2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, 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.
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.
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.
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.
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.
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.
36 * Note: this file was generated by the GROMACS sse2_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "gromacs/simd/math_x86_sse2_double.h"
48 #include "kernelutil_x86_sse2_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_sse2_double
52 * Electrostatics interaction: Ewald
53 * VdW interaction: None
54 * Geometry: Water3-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_sse2_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
83 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
86 int vdwjidx0A,vdwjidx0B;
87 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
88 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
89 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
90 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
91 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
94 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
96 __m128d dummy_mask,cutoff_mask;
97 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
98 __m128d one = _mm_set1_pd(1.0);
99 __m128d two = _mm_set1_pd(2.0);
105 jindex = nlist->jindex;
107 shiftidx = nlist->shift;
109 shiftvec = fr->shift_vec[0];
110 fshift = fr->fshift[0];
111 facel = _mm_set1_pd(fr->epsfac);
112 charge = mdatoms->chargeA;
114 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
115 ewtab = fr->ic->tabq_coul_FDV0;
116 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
117 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
119 /* Setup water-specific parameters */
120 inr = nlist->iinr[0];
121 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
122 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
123 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
125 /* Avoid stupid compiler warnings */
133 /* Start outer loop over neighborlists */
134 for(iidx=0; iidx<nri; iidx++)
136 /* Load shift vector for this list */
137 i_shift_offset = DIM*shiftidx[iidx];
139 /* Load limits for loop over neighbors */
140 j_index_start = jindex[iidx];
141 j_index_end = jindex[iidx+1];
143 /* Get outer coordinate index */
145 i_coord_offset = DIM*inr;
147 /* Load i particle coords and add shift vector */
148 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
149 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
151 fix0 = _mm_setzero_pd();
152 fiy0 = _mm_setzero_pd();
153 fiz0 = _mm_setzero_pd();
154 fix1 = _mm_setzero_pd();
155 fiy1 = _mm_setzero_pd();
156 fiz1 = _mm_setzero_pd();
157 fix2 = _mm_setzero_pd();
158 fiy2 = _mm_setzero_pd();
159 fiz2 = _mm_setzero_pd();
161 /* Reset potential sums */
162 velecsum = _mm_setzero_pd();
164 /* Start inner kernel loop */
165 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
168 /* Get j neighbor index, and coordinate index */
171 j_coord_offsetA = DIM*jnrA;
172 j_coord_offsetB = DIM*jnrB;
174 /* load j atom coordinates */
175 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
178 /* Calculate displacement vector */
179 dx00 = _mm_sub_pd(ix0,jx0);
180 dy00 = _mm_sub_pd(iy0,jy0);
181 dz00 = _mm_sub_pd(iz0,jz0);
182 dx10 = _mm_sub_pd(ix1,jx0);
183 dy10 = _mm_sub_pd(iy1,jy0);
184 dz10 = _mm_sub_pd(iz1,jz0);
185 dx20 = _mm_sub_pd(ix2,jx0);
186 dy20 = _mm_sub_pd(iy2,jy0);
187 dz20 = _mm_sub_pd(iz2,jz0);
189 /* Calculate squared distance and things based on it */
190 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
191 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
192 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
194 rinv00 = gmx_mm_invsqrt_pd(rsq00);
195 rinv10 = gmx_mm_invsqrt_pd(rsq10);
196 rinv20 = gmx_mm_invsqrt_pd(rsq20);
198 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
199 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
200 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
202 /* Load parameters for j particles */
