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_ElecCoul_VdwCSTab_GeomW3P1_VF_sse4_1_double
38 * Electrostatics interaction: Coulomb
39 * VdW interaction: CubicSplineTable
40 * Geometry: Water3-Particle
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCoul_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 nvdwtype = fr->ntype;
108 vdwtype = mdatoms->typeA;
110 vftab = kernel_data->table_vdw->data;
111 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
113 /* Setup water-specific parameters */
114 inr = nlist->iinr[0];
115 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
116 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
117 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
118 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
120 /* Avoid stupid compiler warnings */
128 /* Start outer loop over neighborlists */
129 for(iidx=0; iidx<nri; iidx++)
131 /* Load shift vector for this list */
132 i_shift_offset = DIM*shiftidx[iidx];
134 /* Load limits for loop over neighbors */
135 j_index_start = jindex[iidx];
136 j_index_end = jindex[iidx+1];
138 /* Get outer coordinate index */
140 i_coord_offset = DIM*inr;
142 /* Load i particle coords and add shift vector */
143 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
144 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
146 fix0 = _mm_setzero_pd();
147 fiy0 = _mm_setzero_pd();
148 fiz0 = _mm_setzero_pd();
149 fix1 = _mm_setzero_pd();
150 fiy1 = _mm_setzero_pd();
151 fiz1 = _mm_setzero_pd();
152 fix2 = _mm_setzero_pd();
153 fiy2 = _mm_setzero_pd();
154 fiz2 = _mm_setzero_pd();
156 /* Reset potential sums */
157 velecsum = _mm_setzero_pd();
158 vvdwsum = _mm_setzero_pd();
160 /* Start inner kernel loop */
161 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
164 /* Get j neighbor index, and coordinate index */
167 j_coord_offsetA = DIM*jnrA;
168 j_coord_offsetB = DIM*jnrB;
170 /* load j atom coordinates */
171 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
174 /* Calculate displacement vector */
175 dx00 = _mm_sub_pd(ix0,jx0);
176 dy00 = _mm_sub_pd(iy0,jy0);
177 dz00 = _mm_sub_pd(iz0,jz0);
178 dx10 = _mm_sub_pd(ix1,jx0);
179 dy10 = _mm_sub_pd(iy1,jy0);
180 dz10 = _mm_sub_pd(iz1,jz0);
181 dx20 = _mm_sub_pd(ix2,jx0);
182 dy20 = _mm_sub_pd(iy2,jy0);
183 dz20 = _mm_sub_pd(iz2,jz0);
185 /* Calculate squared distance and things based on it */
186 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
187 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
188 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
190 rinv00 = gmx_mm_invsqrt_pd(rsq00);
191 rinv10 = gmx_mm_invsqrt_pd(rsq10);
192 rinv20 = gmx_mm_invsqrt_pd(rsq20);
194 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
195 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
196 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
198 /* Load parameters for j particles */
199 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
200 vdwjidx0A = 2*vdwtype[jnrA+0];
201 vdwjidx0B = 2*vdwtype[jnrB+0];
203 fjx0 = _mm_setzero_pd();
204 fjy0 = _mm_setzero_pd();
205 fjz0 = _mm_setzero_pd();
207 /**************************
208 * CALCULATE INTERACTIONS *
209 **************************/
211 r00 = _mm_mul_pd(rsq00,rinv00);
213 /* Compute parameters for interactions between i and j atoms */
214 qq00 = _mm_mul_pd(iq0,jq0);
215 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
216 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
218 /* Calculate table index by multiplying r with table scale and truncate to integer */
219 rt = _mm_mul_pd(r00,vftabscale);
220 vfitab = _mm_cvttpd_epi32(rt);
221 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
