2 * Note: this file was generated by the Gromacs sse2_double kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_sse2_double.h"
34 #include "kernelutil_x86_sse2_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_sse2_double
38 * Electrostatics interaction: Ewald
39 * VdW interaction: None
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_sse2_double
45 (t_nblist * gmx_restrict nlist,
46 rvec * gmx_restrict xx,
47 rvec * gmx_restrict ff,
48 t_forcerec * gmx_restrict fr,
49 t_mdatoms * gmx_restrict mdatoms,
50 nb_kernel_data_t * gmx_restrict kernel_data,
51 t_nrnb * gmx_restrict nrnb)
53 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
54 * just 0 for non-waters.
55 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
56 * jnr indices corresponding to data put in the four positions in the SIMD register.
58 int i_shift_offset,i_coord_offset,outeriter,inneriter;
59 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
61 int j_coord_offsetA,j_coord_offsetB;
62 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
64 real *shiftvec,*fshift,*x,*f;
65 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
67 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
69 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
71 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
72 int vdwjidx0A,vdwjidx0B;
73 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
74 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
75 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
76 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
77 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
80 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
82 __m128d dummy_mask,cutoff_mask;
83 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
84 __m128d one = _mm_set1_pd(1.0);
85 __m128d two = _mm_set1_pd(2.0);
91 jindex = nlist->jindex;
93 shiftidx = nlist->shift;
95 shiftvec = fr->shift_vec[0];
96 fshift = fr->fshift[0];
97 facel = _mm_set1_pd(fr->epsfac);
98 charge = mdatoms->chargeA;
100 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
101 ewtab = fr->ic->tabq_coul_FDV0;
102 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
103 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
105 /* Setup water-specific parameters */
106 inr = nlist->iinr[0];
107 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
108 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
109 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
111 /* Avoid stupid compiler warnings */
119 /* Start outer loop over neighborlists */
120 for(iidx=0; iidx<nri; iidx++)
122 /* Load shift vector for this list */
123 i_shift_offset = DIM*shiftidx[iidx];
125 /* Load limits for loop over neighbors */
126 j_index_start = jindex[iidx];
127 j_index_end = jindex[iidx+1];
129 /* Get outer coordinate index */
131 i_coord_offset = DIM*inr;
133 /* Load i particle coords and add shift vector */
134 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
135 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
137 fix0 = _mm_setzero_pd();
138 fiy0 = _mm_setzero_pd();
139 fiz0 = _mm_setzero_pd();
140 fix1 = _mm_setzero_pd();
141 fiy1 = _mm_setzero_pd();
142 fiz1 = _mm_setzero_pd();
143 fix2 = _mm_setzero_pd();
144 fiy2 = _mm_setzero_pd();
145 fiz2 = _mm_setzero_pd();
147 /* Reset potential sums */
148 velecsum = _mm_setzero_pd();
150 /* Start inner kernel loop */
151 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
154 /* Get j neighbor index, and coordinate index */
157 j_coord_offsetA = DIM*jnrA;
158 j_coord_offsetB = DIM*jnrB;
160 /* load j atom coordinates */
161 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
164 /* Calculate displacement vector */
165 dx00 = _mm_sub_pd(ix0,jx0);
166 dy00 = _mm_sub_pd(iy0,jy0);
167 dz00 = _mm_sub_pd(iz0,jz0);
168 dx10 = _mm_sub_pd(ix1,jx0);
169 dy10 = _mm_sub_pd(iy1,jy0);
170 dz10 = _mm_sub_pd(iz1,jz0);
171 dx20 = _mm_sub_pd(ix2,jx0);
172 dy20 = _mm_sub_pd(iy2,jy0);
173 dz20 = _mm_sub_pd(iz2,jz0);
175 /* Calculate squared distance and things based on it */
176 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
177 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
178 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
