2 * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_double.h"
34 #include "kernelutil_x86_sse4_1_double.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse4_1_double
38 * Electrostatics interaction: ReactionField
39 * VdW interaction: CubicSplineTable
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse4_1_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 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
83 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
84 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
86 __m128i ifour = _mm_set1_epi32(4);
87 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
89 __m128d dummy_mask,cutoff_mask;
90 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
91 __m128d one = _mm_set1_pd(1.0);
92 __m128d two = _mm_set1_pd(2.0);
98 jindex = nlist->jindex;
100 shiftidx = nlist->shift;
102 shiftvec = fr->shift_vec[0];
103 fshift = fr->fshift[0];
104 facel = _mm_set1_pd(fr->epsfac);
105 charge = mdatoms->chargeA;
106 krf = _mm_set1_pd(fr->ic->k_rf);
107 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
108 crf = _mm_set1_pd(fr->ic->c_rf);
109 nvdwtype = fr->ntype;
111 vdwtype = mdatoms->typeA;
113 vftab = kernel_data->table_vdw->data;
114 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
116 /* Setup water-specific parameters */
117 inr = nlist->iinr[0];
118 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
119 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
120 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
121 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
123 /* Avoid stupid compiler warnings */
131 /* Start outer loop over neighborlists */
132 for(iidx=0; iidx<nri; iidx++)
134 /* Load shift vector for this list */
135 i_shift_offset = DIM*shiftidx[iidx];
137 /* Load limits for loop over neighbors */
138 j_index_start = jindex[iidx];
139 j_index_end = jindex[iidx+1];
141 /* Get outer coordinate index */
143 i_coord_offset = DIM*inr;
145 /* Load i particle coords and add shift vector */
146 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
147 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
149 fix0 = _mm_setzero_pd();
150 fiy0 = _mm_setzero_pd();
151 fiz0 = _mm_setzero_pd();
152 fix1 = _mm_setzero_pd();
153 fiy1 = _mm_setzero_pd();
154 fiz1 = _mm_setzero_pd();
155 fix2 = _mm_setzero_pd();
156 fiy2 = _mm_setzero_pd();
157 fiz2 = _mm_setzero_pd();
159 /* Reset potential sums */
160 velecsum = _mm_setzero_pd();
161 vvdwsum = _mm_setzero_pd();
163 /* Start inner kernel loop */
164 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
167 /* Get j neighbor index, and coordinate index */
170 j_coord_offsetA = DIM*jnrA;
171 j_coord_offsetB = DIM*jnrB;
173 /* load j atom coordinates */
174 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
177 /* Calculate displacement vector */
178 dx00 = _mm_sub_pd(ix0,jx0);
179 dy00 = _mm_sub_pd(iy0,jy0);
180 dz00 = _mm_sub_pd(iz0,jz0);
181 dx10 = _mm_sub_pd(ix1,jx0);
182 dy10 = _mm_sub_pd(iy1,jy0);
183 dz10 = _mm_sub_pd(iz1,jz0);
184 dx20 = _mm_sub_pd(ix2,jx0);
185 dy20 = _mm_sub_pd(iy2,jy0);
186 dz20 = _mm_sub_pd(iz2,jz0);
188 /* Calculate squared distance and things based on it */
189 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
190 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
191 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
193 rinv00 = gmx_mm_invsqrt_pd(rsq00);
194 rinv10 = gmx_mm_invsqrt_pd(rsq10);
195 rinv20 = gmx_mm_invsqrt_pd(rsq20);
197 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
198 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
199 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
201 /* Load parameters for j particles */
202 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
203 vdwjidx0A = 2*vdwtype[jnrA+0];
204 vdwjidx0B = 2*vdwtype[jnrB+0];
206 fjx0 = _mm_setzero_pd();
207 fjy0 = _mm_setzero_pd();
208 fjz0 = _mm_setzero_pd();
