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 avx_128_fma_double kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
50 #include "kernelutil_x86_avx_128_fma_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
54 * Electrostatics interaction: Ewald
55 * VdW interaction: CubicSplineTable
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_avx_128_fma_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_unused * 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 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
93 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
94 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
96 __m128i ifour = _mm_set1_epi32(4);
97 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
100 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
102 __m128d dummy_mask,cutoff_mask;
103 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
104 __m128d one = _mm_set1_pd(1.0);
105 __m128d two = _mm_set1_pd(2.0);
111 jindex = nlist->jindex;
113 shiftidx = nlist->shift;
115 shiftvec = fr->shift_vec[0];
116 fshift = fr->fshift[0];
117 facel = _mm_set1_pd(fr->epsfac);
118 charge = mdatoms->chargeA;
119 nvdwtype = fr->ntype;
121 vdwtype = mdatoms->typeA;
123 vftab = kernel_data->table_vdw->data;
124 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
126 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
127 ewtab = fr->ic->tabq_coul_FDV0;
128 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
129 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
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 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
164 /* Reset potential sums */
165 velecsum = _mm_setzero_pd();
166 vvdwsum = _mm_setzero_pd();
168 /* Start inner kernel loop */
169 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
172 /* Get j neighbor index, and coordinate index */
175 j_coord_offsetA = DIM*jnrA;
176 j_coord_offsetB = DIM*jnrB;
178 /* load j atom coordinates */
179 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
182 /* Calculate displacement vector */
183 dx00 = _mm_sub_pd(ix0,jx0);
184 dy00 = _mm_sub_pd(iy0,jy0);
185 dz00 = _mm_sub_pd(iz0,jz0);
187 /* Calculate squared distance and things based on it */
188 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
190 rinv00 = gmx_mm_invsqrt_pd(rsq00);
192 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
194 /* Load parameters for j particles */
195 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
196 vdwjidx0A = 2*vdwtype[jnrA+0];
197 vdwjidx0B = 2*vdwtype[jnrB+0];
199 /**************************
200 * CALCULATE INTERACTIONS *
201 **************************/
203 r00 = _mm_mul_pd(rsq00,rinv00);
205 /* Compute parameters for interactions between i and j atoms */
206 qq00 = _mm_mul_pd(iq0,jq0);
207 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
208 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
210 /* Calculate table index by multiplying r with table scale and truncate to integer */
211 rt = _mm_mul_pd(r00,vftabscale);
212 vfitab = _mm_cvttpd_epi32(rt);
214 vfeps = _mm_frcz_pd(rt);
216 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
218 twovfeps = _mm_add_pd(vfeps,vfeps);
219 vfitab = _mm_slli_epi32(vfitab,3);
221 /* EWALD ELECTROSTATICS */
223 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
224 ewrt = _mm_mul_pd(r00,ewtabscale);
225 ewitab = _mm_cvttpd_epi32(ewrt);
227 eweps = _mm_frcz_pd(ewrt);
229 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
231 twoeweps = _mm_add_pd(eweps,eweps);
232 ewitab = _mm_slli_epi32(ewitab,2);
233 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
234 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
235 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
236 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
237 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
238 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
239 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
240 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
241 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
242 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
244 /* CUBIC SPLINE TABLE DISPERSION */
245 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
246 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
247 GMX_MM_TRANSPOSE2_PD(Y,F);
248 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
249 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
250 GMX_MM_TRANSPOSE2_PD(G,H);
251 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
252 VV = _mm_macc_pd(vfeps,Fp,Y);
253 vvdw6 = _mm_mul_pd(c6_00,VV);
254 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
255 fvdw6 = _mm_mul_pd(c6_00,FF);
257 /* CUBIC SPLINE TABLE REPULSION */
258 vfitab = _mm_add_epi32(vfitab,ifour);
259 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
260 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
261 GMX_MM_TRANSPOSE2_PD(Y,F);
262 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
263 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
264 GMX_MM_TRANSPOSE2_PD(G,H);
265 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
266 VV = _mm_macc_pd(vfeps,Fp,Y);
267 vvdw12 = _mm_mul_pd(c12_00,VV);
268 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
269 fvdw12 = _mm_mul_pd(c12_00,FF);
270 vvdw = _mm_add_pd(vvdw12,vvdw6);
271 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
273 /* Update potential sum for this i atom from the interaction with this j atom. */
274 velecsum = _mm_add_pd(velecsum,velec);
275 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
