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_VdwLJEw_GeomW4P1_VF_avx_128_fma_double
54 * Electrostatics interaction: Ewald
55 * VdW interaction: LJEwald
56 * Geometry: Water4-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_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;
85 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
87 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
89 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
90 int vdwjidx0A,vdwjidx0B;
91 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
92 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
93 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
94 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
95 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
96 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
99 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
102 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
103 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
109 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
110 __m128d one_half = _mm_set1_pd(0.5);
111 __m128d minus_one = _mm_set1_pd(-1.0);
113 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
115 __m128d dummy_mask,cutoff_mask;
116 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
117 __m128d one = _mm_set1_pd(1.0);
118 __m128d two = _mm_set1_pd(2.0);
124 jindex = nlist->jindex;
126 shiftidx = nlist->shift;
128 shiftvec = fr->shift_vec[0];
129 fshift = fr->fshift[0];
130 facel = _mm_set1_pd(fr->epsfac);
131 charge = mdatoms->chargeA;
132 nvdwtype = fr->ntype;
134 vdwtype = mdatoms->typeA;
135 vdwgridparam = fr->ljpme_c6grid;
136 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
137 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
138 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
140 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
141 ewtab = fr->ic->tabq_coul_FDV0;
142 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
143 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
145 /* Setup water-specific parameters */
146 inr = nlist->iinr[0];
147 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
148 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
149 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
150 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
152 /* Avoid stupid compiler warnings */
160 /* Start outer loop over neighborlists */
161 for(iidx=0; iidx<nri; iidx++)
163 /* Load shift vector for this list */
164 i_shift_offset = DIM*shiftidx[iidx];
166 /* Load limits for loop over neighbors */
167 j_index_start = jindex[iidx];
168 j_index_end = jindex[iidx+1];
170 /* Get outer coordinate index */
172 i_coord_offset = DIM*inr;
174 /* Load i particle coords and add shift vector */
175 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
176 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
178 fix0 = _mm_setzero_pd();
179 fiy0 = _mm_setzero_pd();
180 fiz0 = _mm_setzero_pd();
181 fix1 = _mm_setzero_pd();
182 fiy1 = _mm_setzero_pd();
183 fiz1 = _mm_setzero_pd();
184 fix2 = _mm_setzero_pd();
185 fiy2 = _mm_setzero_pd();
186 fiz2 = _mm_setzero_pd();
187 fix3 = _mm_setzero_pd();
188 fiy3 = _mm_setzero_pd();
189 fiz3 = _mm_setzero_pd();
191 /* Reset potential sums */
192 velecsum = _mm_setzero_pd();
193 vvdwsum = _mm_setzero_pd();
195 /* Start inner kernel loop */
196 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
199 /* Get j neighbor index, and coordinate index */
202 j_coord_offsetA = DIM*jnrA;
203 j_coord_offsetB = DIM*jnrB;
205 /* load j atom coordinates */
206 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
209 /* Calculate displacement vector */
210 dx00 = _mm_sub_pd(ix0,jx0);
211 dy00 = _mm_sub_pd(iy0,jy0);
212 dz00 = _mm_sub_pd(iz0,jz0);
213 dx10 = _mm_sub_pd(ix1,jx0);
214 dy10 = _mm_sub_pd(iy1,jy0);
215 dz10 = _mm_sub_pd(iz1,jz0);
216 dx20 = _mm_sub_pd(ix2,jx0);
217 dy20 = _mm_sub_pd(iy2,jy0);
218 dz20 = _mm_sub_pd(iz2,jz0);
219 dx30 = _mm_sub_pd(ix3,jx0);
220 dy30 = _mm_sub_pd(iy3,jy0);
221 dz30 = _mm_sub_pd(iz3,jz0);
223 /* Calculate squared distance and things based on it */
224 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
225 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
226 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
227 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
229 rinv00 = gmx_mm_invsqrt_pd(rsq00);
230 rinv10 = gmx_mm_invsqrt_pd(rsq10);
231 rinv20 = gmx_mm_invsqrt_pd(rsq20);
232 rinv30 = gmx_mm_invsqrt_pd(rsq30);
234 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
235 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
236 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
237 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
239 /* Load parameters for j particles */
240 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
241 vdwjidx0A = 2*vdwtype[jnrA+0];
242 vdwjidx0B = 2*vdwtype[jnrB+0];
244 fjx0 = _mm_setzero_pd();
245 fjy0 = _mm_setzero_pd();
246 fjz0 = _mm_setzero_pd();
248 /**************************
249 * CALCULATE INTERACTIONS *
250 **************************/
252 r00 = _mm_mul_pd(rsq00,rinv00);
254 /* Compute parameters for interactions between i and j atoms */
255 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
256 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
