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.
42 #include "../nb_kernel.h"
43 #include "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
48 #include "kernelutil_x86_avx_128_fma_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
52 * Electrostatics interaction: Ewald
53 * VdW interaction: CubicSplineTable
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_avx_128_fma_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 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
91 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
92 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
94 __m128i ifour = _mm_set1_epi32(4);
95 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
98 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
100 __m128d dummy_mask,cutoff_mask;
101 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
102 __m128d one = _mm_set1_pd(1.0);
103 __m128d two = _mm_set1_pd(2.0);
109 jindex = nlist->jindex;
111 shiftidx = nlist->shift;
113 shiftvec = fr->shift_vec[0];
114 fshift = fr->fshift[0];
115 facel = _mm_set1_pd(fr->epsfac);
116 charge = mdatoms->chargeA;
117 nvdwtype = fr->ntype;
119 vdwtype = mdatoms->typeA;
121 vftab = kernel_data->table_vdw->data;
122 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
124 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
125 ewtab = fr->ic->tabq_coul_FDV0;
126 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
127 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
129 /* Avoid stupid compiler warnings */
137 /* Start outer loop over neighborlists */
138 for(iidx=0; iidx<nri; iidx++)
140 /* Load shift vector for this list */
141 i_shift_offset = DIM*shiftidx[iidx];
143 /* Load limits for loop over neighbors */
144 j_index_start = jindex[iidx];
145 j_index_end = jindex[iidx+1];
147 /* Get outer coordinate index */
149 i_coord_offset = DIM*inr;
151 /* Load i particle coords and add shift vector */
152 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
154 fix0 = _mm_setzero_pd();
155 fiy0 = _mm_setzero_pd();
156 fiz0 = _mm_setzero_pd();
158 /* Load parameters for i particles */
159 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
160 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
162 /* Reset potential sums */
163 velecsum = _mm_setzero_pd();
164 vvdwsum = _mm_setzero_pd();
166 /* Start inner kernel loop */
167 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
170 /* Get j neighbor index, and coordinate index */
173 j_coord_offsetA = DIM*jnrA;
174 j_coord_offsetB = DIM*jnrB;
176 /* load j atom coordinates */
177 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
180 /* Calculate displacement vector */
181 dx00 = _mm_sub_pd(ix0,jx0);
182 dy00 = _mm_sub_pd(iy0,jy0);
183 dz00 = _mm_sub_pd(iz0,jz0);
185 /* Calculate squared distance and things based on it */
186 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
188 rinv00 = gmx_mm_invsqrt_pd(rsq00);
190 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
192 /* Load parameters for j particles */
193 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
194 vdwjidx0A = 2*vdwtype[jnrA+0];
195 vdwjidx0B = 2*vdwtype[jnrB+0];
197 /**************************
198 * CALCULATE INTERACTIONS *
199 **************************/
201 r00 = _mm_mul_pd(rsq00,rinv00);
203 /* Compute parameters for interactions between i and j atoms */
204 qq00 = _mm_mul_pd(iq0,jq0);
205 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
206 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
208 /* Calculate table index by multiplying r with table scale and truncate to integer */
209 rt = _mm_mul_pd(r00,vftabscale);
210 vfitab = _mm_cvttpd_epi32(rt);
212 vfeps = _mm_frcz_pd(rt);
214 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
216 twovfeps = _mm_add_pd(vfeps,vfeps);
217 vfitab = _mm_slli_epi32(vfitab,3);
219 /* EWALD ELECTROSTATICS */
221 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
222 ewrt = _mm_mul_pd(r00,ewtabscale);
223 ewitab = _mm_cvttpd_epi32(ewrt);
225 eweps = _mm_frcz_pd(ewrt);
227 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
229 twoeweps = _mm_add_pd(eweps,eweps);
230 ewitab = _mm_slli_epi32(ewitab,2);
231 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
232 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
233 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
234 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
235 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
236 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
237 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
238 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
239 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
240 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
242 /* CUBIC SPLINE TABLE DISPERSION */
243 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
244 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
245 GMX_MM_TRANSPOSE2_PD(Y,F);
246 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
247 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
248 GMX_MM_TRANSPOSE2_PD(G,H);
249 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
250 VV = _mm_macc_pd(vfeps,Fp,Y);
251 vvdw6 = _mm_mul_pd(c6_00,VV);
252 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
253 fvdw6 = _mm_mul_pd(c6_00,FF);
255 /* CUBIC SPLINE TABLE REPULSION */
256 vfitab = _mm_add_epi32(vfitab,ifour);
257 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
258 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
259 GMX_MM_TRANSPOSE2_PD(Y,F);
260 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
261 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
262 GMX_MM_TRANSPOSE2_PD(G,H);
263 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
264 VV = _mm_macc_pd(vfeps,Fp,Y);
265 vvdw12 = _mm_mul_pd(c12_00,VV);
266 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