277 fscal = _mm_add_pd(felec,fvdw);
279 /* Update vectorial force */
280 fix0 = _mm_macc_pd(dx00,fscal,fix0);
281 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
282 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
284 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
285 _mm_mul_pd(dx00,fscal),
286 _mm_mul_pd(dy00,fscal),
287 _mm_mul_pd(dz00,fscal));
289 /* Inner loop uses 78 flops */
296 j_coord_offsetA = DIM*jnrA;
298 /* load j atom coordinates */
299 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
302 /* Calculate displacement vector */
303 dx00 = _mm_sub_pd(ix0,jx0);
304 dy00 = _mm_sub_pd(iy0,jy0);
305 dz00 = _mm_sub_pd(iz0,jz0);
307 /* Calculate squared distance and things based on it */
308 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
310 rinv00 = gmx_mm_invsqrt_pd(rsq00);
312 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
314 /* Load parameters for j particles */
315 jq0 = _mm_load_sd(charge+jnrA+0);
316 vdwjidx0A = 2*vdwtype[jnrA+0];
318 /**************************
319 * CALCULATE INTERACTIONS *
320 **************************/
322 r00 = _mm_mul_pd(rsq00,rinv00);
324 /* Compute parameters for interactions between i and j atoms */
325 qq00 = _mm_mul_pd(iq0,jq0);
326 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
328 /* Calculate table index by multiplying r with table scale and truncate to integer */
329 rt = _mm_mul_pd(r00,vftabscale);
330 vfitab = _mm_cvttpd_epi32(rt);
332 vfeps = _mm_frcz_pd(rt);
334 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
336 twovfeps = _mm_add_pd(vfeps,vfeps);
337 vfitab = _mm_slli_epi32(vfitab,3);
339 /* EWALD ELECTROSTATICS */
341 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
342 ewrt = _mm_mul_pd(r00,ewtabscale);
343 ewitab = _mm_cvttpd_epi32(ewrt);
345 eweps = _mm_frcz_pd(ewrt);
347 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
349 twoeweps = _mm_add_pd(eweps,eweps);
350 ewitab = _mm_slli_epi32(ewitab,2);
351 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
352 ewtabD = _mm_setzero_pd();
353 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
354 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
355 ewtabFn = _mm_setzero_pd();
356 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
357 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
358 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
359 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
360 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
362 /* CUBIC SPLINE TABLE DISPERSION */
363 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
364 F = _mm_setzero_pd();
365 GMX_MM_TRANSPOSE2_PD(Y,F);
366 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
367 H = _mm_setzero_pd();
368 GMX_MM_TRANSPOSE2_PD(G,H);
369 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
370 VV = _mm_macc_pd(vfeps,Fp,Y);
371 vvdw6 = _mm_mul_pd(c6_00,VV);
372 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
373 fvdw6 = _mm_mul_pd(c6_00,FF);
375 /* CUBIC SPLINE TABLE REPULSION */
376 vfitab = _mm_add_epi32(vfitab,ifour);
377 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
378 F = _mm_setzero_pd();
379 GMX_MM_TRANSPOSE2_PD(Y,F);
380 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
381 H = _mm_setzero_pd();
382 GMX_MM_TRANSPOSE2_PD(G,H);
383 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
384 VV = _mm_macc_pd(vfeps,Fp,Y);
385 vvdw12 = _mm_mul_pd(c12_00,VV);
386 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
387 fvdw12 = _mm_mul_pd(c12_00,FF);
388 vvdw = _mm_add_pd(vvdw12,vvdw6);
389 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
391 /* Update potential sum for this i atom from the interaction with this j atom. */
392 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
393 velecsum = _mm_add_pd(velecsum,velec);
394 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
395 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
397 fscal = _mm_add_pd(felec,fvdw);
399 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
401 /* Update vectorial force */
402 fix0 = _mm_macc_pd(dx00,fscal,fix0);
403 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
404 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
406 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
407 _mm_mul_pd(dx00,fscal),
408 _mm_mul_pd(dy00,fscal),
409 _mm_mul_pd(dz00,fscal));
411 /* Inner loop uses 78 flops */
414 /* End of innermost loop */
416 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
417 f+i_coord_offset,fshift+i_shift_offset);
420 /* Update potential energies */
421 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
422 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
424 /* Increment number of inner iterations */
425 inneriter += j_index_end - j_index_start;
427 /* Outer loop uses 9 flops */
430 /* Increment number of outer iterations */
433 /* Update outer/inner flops */
435 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*78);
438 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_128_fma_double
439 * Electrostatics interaction: Ewald
440 * VdW interaction: CubicSplineTable
441 * Geometry: Particle-Particle
442 * Calculate force/pot: Force
445 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_128_fma_double
446 (t_nblist * gmx_restrict nlist,
447 rvec * gmx_restrict xx,
448 rvec * gmx_restrict ff,
449 t_forcerec * gmx_restrict fr,
450 t_mdatoms * gmx_restrict mdatoms,
451 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
452 t_nrnb * gmx_restrict nrnb)
454 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
455 * just 0 for non-waters.