203 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
205 fjx0 = _mm_setzero_pd();
206 fjy0 = _mm_setzero_pd();
207 fjz0 = _mm_setzero_pd();
209 /**************************
210 * CALCULATE INTERACTIONS *
211 **************************/
213 r00 = _mm_mul_pd(rsq00,rinv00);
215 /* Compute parameters for interactions between i and j atoms */
216 qq00 = _mm_mul_pd(iq0,jq0);
218 /* EWALD ELECTROSTATICS */
220 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
221 ewrt = _mm_mul_pd(r00,ewtabscale);
222 ewitab = _mm_cvttpd_epi32(ewrt);
223 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
224 ewitab = _mm_slli_epi32(ewitab,2);
225 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
226 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
227 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
228 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
229 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
230 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
231 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
232 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
233 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
234 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
236 /* Update potential sum for this i atom from the interaction with this j atom. */
237 velecsum = _mm_add_pd(velecsum,velec);
241 /* Calculate temporary vectorial force */
242 tx = _mm_mul_pd(fscal,dx00);
243 ty = _mm_mul_pd(fscal,dy00);
244 tz = _mm_mul_pd(fscal,dz00);
246 /* Update vectorial force */
247 fix0 = _mm_add_pd(fix0,tx);
248 fiy0 = _mm_add_pd(fiy0,ty);
249 fiz0 = _mm_add_pd(fiz0,tz);
251 fjx0 = _mm_add_pd(fjx0,tx);
252 fjy0 = _mm_add_pd(fjy0,ty);
253 fjz0 = _mm_add_pd(fjz0,tz);
255 /**************************
256 * CALCULATE INTERACTIONS *
257 **************************/
259 r10 = _mm_mul_pd(rsq10,rinv10);
261 /* Compute parameters for interactions between i and j atoms */
262 qq10 = _mm_mul_pd(iq1,jq0);
264 /* EWALD ELECTROSTATICS */
266 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
267 ewrt = _mm_mul_pd(r10,ewtabscale);
268 ewitab = _mm_cvttpd_epi32(ewrt);
269 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
270 ewitab = _mm_slli_epi32(ewitab,2);
271 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
272 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
273 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
274 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
275 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
276 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
277 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
278 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
279 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
280 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
282 /* Update potential sum for this i atom from the interaction with this j atom. */
283 velecsum = _mm_add_pd(velecsum,velec);
287 /* Calculate temporary vectorial force */
288 tx = _mm_mul_pd(fscal,dx10);
289 ty = _mm_mul_pd(fscal,dy10);
290 tz = _mm_mul_pd(fscal,dz10);
292 /* Update vectorial force */
293 fix1 = _mm_add_pd(fix1,tx);
294 fiy1 = _mm_add_pd(fiy1,ty);
295 fiz1 = _mm_add_pd(fiz1,tz);
297 fjx0 = _mm_add_pd(fjx0,tx);
298 fjy0 = _mm_add_pd(fjy0,ty);
299 fjz0 = _mm_add_pd(fjz0,tz);
301 /**************************
302 * CALCULATE INTERACTIONS *
303 **************************/
305 r20 = _mm_mul_pd(rsq20,rinv20);
307 /* Compute parameters for interactions between i and j atoms */
308 qq20 = _mm_mul_pd(iq2,jq0);
310 /* EWALD ELECTROSTATICS */
312 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
313 ewrt = _mm_mul_pd(r20,ewtabscale);
314 ewitab = _mm_cvttpd_epi32(ewrt);
315 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
316 ewitab = _mm_slli_epi32(ewitab,2);
317 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
318 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
319 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
320 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
321 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
322 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
