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_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_128_fma_double
52 * Electrostatics interaction: Ewald
53 * VdW interaction: LJEwald
54 * Geometry: Water3-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_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;
83 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
85 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
86 int vdwjidx0A,vdwjidx0B;
87 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
88 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
89 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
90 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
91 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
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);
103 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
104 __m128d one_half = _mm_set1_pd(0.5);
105 __m128d minus_one = _mm_set1_pd(-1.0);
107 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
109 __m128d dummy_mask,cutoff_mask;
110 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
111 __m128d one = _mm_set1_pd(1.0);
112 __m128d two = _mm_set1_pd(2.0);
118 jindex = nlist->jindex;
120 shiftidx = nlist->shift;
122 shiftvec = fr->shift_vec[0];
123 fshift = fr->fshift[0];
124 facel = _mm_set1_pd(fr->epsfac);
125 charge = mdatoms->chargeA;
126 nvdwtype = fr->ntype;
128 vdwtype = mdatoms->typeA;
129 vdwgridparam = fr->ljpme_c6grid;
130 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
131 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
132 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
134 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
135 ewtab = fr->ic->tabq_coul_FDV0;
136 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
137 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
139 /* Setup water-specific parameters */
140 inr = nlist->iinr[0];
141 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
142 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
143 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
144 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
146 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
147 rcutoff_scalar = fr->rcoulomb;
148 rcutoff = _mm_set1_pd(rcutoff_scalar);
149 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
151 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
152 rvdw = _mm_set1_pd(fr->rvdw);
154 /* Avoid stupid compiler warnings */
162 /* Start outer loop over neighborlists */
163 for(iidx=0; iidx<nri; iidx++)
165 /* Load shift vector for this list */
166 i_shift_offset = DIM*shiftidx[iidx];
168 /* Load limits for loop over neighbors */
169 j_index_start = jindex[iidx];
170 j_index_end = jindex[iidx+1];
172 /* Get outer coordinate index */
174 i_coord_offset = DIM*inr;
176 /* Load i particle coords and add shift vector */
177 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
178 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
180 fix0 = _mm_setzero_pd();
181 fiy0 = _mm_setzero_pd();
182 fiz0 = _mm_setzero_pd();
183 fix1 = _mm_setzero_pd();
184 fiy1 = _mm_setzero_pd();
185 fiz1 = _mm_setzero_pd();
186 fix2 = _mm_setzero_pd();
187 fiy2 = _mm_setzero_pd();
188 fiz2 = _mm_setzero_pd();
190 /* Reset potential sums */
191 velecsum = _mm_setzero_pd();
192 vvdwsum = _mm_setzero_pd();
194 /* Start inner kernel loop */
195 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
198 /* Get j neighbor index, and coordinate index */
201 j_coord_offsetA = DIM*jnrA;
202 j_coord_offsetB = DIM*jnrB;
204 /* load j atom coordinates */
205 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
208 /* Calculate displacement vector */
209 dx00 = _mm_sub_pd(ix0,jx0);
210 dy00 = _mm_sub_pd(iy0,jy0);
211 dz00 = _mm_sub_pd(iz0,jz0);
212 dx10 = _mm_sub_pd(ix1,jx0);
213 dy10 = _mm_sub_pd(iy1,jy0);
214 dz10 = _mm_sub_pd(iz1,jz0);
215 dx20 = _mm_sub_pd(ix2,jx0);
216 dy20 = _mm_sub_pd(iy2,jy0);
217 dz20 = _mm_sub_pd(iz2,jz0);
219 /* Calculate squared distance and things based on it */
220 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
221 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
222 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
224 rinv00 = gmx_mm_invsqrt_pd(rsq00);
225 rinv10 = gmx_mm_invsqrt_pd(rsq10);
226 rinv20 = gmx_mm_invsqrt_pd(rsq20);
228 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
229 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
230 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
232 /* Load parameters for j particles */
233 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
234 vdwjidx0A = 2*vdwtype[jnrA+0];
235 vdwjidx0B = 2*vdwtype[jnrB+0];
237 fjx0 = _mm_setzero_pd();
238 fjy0 = _mm_setzero_pd();
239 fjz0 = _mm_setzero_pd();
241 /**************************
242 * CALCULATE INTERACTIONS *
243 **************************/
245 if (gmx_mm_any_lt(rsq00,rcutoff2))
248 r00 = _mm_mul_pd(rsq00,rinv00);
250 /* Compute parameters for interactions between i and j atoms */
251 qq00 = _mm_mul_pd(iq0,jq0);
252 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
253 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
254 c6grid_00 = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
255 vdwgridparam+vdwioffset0+vdwjidx0B);
257 /* EWALD ELECTROSTATICS */
259 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
260 ewrt = _mm_mul_pd(r00,ewtabscale);
261 ewitab = _mm_cvttpd_epi32(ewrt);
263 eweps = _mm_frcz_pd(ewrt);
265 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
