2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012, by the GROMACS development team, led by
5 * David van der Spoel, Berk Hess, Erik Lindahl, and including many
6 * others, as listed in the AUTHORS file in the top-level source
7 * 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.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
49 #include "gmx_math_x86_sse2_double.h"
50 #include "kernelutil_x86_sse2_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_double
54 * Electrostatics interaction: GeneralizedBorn
55 * VdW interaction: CubicSplineTable
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_VF_sse2_double
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
77 int j_coord_offsetA,j_coord_offsetB;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
83 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
84 int vdwjidx0A,vdwjidx0B;
85 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
86 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
87 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
90 __m128d vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
91 __m128d minushalf = _mm_set1_pd(-0.5);
92 real *invsqrta,*dvda,*gbtab;
94 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
97 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
98 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
100 __m128i ifour = _mm_set1_epi32(4);
101 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
103 __m128d dummy_mask,cutoff_mask;
104 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
105 __m128d one = _mm_set1_pd(1.0);
106 __m128d two = _mm_set1_pd(2.0);
112 jindex = nlist->jindex;
114 shiftidx = nlist->shift;
116 shiftvec = fr->shift_vec[0];
117 fshift = fr->fshift[0];
118 facel = _mm_set1_pd(fr->epsfac);
119 charge = mdatoms->chargeA;
120 nvdwtype = fr->ntype;
122 vdwtype = mdatoms->typeA;
124 vftab = kernel_data->table_vdw->data;
125 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
127 invsqrta = fr->invsqrta;
129 gbtabscale = _mm_set1_pd(fr->gbtab.scale);
130 gbtab = fr->gbtab.data;
131 gbinvepsdiff = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
133 /* Avoid stupid compiler warnings */
141 /* Start outer loop over neighborlists */
142 for(iidx=0; iidx<nri; iidx++)
144 /* Load shift vector for this list */
145 i_shift_offset = DIM*shiftidx[iidx];
147 /* Load limits for loop over neighbors */
148 j_index_start = jindex[iidx];
149 j_index_end = jindex[iidx+1];
151 /* Get outer coordinate index */
153 i_coord_offset = DIM*inr;
155 /* Load i particle coords and add shift vector */
156 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
158 fix0 = _mm_setzero_pd();
159 fiy0 = _mm_setzero_pd();
160 fiz0 = _mm_setzero_pd();
162 /* Load parameters for i particles */
163 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
164 isai0 = _mm_load1_pd(invsqrta+inr+0);
165 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
167 /* Reset potential sums */
168 velecsum = _mm_setzero_pd();
169 vgbsum = _mm_setzero_pd();
170 vvdwsum = _mm_setzero_pd();
171 dvdasum = _mm_setzero_pd();
173 /* Start inner kernel loop */
174 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
177 /* Get j neighbor index, and coordinate index */
180 j_coord_offsetA = DIM*jnrA;
181 j_coord_offsetB = DIM*jnrB;
183 /* load j atom coordinates */
184 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
187 /* Calculate displacement vector */
188 dx00 = _mm_sub_pd(ix0,jx0);
189 dy00 = _mm_sub_pd(iy0,jy0);
190 dz00 = _mm_sub_pd(iz0,jz0);
192 /* Calculate squared distance and things based on it */
193 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
195 rinv00 = gmx_mm_invsqrt_pd(rsq00);
197 /* Load parameters for j particles */
198 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
199 isaj0 = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
200 vdwjidx0A = 2*vdwtype[jnrA+0];
201 vdwjidx0B = 2*vdwtype[jnrB+0];
203 /**************************
204 * CALCULATE INTERACTIONS *
205 **************************/
207 r00 = _mm_mul_pd(rsq00,rinv00);
209 /* Compute parameters for interactions between i and j atoms */
210 qq00 = _mm_mul_pd(iq0,jq0);
211 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
212 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
214 /* Calculate table index by multiplying r with table scale and truncate to integer */
215 rt = _mm_mul_pd(r00,vftabscale);
216 vfitab = _mm_cvttpd_epi32(rt);
217 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
218 vfitab = _mm_slli_epi32(vfitab,3);
220 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
221 isaprod = _mm_mul_pd(isai0,isaj0);
222 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
223 gbscale = _mm_mul_pd(isaprod,gbtabscale);
225 /* Calculate generalized born table index - this is a separate table from the normal one,
226 * but we use the same procedure by multiplying r with scale and truncating to integer.