222 vfitab = _mm_slli_epi32(vfitab,3);
224 /* COULOMB ELECTROSTATICS */
225 velec = _mm_mul_pd(qq00,rinv00);
226 felec = _mm_mul_pd(velec,rinvsq00);
228 /* CUBIC SPLINE TABLE DISPERSION */
229 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
230 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
231 GMX_MM_TRANSPOSE2_PD(Y,F);
232 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
233 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
234 GMX_MM_TRANSPOSE2_PD(G,H);
235 Heps = _mm_mul_pd(vfeps,H);
236 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
237 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
238 vvdw6 = _mm_mul_pd(c6_00,VV);
239 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
240 fvdw6 = _mm_mul_pd(c6_00,FF);
242 /* CUBIC SPLINE TABLE REPULSION */
243 vfitab = _mm_add_epi32(vfitab,ifour);
244 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
245 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
246 GMX_MM_TRANSPOSE2_PD(Y,F);
247 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
248 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
249 GMX_MM_TRANSPOSE2_PD(G,H);
250 Heps = _mm_mul_pd(vfeps,H);
251 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
252 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
253 vvdw12 = _mm_mul_pd(c12_00,VV);
254 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
255 fvdw12 = _mm_mul_pd(c12_00,FF);
256 vvdw = _mm_add_pd(vvdw12,vvdw6);
257 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
259 /* Update potential sum for this i atom from the interaction with this j atom. */
260 velecsum = _mm_add_pd(velecsum,velec);
261 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
263 fscal = _mm_add_pd(felec,fvdw);
265 /* Calculate temporary vectorial force */
266 tx = _mm_mul_pd(fscal,dx00);
267 ty = _mm_mul_pd(fscal,dy00);
268 tz = _mm_mul_pd(fscal,dz00);
270 /* Update vectorial force */
271 fix0 = _mm_add_pd(fix0,tx);
272 fiy0 = _mm_add_pd(fiy0,ty);
273 fiz0 = _mm_add_pd(fiz0,tz);
275 fjx0 = _mm_add_pd(fjx0,tx);
276 fjy0 = _mm_add_pd(fjy0,ty);
277 fjz0 = _mm_add_pd(fjz0,tz);
279 /**************************
280 * CALCULATE INTERACTIONS *
281 **************************/
283 /* Compute parameters for interactions between i and j atoms */
284 qq10 = _mm_mul_pd(iq1,jq0);
286 /* COULOMB ELECTROSTATICS */
287 velec = _mm_mul_pd(qq10,rinv10);
288 felec = _mm_mul_pd(velec,rinvsq10);
290 /* Update potential sum for this i atom from the interaction with this j atom. */
291 velecsum = _mm_add_pd(velecsum,velec);
295 /* Calculate temporary vectorial force */
296 tx = _mm_mul_pd(fscal,dx10);
297 ty = _mm_mul_pd(fscal,dy10);
298 tz = _mm_mul_pd(fscal,dz10);
300 /* Update vectorial force */
301 fix1 = _mm_add_pd(fix1,tx);
302 fiy1 = _mm_add_pd(fiy1,ty);
303 fiz1 = _mm_add_pd(fiz1,tz);
305 fjx0 = _mm_add_pd(fjx0,tx);
306 fjy0 = _mm_add_pd(fjy0,ty);
307 fjz0 = _mm_add_pd(fjz0,tz);
309 /**************************
310 * CALCULATE INTERACTIONS *
311 **************************/
313 /* Compute parameters for interactions between i and j atoms */
314 qq20 = _mm_mul_pd(iq2,jq0);
316 /* COULOMB ELECTROSTATICS */
317 velec = _mm_mul_pd(qq20,rinv20);
318 felec = _mm_mul_pd(velec,rinvsq20);
320 /* Update potential sum for this i atom from the interaction with this j atom. */
321 velecsum = _mm_add_pd(velecsum,velec);
325 /* Calculate temporary vectorial force */
326 tx = _mm_mul_pd(fscal,dx20);
327 ty = _mm_mul_pd(fscal,dy20);
328 tz = _mm_mul_pd(fscal,dz20);
330 /* Update vectorial force */
331 fix2 = _mm_add_pd(fix2,tx);
332 fiy2 = _mm_add_pd(fiy2,ty);
333 fiz2 = _mm_add_pd(fiz2,tz);
