setzero_ps();
228 fjy0 = _mm_setzero_ps();
229 fjz0 = _mm_setzero_ps();
230
231 /**************************
232 * CALCULATE INTERACTIONS *
233 **************************/
234
235 if (gmx_mm_any_lt(rsq00,rcutoff2))
236 {
237
238 r00 = _mm_mul_ps(rsq00,rinv00);
239
240 /* Compute parameters for interactions between i and j atoms */
241 qq00 = _mm_mul_ps(iq0,jq0);
242
243 /* EWALD ELECTROSTATICS */
244
245 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
246 ewrt = _mm_mul_ps(r00,ewtabscale);
247 ewitab = _mm_cvttps_epi32(ewrt);
248 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
249 ewitab = _mm_slli_epi32(ewitab,2);
250 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
251 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
252 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
253 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
254 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
255 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
256 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
257 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
258 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
259
260 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
261
262 /* Update potential sum for this i atom from the interaction with this j atom. */
263 velec = _mm_and_ps(velec,cutoff_mask);
264 velecsum = _mm_add_ps(velecsum,velec);
265
266 fscal = felec;
267
268 fscal = _mm_and_ps(fscal,cutoff_mask);
269
270 /* Calculate temporary vectorial force */
271 tx = _mm_mul_ps(fscal,dx00);
272 ty = _mm_mul_ps(fscal,dy00);
273 tz = _mm_mul_ps(fscal,dz00);
274
275 /* Update vectorial force */
276 fix0 = _mm_add_ps(fix0,tx);
277 fiy0 = _mm_add_ps(fiy0,ty);
278 fiz0 = _mm_add_ps(fiz0,tz);
279
280 fjx0 = _mm_add_ps(fjx0,tx);
281 fjy0 = _mm_add_ps(fjy0,ty);
282 fjz0 = _mm_add_ps(fjz0,tz);
283
284 }
285
286 /**************************
287 * CALCULATE INTERACTIONS *
288 **************************/
289
290 if (gmx_mm_any_lt(rsq10,rcutoff2))
291 {
292
293 r10 = _mm_mul_ps(rsq10,rinv10);
294
295 /* Compute parameters for interactions between i and j atoms */
296 qq10 = _mm_mul_ps(iq1,jq0);
297
298 /* EWALD ELECTROSTATICS */
299
300 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
301 ewrt = _mm_mul_ps(r10,ewtabscale);
302 ewitab = _mm_cvttps_epi32(ewrt);
303 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
304 ewitab = _mm_slli_epi32(ewitab,2);
305 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
306 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
307 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
308 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
309 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
310 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
311 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
312 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
313 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
314
315 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
316
317 /* Update potential sum for this i atom from the interaction with this j atom. */
318 velec = _mm_and_ps(velec,cutoff_mask);
319 velecsum = _mm_add_ps(velecsum,velec);
320
321 fscal = felec;
322
323 fscal = _mm_and_ps(fscal,cutoff_mask);
324
325 /* Calculate temporary vectorial force */
326 tx = _mm_mul_ps(fscal,dx10);
327 ty = _mm_mul_ps(fscal,dy10);
328 tz = _mm_mul_ps(fscal,dz10);
329
330 /* Update vectorial force */
331 fix1 = _mm_add_ps(fix1,tx);
332 fiy1 = _mm_add_ps(fiy1,ty);
333 fiz1 = _mm_add_ps(fiz1,tz);
334
335 fjx0 = _mm_add_ps(fjx0,tx);
336 fjy0 = _mm_add_ps(fjy0,ty);
337 fjz0 = _mm_add_ps(fjz0,tz);
338
339 }
340
341 /**************************
342 * CALCULATE INTERACTIONS *
343 **************************/
344
345 if (gmx_mm_any_lt(rsq20,rcutoff2))
346 {
347
348 r20 = _mm_mul_ps(rsq20,rinv20);
349
350 /* Compute parameters for interactions between i and j atoms */
351 qq20 = _mm_mul_ps(iq2,jq0);
352
353 /* EWALD ELECTROSTATICS */
354
355 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