228 rt = _mm_mul_pd(r00,gbscale);
229 gbitab = _mm_cvttpd_epi32(rt);
230 gbeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
231 gbitab = _mm_slli_epi32(gbitab,2);
233 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
234 F = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
235 GMX_MM_TRANSPOSE2_PD(Y,F);
236 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
237 H = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
238 GMX_MM_TRANSPOSE2_PD(G,H);
239 Heps = _mm_mul_pd(gbeps,H);
240 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
241 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
242 vgb = _mm_mul_pd(gbqqfactor,VV);
244 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
245 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
246 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
247 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
248 gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
249 velec = _mm_mul_pd(qq00,rinv00);
250 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
252 /* CUBIC SPLINE TABLE DISPERSION */
253 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
254 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
255 GMX_MM_TRANSPOSE2_PD(Y,F);
256 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
257 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
258 GMX_MM_TRANSPOSE2_PD(G,H);
259 Heps = _mm_mul_pd(vfeps,H);
260 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
261 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
262 vvdw6 = _mm_mul_pd(c6_00,VV);
263 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
264 fvdw6 = _mm_mul_pd(c6_00,FF);
266 /* CUBIC SPLINE TABLE REPULSION */
267 vfitab = _mm_add_epi32(vfitab,ifour);
268 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
269 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
270 GMX_MM_TRANSPOSE2_PD(Y,F);
271 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
272 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
273 GMX_MM_TRANSPOSE2_PD(G,H);
274 Heps = _mm_mul_pd(vfeps,H);
275 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
276 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
277 vvdw12 = _mm_mul_pd(c12_00,VV);
278 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
279 fvdw12 = _mm_mul_pd(c12_00,FF);
280 vvdw = _mm_add_pd(vvdw12,vvdw6);
281 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
283 /* Update potential sum for this i atom from the interaction with this j atom. */
284 velecsum = _mm_add_pd(velecsum,velec);
285 vgbsum = _mm_add_pd(vgbsum,vgb);
286 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
288 fscal = _mm_add_pd(felec,fvdw);
290 /* Calculate temporary vectorial force */
291 tx = _mm_mul_pd(fscal,dx00);
292 ty = _mm_mul_pd(fscal,dy00);
293 tz = _mm_mul_pd(fscal,dz00);
295 /* Update vectorial force */
296 fix0 = _mm_add_pd(fix0,tx);
297 fiy0 = _mm_add_pd(fiy0,ty);
298 fiz0 = _mm_add_pd(fiz0,tz);
300 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
302 /* Inner loop uses 92 flops */
309 j_coord_offsetA = DIM*jnrA;
311 /* load j atom coordinates */
312 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
315 /* Calculate displacement vector */
316 dx00 = _mm_sub_pd(ix0,jx0);
317 dy00 = _mm_sub_pd(iy0,jy0);
318 dz00 = _mm_sub_pd(iz0,jz0);
320 /* Calculate squared distance and things based on it */
321 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
323 rinv00 = gmx_mm_invsqrt_pd(rsq00);
325 /* Load parameters for j particles */
326 jq0 = _mm_load_sd(charge+jnrA+0);
327 isaj0 = _mm_load_sd(invsqrta+jnrA+0);
328 vdwjidx0A = 2*vdwtype[jnrA+0];
330 /**************************
331 * CALCULATE INTERACTIONS *
332 **************************/
334 r00 = _mm_mul_pd(rsq00,rinv00);
336 /* Compute parameters for interactions between i and j atoms */
337 qq00 = _mm_mul_pd(iq0,jq0);
338 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
340 /* Calculate table index by multiplying r with table scale and truncate to integer */
341 rt = _mm_mul_pd(r00,vftabscale);
342 vfitab = _mm_cvttpd_epi32(rt);
343 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
344 vfitab = _mm_slli_epi32(vfitab,3);
346 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
347 isaprod = _mm_mul_pd(isai0,isaj0);
348 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
349 gbscale = _mm_mul_pd(isaprod,gbtabscale);
351 /* Calculate generalized born table index - this is a separate table from the normal one,
352 * but we use the same procedure by multiplying r with scale and truncating to integer.