323 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
324 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
325 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
326 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
328 /* Update potential sum for this i atom from the interaction with this j atom. */
329 velecsum = _mm_add_pd(velecsum,velec);
333 /* Calculate temporary vectorial force */
334 tx = _mm_mul_pd(fscal,dx20);
335 ty = _mm_mul_pd(fscal,dy20);
336 tz = _mm_mul_pd(fscal,dz20);
338 /* Update vectorial force */
339 fix2 = _mm_add_pd(fix2,tx);
340 fiy2 = _mm_add_pd(fiy2,ty);
341 fiz2 = _mm_add_pd(fiz2,tz);
343 fjx0 = _mm_add_pd(fjx0,tx);
344 fjy0 = _mm_add_pd(fjy0,ty);
345 fjz0 = _mm_add_pd(fjz0,tz);
347 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
349 /* Inner loop uses 126 flops */
356 j_coord_offsetA = DIM*jnrA;
358 /* load j atom coordinates */
359 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
362 /* Calculate displacement vector */
363 dx00 = _mm_sub_pd(ix0,jx0);
364 dy00 = _mm_sub_pd(iy0,jy0);
365 dz00 = _mm_sub_pd(iz0,jz0);
366 dx10 = _mm_sub_pd(ix1,jx0);
367 dy10 = _mm_sub_pd(iy1,jy0);
368 dz10 = _mm_sub_pd(iz1,jz0);
369 dx20 = _mm_sub_pd(ix2,jx0);
370 dy20 = _mm_sub_pd(iy2,jy0);
371 dz20 = _mm_sub_pd(iz2,jz0);
373 /* Calculate squared distance and things based on it */
374 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
375 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
376 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
378 rinv00 = gmx_mm_invsqrt_pd(rsq00);
379 rinv10 = gmx_mm_invsqrt_pd(rsq10);
380 rinv20 = gmx_mm_invsqrt_pd(rsq20);
382 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
383 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
384 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
386 /* Load parameters for j particles */
387 jq0 = _mm_load_sd(charge+jnrA+0);
389 fjx0 = _mm_setzero_pd();
390 fjy0 = _mm_setzero_pd();
391 fjz0 = _mm_setzero_pd();
393 /**************************
394 * CALCULATE INTERACTIONS *
395 **************************/
397 r00 = _mm_mul_pd(rsq00,rinv00);
399 /* Compute parameters for interactions between i and j atoms */
400 qq00 = _mm_mul_pd(iq0,jq0);
402 /* EWALD ELECTROSTATICS */
404 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
405 ewrt = _mm_mul_pd(r00,ewtabscale);
406 ewitab = _mm_cvttpd_epi32(ewrt);
407 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
408 ewitab = _mm_slli_epi32(ewitab,2);
409 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
410 ewtabD = _mm_setzero_pd();
411 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
412 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
413 ewtabFn = _mm_setzero_pd();
414 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
415 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
416 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
417 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
418 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
420 /* Update potential sum for this i atom from the interaction with this j atom. */
421 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
422 velecsum = _mm_add_pd(velecsum,velec);
426 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
428 /* Calculate temporary vectorial force */
429 tx = _mm_mul_pd(fscal,dx00);
430 ty = _mm_mul_pd(fscal,dy00);
431 tz = _mm_mul_pd(fscal,dz00);
433 /* Update vectorial force */
434 fix0 = _mm_add_pd(fix0,tx);
435 fiy0 = _mm_add_pd(fiy0,ty);
436 fiz0 = _mm_add_pd(fiz0,tz);
438 fjx0 = _mm_add_pd(fjx0,tx);
439 fjy0 = _mm_add_pd(fjy0,ty);
440 fjz0 = _mm_add_pd(fjz0,tz);
442 /**************************
443 * CALCULATE INTERACTIONS *
444 **************************/
446 r10 = _mm_mul_pd(rsq10,rinv10);
448 /* Compute parameters for interactions between i and j atoms */
449 qq10 = _mm_mul_pd(iq1,jq0);
451 /* EWALD ELECTROSTATICS */
453 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
454 ewrt = _mm_mul_pd(r10,ewtabscale);
455 ewitab = _mm_cvttpd_epi32(ewrt);
456 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
457 ewitab = _mm_slli_epi32(ewitab,2);
