2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS sse2_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_sse2_double.h"
48 #include "kernelutil_x86_sse2_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_double
52 * Electrostatics interaction: GeneralizedBorn
53 * VdW interaction: CubicSplineTable
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
82 int vdwjidx0A,vdwjidx0B;
83 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
84 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
85 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
88 __m128d vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
89 __m128d minushalf = _mm_set1_pd(-0.5);
90 real *invsqrta,*dvda,*gbtab;
92 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
95 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
96 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
98 __m128i ifour = _mm_set1_epi32(4);
99 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
101 __m128d dummy_mask,cutoff_mask;
102 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
103 __m128d one = _mm_set1_pd(1.0);
104 __m128d two = _mm_set1_pd(2.0);
110 jindex = nlist->jindex;
112 shiftidx = nlist->shift;
114 shiftvec = fr->shift_vec[0];
115 fshift = fr->fshift[0];
116 facel = _mm_set1_pd(fr->epsfac);
117 charge = mdatoms->chargeA;
118 nvdwtype = fr->ntype;
120 vdwtype = mdatoms->typeA;
122 vftab = kernel_data->table_vdw->data;
123 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
125 invsqrta = fr->invsqrta;
127 gbtabscale = _mm_set1_pd(fr->gbtab.scale);
128 gbtab = fr->gbtab.data;
129 gbinvepsdiff = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
131 /* Avoid stupid compiler warnings */
139 /* Start outer loop over neighborlists */
140 for(iidx=0; iidx<nri; iidx++)
142 /* Load shift vector for this list */
143 i_shift_offset = DIM*shiftidx[iidx];
145 /* Load limits for loop over neighbors */
146 j_index_start = jindex[iidx];
147 j_index_end = jindex[iidx+1];
149 /* Get outer coordinate index */
151 i_coord_offset = DIM*inr;
153 /* Load i particle coords and add shift vector */
154 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
156 fix0 = _mm_setzero_pd();
157 fiy0 = _mm_setzero_pd();
158 fiz0 = _mm_setzero_pd();
160 /* Load parameters for i particles */
161 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
162 isai0 = _mm_load1_pd(invsqrta+inr+0);
163 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
165 /* Reset potential sums */
166 velecsum = _mm_setzero_pd();
167 vgbsum = _mm_setzero_pd();
168 vvdwsum = _mm_setzero_pd();
169 dvdasum = _mm_setzero_pd();
171 /* Start inner kernel loop */
172 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
175 /* Get j neighbor index, and coordinate index */
178 j_coord_offsetA = DIM*jnrA;
179 j_coord_offsetB = DIM*jnrB;
181 /* load j atom coordinates */
182 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
185 /* Calculate displacement vector */
186 dx00 = _mm_sub_pd(ix0,jx0);
187 dy00 = _mm_sub_pd(iy0,jy0);
188 dz00 = _mm_sub_pd(iz0,jz0);
190 /* Calculate squared distance and things based on it */
191 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
193 rinv00 = gmx_mm_invsqrt_pd(rsq00);
195 /* Load parameters for j particles */
196 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
197 isaj0 = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
198 vdwjidx0A = 2*vdwtype[jnrA+0];
199 vdwjidx0B = 2*vdwtype[jnrB+0];
201 /**************************
202 * CALCULATE INTERACTIONS *
203 **************************/
205 r00 = _mm_mul_pd(rsq00,rinv00);
207 /* Compute parameters for interactions between i and j atoms */
208 qq00 = _mm_mul_pd(iq0,jq0);
209 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
210 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
212 /* Calculate table index by multiplying r with table scale and truncate to integer */
213 rt = _mm_mul_pd(r00,vftabscale);
214 vfitab = _mm_cvttpd_epi32(rt);
215 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
216 vfitab = _mm_slli_epi32(vfitab,3);
218 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
219 isaprod = _mm_mul_pd(isai0,isaj0);
220 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
221 gbscale = _mm_mul_pd(isaprod,gbtabscale);
223 /* Calculate generalized born table index - this is a separate table from the normal one,
224 * but we use the same procedure by multiplying r with scale and truncating to integer.