456 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
457 * jnr indices corresponding to data put in the four positions in the SIMD register.
459 int i_shift_offset,i_coord_offset,outeriter,inneriter;
460 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
462 int j_coord_offsetA,j_coord_offsetB;
463 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
465 real *shiftvec,*fshift,*x,*f;
466 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
468 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
469 int vdwjidx0A,vdwjidx0B;
470 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
471 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
472 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
475 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
478 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
479 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
481 __m128i ifour = _mm_set1_epi32(4);
482 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
485 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
487 __m128d dummy_mask,cutoff_mask;
488 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
489 __m128d one = _mm_set1_pd(1.0);
490 __m128d two = _mm_set1_pd(2.0);
496 jindex = nlist->jindex;
498 shiftidx = nlist->shift;
500 shiftvec = fr->shift_vec[0];
501 fshift = fr->fshift[0];
502 facel = _mm_set1_pd(fr->epsfac);
503 charge = mdatoms->chargeA;
504 nvdwtype = fr->ntype;
506 vdwtype = mdatoms->typeA;
508 vftab = kernel_data->table_vdw->data;
509 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
511 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
512 ewtab = fr->ic->tabq_coul_F;
513 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
514 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
516 /* Avoid stupid compiler warnings */
524 /* Start outer loop over neighborlists */
525 for(iidx=0; iidx<nri; iidx++)
527 /* Load shift vector for this list */
528 i_shift_offset = DIM*shiftidx[iidx];
530 /* Load limits for loop over neighbors */
531 j_index_start = jindex[iidx];
532 j_index_end = jindex[iidx+1];
534 /* Get outer coordinate index */
536 i_coord_offset = DIM*inr;
538 /* Load i particle coords and add shift vector */
539 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
541 fix0 = _mm_setzero_pd();
542 fiy0 = _mm_setzero_pd();
543 fiz0 = _mm_setzero_pd();
545 /* Load parameters for i particles */
546 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
547 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
549 /* Start inner kernel loop */
550 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
553 /* Get j neighbor index, and coordinate index */
556 j_coord_offsetA = DIM*jnrA;
557 j_coord_offsetB = DIM*jnrB;
559 /* load j atom coordinates */
560 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
563 /* Calculate displacement vector */
564 dx00 = _mm_sub_pd(ix0,jx0);
565 dy00 = _mm_sub_pd(iy0,jy0);
566 dz00 = _mm_sub_pd(iz0,jz0);
568 /* Calculate squared distance and things based on it */
569 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
571 rinv00 = gmx_mm_invsqrt_pd(rsq00);
573 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
575 /* Load parameters for j particles */
576 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
577 vdwjidx0A = 2*vdwtype[jnrA+0];
578 vdwjidx0B = 2*vdwtype[jnrB+0];
580 /**************************
581 * CALCULATE INTERACTIONS *
582 **************************/
584 r00 = _mm_mul_pd(rsq00,rinv00);
586 /* Compute parameters for interactions between i and j atoms */
587 qq00 = _mm_mul_pd(iq0,jq0);
588 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
589 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
591 /* Calculate table index by multiplying r with table scale and truncate to integer */
592 rt = _mm_mul_pd(r00,vftabscale);
593 vfitab = _mm_cvttpd_epi32(rt);
595 vfeps = _mm_frcz_pd(rt);
597 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
599 twovfeps = _mm_add_pd(vfeps,vfeps);
600 vfitab = _mm_slli_epi32(vfitab,3);
602 /* EWALD ELECTROSTATICS */
604 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
605 ewrt = _mm_mul_pd(r00,ewtabscale);
606 ewitab = _mm_cvttpd_epi32(ewrt);
608 eweps = _mm_frcz_pd(ewrt);
610 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
612 twoeweps = _mm_add_pd(eweps,eweps);
613 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
615 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
616 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
618 /* CUBIC SPLINE TABLE DISPERSION */
619 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
620 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
621 GMX_MM_TRANSPOSE2_PD(Y,F);
622 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