210 /**************************
211 * CALCULATE INTERACTIONS *
212 **************************/
214 r00 = _mm_mul_pd(rsq00,rinv00);
216 /* Compute parameters for interactions between i and j atoms */
217 qq00 = _mm_mul_pd(iq0,jq0);
218 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
219 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
221 /* Calculate table index by multiplying r with table scale and truncate to integer */
222 rt = _mm_mul_pd(r00,vftabscale);
223 vfitab = _mm_cvttpd_epi32(rt);
224 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
225 vfitab = _mm_slli_epi32(vfitab,3);
227 /* REACTION-FIELD ELECTROSTATICS */
228 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
229 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
231 /* CUBIC SPLINE TABLE DISPERSION */
232 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
233 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
234 GMX_MM_TRANSPOSE2_PD(Y,F);
235 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
236 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
237 GMX_MM_TRANSPOSE2_PD(G,H);
238 Heps = _mm_mul_pd(vfeps,H);
239 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
240 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
241 vvdw6 = _mm_mul_pd(c6_00,VV);
242 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
243 fvdw6 = _mm_mul_pd(c6_00,FF);
245 /* CUBIC SPLINE TABLE REPULSION */
246 vfitab = _mm_add_epi32(vfitab,ifour);
247 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
248 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
249 GMX_MM_TRANSPOSE2_PD(Y,F);
250 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
251 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
252 GMX_MM_TRANSPOSE2_PD(G,H);
253 Heps = _mm_mul_pd(vfeps,H);
254 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
255 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
256 vvdw12 = _mm_mul_pd(c12_00,VV);
257 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
258 fvdw12 = _mm_mul_pd(c12_00,FF);
259 vvdw = _mm_add_pd(vvdw12,vvdw6);
260 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
262 /* Update potential sum for this i atom from the interaction with this j atom. */
263 velecsum = _mm_add_pd(velecsum,velec);
264 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
266 fscal = _mm_add_pd(felec,fvdw);
268 /* Calculate temporary vectorial force */
269 tx = _mm_mul_pd(fscal,dx00);
270 ty = _mm_mul_pd(fscal,dy00);
271 tz = _mm_mul_pd(fscal,dz00);
273 /* Update vectorial force */
274 fix0 = _mm_add_pd(fix0,tx);
275 fiy0 = _mm_add_pd(fiy0,ty);
276 fiz0 = _mm_add_pd(fiz0,tz);
278 fjx0 = _mm_add_pd(fjx0,tx);
279 fjy0 = _mm_add_pd(fjy0,ty);
280 fjz0 = _mm_add_pd(fjz0,tz);
282 /**************************
283 * CALCULATE INTERACTIONS *
284 **************************/
286 /* Compute parameters for interactions between i and j atoms */
287 qq10 = _mm_mul_pd(iq1,jq0);
289 /* REACTION-FIELD ELECTROSTATICS */
290 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
291 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
293 /* Update potential sum for this i atom from the interaction with this j atom. */
294 velecsum = _mm_add_pd(velecsum,velec);
298 /* Calculate temporary vectorial force */
299 tx = _mm_mul_pd(fscal,dx10);
300 ty = _mm_mul_pd(fscal,dy10);
301 tz = _mm_mul_pd(fscal,dz10);
303 /* Update vectorial force */
304 fix1 = _mm_add_pd(fix1,tx);
305 fiy1 = _mm_add_pd(fiy1,ty);
306 fiz1 = _mm_add_pd(fiz1,tz);
308 fjx0 = _mm_add_pd(fjx0,tx);
309 fjy0 = _mm_add_pd(fjy0,ty);
310 fjz0 = _mm_add_pd(fjz0,tz);
312 /**************************
313 * CALCULATE INTERACTIONS *
314 **************************/
316 /* Compute parameters for interactions between i and j atoms */
317 qq20 = _mm_mul_pd(iq2,jq0);
319 /* REACTION-FIELD ELECTROSTATICS */
320 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
321 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