226 rt = _mm_mul_pd(r00,gbscale);
227 gbitab = _mm_cvttpd_epi32(rt);
228 gbeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
229 gbitab = _mm_slli_epi32(gbitab,2);
231 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
232 F = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
233 GMX_MM_TRANSPOSE2_PD(Y,F);
234 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
235 H = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
236 GMX_MM_TRANSPOSE2_PD(G,H);
237 Heps = _mm_mul_pd(gbeps,H);
238 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
239 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
240 vgb = _mm_mul_pd(gbqqfactor,VV);
242 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
243 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
244 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
245 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
246 gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
247 velec = _mm_mul_pd(qq00,rinv00);
248 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
250 /* CUBIC SPLINE TABLE DISPERSION */
251 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
252 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
253 GMX_MM_TRANSPOSE2_PD(Y,F);
254 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
255 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
256 GMX_MM_TRANSPOSE2_PD(G,H);
257 Heps = _mm_mul_pd(vfeps,H);
258 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
259 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
260 vvdw6 = _mm_mul_pd(c6_00,VV);
261 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
262 fvdw6 = _mm_mul_pd(c6_00,FF);
264 /* CUBIC SPLINE TABLE REPULSION */
265 vfitab = _mm_add_epi32(vfitab,ifour);
266 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
267 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
268 GMX_MM_TRANSPOSE2_PD(Y,F);
269 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
270 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
271 GMX_MM_TRANSPOSE2_PD(G,H);
272 Heps = _mm_mul_pd(vfeps,H);
273 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
274 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
275 vvdw12 = _mm_mul_pd(c12_00,VV);
276 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
277 fvdw12 = _mm_mul_pd(c12_00,FF);
278 vvdw = _mm_add_pd(vvdw12,vvdw6);
279 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
281 /* Update potential sum for this i atom from the interaction with this j atom. */
282 velecsum = _mm_add_pd(velecsum,velec);
283 vgbsum = _mm_add_pd(vgbsum,vgb);
284 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
286 fscal = _mm_add_pd(felec,fvdw);
288 /* Calculate temporary vectorial force */
289 tx = _mm_mul_pd(fscal,dx00);
290 ty = _mm_mul_pd(fscal,dy00);
291 tz = _mm_mul_pd(fscal,dz00);
293 /* Update vectorial force */
294 fix0 = _mm_add_pd(fix0,tx);
295 fiy0 = _mm_add_pd(fiy0,ty);
296 fiz0 = _mm_add_pd(fiz0,tz);
298 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
300 /* Inner loop uses 92 flops */
307 j_coord_offsetA = DIM*jnrA;
309 /* load j atom coordinates */
310 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
313 /* Calculate displacement vector */
314 dx00 = _mm_sub_pd(ix0,jx0);
315 dy00 = _mm_sub_pd(iy0,jy0);
316 dz00 = _mm_sub_pd(iz0,jz0);
318 /* Calculate squared distance and things based on it */
319 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
321 rinv00 = gmx_mm_invsqrt_pd(rsq00);
323 /* Load parameters for j particles */
324 jq0 = _mm_load_sd(charge+jnrA+0);
325 isaj0 = _mm_load_sd(invsqrta+jnrA+0);
326 vdwjidx0A = 2*vdwtype[jnrA+0];
328 /**************************
329 * CALCULATE INTERACTIONS *
330 **************************/
332 r00 = _mm_mul_pd(rsq00,rinv00);
334 /* Compute parameters for interactions between i and j atoms */
335 qq00 = _mm_mul_pd(iq0,jq0);
336 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
338 /* Calculate table index by multiplying r with table scale and truncate to integer */
339 rt = _mm_mul_pd(r00,vftabscale);
340 vfitab = _mm_cvttpd_epi32(rt);
341 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
342 vfitab = _mm_slli_epi32(vfitab,3);
344 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
345 isaprod = _mm_mul_pd(isai0,isaj0);
346 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
347 gbscale = _mm_mul_pd(isaprod,gbtabscale);
349 /* Calculate generalized born table index - this is a separate table from the normal one,
350 * but we use the same procedure by multiplying r with scale and truncating to integer.