267 twoeweps = _mm_add_pd(eweps,eweps);
268 ewitab = _mm_slli_epi32(ewitab,2);
269 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
270 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
271 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
272 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
273 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
274 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
275 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
276 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
277 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
278 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
280 /* Analytical LJ-PME */
281 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
282 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
283 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
284 exponent = gmx_simd_exp_d(ewcljrsq);
285 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
286 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
287 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
288 vvdw6 = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
289 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
290 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
291 _mm_mul_pd(_mm_sub_pd(vvdw6,_mm_macc_pd(c6grid_00,sh_lj_ewald,_mm_mul_pd(c6_00,sh_vdw_invrcut6))),one_sixth));
292 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
293 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);
295 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
297 /* Update potential sum for this i atom from the interaction with this j atom. */
298 velec = _mm_and_pd(velec,cutoff_mask);
299 velecsum = _mm_add_pd(velecsum,velec);
300 vvdw = _mm_and_pd(vvdw,cutoff_mask);
301 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
303 fscal = _mm_add_pd(felec,fvdw);
305 fscal = _mm_and_pd(fscal,cutoff_mask);
307 /* Update vectorial force */
308 fix0 = _mm_macc_pd(dx00,fscal,fix0);
309 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
310 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
312 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
313 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
314 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
318 /**************************
319 * CALCULATE INTERACTIONS *
320 **************************/
322 if (gmx_mm_any_lt(rsq10,rcutoff2))
325 r10 = _mm_mul_pd(rsq10,rinv10);
327 /* Compute parameters for interactions between i and j atoms */
328 qq10 = _mm_mul_pd(iq1,jq0);
330 /* EWALD ELECTROSTATICS */
332 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
333 ewrt = _mm_mul_pd(r10,ewtabscale);
334 ewitab = _mm_cvttpd_epi32(ewrt);
336 eweps = _mm_frcz_pd(ewrt);
338 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
340 twoeweps = _mm_add_pd(eweps,eweps);
341 ewitab = _mm_slli_epi32(ewitab,2);
342 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
343 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
344 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
345 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
346 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
347 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
348 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
349 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
350 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
351 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
353 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
355 /* Update potential sum for this i atom from the interaction with this j atom. */
356 velec = _mm_and_pd(velec,cutoff_mask);
357 velecsum = _mm_add_pd(velecsum,velec);
361 fscal = _mm_and_pd(fscal,cutoff_mask);
363 /* Update vectorial force */
364 fix1 = _mm_macc_pd(dx10,fscal,fix1);
365 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
366 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
368 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
369 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
370 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
374 /**************************
375 * CALCULATE INTERACTIONS *
376 **************************/
378 if (gmx_mm_any_lt(rsq20,rcutoff2))
381 r20 = _mm_mul_pd(rsq20,rinv20);
383 /* Compute parameters for interactions between i and j atoms */
384 qq20 = _mm_mul_pd(iq2,jq0);
386 /* EWALD ELECTROSTATICS */
388 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
389 ewrt = _mm_mul_pd(r20,ewtabscale);
390 ewitab = _mm_cvttpd_epi32(ewrt);
392 eweps = _mm_frcz_pd(ewrt);
394 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
396 twoeweps = _mm_add_pd(eweps,eweps);
397 ewitab = _mm_slli_epi32(ewitab,2);
398 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
399 ewtabD = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
400 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
401 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
402 ewtabFn = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
403 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
404 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
405 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
406 velec = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
407 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
409 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
411 /* Update potential sum for this i atom from the interaction with this j atom. */
412 velec = _mm_and_pd(velec,cutoff_mask);
413 velecsum = _mm_add_pd(velecsum,velec);