335 fjx0 = _mm_add_pd(fjx0,tx);
336 fjy0 = _mm_add_pd(fjy0,ty);
337 fjz0 = _mm_add_pd(fjz0,tz);
339 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
341 /* Inner loop uses 122 flops */
348 j_coord_offsetA = DIM*jnrA;
350 /* load j atom coordinates */
351 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
354 /* Calculate displacement vector */
355 dx00 = _mm_sub_pd(ix0,jx0);
356 dy00 = _mm_sub_pd(iy0,jy0);
357 dz00 = _mm_sub_pd(iz0,jz0);
358 dx10 = _mm_sub_pd(ix1,jx0);
359 dy10 = _mm_sub_pd(iy1,jy0);
360 dz10 = _mm_sub_pd(iz1,jz0);
361 dx20 = _mm_sub_pd(ix2,jx0);
362 dy20 = _mm_sub_pd(iy2,jy0);
363 dz20 = _mm_sub_pd(iz2,jz0);
365 /* Calculate squared distance and things based on it */
366 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
367 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
368 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
370 rinv00 = gmx_mm_invsqrt_pd(rsq00);
371 rinv10 = gmx_mm_invsqrt_pd(rsq10);
372 rinv20 = gmx_mm_invsqrt_pd(rsq20);
374 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
375 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
376 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
378 /* Load parameters for j particles */
379 jq0 = _mm_load_sd(charge+jnrA+0);
380 vdwjidx0A = 2*vdwtype[jnrA+0];
382 fjx0 = _mm_setzero_pd();
383 fjy0 = _mm_setzero_pd();
384 fjz0 = _mm_setzero_pd();
386 /**************************
387 * CALCULATE INTERACTIONS *
388 **************************/
390 r00 = _mm_mul_pd(rsq00,rinv00);
392 /* Compute parameters for interactions between i and j atoms */
393 qq00 = _mm_mul_pd(iq0,jq0);
394 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
396 /* Calculate table index by multiplying r with table scale and truncate to integer */
397 rt = _mm_mul_pd(r00,vftabscale);
398 vfitab = _mm_cvttpd_epi32(rt);
399 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
400 vfitab = _mm_slli_epi32(vfitab,3);
402 /* COULOMB ELECTROSTATICS */
403 velec = _mm_mul_pd(qq00,rinv00);
404 felec = _mm_mul_pd(velec,rinvsq00);
406 /* CUBIC SPLINE TABLE DISPERSION */
407 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
408 F = _mm_setzero_pd();
409 GMX_MM_TRANSPOSE2_PD(Y,F);
410 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
411 H = _mm_setzero_pd();
412 GMX_MM_TRANSPOSE2_PD(G,H);
413 Heps = _mm_mul_pd(vfeps,H);
414 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
415 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
416 vvdw6 = _mm_mul_pd(c6_00,VV);
417 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
418 fvdw6 = _mm_mul_pd(c6_00,FF);
420 /* CUBIC SPLINE TABLE REPULSION */
421 vfitab = _mm_add_epi32(vfitab,ifour);
422 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
423 F = _mm_setzero_pd();
424 GMX_MM_TRANSPOSE2_PD(Y,F);
425 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
426 H = _mm_setzero_pd();
427 GMX_MM_TRANSPOSE2_PD(G,H);
428 Heps = _mm_mul_pd(vfeps,H);
429 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
430 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
431 vvdw12 = _mm_mul_pd(c12_00,VV);
432 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
433 fvdw12 = _mm_mul_pd(c12_00,FF);
434 vvdw = _mm_add_pd(vvdw12,vvdw6);
435 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
437 /* Update potential sum for this i atom from the interaction with this j atom. */
438 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
439 velecsum = _mm_add_pd(velecsum,velec);
440 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
441 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
443 fscal = _mm_add_pd(felec,fvdw);
445 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