180 rinv00 = gmx_mm_invsqrt_pd(rsq00);
181 rinv10 = gmx_mm_invsqrt_pd(rsq10);
182 rinv20 = gmx_mm_invsqrt_pd(rsq20);
184 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
185 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
186 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
188 /* Load parameters for j particles */
189 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
191 fjx0 = _mm_setzero_pd();
192 fjy0 = _mm_setzero_pd();
193 fjz0 = _mm_setzero_pd();
195 /**************************
196 * CALCULATE INTERACTIONS *
197 **************************/
199 r00 = _mm_mul_pd(rsq00,rinv00);
201 /* Compute parameters for interactions between i and j atoms */
202 qq00 = _mm_mul_pd(iq0,jq0);
204 /* EWALD ELECTROSTATICS */
206 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
207 ewrt = _mm_mul_pd(r00,ewtabscale);
208 ewitab = _mm_cvttpd_epi32(ewrt);
209 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
210 ewitab = _mm_slli_epi32(ewitab,2);
211 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
212 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
213 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
214 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
215 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
216 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
217 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
218 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
219 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
220 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
222 /* Update potential sum for this i atom from the interaction with this j atom. */
223 velecsum = _mm_add_pd(velecsum,velec);
227 /* Calculate temporary vectorial force */
228 tx = _mm_mul_pd(fscal,dx00);
229 ty = _mm_mul_pd(fscal,dy00);
230 tz = _mm_mul_pd(fscal,dz00);
232 /* Update vectorial force */
233 fix0 = _mm_add_pd(fix0,tx);
234 fiy0 = _mm_add_pd(fiy0,ty);
235 fiz0 = _mm_add_pd(fiz0,tz);
237 fjx0 = _mm_add_pd(fjx0,tx);
238 fjy0 = _mm_add_pd(fjy0,ty);
239 fjz0 = _mm_add_pd(fjz0,tz);
241 /**************************
242 * CALCULATE INTERACTIONS *
243 **************************/
245 r10 = _mm_mul_pd(rsq10,rinv10);
247 /* Compute parameters for interactions between i and j atoms */
248 qq10 = _mm_mul_pd(iq1,jq0);
250 /* EWALD ELECTROSTATICS */
252 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
253 ewrt = _mm_mul_pd(r10,ewtabscale);
254 ewitab = _mm_cvttpd_epi32(ewrt);
255 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
256 ewitab = _mm_slli_epi32(ewitab,2);
257 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
258 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
259 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
260 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
261 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
262 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
263 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
264 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
265 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
266 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
268 /* Update potential sum for this i atom from the interaction with this j atom. */
269 velecsum = _mm_add_pd(velecsum,velec);
273 /* Calculate temporary vectorial force */
274 tx = _mm_mul_pd(fscal,dx10);
275 ty = _mm_mul_pd(fscal,dy10);
276 tz = _mm_mul_pd(fscal,dz10);
278 /* Update vectorial force */
279 fix1 = _mm_add_pd(fix1,tx);
280 fiy1 = _mm_add_pd(fiy1,ty);
281 fiz1 = _mm_add_pd(fiz1,tz);
283 fjx0 = _mm_add_pd(fjx0,tx);
284 fjy0 = _mm_add_pd(fjy0,ty);
285 fjz0 = _mm_add_pd(fjz0,tz);
287 /**************************
288 * CALCULATE INTERACTIONS *
289 **************************/
291 r20 = _mm_mul_pd(rsq20,rinv20);
293 /* Compute parameters for interactions between i and j atoms */
294 qq20 = _mm_mul_pd(iq2,jq0);
296 /* EWALD ELECTROSTATICS */
298 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
299 ewrt = _mm_mul_pd(r20,ewtabscale);
300 ewitab = _mm_cvttpd_epi32(ewrt);
301 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
302 ewitab = _mm_slli_epi32(ewitab,2);
303 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
304 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