267 fvdw12 = _mm_mul_pd(c12_00,FF);
268 vvdw = _mm_add_pd(vvdw12,vvdw6);
269 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
271 /* Update potential sum for this i atom from the interaction with this j atom. */
272 velecsum = _mm_add_pd(velecsum,velec);
273 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
275 fscal = _mm_add_pd(felec,fvdw);
277 /* Update vectorial force */
278 fix0 = _mm_macc_pd(dx00,fscal,fix0);
279 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
280 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
282 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
283 _mm_mul_pd(dx00,fscal),
284 _mm_mul_pd(dy00,fscal),
285 _mm_mul_pd(dz00,fscal));
287 /* Inner loop uses 78 flops */
294 j_coord_offsetA = DIM*jnrA;
296 /* load j atom coordinates */
297 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
300 /* Calculate displacement vector */
301 dx00 = _mm_sub_pd(ix0,jx0);
302 dy00 = _mm_sub_pd(iy0,jy0);
303 dz00 = _mm_sub_pd(iz0,jz0);
305 /* Calculate squared distance and things based on it */
306 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
308 rinv00 = gmx_mm_invsqrt_pd(rsq00);
310 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
312 /* Load parameters for j particles */
313 jq0 = _mm_load_sd(charge+jnrA+0);
314 vdwjidx0A = 2*vdwtype[jnrA+0];
316 /**************************
317 * CALCULATE INTERACTIONS *
318 **************************/
320 r00 = _mm_mul_pd(rsq00,rinv00);
322 /* Compute parameters for interactions between i and j atoms */
323 qq00 = _mm_mul_pd(iq0,jq0);
324 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
326 /* Calculate table index by multiplying r with table scale and truncate to integer */
327 rt = _mm_mul_pd(r00,vftabscale);
328 vfitab = _mm_cvttpd_epi32(rt);
330 vfeps = _mm_frcz_pd(rt);
332 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
334 twovfeps = _mm_add_pd(vfeps,vfeps);
335 vfitab = _mm_slli_epi32(vfitab,3);
337 /* EWALD ELECTROSTATICS */
339 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
340 ewrt = _mm_mul_pd(r00,ewtabscale);
341 ewitab = _mm_cvttpd_epi32(ewrt);
343 eweps = _mm_frcz_pd(ewrt);
345 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
347 twoeweps = _mm_add_pd(eweps,eweps);
348 ewitab = _mm_slli_epi32(ewitab,2);
349 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
350 ewtabD = _mm_setzero_pd();
351 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
352 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
353 ewtabFn = _mm_setzero_pd();
354 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
355 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
356 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
357 velec = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
358 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
360 /* CUBIC SPLINE TABLE DISPERSION */
361 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
362 F = _mm_setzero_pd();
363 GMX_MM_TRANSPOSE2_PD(Y,F);
364 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
365 H = _mm_setzero_pd();
366 GMX_MM_TRANSPOSE2_PD(G,H);
367 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
368 VV = _mm_macc_pd(vfeps,Fp,Y);
369 vvdw6 = _mm_mul_pd(c6_00,VV);
370 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
371 fvdw6 = _mm_mul_pd(c6_00,FF);
373 /* CUBIC SPLINE TABLE REPULSION */
374 vfitab = _mm_add_epi32(vfitab,ifour);
375 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
376 F = _mm_setzero_pd();
377 GMX_MM_TRANSPOSE2_PD(Y,F);
378 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
379 H = _mm_setzero_pd();
380 GMX_MM_TRANSPOSE2_PD(G,H);
381 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
382 VV = _mm_macc_pd(vfeps,Fp,Y);
383 vvdw12 = _mm_mul_pd(c12_00,VV);
384 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
385 fvdw12 = _mm_mul_pd(c12_00,FF);
386 vvdw = _mm_add_pd(vvdw12,vvdw6);
387 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
389 /* Update potential sum for this i atom from the interaction with this j atom. */
390 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
391 velecsum = _mm_add_pd(velecsum,velec);
392 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
393 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
395 fscal = _mm_add_pd(felec,fvdw);
397 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
399 /* Update vectorial force */
400 fix0 = _mm_macc_pd(dx00,fscal,fix0);
401 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
402 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
404 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
405 _mm_mul_pd(dx00,fscal),
406 _mm_mul_pd(dy00,fscal),
407 _mm_mul_pd(dz00,fscal));
409 /* Inner loop uses 78 flops */
412 /* End of innermost loop */
414 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
415 f+i_coord_offset,fshift+i_shift_offset);
418 /* Update potential energies */
419 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
420 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
422 /* Increment number of inner iterations */
423 inneriter += j_index_end - j_index_start;
425 /* Outer loop uses 9 flops */
428 /* Increment number of outer iterations */
431 /* Update outer/inner flops */
433 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*78);
436 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_128_fma_double
437 * Electrostatics interaction: Ewald
438 * VdW interaction: CubicSplineTable
439 * Geometry: Particle-Particle
440 * Calculate force/pot: Force
443 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_128_fma_double
444 (t_nblist * gmx_restrict nlist,
445 rvec * gmx_restrict xx,
446 rvec * gmx_restrict ff,
447 t_forcerec * gmx_restrict fr,
448 t_mdatoms * gmx_restrict mdatoms,
449 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
450 t_nrnb * gmx_restrict nrnb)
452 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
453 * just 0 for non-waters.