417 fscal = _mm_and_pd(fscal,cutoff_mask);
419 /* Update vectorial force */
420 fix2 = _mm_macc_pd(dx20,fscal,fix2);
421 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
422 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
424 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
425 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
426 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
430 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
432 /* Inner loop uses 179 flops */
439 j_coord_offsetA = DIM*jnrA;
441 /* load j atom coordinates */
442 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
445 /* Calculate displacement vector */
446 dx00 = _mm_sub_pd(ix0,jx0);
447 dy00 = _mm_sub_pd(iy0,jy0);
448 dz00 = _mm_sub_pd(iz0,jz0);
449 dx10 = _mm_sub_pd(ix1,jx0);
450 dy10 = _mm_sub_pd(iy1,jy0);
451 dz10 = _mm_sub_pd(iz1,jz0);
452 dx20 = _mm_sub_pd(ix2,jx0);
453 dy20 = _mm_sub_pd(iy2,jy0);
454 dz20 = _mm_sub_pd(iz2,jz0);
456 /* Calculate squared distance and things based on it */
457 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
458 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
459 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
461 rinv00 = gmx_mm_invsqrt_pd(rsq00);
462 rinv10 = gmx_mm_invsqrt_pd(rsq10);
463 rinv20 = gmx_mm_invsqrt_pd(rsq20);
465 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
466 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
467 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
469 /* Load parameters for j particles */
470 jq0 = _mm_load_sd(charge+jnrA+0);
471 vdwjidx0A = 2*vdwtype[jnrA+0];
473 fjx0 = _mm_setzero_pd();
474 fjy0 = _mm_setzero_pd();
475 fjz0 = _mm_setzero_pd();
477 /**************************
478 * CALCULATE INTERACTIONS *
479 **************************/
481 if (gmx_mm_any_lt(rsq00,rcutoff2))
484 r00 = _mm_mul_pd(rsq00,rinv00);
486 /* Compute parameters for interactions between i and j atoms */
487 qq00 = _mm_mul_pd(iq0,jq0);
488 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
489 c6grid_00 = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
491 /* EWALD ELECTROSTATICS */
493 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
494 ewrt = _mm_mul_pd(r00,ewtabscale);
495 ewitab = _mm_cvttpd_epi32(ewrt);
497 eweps = _mm_frcz_pd(ewrt);
499 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
501 twoeweps = _mm_add_pd(eweps,eweps);
502 ewitab = _mm_slli_epi32(ewitab,2);
503 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
504 ewtabD = _mm_setzero_pd();
505 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
506 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
507 ewtabFn = _mm_setzero_pd();
508 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
509 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
510 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
511 velec = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
512 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
514 /* Analytical LJ-PME */
515 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
516 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
517 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
518 exponent = gmx_simd_exp_d(ewcljrsq);
519 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
520 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
521 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
522 vvdw6 = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
523 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
524 vvdw = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
525 _mm_mul_pd(_mm_sub_pd(vvdw6,_mm_macc_pd(c6grid_00,sh_lj_ewald,_mm_mul_pd(c6_00,sh_vdw_invrcut6))),one_sixth));
526 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
527 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);
529 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
531 /* Update potential sum for this i atom from the interaction with this j atom. */
532 velec = _mm_and_pd(velec,cutoff_mask);
533 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
534 velecsum = _mm_add_pd(velecsum,velec);
535 vvdw = _mm_and_pd(vvdw,cutoff_mask);
536 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
537 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
539 fscal = _mm_add_pd(felec,fvdw);
541 fscal = _mm_and_pd(fscal,cutoff_mask);
543 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
545 /* Update vectorial force */
546 fix0 = _mm_macc_pd(dx00,fscal,fix0);
547 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
548 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
550 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
551 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
552 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
556 /**************************
557 * CALCULATE INTERACTIONS *
558 **************************/
560 if (gmx_mm_any_lt(rsq10,rcutoff2))
563 r10 = _mm_mul_pd(rsq10,rinv10);
565 /* Compute parameters for interactions between i and j atoms */
566 qq10 = _mm_mul_pd(iq1,jq0);
568 /* EWALD ELECTROSTATICS */
570 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
571 ewrt = _mm_mul_pd(r10,ewtabscale);
572 ewitab = _mm_cvttpd_epi32(ewrt);
574 eweps = _mm_frcz_pd(ewrt);
576 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
578 twoeweps = _mm_add_pd(eweps,eweps);
579 ewitab = _mm_slli_epi32(ewitab,2);
580 ewtabF = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
581 ewtabD = _mm_setzero_pd();
582 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
583 ewtabV = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