305 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
306 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
307 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
308 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
309 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
310 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
311 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
312 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
314 /* Update potential sum for this i atom from the interaction with this j atom. */
315 velecsum = _mm_add_pd(velecsum,velec);
319 /* Calculate temporary vectorial force */
320 tx = _mm_mul_pd(fscal,dx20);
321 ty = _mm_mul_pd(fscal,dy20);
322 tz = _mm_mul_pd(fscal,dz20);
324 /* Update vectorial force */
325 fix2 = _mm_add_pd(fix2,tx);
326 fiy2 = _mm_add_pd(fiy2,ty);
327 fiz2 = _mm_add_pd(fiz2,tz);
329 fjx0 = _mm_add_pd(fjx0,tx);
330 fjy0 = _mm_add_pd(fjy0,ty);
331 fjz0 = _mm_add_pd(fjz0,tz);
333 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
335 /* Inner loop uses 126 flops */
342 j_coord_offsetA = DIM*jnrA;
344 /* load j atom coordinates */
345 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
348 /* Calculate displacement vector */
349 dx00 = _mm_sub_pd(ix0,jx0);
350 dy00 = _mm_sub_pd(iy0,jy0);
351 dz00 = _mm_sub_pd(iz0,jz0);
352 dx10 = _mm_sub_pd(ix1,jx0);
353 dy10 = _mm_sub_pd(iy1,jy0);
354 dz10 = _mm_sub_pd(iz1,jz0);
355 dx20 = _mm_sub_pd(ix2,jx0);
356 dy20 = _mm_sub_pd(iy2,jy0);
357 dz20 = _mm_sub_pd(iz2,jz0);
359 /* Calculate squared distance and things based on it */
360 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
361 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
362 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
364 rinv00 = gmx_mm_invsqrt_pd(rsq00);
365 rinv10 = gmx_mm_invsqrt_pd(rsq10);
366 rinv20 = gmx_mm_invsqrt_pd(rsq20);
368 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
369 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
370 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
372 /* Load parameters for j particles */
373 jq0 = _mm_load_sd(charge+jnrA+0);
375 fjx0 = _mm_setzero_pd();
376 fjy0 = _mm_setzero_pd();
377 fjz0 = _mm_setzero_pd();
379 /**************************
380 * CALCULATE INTERACTIONS *
381 **************************/
383 r00 = _mm_mul_pd(rsq00,rinv00);
385 /* Compute parameters for interactions between i and j atoms */
386 qq00 = _mm_mul_pd(iq0,jq0);
388 /* EWALD ELECTROSTATICS */
390 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
391 ewrt = _mm_mul_pd(r00,ewtabscale);
392 ewitab = _mm_cvttpd_epi32(ewrt);
393 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
394 ewitab = _mm_slli_epi32(ewitab,2);
395 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
396 ewtabD = _mm_setzero_pd();
397 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
398 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
399 ewtabFn = _mm_setzero_pd();
400 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
401 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
402 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
403 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
404 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
406 /* Update potential sum for this i atom from the interaction with this j atom. */
407 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
408 velecsum = _mm_add_pd(velecsum,velec);
412 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
414 /* Calculate temporary vectorial force */
415 tx = _mm_mul_pd(fscal,dx00);
416 ty = _mm_mul_pd(fscal,dy00);
417 tz = _mm_mul_pd(fscal,dz00);
419 /* Update vectorial force */
420 fix0 = _mm_add_pd(fix0,tx);
421 fiy0 = _mm_add_pd(fiy0,ty);
422 fiz0 = _mm_add_pd(fiz0,tz);
424 fjx0 = _mm_add_pd(fjx0,tx);
425 fjy0 = _mm_add_pd(fjy0,ty);
426 fjz0 = _mm_add_pd(fjz0,tz);
428 /**************************
429 * CALCULATE INTERACTIONS *
430 **************************/
432 r10 = _mm_mul_pd(rsq10,rinv10);
434 /* Compute parameters for interactions between i and j atoms */
435 qq10 = _mm_mul_pd(iq1,jq0);
437 /* EWALD ELECTROSTATICS */
439 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
440 ewrt = _mm_mul_pd(r10,ewtabscale);
441 ewitab = _mm_cvttpd_epi32(ewrt);