352 rt = _mm_mul_pd(r00,gbscale);
353 gbitab = _mm_cvttpd_epi32(rt);
354 gbeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
355 gbitab = _mm_slli_epi32(gbitab,2);
357 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
358 F = _mm_setzero_pd();
359 GMX_MM_TRANSPOSE2_PD(Y,F);
360 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
361 H = _mm_setzero_pd();
362 GMX_MM_TRANSPOSE2_PD(G,H);
363 Heps = _mm_mul_pd(gbeps,H);
364 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
365 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
366 vgb = _mm_mul_pd(gbqqfactor,VV);
368 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
369 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
370 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
371 dvdatmp = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
372 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
373 gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
374 velec = _mm_mul_pd(qq00,rinv00);
375 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
377 /* CUBIC SPLINE TABLE DISPERSION */
378 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
379 F = _mm_setzero_pd();
380 GMX_MM_TRANSPOSE2_PD(Y,F);
381 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
382 H = _mm_setzero_pd();
383 GMX_MM_TRANSPOSE2_PD(G,H);
384 Heps = _mm_mul_pd(vfeps,H);
385 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
386 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
387 vvdw6 = _mm_mul_pd(c6_00,VV);
388 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
389 fvdw6 = _mm_mul_pd(c6_00,FF);
391 /* CUBIC SPLINE TABLE REPULSION */
392 vfitab = _mm_add_epi32(vfitab,ifour);
393 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
394 F = _mm_setzero_pd();
395 GMX_MM_TRANSPOSE2_PD(Y,F);
396 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
397 H = _mm_setzero_pd();
398 GMX_MM_TRANSPOSE2_PD(G,H);
399 Heps = _mm_mul_pd(vfeps,H);
400 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
401 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
402 vvdw12 = _mm_mul_pd(c12_00,VV);
403 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
404 fvdw12 = _mm_mul_pd(c12_00,FF);
405 vvdw = _mm_add_pd(vvdw12,vvdw6);
406 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
408 /* Update potential sum for this i atom from the interaction with this j atom. */
409 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
410 velecsum = _mm_add_pd(velecsum,velec);
411 vgb = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
412 vgbsum = _mm_add_pd(vgbsum,vgb);
413 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
414 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
416 fscal = _mm_add_pd(felec,fvdw);
418 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
420 /* Calculate temporary vectorial force */
421 tx = _mm_mul_pd(fscal,dx00);
422 ty = _mm_mul_pd(fscal,dy00);
423 tz = _mm_mul_pd(fscal,dz00);
425 /* Update vectorial force */
426 fix0 = _mm_add_pd(fix0,tx);
427 fiy0 = _mm_add_pd(fiy0,ty);
428 fiz0 = _mm_add_pd(fiz0,tz);
430 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
432 /* Inner loop uses 92 flops */
435 /* End of innermost loop */
437 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
438 f+i_coord_offset,fshift+i_shift_offset);
441 /* Update potential energies */
442 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
443 gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
444 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
445 dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
446 gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
448 /* Increment number of inner iterations */
449 inneriter += j_index_end - j_index_start;
451 /* Outer loop uses 10 flops */
454 /* Increment number of outer iterations */
457 /* Update outer/inner flops */
459 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*92);
462 * Gromacs nonbonded kernel: nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_double
463 * Electrostatics interaction: GeneralizedBorn
464 * VdW interaction: CubicSplineTable
465 * Geometry: Particle-Particle
466 * Calculate force/pot: Force
469 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_double
470 (t_nblist * gmx_restrict nlist,
471 rvec * gmx_restrict xx,
472 rvec * gmx_restrict ff,
473 t_forcerec * gmx_restrict fr,
474 t_mdatoms * gmx_restrict mdatoms,
475 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
476 t_nrnb * gmx_restrict nrnb)
478 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
479 * just 0 for non-waters.
480 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
481 * jnr indices corresponding to data put in the four positions in the SIMD register.