354 rt = _mm_mul_pd(r00,gbscale);
355 gbitab = _mm_cvttpd_epi32(rt);
356 gbeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
357 gbitab = _mm_slli_epi32(gbitab,2);
359 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
360 F = _mm_setzero_pd();
361 GMX_MM_TRANSPOSE2_PD(Y,F);
362 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
363 H = _mm_setzero_pd();
364 GMX_MM_TRANSPOSE2_PD(G,H);
365 Heps = _mm_mul_pd(gbeps,H);
366 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
367 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
368 vgb = _mm_mul_pd(gbqqfactor,VV);
370 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
371 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
372 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
373 dvdatmp = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
374 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
375 gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
376 velec = _mm_mul_pd(qq00,rinv00);
377 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
379 /* CUBIC SPLINE TABLE DISPERSION */
380 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
381 F = _mm_setzero_pd();
382 GMX_MM_TRANSPOSE2_PD(Y,F);
383 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
384 H = _mm_setzero_pd();
385 GMX_MM_TRANSPOSE2_PD(G,H);
386 Heps = _mm_mul_pd(vfeps,H);
387 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
388 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
389 vvdw6 = _mm_mul_pd(c6_00,VV);
390 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
391 fvdw6 = _mm_mul_pd(c6_00,FF);
393 /* CUBIC SPLINE TABLE REPULSION */
394 vfitab = _mm_add_epi32(vfitab,ifour);
395 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
396 F = _mm_setzero_pd();
397 GMX_MM_TRANSPOSE2_PD(Y,F);
398 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
399 H = _mm_setzero_pd();
400 GMX_MM_TRANSPOSE2_PD(G,H);
401 Heps = _mm_mul_pd(vfeps,H);
402 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
403 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
404 vvdw12 = _mm_mul_pd(c12_00,VV);
405 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
406 fvdw12 = _mm_mul_pd(c12_00,FF);
407 vvdw = _mm_add_pd(vvdw12,vvdw6);
408 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
410 /* Update potential sum for this i atom from the interaction with this j atom. */
411 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
412 velecsum = _mm_add_pd(velecsum,velec);
413 vgb = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
414 vgbsum = _mm_add_pd(vgbsum,vgb);
415 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
416 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
418 fscal = _mm_add_pd(felec,fvdw);
420 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
422 /* Calculate temporary vectorial force */
423 tx = _mm_mul_pd(fscal,dx00);
424 ty = _mm_mul_pd(fscal,dy00);
425 tz = _mm_mul_pd(fscal,dz00);
427 /* Update vectorial force */
428 fix0 = _mm_add_pd(fix0,tx);
429 fiy0 = _mm_add_pd(fiy0,ty);
430 fiz0 = _mm_add_pd(fiz0,tz);
432 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
434 /* Inner loop uses 92 flops */
437 /* End of innermost loop */
439 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
440 f+i_coord_offset,fshift+i_shift_offset);
443 /* Update potential energies */
444 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
445 gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
446 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
447 dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
448 gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
450 /* Increment number of inner iterations */
451 inneriter += j_index_end - j_index_start;
453 /* Outer loop uses 10 flops */
456 /* Increment number of outer iterations */
459 /* Update outer/inner flops */
461 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*92);
464 * Gromacs nonbonded kernel: nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_double
465 * Electrostatics interaction: GeneralizedBorn
466 * VdW interaction: CubicSplineTable
467 * Geometry: Particle-Particle
468 * Calculate force/pot: Force
471 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_sse2_double
472 (t_nblist * gmx_restrict nlist,
473 rvec * gmx_restrict xx,
474 rvec * gmx_restrict ff,
475 t_forcerec * gmx_restrict fr,
476 t_mdatoms * gmx_restrict mdatoms,
477 nb_kernel_data_t * gmx_restrict kernel_data,
478 t_nrnb * gmx_restrict nrnb)
480 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
481 * just 0 for non-waters.
482 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
483 * jnr indices corresponding to data put in the four positions in the SIMD register.