323 /* Update potential sum for this i atom from the interaction with this j atom. */
324 velecsum = _mm_add_pd(velecsum,velec);
328 /* Calculate temporary vectorial force */
329 tx = _mm_mul_pd(fscal,dx20);
330 ty = _mm_mul_pd(fscal,dy20);
331 tz = _mm_mul_pd(fscal,dz20);
333 /* Update vectorial force */
334 fix2 = _mm_add_pd(fix2,tx);
335 fiy2 = _mm_add_pd(fiy2,ty);
336 fiz2 = _mm_add_pd(fiz2,tz);
338 fjx0 = _mm_add_pd(fjx0,tx);
339 fjy0 = _mm_add_pd(fjy0,ty);
340 fjz0 = _mm_add_pd(fjz0,tz);
342 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
344 /* Inner loop uses 134 flops */
351 j_coord_offsetA = DIM*jnrA;
353 /* load j atom coordinates */
354 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
357 /* Calculate displacement vector */
358 dx00 = _mm_sub_pd(ix0,jx0);
359 dy00 = _mm_sub_pd(iy0,jy0);
360 dz00 = _mm_sub_pd(iz0,jz0);
361 dx10 = _mm_sub_pd(ix1,jx0);
362 dy10 = _mm_sub_pd(iy1,jy0);
363 dz10 = _mm_sub_pd(iz1,jz0);
364 dx20 = _mm_sub_pd(ix2,jx0);
365 dy20 = _mm_sub_pd(iy2,jy0);
366 dz20 = _mm_sub_pd(iz2,jz0);
368 /* Calculate squared distance and things based on it */
369 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
370 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
371 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
373 rinv00 = gmx_mm_invsqrt_pd(rsq00);
374 rinv10 = gmx_mm_invsqrt_pd(rsq10);
375 rinv20 = gmx_mm_invsqrt_pd(rsq20);
377 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
378 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
379 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
381 /* Load parameters for j particles */
382 jq0 = _mm_load_sd(charge+jnrA+0);
383 vdwjidx0A = 2*vdwtype[jnrA+0];
385 fjx0 = _mm_setzero_pd();
386 fjy0 = _mm_setzero_pd();
387 fjz0 = _mm_setzero_pd();
389 /**************************
390 * CALCULATE INTERACTIONS *
391 **************************/
393 r00 = _mm_mul_pd(rsq00,rinv00);
395 /* Compute parameters for interactions between i and j atoms */
396 qq00 = _mm_mul_pd(iq0,jq0);
397 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
399 /* Calculate table index by multiplying r with table scale and truncate to integer */
400 rt = _mm_mul_pd(r00,vftabscale);
401 vfitab = _mm_cvttpd_epi32(rt);
402 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
403 vfitab = _mm_slli_epi32(vfitab,3);
405 /* REACTION-FIELD ELECTROSTATICS */
406 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
407 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
409 /* CUBIC SPLINE TABLE DISPERSION */
410 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
411 F = _mm_setzero_pd();
412 GMX_MM_TRANSPOSE2_PD(Y,F);
413 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
414 H = _mm_setzero_pd();
415 GMX_MM_TRANSPOSE2_PD(G,H);
416 Heps = _mm_mul_pd(vfeps,H);
417 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
418 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
419 vvdw6 = _mm_mul_pd(c6_00,VV);
420 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
421 fvdw6 = _mm_mul_pd(c6_00,FF);
423 /* CUBIC SPLINE TABLE REPULSION */
424 vfitab = _mm_add_epi32(vfitab,ifour);
425 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
426 F = _mm_setzero_pd();
427 GMX_MM_TRANSPOSE2_PD(Y,F);
428 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
429 H = _mm_setzero_pd();
430 GMX_MM_TRANSPOSE2_PD(G,H);
431 Heps = _mm_mul_pd(vfeps,H);
432 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
433 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
434 vvdw12 = _mm_mul_pd(c12_00,VV);
435 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
436 fvdw12 = _mm_mul_pd(c12_00,FF);
437 vvdw = _mm_add_pd(vvdw12,vvdw6);
438 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
440 /* Update potential sum for this i atom from the interaction with this j atom. */
441 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
442 velecsum = _mm_add_pd(velecsum,velec);