458 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
459 ewtabD = _mm_setzero_pd();
460 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
461 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
462 ewtabFn = _mm_setzero_pd();
463 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
464 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
465 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
466 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
467 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
469 /* Update potential sum for this i atom from the interaction with this j atom. */
470 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
471 velecsum = _mm_add_pd(velecsum,velec);
475 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
477 /* Calculate temporary vectorial force */
478 tx = _mm_mul_pd(fscal,dx10);
479 ty = _mm_mul_pd(fscal,dy10);
480 tz = _mm_mul_pd(fscal,dz10);
482 /* Update vectorial force */
483 fix1 = _mm_add_pd(fix1,tx);
484 fiy1 = _mm_add_pd(fiy1,ty);
485 fiz1 = _mm_add_pd(fiz1,tz);
487 fjx0 = _mm_add_pd(fjx0,tx);
488 fjy0 = _mm_add_pd(fjy0,ty);
489 fjz0 = _mm_add_pd(fjz0,tz);
491 /**************************
492 * CALCULATE INTERACTIONS *
493 **************************/
495 r20 = _mm_mul_pd(rsq20,rinv20);
497 /* Compute parameters for interactions between i and j atoms */
498 qq20 = _mm_mul_pd(iq2,jq0);
500 /* EWALD ELECTROSTATICS */
502 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
503 ewrt = _mm_mul_pd(r20,ewtabscale);
504 ewitab = _mm_cvttpd_epi32(ewrt);
505 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
506 ewitab = _mm_slli_epi32(ewitab,2);
507 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
508 ewtabD = _mm_setzero_pd();
509 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
510 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
511 ewtabFn = _mm_setzero_pd();
512 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
513 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
514 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
515 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
516 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
518 /* Update potential sum for this i atom from the interaction with this j atom. */
519 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
520 velecsum = _mm_add_pd(velecsum,velec);
524 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
526 /* Calculate temporary vectorial force */
527 tx = _mm_mul_pd(fscal,dx20);
528 ty = _mm_mul_pd(fscal,dy20);
529 tz = _mm_mul_pd(fscal,dz20);
531 /* Update vectorial force */
532 fix2 = _mm_add_pd(fix2,tx);
533 fiy2 = _mm_add_pd(fiy2,ty);
534 fiz2 = _mm_add_pd(fiz2,tz);
536 fjx0 = _mm_add_pd(fjx0,tx);
537 fjy0 = _mm_add_pd(fjy0,ty);
538 fjz0 = _mm_add_pd(fjz0,tz);
540 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
542 /* Inner loop uses 126 flops */
545 /* End of innermost loop */
547 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
548 f+i_coord_offset,fshift+i_shift_offset);
551 /* Update potential energies */
552 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
554 /* Increment number of inner iterations */
555 inneriter += j_index_end - j_index_start;
557 /* Outer loop uses 19 flops */
560 /* Increment number of outer iterations */
563 /* Update outer/inner flops */
565 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*126);
568 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3P1_F_sse2_double
569 * Electrostatics interaction: Ewald
570 * VdW interaction: None
571 * Geometry: Water3-Particle
572 * Calculate force/pot: Force
575 nb_kernel_ElecEw_VdwNone_GeomW3P1_F_sse2_double
576 (t_nblist * gmx_restrict nlist,
577 rvec * gmx_restrict xx,
578 rvec * gmx_restrict ff,
579 t_forcerec * gmx_restrict fr,
580 t_mdatoms * gmx_restrict mdatoms,
581 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
582 t_nrnb * gmx_restrict nrnb)
584 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
585 * just 0 for non-waters.
586 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
587 * jnr indices corresponding to data put in the four positions in the SIMD register.