257 c6grid_00 = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
258 vdwgridparam+vdwioffset0+vdwjidx0B);
260 /* Analytical LJ-PME */
261 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
262 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
263 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
264 exponent = gmx_simd_exp_d(ewcljrsq);
265 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
266 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
267 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
268 vvdw6 = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
269 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
270 vvdw = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
271 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
272 fvdw = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
274 /* Update potential sum for this i atom from the interaction with this j atom. */
275 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
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 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
285 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
286 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
288 /**************************
289 * CALCULATE INTERACTIONS *
290 **************************/
292 r10 = _mm_mul_pd(rsq10,rinv10);
294 /* Compute parameters for interactions between i and j atoms */
295 qq10 = _mm_mul_pd(iq1,jq0);
297 /* EWALD ELECTROSTATICS */
299 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
300 ewrt = _mm_mul_pd(r10,ewtabscale);
301 ewitab = _mm_cvttpd_epi32(ewrt);
303 eweps = _mm_frcz_pd(ewrt);
305 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
307 twoeweps = _mm_add_pd(eweps,eweps);
308 ewitab = _mm_slli_epi32(ewitab,2);
309 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
310 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
311 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
312 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
313 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
314 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
315 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
316 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
317 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
318 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
320 /* Update potential sum for this i atom from the interaction with this j atom. */
321 velecsum = _mm_add_pd(velecsum,velec);
325 /* Update vectorial force */
326 fix1 = _mm_macc_pd(dx10,fscal,fix1);
327 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
328 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
330 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
331 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
332 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
334 /**************************
335 * CALCULATE INTERACTIONS *
336 **************************/
338 r20 = _mm_mul_pd(rsq20,rinv20);
340 /* Compute parameters for interactions between i and j atoms */
341 qq20 = _mm_mul_pd(iq2,jq0);
343 /* EWALD ELECTROSTATICS */
345 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
346 ewrt = _mm_mul_pd(r20,ewtabscale);
347 ewitab = _mm_cvttpd_epi32(ewrt);
349 eweps = _mm_frcz_pd(ewrt);
351 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
353 twoeweps = _mm_add_pd(eweps,eweps);
354 ewitab = _mm_slli_epi32(ewitab,2);
355 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
356 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
357 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
358 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
359 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
360 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
361 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
362 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
363 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
364 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
366 /* Update potential sum for this i atom from the interaction with this j atom. */
367 velecsum = _mm_add_pd(velecsum,velec);
371 /* Update vectorial force */
372 fix2 = _mm_macc_pd(dx20,fscal,fix2);
373 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
374 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
376 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
377 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
378 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
380 /**************************
381 * CALCULATE INTERACTIONS *
382 **************************/
384 r30 = _mm_mul_pd(rsq30,rinv30);
386 /* Compute parameters for interactions between i and j atoms */
387 qq30 = _mm_mul_pd(iq3,jq0);
389 /* EWALD ELECTROSTATICS */
391 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
392 ewrt = _mm_mul_pd(r30,ewtabscale);
393 ewitab = _mm_cvttpd_epi32(ewrt);
395 eweps = _mm_frcz_pd(ewrt);
397 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
399 twoeweps = _mm_add_pd(eweps,eweps);
400 ewitab = _mm_slli_epi32(ewitab,2);
401 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
402 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
403 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
404 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