584 ewtabFn = _mm_setzero_pd();
585 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
586 felec = _mm_macc_pd(eweps,ewtabD,ewtabF);
587 velec = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
588 velec = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
589 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
591 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
593 /* Update potential sum for this i atom from the interaction with this j atom. */
594 velec = _mm_and_pd(velec,cutoff_mask);
595 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
596 velecsum = _mm_add_pd(velecsum,velec);
600 fscal = _mm_and_pd(fscal,cutoff_mask);
602 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
604 /* Update vectorial force */
605 fix1 = _mm_macc_pd(dx10,fscal,fix1);
606 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
607 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
609 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
610 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
611 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
615 /**************************
616 * CALCULATE INTERACTIONS *
617 **************************/
619 if (gmx_mm_any_lt(rsq20,rcutoff2))
622 r20 = _mm_mul_pd(rsq20,rinv20);
624 /* Compute parameters for interactions between i and j atoms */
625 qq20 = _mm_mul_pd(iq2,jq0);
627 /* EWALD ELECTROSTATICS */
629 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
630 ewrt = _mm_mul_pd(r20,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(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
648 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
650 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
652 /* Update potential sum for this i atom from the interaction with this j atom. */
653 velec = _mm_and_pd(velec,cutoff_mask);
654 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
655 velecsum = _mm_add_pd(velecsum,velec);
659 fscal = _mm_and_pd(fscal,cutoff_mask);
661 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
663 /* Update vectorial force */
664 fix2 = _mm_macc_pd(dx20,fscal,fix2);
665 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
666 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
668 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
669 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
670 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
674 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
676 /* Inner loop uses 179 flops */
679 /* End of innermost loop */
681 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
682 f+i_coord_offset,fshift+i_shift_offset);
685 /* Update potential energies */
686 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
687 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
689 /* Increment number of inner iterations */
690 inneriter += j_index_end - j_index_start;
692 /* Outer loop uses 20 flops */
695 /* Increment number of outer iterations */
698 /* Update outer/inner flops */
700 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*179);
703 * Gromacs nonbonded kernel: nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_128_fma_double
704 * Electrostatics interaction: Ewald
705 * VdW interaction: LJEwald
706 * Geometry: Water3-Particle
707 * Calculate force/pot: Force
710 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_128_fma_double
711 (t_nblist * gmx_restrict nlist,
712 rvec * gmx_restrict xx,
713 rvec * gmx_restrict ff,
714 t_forcerec * gmx_restrict fr,
715 t_mdatoms * gmx_restrict mdatoms,
716 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
717 t_nrnb * gmx_restrict nrnb)
719 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
720 * just 0 for non-waters.
721 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
722 * jnr indices corresponding to data put in the four positions in the SIMD register.
724 int i_shift_offset,i_coord_offset,outeriter,inneriter;
725 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
727 int j_coord_offsetA,j_coord_offsetB;
728 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
730 real *shiftvec,*fshift,*x,*f;
731 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
733 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
735 __m128d ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
737 __m128d ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
738 int vdwjidx0A,vdwjidx0B;
739 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
740 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
741 __m128d dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
742 __m128d dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
743 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
746 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
749 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
750 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
755 __m128d ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
756 __m128d one_half = _mm_set1_pd(0.5);
757 __m128d minus_one = _mm_set1_pd(-1.0);
759 __m128d ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
761 __m128d dummy_mask,cutoff_mask;
762 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
763 __m128d one = _mm_set1_pd(1.0);
764 __m128d two = _mm_set1_pd(2.0);
770 jindex = nlist->jindex;
772 shiftidx = nlist->shift;
774 shiftvec = fr->shift_vec[0];
775 fshift = fr->fshift[0];
776 facel = _mm_set1_pd(fr->epsfac);
777 charge = mdatoms->chargeA;
778 nvdwtype = fr->ntype;
780 vdwtype = mdatoms->typeA;
781 vdwgridparam = fr->ljpme_c6grid;