356 ewrt = _mm_mul_ps(r20,ewtabscale);
357 ewitab = _mm_cvttps_epi32(ewrt);
358 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
359 ewitab = _mm_slli_epi32(ewitab,2);
360 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
361 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
362 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
363 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
364 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
365 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
366 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
367 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
368 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
369
370 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
371
372 /* Update potential sum for this i atom from the interaction with this j atom. */
373 velec = _mm_and_ps(velec,cutoff_mask);
374 velecsum = _mm_add_ps(velecsum,velec);
375
376 fscal = felec;
377
378 fscal = _mm_and_ps(fscal,cutoff_mask);
379
380 /* Calculate temporary vectorial force */
381 tx = _mm_mul_ps(fscal,dx20);
382 ty = _mm_mul_ps(fscal,dy20);
383 tz = _mm_mul_ps(fscal,dz20);
384
385 /* Update vectorial force */
386 fix2 = _mm_add_ps(fix2,tx);
387 fiy2 = _mm_add_ps(fiy2,ty);
388 fiz2 = _mm_add_ps(fiz2,tz);
389
390 fjx0 = _mm_add_ps(fjx0,tx);
391 fjy0 = _mm_add_ps(fjy0,ty);
392 fjz0 = _mm_add_ps(fjz0,tz);
393
394 }
395
396 fjptrA = f+j_coord_offsetA;
397 fjptrB = f+j_coord_offsetB;
398 fjptrC = f+j_coord_offsetC;
399 fjptrD = f+j_coord_offsetD;
400
401 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
402
403 /* Inner loop uses 138 flops */
404 }
405
406 if(jidx<j_index_end)
407 {
408
409 /* Get j neighbor index, and coordinate index */
410 jnrlistA = jjnr[jidx];
411 jnrlistB = jjnr[jidx+1];
412 jnrlistC = jjnr[jidx+2];
413 jnrlistD = jjnr[jidx+3];
414 /* Sign of each element will be negative for non-real atoms.
415 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
416 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
417 */
418 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
419 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
420 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
421 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
422 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
423 j_coord_offsetA = DIM*jnrA;
424 j_coord_offsetB = DIM*jnrB;
425 j_coord_offsetC = DIM*jnrC;
426 j_coord_offsetD = DIM*jnrD;
427
428 /* load j atom coordinates */
429 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
430 x+j_coord_offsetC,x+j_coord_offsetD,
431 &jx0,&jy0,&jz0);
432
433 /* Calculate displacement vector */
434 dx00 = _mm_sub_ps(ix0,jx0);
435 dy00 = _mm_sub_ps(iy0,jy0);
436 dz00 = _mm_sub_ps(iz0,jz0);
437 dx10 = _mm_sub_ps(ix1,jx0);
438 dy10 = _mm_sub_ps(iy1,jy0);
439 dz10 = _mm_sub_ps(iz1,jz0);
440 dx20 = _mm_sub_ps(ix2,jx0);
441 dy20 = _mm_sub_ps(iy2,jy0);
442 dz20 = _mm_sub_ps(iz2,jz0);
443
444 /* Calculate squared distance and things based on it */
445 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
446 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
447 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
448
449 rinv00 = gmx_mm_invsqrt_ps(rsq00);
450 rinv10 = gmx_mm_invsqrt_ps(rsq10);
451 rinv20 = gmx_mm_invsqrt_ps(rsq20);
452
453 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
454 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
455 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
456
457 /* Load parameters for j particles */
458 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
459 charge+jnrC+0,charge+jnrD+0);
460
461 fjx0 = _mm_setzero_ps();
462 fjy0 = _mm_setzero_ps();
463 fjz0 = _mm_setzero_ps();
464
465 /**************************
466 * CALCULATE INTERACTIONS *
467 **************************/
468
469 if (gmx_mm_any_lt(rsq00,rcutoff2))
470 {
471
472 r00 = _mm_mul_ps(rsq00,rinv00);
473 r00 = _mm_andnot_ps(dummy_mask,r00);
474
475 /* Compute parameters for interactions between i and j atoms */
476 qq00 = _mm_mul_ps(iq0,jq0);
477
478 /* EWALD ELECTROSTATICS */