454 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
455 * jnr indices corresponding to data put in the four positions in the SIMD register.
457 int i_shift_offset,i_coord_offset,outeriter,inneriter;
458 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
460 int j_coord_offsetA,j_coord_offsetB;
461 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
463 real *shiftvec,*fshift,*x,*f;
464 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
466 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
467 int vdwjidx0A,vdwjidx0B;
468 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
469 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
470 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
473 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
476 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
477 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
479 __m128i ifour = _mm_set1_epi32(4);
480 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
483 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
485 __m128d dummy_mask,cutoff_mask;
486 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
487 __m128d one = _mm_set1_pd(1.0);
488 __m128d two = _mm_set1_pd(2.0);
494 jindex = nlist->jindex;
496 shiftidx = nlist->shift;
498 shiftvec = fr->shift_vec[0];
499 fshift = fr->fshift[0];
500 facel = _mm_set1_pd(fr->epsfac);
501 charge = mdatoms->chargeA;
502 nvdwtype = fr->ntype;
504 vdwtype = mdatoms->typeA;
506 vftab = kernel_data->table_vdw->data;
507 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
509 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
510 ewtab = fr->ic->tabq_coul_F;
511 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
512 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
514 /* Avoid stupid compiler warnings */
522 /* Start outer loop over neighborlists */
523 for(iidx=0; iidx<nri; iidx++)
525 /* Load shift vector for this list */
526 i_shift_offset = DIM*shiftidx[iidx];
528 /* Load limits for loop over neighbors */
529 j_index_start = jindex[iidx];
530 j_index_end = jindex[iidx+1];
532 /* Get outer coordinate index */
534 i_coord_offset = DIM*inr;
536 /* Load i particle coords and add shift vector */
537 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
539 fix0 = _mm_setzero_pd();
540 fiy0 = _mm_setzero_pd();
541 fiz0 = _mm_setzero_pd();
543 /* Load parameters for i particles */
544 iq0 = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
545 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
547 /* Start inner kernel loop */
548 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
551 /* Get j neighbor index, and coordinate index */
554 j_coord_offsetA = DIM*jnrA;
555 j_coord_offsetB = DIM*jnrB;
557 /* load j atom coordinates */
558 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
561 /* Calculate displacement vector */
562 dx00 = _mm_sub_pd(ix0,jx0);
563 dy00 = _mm_sub_pd(iy0,jy0);
564 dz00 = _mm_sub_pd(iz0,jz0);
566 /* Calculate squared distance and things based on it */
567 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
569 rinv00 = gmx_mm_invsqrt_pd(rsq00);
571 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
573 /* Load parameters for j particles */
574 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
575 vdwjidx0A = 2*vdwtype[jnrA+0];
576 vdwjidx0B = 2*vdwtype[jnrB+0];
578 /**************************
579 * CALCULATE INTERACTIONS *
580 **************************/
582 r00 = _mm_mul_pd(rsq00,rinv00);
584 /* Compute parameters for interactions between i and j atoms */
585 qq00 = _mm_mul_pd(iq0,jq0);
586 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
587 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
589 /* Calculate table index by multiplying r with table scale and truncate to integer */
590 rt = _mm_mul_pd(r00,vftabscale);
591 vfitab = _mm_cvttpd_epi32(rt);
593 vfeps = _mm_frcz_pd(rt);
595 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
597 twovfeps = _mm_add_pd(vfeps,vfeps);
598 vfitab = _mm_slli_epi32(vfitab,3);
600 /* EWALD ELECTROSTATICS */
602 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
603 ewrt = _mm_mul_pd(r00,ewtabscale);
604 ewitab = _mm_cvttpd_epi32(ewrt);
606 eweps = _mm_frcz_pd(ewrt);
608 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
610 twoeweps = _mm_add_pd(eweps,eweps);
611 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
613 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
614 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
616 /* CUBIC SPLINE TABLE DISPERSION */
617 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
618 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
619 GMX_MM_TRANSPOSE2_PD(Y,F);
620 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