443 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
444 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
446 fscal = _mm_add_pd(felec,fvdw);
448 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
450 /* Calculate temporary vectorial force */
451 tx = _mm_mul_pd(fscal,dx00);
452 ty = _mm_mul_pd(fscal,dy00);
453 tz = _mm_mul_pd(fscal,dz00);
455 /* Update vectorial force */
456 fix0 = _mm_add_pd(fix0,tx);
457 fiy0 = _mm_add_pd(fiy0,ty);
458 fiz0 = _mm_add_pd(fiz0,tz);
460 fjx0 = _mm_add_pd(fjx0,tx);
461 fjy0 = _mm_add_pd(fjy0,ty);
462 fjz0 = _mm_add_pd(fjz0,tz);
464 /**************************
465 * CALCULATE INTERACTIONS *
466 **************************/
468 /* Compute parameters for interactions between i and j atoms */
469 qq10 = _mm_mul_pd(iq1,jq0);
471 /* REACTION-FIELD ELECTROSTATICS */
472 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
473 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
475 /* Update potential sum for this i atom from the interaction with this j atom. */
476 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
477 velecsum = _mm_add_pd(velecsum,velec);
481 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
483 /* Calculate temporary vectorial force */
484 tx = _mm_mul_pd(fscal,dx10);
485 ty = _mm_mul_pd(fscal,dy10);
486 tz = _mm_mul_pd(fscal,dz10);
488 /* Update vectorial force */
489 fix1 = _mm_add_pd(fix1,tx);
490 fiy1 = _mm_add_pd(fiy1,ty);
491 fiz1 = _mm_add_pd(fiz1,tz);
493 fjx0 = _mm_add_pd(fjx0,tx);
494 fjy0 = _mm_add_pd(fjy0,ty);
495 fjz0 = _mm_add_pd(fjz0,tz);
497 /**************************
498 * CALCULATE INTERACTIONS *
499 **************************/
501 /* Compute parameters for interactions between i and j atoms */
502 qq20 = _mm_mul_pd(iq2,jq0);
504 /* REACTION-FIELD ELECTROSTATICS */
505 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
506 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
508 /* Update potential sum for this i atom from the interaction with this j atom. */
509 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
510 velecsum = _mm_add_pd(velecsum,velec);
514 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
516 /* Calculate temporary vectorial force */
517 tx = _mm_mul_pd(fscal,dx20);
518 ty = _mm_mul_pd(fscal,dy20);
519 tz = _mm_mul_pd(fscal,dz20);
521 /* Update vectorial force */
522 fix2 = _mm_add_pd(fix2,tx);
523 fiy2 = _mm_add_pd(fiy2,ty);
524 fiz2 = _mm_add_pd(fiz2,tz);
526 fjx0 = _mm_add_pd(fjx0,tx);
527 fjy0 = _mm_add_pd(fjy0,ty);
528 fjz0 = _mm_add_pd(fjz0,tz);
530 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
532 /* Inner loop uses 134 flops */
535 /* End of innermost loop */
537 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
538 f+i_coord_offset,fshift+i_shift_offset);
541 /* Update potential energies */
542 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
543 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
545 /* Increment number of inner iterations */
546 inneriter += j_index_end - j_index_start;
548 /* Outer loop uses 20 flops */
551 /* Increment number of outer iterations */
554 /* Update outer/inner flops */
556 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*134);
559 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse4_1_double
560 * Electrostatics interaction: ReactionField
561 * VdW interaction: CubicSplineTable
562 * Geometry: Water3-Particle
563 * Calculate force/pot: Force
566 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse4_1_double
567 (t_nblist * gmx_restrict nlist,
568 rvec * gmx_restrict xx,
569 rvec * gmx_restrict ff,
570 t_forcerec * gmx_restrict fr,
571 t_mdatoms * gmx_restrict mdatoms,
572 nb_kernel_data_t * gmx_restrict kernel_data,
573 t_nrnb * gmx_restrict nrnb)
575 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
576 * just 0 for non-waters.
577 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
578 * jnr indices corresponding to data put in the four positions in the SIMD register.