479
480 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
481 ewrt = _mm_mul_ps(r00,ewtabscale);
482 ewitab = _mm_cvttps_epi32(ewrt);
483 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
484 ewitab = _mm_slli_epi32(ewitab,2);
485 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
486 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
487 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
488 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
489 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
490 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
491 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
492 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
493 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
494
495 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
496
497 /* Update potential sum for this i atom from the interaction with this j atom. */
498 velec = _mm_and_ps(velec,cutoff_mask);
499 velec = _mm_andnot_ps(dummy_mask,velec);
500 velecsum = _mm_add_ps(velecsum,velec);
501
502 fscal = felec;
503
504 fscal = _mm_and_ps(fscal,cutoff_mask);
505
506 fscal = _mm_andnot_ps(dummy_mask,fscal);
507
508 /* Calculate temporary vectorial force */
509 tx = _mm_mul_ps(fscal,dx00);
510 ty = _mm_mul_ps(fscal,dy00);
511 tz = _mm_mul_ps(fscal,dz00);
512
513 /* Update vectorial force */
514 fix0 = _mm_add_ps(fix0,tx);
515 fiy0 = _mm_add_ps(fiy0,ty);
516 fiz0 = _mm_add_ps(fiz0,tz);
517
518 fjx0 = _mm_add_ps(fjx0,tx);
519 fjy0 = _mm_add_ps(fjy0,ty);
520 fjz0 = _mm_add_ps(fjz0,tz);
521
522 }
523
524 /**************************
525 * CALCULATE INTERACTIONS *
526 **************************/
527
528 if (gmx_mm_any_lt(rsq10,rcutoff2))
529 {
530
531 r10 = _mm_mul_ps(rsq10,rinv10);
532 r10 = _mm_andnot_ps(dummy_mask,r10);
533
534 /* Compute parameters for interactions between i and j atoms */
535 qq10 = _mm_mul_ps(iq1,jq0);
536
537 /* EWALD ELECTROSTATICS */
538
539 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
540 ewrt = _mm_mul_ps(r10,ewtabscale);
541 ewitab = _mm_cvttps_epi32(ewrt);
542 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
543 ewitab = _mm_slli_epi32(ewitab,2);
544 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
545 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
546 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
547 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
548 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
549 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
550 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
551 velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
552 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
553
554 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
555
556 /* Update potential sum for this i atom from the interaction with this j atom. */
557 velec = _mm_and_ps(velec,cutoff_mask);
558 velec = _mm_andnot_ps(dummy_mask,velec);
559 velecsum = _mm_add_ps(velecsum,velec);
560
561 fscal = felec;
562
563 fscal = _mm_and_ps(fscal,cutoff_mask);
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
583 /**************************
584 * CALCULATE INTERACTIONS *
585 **************************/
586
587 if (gmx_mm_any_lt(rsq20,rcutoff2))
588 {
589
590 r20 = _mm_mul_ps(rsq20,rinv20);
591 r20 = _mm_andnot_ps(dummy_mask,r20);
592
593 /* Compute parameters for interactions between i and j atoms */
594 qq20 = _mm_mul_ps(iq2,jq0);
595
596 /* EWALD ELECTROSTATICS */
597
598 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
599 ewrt = _mm_mul_ps(r20,ewtabscale);
600 ewitab = _mm_cvttps_epi32(ewrt);
601 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
602 ewitab = _mm_slli_epi32(ewitab,2);
603 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
604 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
605 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
606 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
607 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
608 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