447 /* Calculate temporary vectorial force */
448 tx = _mm_mul_pd(fscal,dx00);
449 ty = _mm_mul_pd(fscal,dy00);
450 tz = _mm_mul_pd(fscal,dz00);
452 /* Update vectorial force */
453 fix0 = _mm_add_pd(fix0,tx);
454 fiy0 = _mm_add_pd(fiy0,ty);
455 fiz0 = _mm_add_pd(fiz0,tz);
457 fjx0 = _mm_add_pd(fjx0,tx);
458 fjy0 = _mm_add_pd(fjy0,ty);
459 fjz0 = _mm_add_pd(fjz0,tz);
461 /**************************
462 * CALCULATE INTERACTIONS *
463 **************************/
465 /* Compute parameters for interactions between i and j atoms */
466 qq10 = _mm_mul_pd(iq1,jq0);
468 /* COULOMB ELECTROSTATICS */
469 velec = _mm_mul_pd(qq10,rinv10);
470 felec = _mm_mul_pd(velec,rinvsq10);
472 /* Update potential sum for this i atom from the interaction with this j atom. */
473 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
474 velecsum = _mm_add_pd(velecsum,velec);
478 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
480 /* Calculate temporary vectorial force */
481 tx = _mm_mul_pd(fscal,dx10);
482 ty = _mm_mul_pd(fscal,dy10);
483 tz = _mm_mul_pd(fscal,dz10);
485 /* Update vectorial force */
486 fix1 = _mm_add_pd(fix1,tx);
487 fiy1 = _mm_add_pd(fiy1,ty);
488 fiz1 = _mm_add_pd(fiz1,tz);
490 fjx0 = _mm_add_pd(fjx0,tx);
491 fjy0 = _mm_add_pd(fjy0,ty);
492 fjz0 = _mm_add_pd(fjz0,tz);
494 /**************************
495 * CALCULATE INTERACTIONS *
496 **************************/
498 /* Compute parameters for interactions between i and j atoms */
499 qq20 = _mm_mul_pd(iq2,jq0);
501 /* COULOMB ELECTROSTATICS */
502 velec = _mm_mul_pd(qq20,rinv20);
503 felec = _mm_mul_pd(velec,rinvsq20);
505 /* Update potential sum for this i atom from the interaction with this j atom. */
506 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
507 velecsum = _mm_add_pd(velecsum,velec);
511 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
513 /* Calculate temporary vectorial force */
514 tx = _mm_mul_pd(fscal,dx20);
515 ty = _mm_mul_pd(fscal,dy20);
516 tz = _mm_mul_pd(fscal,dz20);
518 /* Update vectorial force */
519 fix2 = _mm_add_pd(fix2,tx);
520 fiy2 = _mm_add_pd(fiy2,ty);
521 fiz2 = _mm_add_pd(fiz2,tz);
523 fjx0 = _mm_add_pd(fjx0,tx);
524 fjy0 = _mm_add_pd(fjy0,ty);
525 fjz0 = _mm_add_pd(fjz0,tz);
527 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
529 /* Inner loop uses 122 flops */
532 /* End of innermost loop */
534 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
535 f+i_coord_offset,fshift+i_shift_offset);
538 /* Update potential energies */
539 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
540 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
542 /* Increment number of inner iterations */
543 inneriter += j_index_end - j_index_start;
545 /* Outer loop uses 20 flops */
548 /* Increment number of outer iterations */
551 /* Update outer/inner flops */
553 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*122);
556 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse4_1_double
557 * Electrostatics interaction: Coulomb
558 * VdW interaction: CubicSplineTable
559 * Geometry: Water3-Particle
560 * Calculate force/pot: Force
563 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse4_1_double
564 (t_nblist * gmx_restrict nlist,
565 rvec * gmx_restrict xx,
566 rvec * gmx_restrict ff,
567 t_forcerec * gmx_restrict fr,
568 t_mdatoms * gmx_restrict mdatoms,
569 nb_kernel_data_t * gmx_restrict kernel_data,
570 t_nrnb * gmx_restrict nrnb)
572 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
573 * just 0 for non-waters.
574 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
575 * jnr indices corresponding to data put in the four positions in the SIMD register.