580 int i_shift_offset,i_coord_offset,outeriter,inneriter;
581 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
583 int j_coord_offsetA,j_coord_offsetB;
584 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
586 real *shiftvec,*fshift,*x,*f;
587 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
589 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
591 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
593 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
594 int vdwjidx0A,vdwjidx0B;
595 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
596 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
597 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
598 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
599 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
602 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
605 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
606 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
608 __m128i ifour = _mm_set1_epi32(4);
609 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
611 __m128d dummy_mask,cutoff_mask;
612 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
613 __m128d one = _mm_set1_pd(1.0);
614 __m128d two = _mm_set1_pd(2.0);
620 jindex = nlist->jindex;
622 shiftidx = nlist->shift;
624 shiftvec = fr->shift_vec[0];
625 fshift = fr->fshift[0];
626 facel = _mm_set1_pd(fr->epsfac);
627 charge = mdatoms->chargeA;
628 krf = _mm_set1_pd(fr->ic->k_rf);
629 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
630 crf = _mm_set1_pd(fr->ic->c_rf);
631 nvdwtype = fr->ntype;
633 vdwtype = mdatoms->typeA;
635 vftab = kernel_data->table_vdw->data;
636 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
638 /* Setup water-specific parameters */
639 inr = nlist->iinr[0];
640 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
641 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
642 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
643 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
645 /* Avoid stupid compiler warnings */
653 /* Start outer loop over neighborlists */
654 for(iidx=0; iidx<nri; iidx++)
656 /* Load shift vector for this list */
657 i_shift_offset = DIM*shiftidx[iidx];
659 /* Load limits for loop over neighbors */
660 j_index_start = jindex[iidx];
661 j_index_end = jindex[iidx+1];
663 /* Get outer coordinate index */
665 i_coord_offset = DIM*inr;
667 /* Load i particle coords and add shift vector */
668 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
669 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
671 fix0 = _mm_setzero_pd();
672 fiy0 = _mm_setzero_pd();
673 fiz0 = _mm_setzero_pd();
674 fix1 = _mm_setzero_pd();
675 fiy1 = _mm_setzero_pd();
676 fiz1 = _mm_setzero_pd();
677 fix2 = _mm_setzero_pd();
678 fiy2 = _mm_setzero_pd();
679 fiz2 = _mm_setzero_pd();
681 /* Start inner kernel loop */
682 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
685 /* Get j neighbor index, and coordinate index */
688 j_coord_offsetA = DIM*jnrA;
689 j_coord_offsetB = DIM*jnrB;
691 /* load j atom coordinates */
692 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
695 /* Calculate displacement vector */
696 dx00 = _mm_sub_pd(ix0,jx0);
697 dy00 = _mm_sub_pd(iy0,jy0);
698 dz00 = _mm_sub_pd(iz0,jz0);
699 dx10 = _mm_sub_pd(ix1,jx0);
700 dy10 = _mm_sub_pd(iy1,jy0);
701 dz10 = _mm_sub_pd(iz1,jz0);
702 dx20 = _mm_sub_pd(ix2,jx0);
703 dy20 = _mm_sub_pd(iy2,jy0);
704 dz20 = _mm_sub_pd(iz2,jz0);
706 /* Calculate squared distance and things based on it */
707 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
708 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
709 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
711 rinv00 = gmx_mm_invsqrt_pd(rsq00);
712 rinv10 = gmx_mm_invsqrt_pd(rsq10);
713 rinv20 = gmx_mm_invsqrt_pd(rsq20);
715 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
716 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
717 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
719 /* Load parameters for j particles */
720 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
721 vdwjidx0A = 2*vdwtype[jnrA+0];
722 vdwjidx0B = 2*vdwtype[jnrB+0];
724 fjx0 = _mm_setzero_pd();
725 fjy0 = _mm_setzero_pd();
726 fjz0 = _mm_setzero_pd();
728 /**************************
729 * CALCULATE INTERACTIONS *
730 **************************/
732 r00 = _mm_mul_pd(rsq00,rinv00);