609 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
610 velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
611 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
612
613 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
614
615 /* Update potential sum for this i atom from the interaction with this j atom. */
616 velec = _mm_and_ps(velec,cutoff_mask);
617 velec = _mm_andnot_ps(dummy_mask,velec);
618 velecsum = _mm_add_ps(velecsum,velec);
619
620 fscal = felec;
621
622 fscal = _mm_and_ps(fscal,cutoff_mask);
623
624 fscal = _mm_andnot_ps(dummy_mask,fscal);
625
626 /* Calculate temporary vectorial force */
627 tx = _mm_mul_ps(fscal,dx20);
628 ty = _mm_mul_ps(fscal,dy20);
629 tz = _mm_mul_ps(fscal,dz20);
630
631 /* Update vectorial force */
632 fix2 = _mm_add_ps(fix2,tx);
633 fiy2 = _mm_add_ps(fiy2,ty);
634 fiz2 = _mm_add_ps(fiz2,tz);
635
636 fjx0 = _mm_add_ps(fjx0,tx);
637 fjy0 = _mm_add_ps(fjy0,ty);
638 fjz0 = _mm_add_ps(fjz0,tz);
639
640 }
641
642 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
643 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
644 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
645 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
646
647 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
648
649 /* Inner loop uses 141 flops */
650 }
651
652 /* End of innermost loop */
653
654 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
655 f+i_coord_offset,fshift+i_shift_offset);
656
657 ggid = gid[iidx];
658 /* Update potential energies */
659 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
660
661 /* Increment number of inner iterations */
662 inneriter += j_index_end - j_index_start;
663
664 /* Outer loop uses 19 flops */
665 }
666
667 /* Increment number of outer iterations */
668 outeriter += nri;
669
670 /* Update outer/inner flops */
671
672 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*141);
673 }
674 /*
675 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_sse2_single
676 * Electrostatics interaction: Ewald
677 * VdW interaction: None
678 * Geometry: Water3-Particle
679 * Calculate force/pot: Force
680 */
681 void
682 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_sse2_single
683 (t_nblist * gmx_restrict nlist,
684 rvec * gmx_restrict xx,
685 rvec * gmx_restrict ff,
686 t_forcerec * gmx_restrict fr,
687 t_mdatoms * gmx_restrict mdatoms,
688 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
689 t_nrnb * gmx_restrict nrnb)
690 {
691 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
692 * just 0 for non-waters.
693 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
694 * jnr indices corresponding to data put in the four positions in the SIMD register.
695 */
696 int i_shift_offset,i_coord_offset,outeriter,inneriter;
697 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
698 int jnrA,jnrB,jnrC,jnrD;
699 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
700 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
701 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
702 real rcutoff_scalar;
703 real *shiftvec,*fshift,*x,*f;
704 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
705 real scratch[4*DIM];
706 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
707 int vdwioffset0;
708 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
709 int vdwioffset1;
710 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
711 int vdwioffset2;
712 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
713 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
714 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
715 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
716 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
717 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
718 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
719 real *charge;
720 __m128i ewitab;
721 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
722 real *ewtab;
723 __m128 dummy_mask,cutoff_mask;
724 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
725 __m128 one = _mm_set1_ps(1.0);