405 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
406 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
407 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
408 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
409 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
410 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
412 /* Update potential sum for this i atom from the interaction with this j atom. */
413 velecsum = _mm_add_pd(velecsum,velec);
417 /* Update vectorial force */
418 fix3 = _mm_macc_pd(dx30,fscal,fix3);
419 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
420 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
422 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
423 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
424 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
426 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
428 /* Inner loop uses 185 flops */
435 j_coord_offsetA = DIM*jnrA;
437 /* load j atom coordinates */
438 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
441 /* Calculate displacement vector */
442 dx00 = _mm_sub_pd(ix0,jx0);
443 dy00 = _mm_sub_pd(iy0,jy0);
444 dz00 = _mm_sub_pd(iz0,jz0);
445 dx10 = _mm_sub_pd(ix1,jx0);
446 dy10 = _mm_sub_pd(iy1,jy0);
447 dz10 = _mm_sub_pd(iz1,jz0);
448 dx20 = _mm_sub_pd(ix2,jx0);
449 dy20 = _mm_sub_pd(iy2,jy0);
450 dz20 = _mm_sub_pd(iz2,jz0);
451 dx30 = _mm_sub_pd(ix3,jx0);
452 dy30 = _mm_sub_pd(iy3,jy0);
453 dz30 = _mm_sub_pd(iz3,jz0);
455 /* Calculate squared distance and things based on it */
456 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
457 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
458 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
459 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
461 rinv00 = gmx_mm_invsqrt_pd(rsq00);
462 rinv10 = gmx_mm_invsqrt_pd(rsq10);
463 rinv20 = gmx_mm_invsqrt_pd(rsq20);
464 rinv30 = gmx_mm_invsqrt_pd(rsq30);
466 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
467 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
468 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
469 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
471 /* Load parameters for j particles */
472 jq0 = _mm_load_sd(charge+jnrA+0);
473 vdwjidx0A = 2*vdwtype[jnrA+0];
475 fjx0 = _mm_setzero_pd();
476 fjy0 = _mm_setzero_pd();
477 fjz0 = _mm_setzero_pd();
479 /**************************
480 * CALCULATE INTERACTIONS *
481 **************************/
483 r00 = _mm_mul_pd(rsq00,rinv00);
485 /* Compute parameters for interactions between i and j atoms */
486 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
487 c6grid_00 = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
489 /* Analytical LJ-PME */
490 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
491 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
492 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
493 exponent = gmx_simd_exp_d(ewcljrsq);
494 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
495 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
496 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
497 vvdw6 = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
498 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
499 vvdw = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
500 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
501 fvdw = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
503 /* Update potential sum for this i atom from the interaction with this j atom. */
504 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
505 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
509 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
511 /* Update vectorial force */
512 fix0 = _mm_macc_pd(dx00,fscal,fix0);
513 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
514 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
516 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
517 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
518 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
520 /**************************
521 * CALCULATE INTERACTIONS *
522 **************************/
524 r10 = _mm_mul_pd(rsq10,rinv10);
526 /* Compute parameters for interactions between i and j atoms */
527 qq10 = _mm_mul_pd(iq1,jq0);
529 /* EWALD ELECTROSTATICS */
531 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
532 ewrt = _mm_mul_pd(r10,ewtabscale);
533 ewitab = _mm_cvttpd_epi32(ewrt);
535 eweps = _mm_frcz_pd(ewrt);
537 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
539 twoeweps = _mm_add_pd(eweps,eweps);
540 ewitab = _mm_slli_epi32(ewitab,2);
541 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
542 ewtabD = _mm_setzero_pd();
543 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
544 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
545 ewtabFn = _mm_setzero_pd();
546 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
547 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
548 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
549 velec = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
550 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
552 /* Update potential sum for this i atom from the interaction with this j atom. */
553 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