442 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
443 ewitab = _mm_slli_epi32(ewitab,2);
444 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
445 ewtabD = _mm_setzero_pd();
446 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
447 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
448 ewtabFn = _mm_setzero_pd();
449 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
450 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
451 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
452 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
453 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
455 /* Update potential sum for this i atom from the interaction with this j atom. */
456 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
457 velecsum = _mm_add_pd(velecsum,velec);
461 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
463 /* Calculate temporary vectorial force */
464 tx = _mm_mul_pd(fscal,dx10);
465 ty = _mm_mul_pd(fscal,dy10);
466 tz = _mm_mul_pd(fscal,dz10);
468 /* Update vectorial force */
469 fix1 = _mm_add_pd(fix1,tx);
470 fiy1 = _mm_add_pd(fiy1,ty);
471 fiz1 = _mm_add_pd(fiz1,tz);
473 fjx0 = _mm_add_pd(fjx0,tx);
474 fjy0 = _mm_add_pd(fjy0,ty);
475 fjz0 = _mm_add_pd(fjz0,tz);
477 /**************************
478 * CALCULATE INTERACTIONS *
479 **************************/
481 r20 = _mm_mul_pd(rsq20,rinv20);
483 /* Compute parameters for interactions between i and j atoms */
484 qq20 = _mm_mul_pd(iq2,jq0);
486 /* EWALD ELECTROSTATICS */
488 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
489 ewrt = _mm_mul_pd(r20,ewtabscale);
490 ewitab = _mm_cvttpd_epi32(ewrt);
491 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
492 ewitab = _mm_slli_epi32(ewitab,2);
493 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
494 ewtabD = _mm_setzero_pd();
495 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
496 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
497 ewtabFn = _mm_setzero_pd();
498 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
499 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
500 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
501 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
502 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
504 /* Update potential sum for this i atom from the interaction with this j atom. */
505 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
506 velecsum = _mm_add_pd(velecsum,velec);
510 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
512 /* Calculate temporary vectorial force */
513 tx = _mm_mul_pd(fscal,dx20);
514 ty = _mm_mul_pd(fscal,dy20);
515 tz = _mm_mul_pd(fscal,dz20);
517 /* Update vectorial force */
518 fix2 = _mm_add_pd(fix2,tx);
519 fiy2 = _mm_add_pd(fiy2,ty);
520 fiz2 = _mm_add_pd(fiz2,tz);
522 fjx0 = _mm_add_pd(fjx0,tx);
523 fjy0 = _mm_add_pd(fjy0,ty);
524 fjz0 = _mm_add_pd(fjz0,tz);
526 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
528 /* Inner loop uses 126 flops */
531 /* End of innermost loop */
533 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
534 f+i_coord_offset,fshift+i_shift_offset);
537 /* Update potential energies */
538 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
540 /* Increment number of inner iterations */
541 inneriter += j_index_end - j_index_start;
543 /* Outer loop uses 19 flops */
546 /* Increment number of outer iterations */
549 /* Update outer/inner flops */
551 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*126);
554 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwNone_GeomW3P1_F_sse2_double
555 * Electrostatics interaction: Ewald
556 * VdW interaction: None
557 * Geometry: Water3-Particle
558 * Calculate force/pot: Force
561 nb_kernel_ElecEw_VdwNone_GeomW3P1_F_sse2_double
562 (t_nblist * gmx_restrict nlist,
563 rvec * gmx_restrict xx,
564 rvec * gmx_restrict ff,
565 t_forcerec * gmx_restrict fr,
566 t_mdatoms * gmx_restrict mdatoms,
567 nb_kernel_data_t * gmx_restrict kernel_data,
568 t_nrnb * gmx_restrict nrnb)
570 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
571 * just 0 for non-waters.
572 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
573 * jnr indices corresponding to data put in the four positions in the SIMD register.