483 int i_shift_offset,i_coord_offset,outeriter,inneriter;
484 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
486 int j_coord_offsetA,j_coord_offsetB;
487 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
489 real *shiftvec,*fshift,*x,*f;
490 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
492 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
493 int vdwjidx0A,vdwjidx0B;
494 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
495 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
496 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
499 __m128d vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
500 __m128d minushalf = _mm_set1_pd(-0.5);
501 real *invsqrta,*dvda,*gbtab;
503 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
506 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
507 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
509 __m128i ifour = _mm_set1_epi32(4);
510 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
512 __m128d dummy_mask,cutoff_mask;
513 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
514 __m128d one = _mm_set1_pd(1.0);
515 __m128d two = _mm_set1_pd(2.0);
521 jindex = nlist->jindex;
523 shiftidx = nlist->shift;
525 shiftvec = fr->shift_vec[0];
526 fshift = fr->fshift[0];
527 facel = _mm_set1_pd(fr->epsfac);
528 charge = mdatoms->chargeA;
529 nvdwtype = fr->ntype;
531 vdwtype = mdatoms->typeA;
533 vftab = kernel_data->table_vdw->data;
534 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
536 invsqrta = fr->invsqrta;
538 gbtabscale = _mm_set1_pd(fr->gbtab.scale);
539 gbtab = fr->gbtab.data;
540 gbinvepsdiff = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
542 /* Avoid stupid compiler warnings */
550 /* Start outer loop over neighborlists */
551 for(iidx=0; iidx<nri; iidx++)
553 /* Load shift vector for this list */
554 i_shift_offset = DIM*shiftidx[iidx];
556 /* Load limits for loop over neighbors */
557 j_index_start = jindex[iidx];
558 j_index_end = jindex[iidx+1];
560 /* Get outer coordinate index */
562 i_coord_offset = DIM*inr;
564 /* Load i particle coords and add shift vector */
565 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
567 fix0 = _mm_setzero_pd();
568 fiy0 = _mm_setzero_pd();
569 fiz0 = _mm_setzero_pd();
571 /* Load parameters for i particles */
572 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
573 isai0 = _mm_load1_pd(invsqrta+inr+0);
574 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
576 dvdasum = _mm_setzero_pd();
578 /* Start inner kernel loop */
579 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
582 /* Get j neighbor index, and coordinate index */
585 j_coord_offsetA = DIM*jnrA;
586 j_coord_offsetB = DIM*jnrB;
588 /* load j atom coordinates */
589 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
592 /* Calculate displacement vector */
593 dx00 = _mm_sub_pd(ix0,jx0);
594 dy00 = _mm_sub_pd(iy0,jy0);
595 dz00 = _mm_sub_pd(iz0,jz0);
597 /* Calculate squared distance and things based on it */
598 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
600 rinv00 = gmx_mm_invsqrt_pd(rsq00);
602 /* Load parameters for j particles */
603 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
604 isaj0 = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
605 vdwjidx0A = 2*vdwtype[jnrA+0];
606 vdwjidx0B = 2*vdwtype[jnrB+0];
608 /**************************
609 * CALCULATE INTERACTIONS *
610 **************************/
612 r00 = _mm_mul_pd(rsq00,rinv00);
614 /* Compute parameters for interactions between i and j atoms */
615 qq00 = _mm_mul_pd(iq0,jq0);
616 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
617 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
619 /* Calculate table index by multiplying r with table scale and truncate to integer */
620 rt = _mm_mul_pd(r00,vftabscale);
621 vfitab = _mm_cvttpd_epi32(rt);
622 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
623 vfitab = _mm_slli_epi32(vfitab,3);
625 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
626 isaprod = _mm_mul_pd(isai0,isaj0);
627 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
628 gbscale = _mm_mul_pd(isaprod,gbtabscale);
630 /* Calculate generalized born table index - this is a separate table from the normal one,
631 * but we use the same procedure by multiplying r with scale and truncating to integer.
633 rt = _mm_mul_pd(r00,gbscale);
634 gbitab = _mm_cvttpd_epi32(rt);
635 gbeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
636 gbitab = _mm_slli_epi32(gbitab,2);
638 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
639 F = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
640 GMX_MM_TRANSPOSE2_PD(Y,F);
641 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
642 H = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
643 GMX_MM_TRANSPOSE2_PD(G,H);
644 Heps = _mm_mul_pd(gbeps,H);
645 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
646 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
647 vgb = _mm_mul_pd(gbqqfactor,VV);
649 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
650 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
651 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
652 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
653 gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
654 velec = _mm_mul_pd(qq00,rinv00);
655 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
657 /* CUBIC SPLINE TABLE DISPERSION */
658 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
659 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
660 GMX_MM_TRANSPOSE2_PD(Y,F);
661 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
662 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
663 GMX_MM_TRANSPOSE2_PD(G,H);
664 Heps = _mm_mul_pd(vfeps,H);