554 velecsum = _mm_add_pd(velecsum,velec);
558 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
560 /* Update vectorial force */
561 fix1 = _mm_macc_pd(dx10,fscal,fix1);
562 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
563 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
565 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
566 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
567 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
569 /**************************
570 * CALCULATE INTERACTIONS *
571 **************************/
573 r20 = _mm_mul_pd(rsq20,rinv20);
575 /* Compute parameters for interactions between i and j atoms */
576 qq20 = _mm_mul_pd(iq2,jq0);
578 /* EWALD ELECTROSTATICS */
580 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
581 ewrt = _mm_mul_pd(r20,ewtabscale);
582 ewitab = _mm_cvttpd_epi32(ewrt);
584 eweps = _mm_frcz_pd(ewrt);
586 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
588 twoeweps = _mm_add_pd(eweps,eweps);
589 ewitab = _mm_slli_epi32(ewitab,2);
590 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
591 ewtabD = _mm_setzero_pd();
592 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
593 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
594 ewtabFn = _mm_setzero_pd();
595 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
596 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
597 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
598 velec = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
599 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
601 /* Update potential sum for this i atom from the interaction with this j atom. */
602 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
603 velecsum = _mm_add_pd(velecsum,velec);
607 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
609 /* Update vectorial force */
610 fix2 = _mm_macc_pd(dx20,fscal,fix2);
611 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
612 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
614 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
615 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
616 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
618 /**************************
619 * CALCULATE INTERACTIONS *
620 **************************/
622 r30 = _mm_mul_pd(rsq30,rinv30);
624 /* Compute parameters for interactions between i and j atoms */
625 qq30 = _mm_mul_pd(iq3,jq0);
627 /* EWALD ELECTROSTATICS */
629 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
630 ewrt = _mm_mul_pd(r30,ewtabscale);
631 ewitab = _mm_cvttpd_epi32(ewrt);
633 eweps = _mm_frcz_pd(ewrt);
635 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
637 twoeweps = _mm_add_pd(eweps,eweps);
638 ewitab = _mm_slli_epi32(ewitab,2);
639 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
640 ewtabD = _mm_setzero_pd();
641 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
642 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
643 ewtabFn = _mm_setzero_pd();
644 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
645 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
646 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
647 velec = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
648 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
650 /* Update potential sum for this i atom from the interaction with this j atom. */
651 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
652 velecsum = _mm_add_pd(velecsum,velec);
656 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
658 /* Update vectorial force */
659 fix3 = _mm_macc_pd(dx30,fscal,fix3);
660 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
661 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
663 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
664 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
665 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
667 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
669 /* Inner loop uses 185 flops */
672 /* End of innermost loop */
674 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
675 f+i_coord_offset,fshift+i_shift_offset);
678 /* Update potential energies */
679 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
680 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
682 /* Increment number of inner iterations */
683 inneriter += j_index_end - j_index_start;
685 /* Outer loop uses 26 flops */
688 /* Increment number of outer iterations */
691 /* Update outer/inner flops */
693 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*185);
696 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_double
697 * Electrostatics interaction: Ewald
698 * VdW interaction: LJEwald
699 * Geometry: Water4-Particle
700 * Calculate force/pot: Force
703 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_double
704 (t_nblist * gmx_restrict nlist,
705 rvec * gmx_restrict xx,
706 rvec * gmx_restrict ff,
707 t_forcerec * gmx_restrict fr,
708 t_mdatoms * gmx_restrict mdatoms,
709 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
710 t_nrnb * gmx_restrict nrnb)
712 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
713 * just 0 for non-waters.
714 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
715 * jnr indices corresponding to data put in the four positions in the SIMD register.