726 __m128 two = _mm_set1_ps(2.0);
727 x = xx[0];
728 f = ff[0];
729
730 nri = nlist->nri;
731 iinr = nlist->iinr;
732 jindex = nlist->jindex;
733 jjnr = nlist->jjnr;
734 shiftidx = nlist->shift;
735 gid = nlist->gid;
736 shiftvec = fr->shift_vec[0];
737 fshift = fr->fshift[0];
738 facel = _mm_set1_ps(fr->epsfac);
739 charge = mdatoms->chargeA;
740
741 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
742 ewtab = fr->ic->tabq_coul_F;
743 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
744 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
745
746 /* Setup water-specific parameters */
747 inr = nlist->iinr[0];
748 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
749 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
750 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
751
752 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
753 rcutoff_scalar = fr->rcoulomb;
754 rcutoff = _mm_set1_ps(rcutoff_scalar);
755 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
756
757 /* Avoid stupid compiler warnings */
758 jnrA = jnrB = jnrC = jnrD = 0;
759 j_coord_offsetA = 0;
760 j_coord_offsetB = 0;
761 j_coord_offsetC = 0;
762 j_coord_offsetD = 0;
763
764 outeriter = 0;
765 inneriter = 0;
766
767 for(iidx=0;iidx<4*DIM;iidx++)
768 {
769 scratch[iidx] = 0.0;
770 }
771
772 /* Start outer loop over neighborlists */
773 for(iidx=0; iidx<nri; iidx++)
774 {
775 /* Load shift vector for this list */
776 i_shift_offset = DIM*shiftidx[iidx];
777
778 /* Load limits for loop over neighbors */
779 j_index_start = jindex[iidx];
780 j_index_end = jindex[iidx+1];
781
782 /* Get outer coordinate index */
783 inr = iinr[iidx];
784 i_coord_offset = DIM*inr;
785
786 /* Load i particle coords and add shift vector */
787 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
788 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
789
790 fix0 = _mm_setzero_ps();
791 fiy0 = _mm_setzero_ps();
792 fiz0 = _mm_setzero_ps();
793 fix1 = _mm_setzero_ps();
794 fiy1 = _mm_setzero_ps();
795 fiz1 = _mm_setzero_ps();
796 fix2 = _mm_setzero_ps();
797 fiy2 = _mm_setzero_ps();
798 fiz2 = _mm_setzero_ps();
799
800 /* Start inner kernel loop */
801 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
802 {
803
804 /* Get j neighbor index, and coordinate index */
805 jnrA = jjnr[jidx];
806 jnrB = jjnr[jidx+1];
807 jnrC = jjnr[jidx+2];
808 jnrD = jjnr[jidx+3];
809 j_coord_offsetA = DIM*jnrA;
810 j_coord_offsetB = DIM*jnrB;
811 j_coord_offsetC = DIM*jnrC;
812 j_coord_offsetD = DIM*jnrD;
813
814 /* load j atom coordinates */
815 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
816 x+j_coord_offsetC,x+j_coord_offsetD,
817 &jx0,&jy0,&jz0);
818
819 /* Calculate displacement vector */
820 dx00 = _mm_sub_ps(ix0,jx0);
821 dy00 = _mm_sub_ps(iy0,jy0);
822 dz00 = _mm_sub_ps(iz0,jz0);
823 dx10 = _mm_sub_ps(ix1,jx0);
824 dy10 = _mm_sub_ps(iy1,jy0);
825 dz10 = _mm_sub_ps(iz1,jz0);
826 dx20 = _mm_sub_ps(ix2,jx0);
827 dy20 = _mm_sub_ps(iy2,jy0);
828 dz20 = _mm_sub_ps(iz2,jz0);
829
830 /* Calculate squared distance and things based on it */
831 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
832 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
833 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
834
835 rinv00 = gmx_mm_invsqrt_ps(rsq00);
836 rinv10 = gmx_mm_invsqrt_ps(rsq10);
837 rinv20 = gmx_mm_invsqrt_ps(rsq20);
838
839 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
840 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
841 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
842
843 /* Load parameters for j particles */
844 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
845 charge+jnrC+0,charge+jnrD+0);
846
847 fjx0 = _mm_setzero_ps();
848 fjy0 = _mm_setzero_ps();
849 fjz0 = _mm_setzero_ps();
850
851 /**************************
852 * CALCULATE INTERACTIONS *
853 **************************/
854
855 if (gmx_mm_any_lt(rsq00,rcutoff2))
856 {
857
858 r00 = _mm_mul_ps(rsq00,rinv00);
859
860 /* Compute parameters for interactions between i and j atoms */
861 qq00 = _mm_mul_ps(iq0,jq0);