665 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
666 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
667 fvdw6 = _mm_mul_pd(c6_00,FF);
669 /* CUBIC SPLINE TABLE REPULSION */
670 vfitab = _mm_add_epi32(vfitab,ifour);
671 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
672 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
673 GMX_MM_TRANSPOSE2_PD(Y,F);
674 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
675 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
676 GMX_MM_TRANSPOSE2_PD(G,H);
677 Heps = _mm_mul_pd(vfeps,H);
678 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
679 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
680 fvdw12 = _mm_mul_pd(c12_00,FF);
681 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
683 fscal = _mm_add_pd(felec,fvdw);
685 /* Calculate temporary vectorial force */
686 tx = _mm_mul_pd(fscal,dx00);
687 ty = _mm_mul_pd(fscal,dy00);
688 tz = _mm_mul_pd(fscal,dz00);
690 /* Update vectorial force */
691 fix0 = _mm_add_pd(fix0,tx);
692 fiy0 = _mm_add_pd(fiy0,ty);
693 fiz0 = _mm_add_pd(fiz0,tz);
695 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
697 /* Inner loop uses 82 flops */
704 j_coord_offsetA = DIM*jnrA;
706 /* load j atom coordinates */
707 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
710 /* Calculate displacement vector */
711 dx00 = _mm_sub_pd(ix0,jx0);
712 dy00 = _mm_sub_pd(iy0,jy0);
713 dz00 = _mm_sub_pd(iz0,jz0);
715 /* Calculate squared distance and things based on it */
716 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
718 rinv00 = gmx_mm_invsqrt_pd(rsq00);
720 /* Load parameters for j particles */
721 jq0 = _mm_load_sd(charge+jnrA+0);
722 isaj0 = _mm_load_sd(invsqrta+jnrA+0);
723 vdwjidx0A = 2*vdwtype[jnrA+0];
725 /**************************
726 * CALCULATE INTERACTIONS *
727 **************************/
729 r00 = _mm_mul_pd(rsq00,rinv00);
731 /* Compute parameters for interactions between i and j atoms */
732 qq00 = _mm_mul_pd(iq0,jq0);
733 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
735 /* Calculate table index by multiplying r with table scale and truncate to integer */
736 rt = _mm_mul_pd(r00,vftabscale);
737 vfitab = _mm_cvttpd_epi32(rt);
738 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
739 vfitab = _mm_slli_epi32(vfitab,3);
741 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
742 isaprod = _mm_mul_pd(isai0,isaj0);
743 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
744 gbscale = _mm_mul_pd(isaprod,gbtabscale);
746 /* Calculate generalized born table index - this is a separate table from the normal one,
747 * but we use the same procedure by multiplying r with scale and truncating to integer.
749 rt = _mm_mul_pd(r00,gbscale);
750 gbitab = _mm_cvttpd_epi32(rt);
751 gbeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
752 gbitab = _mm_slli_epi32(gbitab,2);
754 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
755 F = _mm_setzero_pd();
756 GMX_MM_TRANSPOSE2_PD(Y,F);
757 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
758 H = _mm_setzero_pd();
759 GMX_MM_TRANSPOSE2_PD(G,H);
760 Heps = _mm_mul_pd(gbeps,H);
761 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
762 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
763 vgb = _mm_mul_pd(gbqqfactor,VV);
765 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
766 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
767 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
768 dvdatmp = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
769 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
770 gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
771 velec = _mm_mul_pd(qq00,rinv00);
772 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
774 /* CUBIC SPLINE TABLE DISPERSION */
775 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
776 F = _mm_setzero_pd();
777 GMX_MM_TRANSPOSE2_PD(Y,F);
778 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
779 H = _mm_setzero_pd();
780 GMX_MM_TRANSPOSE2_PD(G,H);
781 Heps = _mm_mul_pd(vfeps,H);
782 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
783 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
784 fvdw6 = _mm_mul_pd(c6_00,FF);
786 /* CUBIC SPLINE TABLE REPULSION */
787 vfitab = _mm_add_epi32(vfitab,ifour);
788 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
789 F = _mm_setzero_pd();
790 GMX_MM_TRANSPOSE2_PD(Y,F);
791 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
792 H = _mm_setzero_pd();
793 GMX_MM_TRANSPOSE2_PD(G,H);
794 Heps = _mm_mul_pd(vfeps,H);
795 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
796 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
797 fvdw12 = _mm_mul_pd(c12_00,FF);
798 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
800 fscal = _mm_add_pd(felec,fvdw);
802 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
804 /* Calculate temporary vectorial force */
805 tx = _mm_mul_pd(fscal,dx00);
806 ty = _mm_mul_pd(fscal,dy00);
807 tz = _mm_mul_pd(fscal,dz00);
809 /* Update vectorial force */
810 fix0 = _mm_add_pd(fix0,tx);
811 fiy0 = _mm_add_pd(fiy0,ty);
812 fiz0 = _mm_add_pd(fiz0,tz);
814 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
816 /* Inner loop uses 82 flops */
819 /* End of innermost loop */
821 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
822 f+i_coord_offset,fshift+i_shift_offset);
824 dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
825 gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
827 /* Increment number of inner iterations */
828 inneriter += j_index_end - j_index_start;
830 /* Outer loop uses 7 flops */
833 /* Increment number of outer iterations */
836 /* Update outer/inner flops */
838 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*82);