485 int i_shift_offset,i_coord_offset,outeriter,inneriter;
486 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
488 int j_coord_offsetA,j_coord_offsetB;
489 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
491 real *shiftvec,*fshift,*x,*f;
492 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
494 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
495 int vdwjidx0A,vdwjidx0B;
496 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
497 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
498 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
501 __m128d vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
502 __m128d minushalf = _mm_set1_pd(-0.5);
503 real *invsqrta,*dvda,*gbtab;
505 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
508 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
509 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
511 __m128i ifour = _mm_set1_epi32(4);
512 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
514 __m128d dummy_mask,cutoff_mask;
515 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
516 __m128d one = _mm_set1_pd(1.0);
517 __m128d two = _mm_set1_pd(2.0);
523 jindex = nlist->jindex;
525 shiftidx = nlist->shift;
527 shiftvec = fr->shift_vec[0];
528 fshift = fr->fshift[0];
529 facel = _mm_set1_pd(fr->epsfac);
530 charge = mdatoms->chargeA;
531 nvdwtype = fr->ntype;
533 vdwtype = mdatoms->typeA;
535 vftab = kernel_data->table_vdw->data;
536 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
538 invsqrta = fr->invsqrta;
540 gbtabscale = _mm_set1_pd(fr->gbtab.scale);
541 gbtab = fr->gbtab.data;
542 gbinvepsdiff = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
544 /* Avoid stupid compiler warnings */
552 /* Start outer loop over neighborlists */
553 for(iidx=0; iidx<nri; iidx++)
555 /* Load shift vector for this list */
556 i_shift_offset = DIM*shiftidx[iidx];
558 /* Load limits for loop over neighbors */
559 j_index_start = jindex[iidx];
560 j_index_end = jindex[iidx+1];
562 /* Get outer coordinate index */
564 i_coord_offset = DIM*inr;
566 /* Load i particle coords and add shift vector */
567 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
569 fix0 = _mm_setzero_pd();
570 fiy0 = _mm_setzero_pd();
571 fiz0 = _mm_setzero_pd();
573 /* Load parameters for i particles */
574 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
575 isai0 = _mm_load1_pd(invsqrta+inr+0);
576 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
578 dvdasum = _mm_setzero_pd();
580 /* Start inner kernel loop */
581 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
584 /* Get j neighbor index, and coordinate index */
587 j_coord_offsetA = DIM*jnrA;
588 j_coord_offsetB = DIM*jnrB;
590 /* load j atom coordinates */
591 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
594 /* Calculate displacement vector */
595 dx00 = _mm_sub_pd(ix0,jx0);
596 dy00 = _mm_sub_pd(iy0,jy0);
597 dz00 = _mm_sub_pd(iz0,jz0);
599 /* Calculate squared distance and things based on it */
600 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
602 rinv00 = gmx_mm_invsqrt_pd(rsq00);
604 /* Load parameters for j particles */
605 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
606 isaj0 = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
607 vdwjidx0A = 2*vdwtype[jnrA+0];
608 vdwjidx0B = 2*vdwtype[jnrB+0];
610 /**************************
611 * CALCULATE INTERACTIONS *
612 **************************/
614 r00 = _mm_mul_pd(rsq00,rinv00);
616 /* Compute parameters for interactions between i and j atoms */
617 qq00 = _mm_mul_pd(iq0,jq0);
618 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
619 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
621 /* Calculate table index by multiplying r with table scale and truncate to integer */
622 rt = _mm_mul_pd(r00,vftabscale);
623 vfitab = _mm_cvttpd_epi32(rt);
624 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
625 vfitab = _mm_slli_epi32(vfitab,3);
627 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
628 isaprod = _mm_mul_pd(isai0,isaj0);
629 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
630 gbscale = _mm_mul_pd(isaprod,gbtabscale);
632 /* Calculate generalized born table index - this is a separate table from the normal one,
633 * but we use the same procedure by multiplying r with scale and truncating to integer.
635 rt = _mm_mul_pd(r00,gbscale);
636 gbitab = _mm_cvttpd_epi32(rt);
637 gbeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
638 gbitab = _mm_slli_epi32(gbitab,2);
640 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
641 F = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
642 GMX_MM_TRANSPOSE2_PD(Y,F);
643 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
644 H = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
645 GMX_MM_TRANSPOSE2_PD(G,H);
646 Heps = _mm_mul_pd(gbeps,H);
647 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
648 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
649 vgb = _mm_mul_pd(gbqqfactor,VV);
651 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
652 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
653 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
654 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
655 gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
656 velec = _mm_mul_pd(qq00,rinv00);
657 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
659 /* CUBIC SPLINE TABLE DISPERSION */
660 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
661 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
662 GMX_MM_TRANSPOSE2_PD(Y,F);
663 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
664 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
665 GMX_MM_TRANSPOSE2_PD(G,H);
666 Heps = _mm_mul_pd(vfeps,H);