734 /* Compute parameters for interactions between i and j atoms */
735 qq00 = _mm_mul_pd(iq0,jq0);
736 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
737 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
739 /* Calculate table index by multiplying r with table scale and truncate to integer */
740 rt = _mm_mul_pd(r00,vftabscale);
741 vfitab = _mm_cvttpd_epi32(rt);
742 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
743 vfitab = _mm_slli_epi32(vfitab,3);
745 /* REACTION-FIELD ELECTROSTATICS */
746 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
748 /* CUBIC SPLINE TABLE DISPERSION */
749 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
750 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
751 GMX_MM_TRANSPOSE2_PD(Y,F);
752 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
753 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
754 GMX_MM_TRANSPOSE2_PD(G,H);
755 Heps = _mm_mul_pd(vfeps,H);
756 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
757 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
758 fvdw6 = _mm_mul_pd(c6_00,FF);
760 /* CUBIC SPLINE TABLE REPULSION */
761 vfitab = _mm_add_epi32(vfitab,ifour);
762 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
763 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
764 GMX_MM_TRANSPOSE2_PD(Y,F);
765 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
766 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
767 GMX_MM_TRANSPOSE2_PD(G,H);
768 Heps = _mm_mul_pd(vfeps,H);
769 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
770 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
771 fvdw12 = _mm_mul_pd(c12_00,FF);
772 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
774 fscal = _mm_add_pd(felec,fvdw);
776 /* Calculate temporary vectorial force */
777 tx = _mm_mul_pd(fscal,dx00);
778 ty = _mm_mul_pd(fscal,dy00);
779 tz = _mm_mul_pd(fscal,dz00);
781 /* Update vectorial force */
782 fix0 = _mm_add_pd(fix0,tx);
783 fiy0 = _mm_add_pd(fiy0,ty);
784 fiz0 = _mm_add_pd(fiz0,tz);
786 fjx0 = _mm_add_pd(fjx0,tx);
787 fjy0 = _mm_add_pd(fjy0,ty);
788 fjz0 = _mm_add_pd(fjz0,tz);
790 /**************************
791 * CALCULATE INTERACTIONS *
792 **************************/
794 /* Compute parameters for interactions between i and j atoms */
795 qq10 = _mm_mul_pd(iq1,jq0);
797 /* REACTION-FIELD ELECTROSTATICS */
798 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
802 /* Calculate temporary vectorial force */
803 tx = _mm_mul_pd(fscal,dx10);
804 ty = _mm_mul_pd(fscal,dy10);
805 tz = _mm_mul_pd(fscal,dz10);
807 /* Update vectorial force */
808 fix1 = _mm_add_pd(fix1,tx);
809 fiy1 = _mm_add_pd(fiy1,ty);
810 fiz1 = _mm_add_pd(fiz1,tz);
812 fjx0 = _mm_add_pd(fjx0,tx);
813 fjy0 = _mm_add_pd(fjy0,ty);
814 fjz0 = _mm_add_pd(fjz0,tz);
816 /**************************
817 * CALCULATE INTERACTIONS *
818 **************************/
820 /* Compute parameters for interactions between i and j atoms */
821 qq20 = _mm_mul_pd(iq2,jq0);
823 /* REACTION-FIELD ELECTROSTATICS */
824 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
828 /* Calculate temporary vectorial force */
829 tx = _mm_mul_pd(fscal,dx20);
830 ty = _mm_mul_pd(fscal,dy20);
831 tz = _mm_mul_pd(fscal,dz20);
833 /* Update vectorial force */
834 fix2 = _mm_add_pd(fix2,tx);
835 fiy2 = _mm_add_pd(fiy2,ty);
836 fiz2 = _mm_add_pd(fiz2,tz);
838 fjx0 = _mm_add_pd(fjx0,tx);
839 fjy0 = _mm_add_pd(fjy0,ty);
840 fjz0 = _mm_add_pd(fjz0,tz);
842 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
844 /* Inner loop uses 111 flops */
851 j_coord_offsetA = DIM*jnrA;
853 /* load j atom coordinates */
854 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
857 /* Calculate displacement vector */
858 dx00 = _mm_sub_pd(ix0,jx0);
859 dy00 = _mm_sub_pd(iy0,jy0);
860 dz00 = _mm_sub_pd(iz0,jz0);
861 dx10 = _mm_sub_pd(ix1,jx0);
862 dy10 = _mm_sub_pd(iy1,jy0);
863 dz10 = _mm_sub_pd(iz1,jz0);
864 dx20 = _mm_sub_pd(ix2,jx0);
865 dy20 = _mm_sub_pd(iy2,jy0);
866 dz20 = _mm_sub_pd(iz2,jz0);
868 /* Calculate squared distance and things based on it */
869 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
870 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
871 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
873 rinv00 = gmx_mm_invsqrt_pd(rsq00);
874 rinv10 = gmx_mm_invsqrt_pd(rsq10);
875 rinv20 = gmx_mm_invsqrt_pd(rsq20);
877 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