575 int i_shift_offset,i_coord_offset,outeriter,inneriter;
576 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
578 int j_coord_offsetA,j_coord_offsetB;
579 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
581 real *shiftvec,*fshift,*x,*f;
582 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
584 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
586 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
588 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
589 int vdwjidx0A,vdwjidx0B;
590 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
591 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
592 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
593 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
594 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
597 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
599 __m128d dummy_mask,cutoff_mask;
600 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
601 __m128d one = _mm_set1_pd(1.0);
602 __m128d two = _mm_set1_pd(2.0);
608 jindex = nlist->jindex;
610 shiftidx = nlist->shift;
612 shiftvec = fr->shift_vec[0];
613 fshift = fr->fshift[0];
614 facel = _mm_set1_pd(fr->epsfac);
615 charge = mdatoms->chargeA;
617 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
618 ewtab = fr->ic->tabq_coul_F;
619 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
620 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
622 /* Setup water-specific parameters */
623 inr = nlist->iinr[0];
624 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
625 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
626 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
628 /* Avoid stupid compiler warnings */
636 /* Start outer loop over neighborlists */
637 for(iidx=0; iidx<nri; iidx++)
639 /* Load shift vector for this list */
640 i_shift_offset = DIM*shiftidx[iidx];
642 /* Load limits for loop over neighbors */
643 j_index_start = jindex[iidx];
644 j_index_end = jindex[iidx+1];
646 /* Get outer coordinate index */
648 i_coord_offset = DIM*inr;
650 /* Load i particle coords and add shift vector */
651 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
652 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
654 fix0 = _mm_setzero_pd();
655 fiy0 = _mm_setzero_pd();
656 fiz0 = _mm_setzero_pd();
657 fix1 = _mm_setzero_pd();
658 fiy1 = _mm_setzero_pd();
659 fiz1 = _mm_setzero_pd();
660 fix2 = _mm_setzero_pd();
661 fiy2 = _mm_setzero_pd();
662 fiz2 = _mm_setzero_pd();
664 /* Start inner kernel loop */
665 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
668 /* Get j neighbor index, and coordinate index */
671 j_coord_offsetA = DIM*jnrA;
672 j_coord_offsetB = DIM*jnrB;
674 /* load j atom coordinates */
675 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
678 /* Calculate displacement vector */
679 dx00 = _mm_sub_pd(ix0,jx0);
680 dy00 = _mm_sub_pd(iy0,jy0);
681 dz00 = _mm_sub_pd(iz0,jz0);
682 dx10 = _mm_sub_pd(ix1,jx0);
683 dy10 = _mm_sub_pd(iy1,jy0);
684 dz10 = _mm_sub_pd(iz1,jz0);
685 dx20 = _mm_sub_pd(ix2,jx0);
686 dy20 = _mm_sub_pd(iy2,jy0);
687 dz20 = _mm_sub_pd(iz2,jz0);
689 /* Calculate squared distance and things based on it */
690 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
691 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
692 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
694 rinv00 = gmx_mm_invsqrt_pd(rsq00);
695 rinv10 = gmx_mm_invsqrt_pd(rsq10);
696 rinv20 = gmx_mm_invsqrt_pd(rsq20);
698 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
699 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
700 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
702 /* Load parameters for j particles */
703 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
705 fjx0 = _mm_setzero_pd();
706 fjy0 = _mm_setzero_pd();
707 fjz0 = _mm_setzero_pd();
709 /**************************
710 * CALCULATE INTERACTIONS *
711 **************************/