862
863 /* EWALD ELECTROSTATICS */
864
865 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
866 ewrt = _mm_mul_ps(r00,ewtabscale);
867 ewitab = _mm_cvttps_epi32(ewrt);
868 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
869 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
870 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
871 &ewtabF,&ewtabFn);
872 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
873 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
874
875 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
876
877 fscal = felec;
878
879 fscal = _mm_and_ps(fscal,cutoff_mask);
880
881 /* Calculate temporary vectorial force */
882 tx = _mm_mul_ps(fscal,dx00);
883 ty = _mm_mul_ps(fscal,dy00);
884 tz = _mm_mul_ps(fscal,dz00);
885
886 /* Update vectorial force */
887 fix0 = _mm_add_ps(fix0,tx);
888 fiy0 = _mm_add_ps(fiy0,ty);
889 fiz0 = _mm_add_ps(fiz0,tz);
890
891 fjx0 = _mm_add_ps(fjx0,tx);
892 fjy0 = _mm_add_ps(fjy0,ty);
893 fjz0 = _mm_add_ps(fjz0,tz);
894
895 }
896
897 /**************************
898 * CALCULATE INTERACTIONS *
899 **************************/
900
901 if (gmx_mm_any_lt(rsq10,rcutoff2))
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 /* EWALD ELECTROSTATICS */
910
911 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
912 ewrt = _mm_mul_ps(r10,ewtabscale);
913 ewitab = _mm_cvttps_epi32(ewrt);
914 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
915 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
916 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
917 &ewtabF,&ewtabFn);
918 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
919 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
920
921 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
922
923 fscal = felec;
924
925 fscal = _mm_and_ps(fscal,cutoff_mask);
926
927 /* Calculate temporary vectorial force */
928 tx = _mm_mul_ps(fscal,dx10);
929 ty = _mm_mul_ps(fscal,dy10);
930 tz = _mm_mul_ps(fscal,dz10);
931
932 /* Update vectorial force */
933 fix1 = _mm_add_ps(fix1,tx);
934 fiy1 = _mm_add_ps(fiy1,ty);
935 fiz1 = _mm_add_ps(fiz1,tz);
936
937 fjx0 = _mm_add_ps(fjx0,tx);
938 fjy0 = _mm_add_ps(fjy0,ty);
939 fjz0 = _mm_add_ps(fjz0,tz);
940
941 }
942
943 /**************************
944 * CALCULATE INTERACTIONS *
945 **************************/
946
947 if (gmx_mm_any_lt(rsq20,rcutoff2))
948 {
949
950 r20 = _mm_mul_ps(rsq20,rinv20);
951
952 /* Compute parameters for interactions between i and j atoms */
953 qq20 = _mm_mul_ps(iq2,jq0);
954
955 /* EWALD ELECTROSTATICS */
956
957 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
958 ewrt = _mm_mul_ps(r20,ewtabscale);
959 ewitab = _mm_cvttps_epi32(ewrt);
960 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
961 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
962 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
963 &ewtabF,&ewtabFn);
964 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
965 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
966
967 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
968
969 fscal = felec;
970
971 fscal = _mm_and_ps(fscal,cutoff_mask);
972
973 /* Calculate temporary vectorial force */
974 tx = _mm_mul_ps(fscal,dx20);
975 ty = _mm_mul_ps(fscal,dy20);
976 tz = _mm_mul_ps(fscal,dz20);
977
978 /* Update vectorial force */
979 fix2 = _mm_add_ps(fix2,tx);
980 fiy2 = _mm_add_ps(fiy2,ty);
981 fiz2 = _mm_add_ps(fiz2,tz);
982
983 fjx0 = _mm_add_ps(fjx0,tx);
984 fjy0 = _mm_add_ps(fjy0,ty);
985 fjz0 = _mm_add_ps(fjz0,tz);
986
987 }
988
989 fjptrA = f+j_coord_offsetA;
990 fjptrB = f+j_coord_offsetB;
991 fjptrC = f+j_coord_offsetC;
992 fjptrD = f+j_coord_offsetD;
993
994 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
995
996 /* Inner loop uses 117 flops */
997 }
998
999 if(jidx<j_index_end)
1000 {
1001
1002 /* Get j neighbor index, and coordinate index */
1003 jnrlistA = jjnr[jidx];
1004 jnrlistB = jjnr[jidx+1];
1005 jnrlistC = jjnr[jidx+2];
1006 jnrlistD = jjnr[jidx+3];
1007 /* Sign of each element will be negative for non-real atoms.