577 int i_shift_offset,i_coord_offset,outeriter,inneriter;
578 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
580 int j_coord_offsetA,j_coord_offsetB;
581 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
583 real *shiftvec,*fshift,*x,*f;
584 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
586 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
588 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
590 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
591 int vdwjidx0A,vdwjidx0B;
592 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
593 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
594 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
595 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
596 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
599 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
602 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
603 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
605 __m128i ifour = _mm_set1_epi32(4);
606 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
608 __m128d dummy_mask,cutoff_mask;
609 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
610 __m128d one = _mm_set1_pd(1.0);
611 __m128d two = _mm_set1_pd(2.0);
617 jindex = nlist->jindex;
619 shiftidx = nlist->shift;
621 shiftvec = fr->shift_vec[0];
622 fshift = fr->fshift[0];
623 facel = _mm_set1_pd(fr->epsfac);
624 charge = mdatoms->chargeA;
625 nvdwtype = fr->ntype;
627 vdwtype = mdatoms->typeA;
629 vftab = kernel_data->table_vdw->data;
630 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
632 /* Setup water-specific parameters */
633 inr = nlist->iinr[0];
634 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
635 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
636 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
637 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
639 /* Avoid stupid compiler warnings */
647 /* Start outer loop over neighborlists */
648 for(iidx=0; iidx<nri; iidx++)
650 /* Load shift vector for this list */
651 i_shift_offset = DIM*shiftidx[iidx];
653 /* Load limits for loop over neighbors */
654 j_index_start = jindex[iidx];
655 j_index_end = jindex[iidx+1];
657 /* Get outer coordinate index */
659 i_coord_offset = DIM*inr;
661 /* Load i particle coords and add shift vector */
662 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
663 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
665 fix0 = _mm_setzero_pd();
666 fiy0 = _mm_setzero_pd();
667 fiz0 = _mm_setzero_pd();
668 fix1 = _mm_setzero_pd();
669 fiy1 = _mm_setzero_pd();
670 fiz1 = _mm_setzero_pd();
671 fix2 = _mm_setzero_pd();
672 fiy2 = _mm_setzero_pd();
673 fiz2 = _mm_setzero_pd();
675 /* Start inner kernel loop */
676 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
679 /* Get j neighbor index, and coordinate index */
682 j_coord_offsetA = DIM*jnrA;
683 j_coord_offsetB = DIM*jnrB;
685 /* load j atom coordinates */
686 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
689 /* Calculate displacement vector */
690 dx00 = _mm_sub_pd(ix0,jx0);
691 dy00 = _mm_sub_pd(iy0,jy0);
692 dz00 = _mm_sub_pd(iz0,jz0);
693 dx10 = _mm_sub_pd(ix1,jx0);
694 dy10 = _mm_sub_pd(iy1,jy0);
695 dz10 = _mm_sub_pd(iz1,jz0);
696 dx20 = _mm_sub_pd(ix2,jx0);
697 dy20 = _mm_sub_pd(iy2,jy0);
698 dz20 = _mm_sub_pd(iz2,jz0);
700 /* Calculate squared distance and things based on it */
701 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
702 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
703 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
705 rinv00 = gmx_mm_invsqrt_pd(rsq00);
706 rinv10 = gmx_mm_invsqrt_pd(rsq10);
707 rinv20 = gmx_mm_invsqrt_pd(rsq20);
709 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
710 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
711 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
713 /* Load parameters for j particles */
714 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
715 vdwjidx0A = 2*vdwtype[jnrA+0];
716 vdwjidx0B = 2*vdwtype[jnrB+0];
718 fjx0 = _mm_setzero_pd();
719 fjy0 = _mm_setzero_pd();