717 int i_shift_offset,i_coord_offset,outeriter,inneriter;
718 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
720 int j_coord_offsetA,j_coord_offsetB;
721 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
723 real *shiftvec,*fshift,*x,*f;
724 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
726 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
728 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
730 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
732 __m128d ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
733 int vdwjidx0A,vdwjidx0B;
734 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
735 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
736 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
737 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
738 __m128d dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
739 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
742 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
745 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
746 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
752 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
753 __m128d one_half = _mm_set1_pd(0.5);
754 __m128d minus_one = _mm_set1_pd(-1.0);
756 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
758 __m128d dummy_mask,cutoff_mask;
759 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
760 __m128d one = _mm_set1_pd(1.0);
761 __m128d two = _mm_set1_pd(2.0);
767 jindex = nlist->jindex;
769 shiftidx = nlist->shift;
771 shiftvec = fr->shift_vec[0];
772 fshift = fr->fshift[0];
773 facel = _mm_set1_pd(fr->epsfac);
774 charge = mdatoms->chargeA;
775 nvdwtype = fr->ntype;
777 vdwtype = mdatoms->typeA;
778 vdwgridparam = fr->ljpme_c6grid;
779 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
780 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
781 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
783 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
784 ewtab = fr->ic->tabq_coul_F;
785 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
786 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
788 /* Setup water-specific parameters */
789 inr = nlist->iinr[0];
790 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
791 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
792 iq3 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
793 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
795 /* Avoid stupid compiler warnings */
803 /* Start outer loop over neighborlists */
804 for(iidx=0; iidx<nri; iidx++)
806 /* Load shift vector for this list */
807 i_shift_offset = DIM*shiftidx[iidx];
809 /* Load limits for loop over neighbors */
810 j_index_start = jindex[iidx];
811 j_index_end = jindex[iidx+1];
813 /* Get outer coordinate index */
815 i_coord_offset = DIM*inr;
817 /* Load i particle coords and add shift vector */
818 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
819 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
821 fix0 = _mm_setzero_pd();
822 fiy0 = _mm_setzero_pd();
823 fiz0 = _mm_setzero_pd();
824 fix1 = _mm_setzero_pd();
825 fiy1 = _mm_setzero_pd();
826 fiz1 = _mm_setzero_pd();
827 fix2 = _mm_setzero_pd();
828 fiy2 = _mm_setzero_pd();
829 fiz2 = _mm_setzero_pd();
830 fix3 = _mm_setzero_pd();
831 fiy3 = _mm_setzero_pd();
832 fiz3 = _mm_setzero_pd();
834 /* Start inner kernel loop */
835 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
838 /* Get j neighbor index, and coordinate index */
841 j_coord_offsetA = DIM*jnrA;
842 j_coord_offsetB = DIM*jnrB;
844 /* load j atom coordinates */
845 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
848 /* Calculate displacement vector */
849 dx00 = _mm_sub_pd(ix0,jx0);
850 dy00 = _mm_sub_pd(iy0,jy0);
851 dz00 = _mm_sub_pd(iz0,jz0);
852 dx10 = _mm_sub_pd(ix1,jx0);
853 dy10 = _mm_sub_pd(iy1,jy0);
854 dz10 = _mm_sub_pd(iz1,jz0);
855 dx20 = _mm_sub_pd(ix2,jx0);
856 dy20 = _mm_sub_pd(iy2,jy0);
857 dz20 = _mm_sub_pd(iz2,jz0);
858 dx30 = _mm_sub_pd(ix3,jx0);
859 dy30 = _mm_sub_pd(iy3,jy0);
860 dz30 = _mm_sub_pd(iz3,jz0);
862 /* Calculate squared distance and things based on it */
863 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
864 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
865 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
866 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
868 rinv00 = gmx_mm_invsqrt_pd(rsq00);
869 rinv10 = gmx_mm_invsqrt_pd(rsq10);
870 rinv20 = gmx_mm_invsqrt_pd(rsq20);
871 rinv30 = gmx_mm_invsqrt_pd(rsq30);
873 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
874 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
875 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
876 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
878 /* Load parameters for j particles */
879 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
880 vdwjidx0A = 2*vdwtype[jnrA+0];
881 vdwjidx0B = 2*vdwtype[jnrB+0];
883 fjx0 = _mm_setzero_pd();
884 fjy0 = _mm_setzero_pd();
885 fjz0 = _mm_setzero_pd();
887 /**************************
888 * CALCULATE INTERACTIONS *
889 **************************/
891 r00 = _mm_mul_pd(rsq00,rinv00);
893 /* Compute parameters for interactions between i and j atoms */
894 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
895 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
896 c6grid_00 = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
897 vdwgridparam+vdwioffset0+vdwjidx0B);
899 /* Analytical LJ-PME */
900 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