1008 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1009 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1010 */
1011 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1012 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
1013 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
1014 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
1015 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
1016 j_coord_offsetA = DIM*jnrA;
1017 j_coord_offsetB = DIM*jnrB;
1018 j_coord_offsetC = DIM*jnrC;
1019 j_coord_offsetD = DIM*jnrD;
1020
1021 /* load j atom coordinates */
1022 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1023 x+j_coord_offsetC,x+j_coord_offsetD,
1024 &jx0,&jy0,&jz0);
1025
1026 /* Calculate displacement vector */
1027 dx00 = _mm_sub_ps(ix0,jx0);
1028 dy00 = _mm_sub_ps(iy0,jy0);
1029 dz00 = _mm_sub_ps(iz0,jz0);
1030 dx10 = _mm_sub_ps(ix1,jx0);
1031 dy10 = _mm_sub_ps(iy1,jy0);
1032 dz10 = _mm_sub_ps(iz1,jz0);
1033 dx20 = _mm_sub_ps(ix2,jx0);
1034 dy20 = _mm_sub_ps(iy2,jy0);
1035 dz20 = _mm_sub_ps(iz2,jz0);
1036
1037 /* Calculate squared distance and things based on it */
1038 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1039 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1040 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1041
1042 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1043 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1044 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1045
1046 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1047 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1048 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1049
1050 /* Load parameters for j particles */
1051 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1052 charge+jnrC+0,charge+jnrD+0);
1053
1054 fjx0 = _mm_setzero_ps();
1055 fjy0 = _mm_setzero_ps();
1056 fjz0 = _mm_setzero_ps();
1057
1058 /**************************
1059 * CALCULATE INTERACTIONS *
1060 **************************/
1061
1062 if (gmx_mm_any_lt(rsq00,rcutoff2))
1063 {
1064
1065 r00 = _mm_mul_ps(rsq00,rinv00);
1066 r00 = _mm_andnot_ps(dummy_mask,r00);
1067
1068 /* Compute parameters for interactions between i and j atoms */
1069 qq00 = _mm_mul_ps(iq0,jq0);
1070
1071 /* EWALD ELECTROSTATICS */
1072
1073 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1074 ewrt = _mm_mul_ps(r00,ewtabscale);
1075 ewitab = _mm_cvttps_epi32(ewrt);
1076 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1077 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1078 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1079 &ewtabF,&ewtabFn);
1080 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1081 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1082
1083 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1084
1085 fscal = felec;
1086
1087 fscal = _mm_and_ps(fscal,cutoff_mask);
1088
1089 fscal = _mm_andnot_ps(dummy_mask,fscal);
1090
1091 /* Calculate temporary vectorial force */
1092 tx = _mm_mul_ps(fscal,dx00);
1093 ty = _mm_mul_ps(fscal,dy00);
1094 tz = _mm_mul_ps(fscal,dz00);
1095
1096 /* Update vectorial force */
1097 fix0 = _mm_add_ps(fix0,tx);
1098 fiy0 = _mm_add_ps(fiy0,ty);
1099 fiz0 = _mm_add_ps(fiz0,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 }
1106
1107 /**************************
1108 * CALCULATE INTERACTIONS *
1109 **************************/
1110
1111 if (gmx_mm_any_lt(rsq10,rcutoff2))