720 fjz0 = _mm_setzero_pd();
722 /**************************
723 * CALCULATE INTERACTIONS *
724 **************************/
726 r00 = _mm_mul_pd(rsq00,rinv00);
728 /* Compute parameters for interactions between i and j atoms */
729 qq00 = _mm_mul_pd(iq0,jq0);
730 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
731 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
733 /* Calculate table index by multiplying r with table scale and truncate to integer */
734 rt = _mm_mul_pd(r00,vftabscale);
735 vfitab = _mm_cvttpd_epi32(rt);
736 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
737 vfitab = _mm_slli_epi32(vfitab,3);
739 /* COULOMB ELECTROSTATICS */
740 velec = _mm_mul_pd(qq00,rinv00);
741 felec = _mm_mul_pd(velec,rinvsq00);
743 /* CUBIC SPLINE TABLE DISPERSION */
744 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
745 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
746 GMX_MM_TRANSPOSE2_PD(Y,F);
747 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
748 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
749 GMX_MM_TRANSPOSE2_PD(G,H);
750 Heps = _mm_mul_pd(vfeps,H);
751 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
752 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
753 fvdw6 = _mm_mul_pd(c6_00,FF);
755 /* CUBIC SPLINE TABLE REPULSION */
756 vfitab = _mm_add_epi32(vfitab,ifour);
757 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
758 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
759 GMX_MM_TRANSPOSE2_PD(Y,F);
760 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
761 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
762 GMX_MM_TRANSPOSE2_PD(G,H);
763 Heps = _mm_mul_pd(vfeps,H);
764 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
765 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
766 fvdw12 = _mm_mul_pd(c12_00,FF);
767 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
769 fscal = _mm_add_pd(felec,fvdw);
771 /* Calculate temporary vectorial force */
772 tx = _mm_mul_pd(fscal,dx00);
773 ty = _mm_mul_pd(fscal,dy00);
774 tz = _mm_mul_pd(fscal,dz00);
776 /* Update vectorial force */
777 fix0 = _mm_add_pd(fix0,tx);
778 fiy0 = _mm_add_pd(fiy0,ty);
779 fiz0 = _mm_add_pd(fiz0,tz);
781 fjx0 = _mm_add_pd(fjx0,tx);
782 fjy0 = _mm_add_pd(fjy0,ty);
783 fjz0 = _mm_add_pd(fjz0,tz);
785 /**************************
786 * CALCULATE INTERACTIONS *
787 **************************/
789 /* Compute parameters for interactions between i and j atoms */
790 qq10 = _mm_mul_pd(iq1,jq0);
792 /* COULOMB ELECTROSTATICS */
793 velec = _mm_mul_pd(qq10,rinv10);
794 felec = _mm_mul_pd(velec,rinvsq10);
798 /* Calculate temporary vectorial force */
799 tx = _mm_mul_pd(fscal,dx10);
800 ty = _mm_mul_pd(fscal,dy10);
801 tz = _mm_mul_pd(fscal,dz10);
803 /* Update vectorial force */
804 fix1 = _mm_add_pd(fix1,tx);
805 fiy1 = _mm_add_pd(fiy1,ty);
806 fiz1 = _mm_add_pd(fiz1,tz);
808 fjx0 = _mm_add_pd(fjx0,tx);
809 fjy0 = _mm_add_pd(fjy0,ty);
810 fjz0 = _mm_add_pd(fjz0,tz);
812 /**************************
813 * CALCULATE INTERACTIONS *
814 **************************/
816 /* Compute parameters for interactions between i and j atoms */
817 qq20 = _mm_mul_pd(iq2,jq0);
819 /* COULOMB ELECTROSTATICS */
820 velec = _mm_mul_pd(qq20,rinv20);
821 felec = _mm_mul_pd(velec,rinvsq20);
825 /* Calculate temporary vectorial force */
826 tx = _mm_mul_pd(fscal,dx20);
827 ty = _mm_mul_pd(fscal,dy20);
828 tz = _mm_mul_pd(fscal,dz20);
830 /* Update vectorial force */
831 fix2 = _mm_add_pd(fix2,tx);
832 fiy2 = _mm_add_pd(fiy2,ty);
833 fiz2 = _mm_add_pd(fiz2,tz);
835 fjx0 = _mm_add_pd(fjx0,tx);
836 fjy0 = _mm_add_pd(fjy0,ty);
837 fjz0 = _mm_add_pd(fjz0,tz);
839 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
841 /* Inner loop uses 111 flops */
848 j_coord_offsetA = DIM*jnrA;
850 /* load j atom coordinates */
851 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
854 /* Calculate displacement vector */
855 dx00 = _mm_sub_pd(ix0,jx0);
856 dy00 = _mm_sub_pd(iy0,jy0);
857 dz00 = _mm_sub_pd(iz0,jz0);
858 dx10 = _mm_sub_pd(ix1,jx0);
859 dy10 = _mm_sub_pd(iy1,jy0);