901 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
902 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
903 exponent = gmx_simd_exp_d(ewcljrsq);
904 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
905 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
906 /* f6A = 6 * C6grid * (1 - poly) */
907 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
908 /* f6B = C6grid * exponent * beta^6 */
909 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
910 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
911 fvdw = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
915 /* Update vectorial force */
916 fix0 = _mm_macc_pd(dx00,fscal,fix0);
917 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
918 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
920 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
921 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
922 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
924 /**************************
925 * CALCULATE INTERACTIONS *
926 **************************/
928 r10 = _mm_mul_pd(rsq10,rinv10);
930 /* Compute parameters for interactions between i and j atoms */
931 qq10 = _mm_mul_pd(iq1,jq0);
933 /* EWALD ELECTROSTATICS */
935 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
936 ewrt = _mm_mul_pd(r10,ewtabscale);
937 ewitab = _mm_cvttpd_epi32(ewrt);
939 eweps = _mm_frcz_pd(ewrt);
941 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
943 twoeweps = _mm_add_pd(eweps,eweps);
944 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
946 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
947 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
951 /* Update vectorial force */
952 fix1 = _mm_macc_pd(dx10,fscal,fix1);
953 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
954 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
956 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
957 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
958 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
960 /**************************
961 * CALCULATE INTERACTIONS *
962 **************************/
964 r20 = _mm_mul_pd(rsq20,rinv20);
966 /* Compute parameters for interactions between i and j atoms */
967 qq20 = _mm_mul_pd(iq2,jq0);
969 /* EWALD ELECTROSTATICS */
971 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
972 ewrt = _mm_mul_pd(r20,ewtabscale);
973 ewitab = _mm_cvttpd_epi32(ewrt);
975 eweps = _mm_frcz_pd(ewrt);
977 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
979 twoeweps = _mm_add_pd(eweps,eweps);
980 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
982 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
983 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
987 /* Update vectorial force */
988 fix2 = _mm_macc_pd(dx20,fscal,fix2);
989 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
990 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
992 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
993 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
994 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
996 /**************************
997 * CALCULATE INTERACTIONS *
998 **************************/
1000 r30 = _mm_mul_pd(rsq30,rinv30);
1002 /* Compute parameters for interactions between i and j atoms */
1003 qq30 = _mm_mul_pd(iq3,jq0);
1005 /* EWALD ELECTROSTATICS */
1007 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1008 ewrt = _mm_mul_pd(r30,ewtabscale);
1009 ewitab = _mm_cvttpd_epi32(ewrt);
1011 eweps = _mm_frcz_pd(ewrt);
1013 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1015 twoeweps = _mm_add_pd(eweps,eweps);
1016 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1018 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1019 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1023 /* Update vectorial force */
1024 fix3 = _mm_macc_pd(dx30,fscal,fix3);
1025 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
1026 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
1028 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
1029 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
1030 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
1032 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1034 /* Inner loop uses 167 flops */
1037 if(jidx<j_index_end)
1041 j_coord_offsetA = DIM*jnrA;
1043 /* load j atom coordinates */
1044 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1047 /* Calculate displacement vector */
1048 dx00 = _mm_sub_pd(ix0,jx0);
1049 dy00 = _mm_sub_pd(iy0,jy0);
1050 dz00 = _mm_sub_pd(iz0,jz0);
1051 dx10 = _mm_sub_pd(ix1,jx0);
1052 dy10 = _mm_sub_pd(iy1,jy0);
1053 dz10 = _mm_sub_pd(iz1,jz0);
1054 dx20 = _mm_sub_pd(ix2,jx0);
1055 dy20 = _mm_sub_pd(iy2,jy0);
1056 dz20 = _mm_sub_pd(iz2,jz0);
1057 dx30 = _mm_sub_pd(ix3,jx0);
1058 dy30 = _mm_sub_pd(iy3,jy0);
1059 dz30 = _mm_sub_pd(iz3,jz0);
1061 /* Calculate squared distance and things based on it */
1062 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1063 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1064 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1065 rsq30 = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1067 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1068 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1069 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1070 rinv30 = gmx_mm_invsqrt_pd(rsq30);
1072 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1073 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1074 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1075 rinvsq30 = _mm_mul_pd(rinv30,rinv30);