621 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
622 GMX_MM_TRANSPOSE2_PD(G,H);
623 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
624 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
625 fvdw6 = _mm_mul_pd(c6_00,FF);
627 /* CUBIC SPLINE TABLE REPULSION */
628 vfitab = _mm_add_epi32(vfitab,ifour);
629 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
630 F = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
631 GMX_MM_TRANSPOSE2_PD(Y,F);
632 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
633 H = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
634 GMX_MM_TRANSPOSE2_PD(G,H);
635 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
636 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
637 fvdw12 = _mm_mul_pd(c12_00,FF);
638 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
640 fscal = _mm_add_pd(felec,fvdw);
642 /* Update vectorial force */
643 fix0 = _mm_macc_pd(dx00,fscal,fix0);
644 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
645 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
647 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
648 _mm_mul_pd(dx00,fscal),
649 _mm_mul_pd(dy00,fscal),
650 _mm_mul_pd(dz00,fscal));
652 /* Inner loop uses 65 flops */
659 j_coord_offsetA = DIM*jnrA;
661 /* load j atom coordinates */
662 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
665 /* Calculate displacement vector */
666 dx00 = _mm_sub_pd(ix0,jx0);
667 dy00 = _mm_sub_pd(iy0,jy0);
668 dz00 = _mm_sub_pd(iz0,jz0);
670 /* Calculate squared distance and things based on it */
671 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
673 rinv00 = gmx_mm_invsqrt_pd(rsq00);
675 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
677 /* Load parameters for j particles */
678 jq0 = _mm_load_sd(charge+jnrA+0);
679 vdwjidx0A = 2*vdwtype[jnrA+0];
681 /**************************
682 * CALCULATE INTERACTIONS *
683 **************************/
685 r00 = _mm_mul_pd(rsq00,rinv00);
687 /* Compute parameters for interactions between i and j atoms */
688 qq00 = _mm_mul_pd(iq0,jq0);
689 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
691 /* Calculate table index by multiplying r with table scale and truncate to integer */
692 rt = _mm_mul_pd(r00,vftabscale);
693 vfitab = _mm_cvttpd_epi32(rt);
695 vfeps = _mm_frcz_pd(rt);
697 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
699 twovfeps = _mm_add_pd(vfeps,vfeps);
700 vfitab = _mm_slli_epi32(vfitab,3);
702 /* EWALD ELECTROSTATICS */
704 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
705 ewrt = _mm_mul_pd(r00,ewtabscale);
706 ewitab = _mm_cvttpd_epi32(ewrt);
708 eweps = _mm_frcz_pd(ewrt);
710 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
712 twoeweps = _mm_add_pd(eweps,eweps);
713 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
714 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
715 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
717 /* CUBIC SPLINE TABLE DISPERSION */
718 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
719 F = _mm_setzero_pd();
720 GMX_MM_TRANSPOSE2_PD(Y,F);
721 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
722 H = _mm_setzero_pd();
723 GMX_MM_TRANSPOSE2_PD(G,H);
724 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
725 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
726 fvdw6 = _mm_mul_pd(c6_00,FF);
728 /* CUBIC SPLINE TABLE REPULSION */
729 vfitab = _mm_add_epi32(vfitab,ifour);
730 Y = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
731 F = _mm_setzero_pd();
732 GMX_MM_TRANSPOSE2_PD(Y,F);
733 G = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
734 H = _mm_setzero_pd();
735 GMX_MM_TRANSPOSE2_PD(G,H);
736 Fp = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
737 FF = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
738 fvdw12 = _mm_mul_pd(c12_00,FF);
739 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
741 fscal = _mm_add_pd(felec,fvdw);
743 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
745 /* Update vectorial force */
746 fix0 = _mm_macc_pd(dx00,fscal,fix0);
747 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
748 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
750 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
751 _mm_mul_pd(dx00,fscal),
752 _mm_mul_pd(dy00,fscal),
753 _mm_mul_pd(dz00,fscal));
755 /* Inner loop uses 65 flops */
758 /* End of innermost loop */
760 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
761 f+i_coord_offset,fshift+i_shift_offset);
763 /* Increment number of inner iterations */
764 inneriter += j_index_end - j_index_start;
766 /* Outer loop uses 7 flops */
769 /* Increment number of outer iterations */
772 /* Update outer/inner flops */
774 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*65);