878 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
879 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
881 /* Load parameters for j particles */
882 jq0 = _mm_load_sd(charge+jnrA+0);
883 vdwjidx0A = 2*vdwtype[jnrA+0];
885 fjx0 = _mm_setzero_pd();
886 fjy0 = _mm_setzero_pd();
887 fjz0 = _mm_setzero_pd();
889 /**************************
890 * CALCULATE INTERACTIONS *
891 **************************/
893 r00 = _mm_mul_pd(rsq00,rinv00);
895 /* Compute parameters for interactions between i and j atoms */
896 qq00 = _mm_mul_pd(iq0,jq0);
897 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
899 /* Calculate table index by multiplying r with table scale and truncate to integer */
900 rt = _mm_mul_pd(r00,vftabscale);
901 vfitab = _mm_cvttpd_epi32(rt);
902 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
903 vfitab = _mm_slli_epi32(vfitab,3);
905 /* REACTION-FIELD ELECTROSTATICS */
906 felec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
908 /* CUBIC SPLINE TABLE DISPERSION */
909 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
910 F = _mm_setzero_pd();
911 GMX_MM_TRANSPOSE2_PD(Y,F);
912 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
913 H = _mm_setzero_pd();
914 GMX_MM_TRANSPOSE2_PD(G,H);
915 Heps = _mm_mul_pd(vfeps,H);
916 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
917 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
918 fvdw6 = _mm_mul_pd(c6_00,FF);
920 /* CUBIC SPLINE TABLE REPULSION */
921 vfitab = _mm_add_epi32(vfitab,ifour);
922 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
923 F = _mm_setzero_pd();
924 GMX_MM_TRANSPOSE2_PD(Y,F);
925 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
926 H = _mm_setzero_pd();
927 GMX_MM_TRANSPOSE2_PD(G,H);
928 Heps = _mm_mul_pd(vfeps,H);
929 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
930 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
931 fvdw12 = _mm_mul_pd(c12_00,FF);
932 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
934 fscal = _mm_add_pd(felec,fvdw);
936 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
938 /* Calculate temporary vectorial force */
939 tx = _mm_mul_pd(fscal,dx00);
940 ty = _mm_mul_pd(fscal,dy00);
941 tz = _mm_mul_pd(fscal,dz00);
943 /* Update vectorial force */
944 fix0 = _mm_add_pd(fix0,tx);
945 fiy0 = _mm_add_pd(fiy0,ty);
946 fiz0 = _mm_add_pd(fiz0,tz);
948 fjx0 = _mm_add_pd(fjx0,tx);
949 fjy0 = _mm_add_pd(fjy0,ty);
950 fjz0 = _mm_add_pd(fjz0,tz);
952 /**************************
953 * CALCULATE INTERACTIONS *
954 **************************/
956 /* Compute parameters for interactions between i and j atoms */
957 qq10 = _mm_mul_pd(iq1,jq0);
959 /* REACTION-FIELD ELECTROSTATICS */
960 felec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
964 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
966 /* Calculate temporary vectorial force */
967 tx = _mm_mul_pd(fscal,dx10);
968 ty = _mm_mul_pd(fscal,dy10);
969 tz = _mm_mul_pd(fscal,dz10);
971 /* Update vectorial force */
972 fix1 = _mm_add_pd(fix1,tx);
973 fiy1 = _mm_add_pd(fiy1,ty);
974 fiz1 = _mm_add_pd(fiz1,tz);
976 fjx0 = _mm_add_pd(fjx0,tx);
977 fjy0 = _mm_add_pd(fjy0,ty);
978 fjz0 = _mm_add_pd(fjz0,tz);
980 /**************************
981 * CALCULATE INTERACTIONS *
982 **************************/
984 /* Compute parameters for interactions between i and j atoms */
985 qq20 = _mm_mul_pd(iq2,jq0);
987 /* REACTION-FIELD ELECTROSTATICS */
988 felec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
992 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
994 /* Calculate temporary vectorial force */
995 tx = _mm_mul_pd(fscal,dx20);
996 ty = _mm_mul_pd(fscal,dy20);
997 tz = _mm_mul_pd(fscal,dz20);
999 /* Update vectorial force */
1000 fix2 = _mm_add_pd(fix2,tx);
1001 fiy2 = _mm_add_pd(fiy2,ty);
1002 fiz2 = _mm_add_pd(fiz2,tz);
1004 fjx0 = _mm_add_pd(fjx0,tx);
1005 fjy0 = _mm_add_pd(fjy0,ty);
1006 fjz0 = _mm_add_pd(fjz0,tz);
1008 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1010 /* Inner loop uses 111 flops */
1013 /* End of innermost loop */
1015 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1016 f+i_coord_offset,fshift+i_shift_offset);
1018 /* Increment number of inner iterations */
1019 inneriter += j_index_end - j_index_start;
1021 /* Outer loop uses 18 flops */
1024 /* Increment number of outer iterations */
1027 /* Update outer/inner flops */
1029 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*111);