1077 /* Load parameters for j particles */
1078 jq0 = _mm_load_sd(charge+jnrA+0);
1079 vdwjidx0A = 2*vdwtype[jnrA+0];
1081 fjx0 = _mm_setzero_pd();
1082 fjy0 = _mm_setzero_pd();
1083 fjz0 = _mm_setzero_pd();
1085 /**************************
1086 * CALCULATE INTERACTIONS *
1087 **************************/
1089 r00 = _mm_mul_pd(rsq00,rinv00);
1091 /* Compute parameters for interactions between i and j atoms */
1092 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1093 c6grid_00 = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1095 /* Analytical LJ-PME */
1096 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1097 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
1098 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1099 exponent = gmx_simd_exp_d(ewcljrsq);
1100 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1101 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1102 /* f6A = 6 * C6grid * (1 - poly) */
1103 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1104 /* f6B = C6grid * exponent * beta^6 */
1105 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1106 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1107 fvdw = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1111 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1113 /* Update vectorial force */
1114 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1115 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1116 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1118 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1119 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1120 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1122 /**************************
1123 * CALCULATE INTERACTIONS *
1124 **************************/
1126 r10 = _mm_mul_pd(rsq10,rinv10);
1128 /* Compute parameters for interactions between i and j atoms */
1129 qq10 = _mm_mul_pd(iq1,jq0);
1131 /* EWALD ELECTROSTATICS */
1133 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1134 ewrt = _mm_mul_pd(r10,ewtabscale);
1135 ewitab = _mm_cvttpd_epi32(ewrt);
1137 eweps = _mm_frcz_pd(ewrt);
1139 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1141 twoeweps = _mm_add_pd(eweps,eweps);
1142 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1143 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1144 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1148 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1150 /* Update vectorial force */
1151 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1152 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1153 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1155 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1156 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1157 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1159 /**************************
1160 * CALCULATE INTERACTIONS *
1161 **************************/
1163 r20 = _mm_mul_pd(rsq20,rinv20);
1165 /* Compute parameters for interactions between i and j atoms */
1166 qq20 = _mm_mul_pd(iq2,jq0);
1168 /* EWALD ELECTROSTATICS */
1170 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1171 ewrt = _mm_mul_pd(r20,ewtabscale);
1172 ewitab = _mm_cvttpd_epi32(ewrt);
1174 eweps = _mm_frcz_pd(ewrt);
1176 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1178 twoeweps = _mm_add_pd(eweps,eweps);
1179 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1180 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1181 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1185 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1187 /* Update vectorial force */
1188 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1189 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1190 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1192 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1193 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1194 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1196 /**************************
1197 * CALCULATE INTERACTIONS *
1198 **************************/
1200 r30 = _mm_mul_pd(rsq30,rinv30);
1202 /* Compute parameters for interactions between i and j atoms */
1203 qq30 = _mm_mul_pd(iq3,jq0);
1205 /* EWALD ELECTROSTATICS */
1207 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1208 ewrt = _mm_mul_pd(r30,ewtabscale);
1209 ewitab = _mm_cvttpd_epi32(ewrt);
1211 eweps = _mm_frcz_pd(ewrt);
1213 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1215 twoeweps = _mm_add_pd(eweps,eweps);
1216 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1217 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1218 felec = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1222 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1224 /* Update vectorial force */
1225 fix3 = _mm_macc_pd(dx30,fscal,fix3);
1226 fiy3 = _mm_macc_pd(dy30,fscal,fiy3);
1227 fiz3 = _mm_macc_pd(dz30,fscal,fiz3);
1229 fjx0 = _mm_macc_pd(dx30,fscal,fjx0);
1230 fjy0 = _mm_macc_pd(dy30,fscal,fjy0);
1231 fjz0 = _mm_macc_pd(dz30,fscal,fjz0);
1233 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1235 /* Inner loop uses 167 flops */
1238 /* End of innermost loop */
1240 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1241 f+i_coord_offset,fshift+i_shift_offset);
1243 /* Increment number of inner iterations */
1244 inneriter += j_index_end - j_index_start;
1246 /* Outer loop uses 24 flops */
1249 /* Increment number of outer iterations */
1252 /* Update outer/inner flops */
1254 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*167);