623 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
624 GMX_MM_TRANSPOSE2_PD(G,H);
625 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
626 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
627 fvdw6 = _mm_mul_pd(c6_00,FF);
629 /* CUBIC SPLINE TABLE REPULSION */
630 vfitab = _mm_add_epi32(vfitab,ifour);
631 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
632 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
633 GMX_MM_TRANSPOSE2_PD(Y,F);
634 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
635 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
636 GMX_MM_TRANSPOSE2_PD(G,H);
637 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
638 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
639 fvdw12 = _mm_mul_pd(c12_00,FF);
640 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
642 fscal = _mm_add_pd(felec,fvdw);
644 /* Update vectorial force */
645 fix0 = _mm_macc_pd(dx00,fscal,fix0);
646 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
647 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
649 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
650 _mm_mul_pd(dx00,fscal),
651 _mm_mul_pd(dy00,fscal),
652 _mm_mul_pd(dz00,fscal));
654 /* Inner loop uses 65 flops */
661 j_coord_offsetA = DIM*jnrA;
663 /* load j atom coordinates */
664 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
667 /* Calculate displacement vector */
668 dx00 = _mm_sub_pd(ix0,jx0);
669 dy00 = _mm_sub_pd(iy0,jy0);
670 dz00 = _mm_sub_pd(iz0,jz0);
672 /* Calculate squared distance and things based on it */
673 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
675 rinv00 = gmx_mm_invsqrt_pd(rsq00);
677 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
679 /* Load parameters for j particles */
680 jq0 = _mm_load_sd(charge+jnrA+0);
681 vdwjidx0A = 2*vdwtype[jnrA+0];
683 /**************************
684 * CALCULATE INTERACTIONS *
685 **************************/
687 r00 = _mm_mul_pd(rsq00,rinv00);
689 /* Compute parameters for interactions between i and j atoms */
690 qq00 = _mm_mul_pd(iq0,jq0);
691 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
693 /* Calculate table index by multiplying r with table scale and truncate to integer */
694 rt = _mm_mul_pd(r00,vftabscale);
695 vfitab = _mm_cvttpd_epi32(rt);
697 vfeps = _mm_frcz_pd(rt);
699 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
701 twovfeps = _mm_add_pd(vfeps,vfeps);
702 vfitab = _mm_slli_epi32(vfitab,3);
704 /* EWALD ELECTROSTATICS */
706 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
707 ewrt = _mm_mul_pd(r00,ewtabscale);
708 ewitab = _mm_cvttpd_epi32(ewrt);
710 eweps = _mm_frcz_pd(ewrt);
712 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
714 twoeweps = _mm_add_pd(eweps,eweps);
715 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
716 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
717 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
719 /* CUBIC SPLINE TABLE DISPERSION */
720 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
721 F = _mm_setzero_pd();
722 GMX_MM_TRANSPOSE2_PD(Y,F);
723 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
724 H = _mm_setzero_pd();
725 GMX_MM_TRANSPOSE2_PD(G,H);
726 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
727 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
728 fvdw6 = _mm_mul_pd(c6_00,FF);
730 /* CUBIC SPLINE TABLE REPULSION */
731 vfitab = _mm_add_epi32(vfitab,ifour);
732 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
733 F = _mm_setzero_pd();
734 GMX_MM_TRANSPOSE2_PD(Y,F);
735 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
736 H = _mm_setzero_pd();
737 GMX_MM_TRANSPOSE2_PD(G,H);
738 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
739 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
740 fvdw12 = _mm_mul_pd(c12_00,FF);
741 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
743 fscal = _mm_add_pd(felec,fvdw);
745 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
747 /* Update vectorial force */
748 fix0 = _mm_macc_pd(dx00,fscal,fix0);
749 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
750 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
752 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
753 _mm_mul_pd(dx00,fscal),
754 _mm_mul_pd(dy00,fscal),
755 _mm_mul_pd(dz00,fscal));
757 /* Inner loop uses 65 flops */
760 /* End of innermost loop */
762 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
763 f+i_coord_offset,fshift+i_shift_offset);
765 /* Increment number of inner iterations */
766 inneriter += j_index_end - j_index_start;
768 /* Outer loop uses 7 flops */
771 /* Increment number of outer iterations */
774 /* Update outer/inner flops */
776 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*65);