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 /* EWALD ELECTROSTATICS */
1121
1122 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1123 ewrt = _mm_mul_ps(r10,ewtabscale);
1124 ewitab = _mm_cvttps_epi32(ewrt);
1125 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1126 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1127 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1128 &ewtabF,&ewtabFn);
1129 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1130 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1131
1132 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1133
1134 fscal = felec;
1135
1136 fscal = _mm_and_ps(fscal,cutoff_mask);
1137
1138 fscal = _mm_andnot_ps(dummy_mask,fscal);
1139
1140 /* Calculate temporary vectorial force */
1141 tx = _mm_mul_ps(fscal,dx10);
1142 ty = _mm_mul_ps(fscal,dy10);
1143 tz = _mm_mul_ps(fscal,dz10);
1144
1145 /* Update vectorial force */
1146 fix1 = _mm_add_ps(fix1,tx);
1147 fiy1 = _mm_add_ps(fiy1,ty);
1148 fiz1 = _mm_add_ps(fiz1,tz);
1149
1150 fjx0 = _mm_add_ps(fjx0,tx);
1151 fjy0 = _mm_add_ps(fjy0,ty);
1152 fjz0 = _mm_add_ps(fjz0,tz);
1153
1154 }
1155
1156 /**************************
1157 * CALCULATE INTERACTIONS *
1158 **************************/
1159
1160 if (gmx_mm_any_lt(rsq20,rcutoff2))
1161 {
1162
1163 r20 = _mm_mul_ps(rsq20,rinv20);
1164 r20 = _mm_andnot_ps(dummy_mask,r20);
1165
1166 /* Compute parameters for interactions between i and j atoms */
1167 qq20 = _mm_mul_ps(iq2,jq0);
1168
1169 /* EWALD ELECTROSTATICS */
1170
1171 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1172 ewrt = _mm_mul_ps(r20,ewtabscale);
1173 ewitab = _mm_cvttps_epi32(ewrt);
1174 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1175 gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1176 ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1177 &ewtabF,&ewtabFn);
1178 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1179 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1180
1181 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1182
1183 fscal = felec;
1184
1185 fscal = _mm_and_ps(fscal,cutoff_mask);
1186
1187 fscal = _mm_andnot_ps(dummy_mask,fscal);
1188
1189 /* Calculate temporary vectorial force */
1190 tx = _mm_mul_ps(fscal,dx20);
1191 ty = _mm_mul_ps(fscal,dy20);
1192 tz = _mm_mul_ps(fscal,dz20);
1193
1194 /* Update vectorial force */
1195 fix2 = _mm_add_ps(fix2,tx);
1196 fiy2 = _mm_add_ps(fiy2,ty);
1197 fiz2 = _mm_add_ps(fiz2,tz);
1198
1199 fjx0 = _mm_add_ps(fjx0,tx);
1200 fjy0 = _mm_add_ps(fjy0,ty);
1201 fjz0 = _mm_add_ps(fjz0,tz);
1202
1203 }
1204
1205 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1206 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1207 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1208 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1209
1210 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1211
1212 /* Inner loop uses 120 flops */
1213 }
1214
1215 /* End of innermost loop */
1216
1217 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1218 f+i_coord_offset,fshift+i_shift_offset);
1219
1220 /* Increment number of inner iterations */
1221 inneriter += j_index_end - j_index_start;
1222
1223 /* Outer loop uses 18 flops */
1224 }
1225
1226 /* Increment number of outer iterations */
1227 outeriter += nri;
1228
1229 /* Update outer/inner flops */
1230
1231 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*120);
1232 }
The ultimate molecular dynamics simulation package