860 dz10 = _mm_sub_pd(iz1,jz0);
861 dx20 = _mm_sub_pd(ix2,jx0);
862 dy20 = _mm_sub_pd(iy2,jy0);
863 dz20 = _mm_sub_pd(iz2,jz0);
865 /* Calculate squared distance and things based on it */
866 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
867 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
868 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
870 rinv00 = gmx_mm_invsqrt_pd(rsq00);
871 rinv10 = gmx_mm_invsqrt_pd(rsq10);
872 rinv20 = gmx_mm_invsqrt_pd(rsq20);
874 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
875 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
876 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
878 /* Load parameters for j particles */
879 jq0 = _mm_load_sd(charge+jnrA+0);
880 vdwjidx0A = 2*vdwtype[jnrA+0];
882 fjx0 = _mm_setzero_pd();
883 fjy0 = _mm_setzero_pd();
884 fjz0 = _mm_setzero_pd();
886 /**************************
887 * CALCULATE INTERACTIONS *
888 **************************/
890 r00 = _mm_mul_pd(rsq00,rinv00);
892 /* Compute parameters for interactions between i and j atoms */
893 qq00 = _mm_mul_pd(iq0,jq0);
894 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
896 /* Calculate table index by multiplying r with table scale and truncate to integer */
897 rt = _mm_mul_pd(r00,vftabscale);
898 vfitab = _mm_cvttpd_epi32(rt);
899 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
900 vfitab = _mm_slli_epi32(vfitab,3);
902 /* COULOMB ELECTROSTATICS */
903 velec = _mm_mul_pd(qq00,rinv00);
904 felec = _mm_mul_pd(velec,rinvsq00);
906 /* CUBIC SPLINE TABLE DISPERSION */
907 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
908 F = _mm_setzero_pd();
909 GMX_MM_TRANSPOSE2_PD(Y,F);
910 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
911 H = _mm_setzero_pd();
912 GMX_MM_TRANSPOSE2_PD(G,H);
913 Heps = _mm_mul_pd(vfeps,H);
914 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
915 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
916 fvdw6 = _mm_mul_pd(c6_00,FF);
918 /* CUBIC SPLINE TABLE REPULSION */
919 vfitab = _mm_add_epi32(vfitab,ifour);
920 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
921 F = _mm_setzero_pd();
922 GMX_MM_TRANSPOSE2_PD(Y,F);
923 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
924 H = _mm_setzero_pd();
925 GMX_MM_TRANSPOSE2_PD(G,H);
926 Heps = _mm_mul_pd(vfeps,H);
927 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
928 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
929 fvdw12 = _mm_mul_pd(c12_00,FF);
930 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
932 fscal = _mm_add_pd(felec,fvdw);
934 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
936 /* Calculate temporary vectorial force */
937 tx = _mm_mul_pd(fscal,dx00);
938 ty = _mm_mul_pd(fscal,dy00);
939 tz = _mm_mul_pd(fscal,dz00);
941 /* Update vectorial force */
942 fix0 = _mm_add_pd(fix0,tx);
943 fiy0 = _mm_add_pd(fiy0,ty);
944 fiz0 = _mm_add_pd(fiz0,tz);
946 fjx0 = _mm_add_pd(fjx0,tx);
947 fjy0 = _mm_add_pd(fjy0,ty);
948 fjz0 = _mm_add_pd(fjz0,tz);
950 /**************************
951 * CALCULATE INTERACTIONS *
952 **************************/
954 /* Compute parameters for interactions between i and j atoms */
955 qq10 = _mm_mul_pd(iq1,jq0);
957 /* COULOMB ELECTROSTATICS */
958 velec = _mm_mul_pd(qq10,rinv10);
959 felec = _mm_mul_pd(velec,rinvsq10);
963 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
965 /* Calculate temporary vectorial force */
966 tx = _mm_mul_pd(fscal,dx10);
967 ty = _mm_mul_pd(fscal,dy10);
968 tz = _mm_mul_pd(fscal,dz10);
970 /* Update vectorial force */
971 fix1 = _mm_add_pd(fix1,tx);
972 fiy1 = _mm_add_pd(fiy1,ty);
973 fiz1 = _mm_add_pd(fiz1,tz);
975 fjx0 = _mm_add_pd(fjx0,tx);
976 fjy0 = _mm_add_pd(fjy0,ty);
977 fjz0 = _mm_add_pd(fjz0,tz);
979 /**************************
980 * CALCULATE INTERACTIONS *
981 **************************/
983 /* Compute parameters for interactions between i and j atoms */
984 qq20 = _mm_mul_pd(iq2,jq0);
986 /* COULOMB ELECTROSTATICS */
987 velec = _mm_mul_pd(qq20,rinv20);
988 felec = _mm_mul_pd(velec,rinvsq20);
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);