713 r00 = _mm_mul_pd(rsq00,rinv00);
715 /* Compute parameters for interactions between i and j atoms */
716 qq00 = _mm_mul_pd(iq0,jq0);
718 /* EWALD ELECTROSTATICS */
720 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
721 ewrt = _mm_mul_pd(r00,ewtabscale);
722 ewitab = _mm_cvttpd_epi32(ewrt);
723 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
724 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
726 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
727 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
731 /* Calculate temporary vectorial force */
732 tx = _mm_mul_pd(fscal,dx00);
733 ty = _mm_mul_pd(fscal,dy00);
734 tz = _mm_mul_pd(fscal,dz00);
736 /* Update vectorial force */
737 fix0 = _mm_add_pd(fix0,tx);
738 fiy0 = _mm_add_pd(fiy0,ty);
739 fiz0 = _mm_add_pd(fiz0,tz);
741 fjx0 = _mm_add_pd(fjx0,tx);
742 fjy0 = _mm_add_pd(fjy0,ty);
743 fjz0 = _mm_add_pd(fjz0,tz);
745 /**************************
746 * CALCULATE INTERACTIONS *
747 **************************/
749 r10 = _mm_mul_pd(rsq10,rinv10);
751 /* Compute parameters for interactions between i and j atoms */
752 qq10 = _mm_mul_pd(iq1,jq0);
754 /* EWALD ELECTROSTATICS */
756 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
757 ewrt = _mm_mul_pd(r10,ewtabscale);
758 ewitab = _mm_cvttpd_epi32(ewrt);
759 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
760 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
762 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
763 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
767 /* Calculate temporary vectorial force */
768 tx = _mm_mul_pd(fscal,dx10);
769 ty = _mm_mul_pd(fscal,dy10);
770 tz = _mm_mul_pd(fscal,dz10);
772 /* Update vectorial force */
773 fix1 = _mm_add_pd(fix1,tx);
774 fiy1 = _mm_add_pd(fiy1,ty);
775 fiz1 = _mm_add_pd(fiz1,tz);
777 fjx0 = _mm_add_pd(fjx0,tx);
778 fjy0 = _mm_add_pd(fjy0,ty);
779 fjz0 = _mm_add_pd(fjz0,tz);
781 /**************************
782 * CALCULATE INTERACTIONS *
783 **************************/
785 r20 = _mm_mul_pd(rsq20,rinv20);
787 /* Compute parameters for interactions between i and j atoms */
788 qq20 = _mm_mul_pd(iq2,jq0);
790 /* EWALD ELECTROSTATICS */
792 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
793 ewrt = _mm_mul_pd(r20,ewtabscale);
794 ewitab = _mm_cvttpd_epi32(ewrt);
795 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
796 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
798 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
799 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
803 /* Calculate temporary vectorial force */
804 tx = _mm_mul_pd(fscal,dx20);
805 ty = _mm_mul_pd(fscal,dy20);
806 tz = _mm_mul_pd(fscal,dz20);
808 /* Update vectorial force */
809 fix2 = _mm_add_pd(fix2,tx);
810 fiy2 = _mm_add_pd(fiy2,ty);
811 fiz2 = _mm_add_pd(fiz2,tz);
813 fjx0 = _mm_add_pd(fjx0,tx);
814 fjy0 = _mm_add_pd(fjy0,ty);
815 fjz0 = _mm_add_pd(fjz0,tz);
817 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
819 /* Inner loop uses 111 flops */
826 j_coord_offsetA = DIM*jnrA;
828 /* load j atom coordinates */
829 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
832 /* Calculate displacement vector */
833 dx00 = _mm_sub_pd(ix0,jx0);
834 dy00 = _mm_sub_pd(iy0,jy0);
835 dz00 = _mm_sub_pd(iz0,jz0);
836 dx10 = _mm_sub_pd(ix1,jx0);
837 dy10 = _mm_sub_pd(iy1,jy0);
838 dz10 = _mm_sub_pd(iz1,jz0);
839 dx20 = _mm_sub_pd(ix2,jx0);
840 dy20 = _mm_sub_pd(iy2,jy0);
841 dz20 = _mm_sub_pd(iz2,jz0);
843 /* Calculate squared distance and things based on it */
844 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
845 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
846 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
848 rinv00 = gmx_mm_invsqrt_pd(rsq00);
849 rinv10 = gmx_mm_invsqrt_pd(rsq10);
850 rinv20 = gmx_mm_invsqrt_pd(rsq20);
852 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
853 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