667 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
668 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
669 fvdw6 = _mm_mul_pd(c6_00,FF);
671 /* CUBIC SPLINE TABLE REPULSION */
672 vfitab = _mm_add_epi32(vfitab,ifour);
673 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
674 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
675 GMX_MM_TRANSPOSE2_PD(Y,F);
676 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
677 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
678 GMX_MM_TRANSPOSE2_PD(G,H);
679 Heps = _mm_mul_pd(vfeps,H);
680 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
681 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
682 fvdw12 = _mm_mul_pd(c12_00,FF);
683 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
685 fscal = _mm_add_pd(felec,fvdw);
687 /* Calculate temporary vectorial force */
688 tx = _mm_mul_pd(fscal,dx00);
689 ty = _mm_mul_pd(fscal,dy00);
690 tz = _mm_mul_pd(fscal,dz00);
692 /* Update vectorial force */
693 fix0 = _mm_add_pd(fix0,tx);
694 fiy0 = _mm_add_pd(fiy0,ty);
695 fiz0 = _mm_add_pd(fiz0,tz);
697 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
699 /* Inner loop uses 82 flops */
706 j_coord_offsetA = DIM*jnrA;
708 /* load j atom coordinates */
709 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
712 /* Calculate displacement vector */
713 dx00 = _mm_sub_pd(ix0,jx0);
714 dy00 = _mm_sub_pd(iy0,jy0);
715 dz00 = _mm_sub_pd(iz0,jz0);
717 /* Calculate squared distance and things based on it */
718 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
720 rinv00 = gmx_mm_invsqrt_pd(rsq00);
722 /* Load parameters for j particles */
723 jq0 = _mm_load_sd(charge+jnrA+0);
724 isaj0 = _mm_load_sd(invsqrta+jnrA+0);
725 vdwjidx0A = 2*vdwtype[jnrA+0];
727 /**************************
728 * CALCULATE INTERACTIONS *
729 **************************/
731 r00 = _mm_mul_pd(rsq00,rinv00);
733 /* Compute parameters for interactions between i and j atoms */
734 qq00 = _mm_mul_pd(iq0,jq0);
735 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
737 /* Calculate table index by multiplying r with table scale and truncate to integer */
738 rt = _mm_mul_pd(r00,vftabscale);
739 vfitab = _mm_cvttpd_epi32(rt);
740 vfeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
741 vfitab = _mm_slli_epi32(vfitab,3);
743 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
744 isaprod = _mm_mul_pd(isai0,isaj0);
745 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
746 gbscale = _mm_mul_pd(isaprod,gbtabscale);
748 /* Calculate generalized born table index - this is a separate table from the normal one,
749 * but we use the same procedure by multiplying r with scale and truncating to integer.
751 rt = _mm_mul_pd(r00,gbscale);
752 gbitab = _mm_cvttpd_epi32(rt);
753 gbeps = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
754 gbitab = _mm_slli_epi32(gbitab,2);
756 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
757 F = _mm_setzero_pd();
758 GMX_MM_TRANSPOSE2_PD(Y,F);
759 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
760 H = _mm_setzero_pd();
761 GMX_MM_TRANSPOSE2_PD(G,H);
762 Heps = _mm_mul_pd(gbeps,H);
763 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
764 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
765 vgb = _mm_mul_pd(gbqqfactor,VV);
767 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
768 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
769 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
770 dvdatmp = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
771 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
772 gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
773 velec = _mm_mul_pd(qq00,rinv00);
774 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
776 /* CUBIC SPLINE TABLE DISPERSION */
777 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
778 F = _mm_setzero_pd();
779 GMX_MM_TRANSPOSE2_PD(Y,F);
780 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
781 H = _mm_setzero_pd();
782 GMX_MM_TRANSPOSE2_PD(G,H);
783 Heps = _mm_mul_pd(vfeps,H);
784 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
785 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
786 fvdw6 = _mm_mul_pd(c6_00,FF);
788 /* CUBIC SPLINE TABLE REPULSION */
789 vfitab = _mm_add_epi32(vfitab,ifour);
790 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
791 F = _mm_setzero_pd();
792 GMX_MM_TRANSPOSE2_PD(Y,F);
793 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
794 H = _mm_setzero_pd();
795 GMX_MM_TRANSPOSE2_PD(G,H);
796 Heps = _mm_mul_pd(vfeps,H);
797 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
798 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
799 fvdw12 = _mm_mul_pd(c12_00,FF);
800 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
802 fscal = _mm_add_pd(felec,fvdw);
804 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
806 /* Calculate temporary vectorial force */
807 tx = _mm_mul_pd(fscal,dx00);
808 ty = _mm_mul_pd(fscal,dy00);
809 tz = _mm_mul_pd(fscal,dz00);
811 /* Update vectorial force */
812 fix0 = _mm_add_pd(fix0,tx);
813 fiy0 = _mm_add_pd(fiy0,ty);
814 fiz0 = _mm_add_pd(fiz0,tz);
816 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
818 /* Inner loop uses 82 flops */
821 /* End of innermost loop */
823 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
824 f+i_coord_offset,fshift+i_shift_offset);
826 dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
827 gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
829 /* Increment number of inner iterations */
830 inneriter += j_index_end - j_index_start;
832 /* Outer loop uses 7 flops */
835 /* Increment number of outer iterations */
838 /* Update outer/inner flops */
840 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*82);