589 int i_shift_offset,i_coord_offset,outeriter,inneriter;
590 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
592 int j_coord_offsetA,j_coord_offsetB;
593 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
595 real *shiftvec,*fshift,*x,*f;
596 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
598 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
600 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
602 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
603 int vdwjidx0A,vdwjidx0B;
604 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
605 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
606 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
607 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
608 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
611 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
613 __m128d dummy_mask,cutoff_mask;
614 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
615 __m128d one = _mm_set1_pd(1.0);
616 __m128d two = _mm_set1_pd(2.0);
622 jindex = nlist->jindex;
624 shiftidx = nlist->shift;
626 shiftvec = fr->shift_vec[0];
627 fshift = fr->fshift[0];
628 facel = _mm_set1_pd(fr->epsfac);
629 charge = mdatoms->chargeA;
631 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
632 ewtab = fr->ic->tabq_coul_F;
633 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
634 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
636 /* Setup water-specific parameters */
637 inr = nlist->iinr[0];
638 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
639 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
640 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
642 /* Avoid stupid compiler warnings */
650 /* Start outer loop over neighborlists */
651 for(iidx=0; iidx<nri; iidx++)
653 /* Load shift vector for this list */
654 i_shift_offset = DIM*shiftidx[iidx];
656 /* Load limits for loop over neighbors */
657 j_index_start = jindex[iidx];
658 j_index_end = jindex[iidx+1];
660 /* Get outer coordinate index */
662 i_coord_offset = DIM*inr;
664 /* Load i particle coords and add shift vector */
665 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
666 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
668 fix0 = _mm_setzero_pd();
669 fiy0 = _mm_setzero_pd();
670 fiz0 = _mm_setzero_pd();
671 fix1 = _mm_setzero_pd();
672 fiy1 = _mm_setzero_pd();
673 fiz1 = _mm_setzero_pd();
674 fix2 = _mm_setzero_pd();
675 fiy2 = _mm_setzero_pd();
676 fiz2 = _mm_setzero_pd();
678 /* Start inner kernel loop */
679 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
682 /* Get j neighbor index, and coordinate index */
685 j_coord_offsetA = DIM*jnrA;
686 j_coord_offsetB = DIM*jnrB;
688 /* load j atom coordinates */
689 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
692 /* Calculate displacement vector */
693 dx00 = _mm_sub_pd(ix0,jx0);
694 dy00 = _mm_sub_pd(iy0,jy0);
695 dz00 = _mm_sub_pd(iz0,jz0);
696 dx10 = _mm_sub_pd(ix1,jx0);
697 dy10 = _mm_sub_pd(iy1,jy0);
698 dz10 = _mm_sub_pd(iz1,jz0);
699 dx20 = _mm_sub_pd(ix2,jx0);
700 dy20 = _mm_sub_pd(iy2,jy0);
701 dz20 = _mm_sub_pd(iz2,jz0);
703 /* Calculate squared distance and things based on it */
704 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
705 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
706 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
708 rinv00 = gmx_mm_invsqrt_pd(rsq00);
709 rinv10 = gmx_mm_invsqrt_pd(rsq10);
710 rinv20 = gmx_mm_invsqrt_pd(rsq20);
712 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
713 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
714 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
716 /* Load parameters for j particles */
717 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
719 fjx0 = _mm_setzero_pd();
720 fjy0 = _mm_setzero_pd();
721 fjz0 = _mm_setzero_pd();
723 /**************************
724 * CALCULATE INTERACTIONS *
725 **************************/