854 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
856 /* Load parameters for j particles */
857 jq0 = _mm_load_sd(charge+jnrA+0);
859 fjx0 = _mm_setzero_pd();
860 fjy0 = _mm_setzero_pd();
861 fjz0 = _mm_setzero_pd();
863 /**************************
864 * CALCULATE INTERACTIONS *
865 **************************/
867 r00 = _mm_mul_pd(rsq00,rinv00);
869 /* Compute parameters for interactions between i and j atoms */
870 qq00 = _mm_mul_pd(iq0,jq0);
872 /* EWALD ELECTROSTATICS */
874 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
875 ewrt = _mm_mul_pd(r00,ewtabscale);
876 ewitab = _mm_cvttpd_epi32(ewrt);
877 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
878 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
879 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
880 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
884 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
886 /* Calculate temporary vectorial force */
887 tx = _mm_mul_pd(fscal,dx00);
888 ty = _mm_mul_pd(fscal,dy00);
889 tz = _mm_mul_pd(fscal,dz00);
891 /* Update vectorial force */
892 fix0 = _mm_add_pd(fix0,tx);
893 fiy0 = _mm_add_pd(fiy0,ty);
894 fiz0 = _mm_add_pd(fiz0,tz);
896 fjx0 = _mm_add_pd(fjx0,tx);
897 fjy0 = _mm_add_pd(fjy0,ty);
898 fjz0 = _mm_add_pd(fjz0,tz);
900 /**************************
901 * CALCULATE INTERACTIONS *
902 **************************/
904 r10 = _mm_mul_pd(rsq10,rinv10);
906 /* Compute parameters for interactions between i and j atoms */
907 qq10 = _mm_mul_pd(iq1,jq0);
909 /* EWALD ELECTROSTATICS */
911 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
912 ewrt = _mm_mul_pd(r10,ewtabscale);
913 ewitab = _mm_cvttpd_epi32(ewrt);
914 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
915 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
916 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
917 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
921 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
923 /* Calculate temporary vectorial force */
924 tx = _mm_mul_pd(fscal,dx10);
925 ty = _mm_mul_pd(fscal,dy10);
926 tz = _mm_mul_pd(fscal,dz10);
928 /* Update vectorial force */
929 fix1 = _mm_add_pd(fix1,tx);
930 fiy1 = _mm_add_pd(fiy1,ty);
931 fiz1 = _mm_add_pd(fiz1,tz);
933 fjx0 = _mm_add_pd(fjx0,tx);
934 fjy0 = _mm_add_pd(fjy0,ty);
935 fjz0 = _mm_add_pd(fjz0,tz);
937 /**************************
938 * CALCULATE INTERACTIONS *
939 **************************/
941 r20 = _mm_mul_pd(rsq20,rinv20);
943 /* Compute parameters for interactions between i and j atoms */
944 qq20 = _mm_mul_pd(iq2,jq0);
946 /* EWALD ELECTROSTATICS */
948 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
949 ewrt = _mm_mul_pd(r20,ewtabscale);
950 ewitab = _mm_cvttpd_epi32(ewrt);
951 eweps = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
952 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
953 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
954 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
958 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
960 /* Calculate temporary vectorial force */
961 tx = _mm_mul_pd(fscal,dx20);
962 ty = _mm_mul_pd(fscal,dy20);
963 tz = _mm_mul_pd(fscal,dz20);
965 /* Update vectorial force */
966 fix2 = _mm_add_pd(fix2,tx);
967 fiy2 = _mm_add_pd(fiy2,ty);
968 fiz2 = _mm_add_pd(fiz2,tz);
970 fjx0 = _mm_add_pd(fjx0,tx);
971 fjy0 = _mm_add_pd(fjy0,ty);
972 fjz0 = _mm_add_pd(fjz0,tz);
974 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
976 /* Inner loop uses 111 flops */
979 /* End of innermost loop */
981 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
982 f+i_coord_offset,fshift+i_shift_offset);
984 /* Increment number of inner iterations */
985 inneriter += j_index_end - j_index_start;
987 /* Outer loop uses 18 flops */
990 /* Increment number of outer iterations */
993 /* Update outer/inner flops */
995 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*111);