727 r00 = _mm_mul_pd(rsq00,rinv00);
729 /* Compute parameters for interactions between i and j atoms */
730 qq00 = _mm_mul_pd(iq0,jq0);
732 /* EWALD ELECTROSTATICS */
734 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
735 ewrt = _mm_mul_pd(r00,ewtabscale);
736 ewitab = _mm_cvttpd_epi32(ewrt);
737 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
738 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
740 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
741 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
745 /* Calculate temporary vectorial force */
746 tx = _mm_mul_pd(fscal,dx00);
747 ty = _mm_mul_pd(fscal,dy00);
748 tz = _mm_mul_pd(fscal,dz00);
750 /* Update vectorial force */
751 fix0 = _mm_add_pd(fix0,tx);
752 fiy0 = _mm_add_pd(fiy0,ty);
753 fiz0 = _mm_add_pd(fiz0,tz);
755 fjx0 = _mm_add_pd(fjx0,tx);
756 fjy0 = _mm_add_pd(fjy0,ty);
757 fjz0 = _mm_add_pd(fjz0,tz);
759 /**************************
760 * CALCULATE INTERACTIONS *
761 **************************/
763 r10 = _mm_mul_pd(rsq10,rinv10);
765 /* Compute parameters for interactions between i and j atoms */
766 qq10 = _mm_mul_pd(iq1,jq0);
768 /* EWALD ELECTROSTATICS */
770 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
771 ewrt = _mm_mul_pd(r10,ewtabscale);
772 ewitab = _mm_cvttpd_epi32(ewrt);
773 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
774 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
776 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
777 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
781 /* Calculate temporary vectorial force */
782 tx = _mm_mul_pd(fscal,dx10);
783 ty = _mm_mul_pd(fscal,dy10);
784 tz = _mm_mul_pd(fscal,dz10);
786 /* Update vectorial force */
787 fix1 = _mm_add_pd(fix1,tx);
788 fiy1 = _mm_add_pd(fiy1,ty);
789 fiz1 = _mm_add_pd(fiz1,tz);
791 fjx0 = _mm_add_pd(fjx0,tx);
792 fjy0 = _mm_add_pd(fjy0,ty);
793 fjz0 = _mm_add_pd(fjz0,tz);
795 /**************************
796 * CALCULATE INTERACTIONS *
797 **************************/
799 r20 = _mm_mul_pd(rsq20,rinv20);
801 /* Compute parameters for interactions between i and j atoms */
802 qq20 = _mm_mul_pd(iq2,jq0);
804 /* EWALD ELECTROSTATICS */
806 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
807 ewrt = _mm_mul_pd(r20,ewtabscale);
808 ewitab = _mm_cvttpd_epi32(ewrt);
809 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
810 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
812 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
813 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
817 /* Calculate temporary vectorial force */
818 tx = _mm_mul_pd(fscal,dx20);
819 ty = _mm_mul_pd(fscal,dy20);
820 tz = _mm_mul_pd(fscal,dz20);
822 /* Update vectorial force */
823 fix2 = _mm_add_pd(fix2,tx);
824 fiy2 = _mm_add_pd(fiy2,ty);
825 fiz2 = _mm_add_pd(fiz2,tz);
827 fjx0 = _mm_add_pd(fjx0,tx);
828 fjy0 = _mm_add_pd(fjy0,ty);
829 fjz0 = _mm_add_pd(fjz0,tz);
831 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
833 /* Inner loop uses 111 flops */
840 j_coord_offsetA = DIM*jnrA;
842 /* load j atom coordinates */
843 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
846 /* Calculate displacement vector */
847 dx00 = _mm_sub_pd(ix0,jx0);
848 dy00 = _mm_sub_pd(iy0,jy0);
849 dz00 = _mm_sub_pd(iz0,jz0);
850 dx10 = _mm_sub_pd(ix1,jx0);
851 dy10 = _mm_sub_pd(iy1,jy0);
852 dz10 = _mm_sub_pd(iz1,jz0);
853 dx20 = _mm_sub_pd(ix2,jx0);
854 dy20 = _mm_sub_pd(iy2,jy0);
855 dz20 = _mm_sub_pd(iz2,jz0);
857 /* Calculate squared distance and things based on it */
858 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
859 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
860 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
862 rinv00 = gmx_mm_invsqrt_pd(rsq00);
863 rinv10 = gmx_mm_invsqrt_pd(rsq10);
864 rinv20 = gmx_mm_invsqrt_pd(rsq20);
866 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
867 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
868 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
870 /* Load parameters for j particles */
871 jq0 = _mm_load_sd(charge+jnrA+0);
873 fjx0 = _mm_setzero_pd();
874 fjy0 = _mm_setzero_pd();
875 fjz0 = _mm_setzero_pd();
877 /**************************
878 * CALCULATE INTERACTIONS *
879 **************************/
881 r00 = _mm_mul_pd(rsq00,rinv00);
883 /* Compute parameters for interactions between i and j atoms */
884 qq00 = _mm_mul_pd(iq0,jq0);
886 /* EWALD ELECTROSTATICS */
888 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
889 ewrt = _mm_mul_pd(r00,ewtabscale);
890 ewitab = _mm_cvttpd_epi32(ewrt);
891 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
892 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
893 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
894 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
898 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
900 /* Calculate temporary vectorial force */
901 tx = _mm_mul_pd(fscal,dx00);
902 ty = _mm_mul_pd(fscal,dy00);
903 tz = _mm_mul_pd(fscal,dz00);
905 /* Update vectorial force */
906 fix0 = _mm_add_pd(fix0,tx);
907 fiy0 = _mm_add_pd(fiy0,ty);
908 fiz0 = _mm_add_pd(fiz0,tz);
910 fjx0 = _mm_add_pd(fjx0,tx);
911 fjy0 = _mm_add_pd(fjy0,ty);
912 fjz0 = _mm_add_pd(fjz0,tz);
914 /**************************
915 * CALCULATE INTERACTIONS *
916 **************************/
918 r10 = _mm_mul_pd(rsq10,rinv10);
920 /* Compute parameters for interactions between i and j atoms */
921 qq10 = _mm_mul_pd(iq1,jq0);
923 /* EWALD ELECTROSTATICS */
925 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
926 ewrt = _mm_mul_pd(r10,ewtabscale);
927 ewitab = _mm_cvttpd_epi32(ewrt);
928 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
929 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
930 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
931 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
935 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
937 /* Calculate temporary vectorial force */
938 tx = _mm_mul_pd(fscal,dx10);
939 ty = _mm_mul_pd(fscal,dy10);
940 tz = _mm_mul_pd(fscal,dz10);
942 /* Update vectorial force */
943 fix1 = _mm_add_pd(fix1,tx);
944 fiy1 = _mm_add_pd(fiy1,ty);
945 fiz1 = _mm_add_pd(fiz1,tz);
947 fjx0 = _mm_add_pd(fjx0,tx);
948 fjy0 = _mm_add_pd(fjy0,ty);
949 fjz0 = _mm_add_pd(fjz0,tz);
951 /**************************
952 * CALCULATE INTERACTIONS *
953 **************************/
955 r20 = _mm_mul_pd(rsq20,rinv20);
957 /* Compute parameters for interactions between i and j atoms */
958 qq20 = _mm_mul_pd(iq2,jq0);
960 /* EWALD ELECTROSTATICS */
962 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
963 ewrt = _mm_mul_pd(r20,ewtabscale);
964 ewitab = _mm_cvttpd_epi32(ewrt);
965 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
966 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
967 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
968 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
972 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
974 /* Calculate temporary vectorial force */
975 tx = _mm_mul_pd(fscal,dx20);
976 ty = _mm_mul_pd(fscal,dy20);
977 tz = _mm_mul_pd(fscal,dz20);
979 /* Update vectorial force */
980 fix2 = _mm_add_pd(fix2,tx);
981 fiy2 = _mm_add_pd(fiy2,ty);
982 fiz2 = _mm_add_pd(fiz2,tz);
984 fjx0 = _mm_add_pd(fjx0,tx);
985 fjy0 = _mm_add_pd(fjy0,ty);
986 fjz0 = _mm_add_pd(fjz0,tz);
988 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
990 /* Inner loop uses 111 flops */
993 /* End of innermost loop */
995 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
996 f+i_coord_offset,fshift+i_shift_offset);
998 /* Increment number of inner iterations */
999 inneriter += j_index_end - j_index_start;
1001 /* Outer loop uses 18 flops */
1004 /* Increment number of outer iterations */
1007 /* Update outer/inner flops */
1009 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*111);