782 sh_lj_ewald = _mm_set1_pd(fr->ic->sh_lj_ewald);
783 ewclj = _mm_set1_pd(fr->ewaldcoeff_lj);
784 ewclj2 = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
786 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
787 ewtab = fr->ic->tabq_coul_F;
788 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
789 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
791 /* Setup water-specific parameters */
792 inr = nlist->iinr[0];
793 iq0 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
794 iq1 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
795 iq2 = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
796 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
798 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
799 rcutoff_scalar = fr->rcoulomb;
800 rcutoff = _mm_set1_pd(rcutoff_scalar);
801 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
803 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
804 rvdw = _mm_set1_pd(fr->rvdw);
806 /* Avoid stupid compiler warnings */
814 /* Start outer loop over neighborlists */
815 for(iidx=0; iidx<nri; iidx++)
817 /* Load shift vector for this list */
818 i_shift_offset = DIM*shiftidx[iidx];
820 /* Load limits for loop over neighbors */
821 j_index_start = jindex[iidx];
822 j_index_end = jindex[iidx+1];
824 /* Get outer coordinate index */
826 i_coord_offset = DIM*inr;
828 /* Load i particle coords and add shift vector */
829 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
830 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
832 fix0 = _mm_setzero_pd();
833 fiy0 = _mm_setzero_pd();
834 fiz0 = _mm_setzero_pd();
835 fix1 = _mm_setzero_pd();
836 fiy1 = _mm_setzero_pd();
837 fiz1 = _mm_setzero_pd();
838 fix2 = _mm_setzero_pd();
839 fiy2 = _mm_setzero_pd();
840 fiz2 = _mm_setzero_pd();
842 /* Start inner kernel loop */
843 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
846 /* Get j neighbor index, and coordinate index */
849 j_coord_offsetA = DIM*jnrA;
850 j_coord_offsetB = DIM*jnrB;
852 /* load j atom coordinates */
853 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
856 /* Calculate displacement vector */
857 dx00 = _mm_sub_pd(ix0,jx0);
858 dy00 = _mm_sub_pd(iy0,jy0);
859 dz00 = _mm_sub_pd(iz0,jz0);
860 dx10 = _mm_sub_pd(ix1,jx0);
861 dy10 = _mm_sub_pd(iy1,jy0);
862 dz10 = _mm_sub_pd(iz1,jz0);
863 dx20 = _mm_sub_pd(ix2,jx0);
864 dy20 = _mm_sub_pd(iy2,jy0);
865 dz20 = _mm_sub_pd(iz2,jz0);
867 /* Calculate squared distance and things based on it */
868 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
869 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
870 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
872 rinv00 = gmx_mm_invsqrt_pd(rsq00);
873 rinv10 = gmx_mm_invsqrt_pd(rsq10);
874 rinv20 = gmx_mm_invsqrt_pd(rsq20);
876 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
877 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
878 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
880 /* Load parameters for j particles */
881 jq0 = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
882 vdwjidx0A = 2*vdwtype[jnrA+0];
883 vdwjidx0B = 2*vdwtype[jnrB+0];
885 fjx0 = _mm_setzero_pd();
886 fjy0 = _mm_setzero_pd();
887 fjz0 = _mm_setzero_pd();
889 /**************************
890 * CALCULATE INTERACTIONS *
891 **************************/
893 if (gmx_mm_any_lt(rsq00,rcutoff2))
896 r00 = _mm_mul_pd(rsq00,rinv00);
898 /* Compute parameters for interactions between i and j atoms */
899 qq00 = _mm_mul_pd(iq0,jq0);
900 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
901 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
902 c6grid_00 = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
903 vdwgridparam+vdwioffset0+vdwjidx0B);
905 /* EWALD ELECTROSTATICS */
907 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
908 ewrt = _mm_mul_pd(r00,ewtabscale);
909 ewitab = _mm_cvttpd_epi32(ewrt);
911 eweps = _mm_frcz_pd(ewrt);
913 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
915 twoeweps = _mm_add_pd(eweps,eweps);
916 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
918 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
919 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
921 /* Analytical LJ-PME */
922 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
923 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
924 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
925 exponent = gmx_simd_exp_d(ewcljrsq);
926 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
927 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
928 /* f6A = 6 * C6grid * (1 - poly) */
929 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
930 /* f6B = C6grid * exponent * beta^6 */
931 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
932 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
933 fvdw = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
935 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
937 fscal = _mm_add_pd(felec,fvdw);
939 fscal = _mm_and_pd(fscal,cutoff_mask);
941 /* Update vectorial force */
942 fix0 = _mm_macc_pd(dx00,fscal,fix0);
943 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
944 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
946 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
947 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
948 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
952 /**************************
953 * CALCULATE INTERACTIONS *
954 **************************/
956 if (gmx_mm_any_lt(rsq10,rcutoff2))
959 r10 = _mm_mul_pd(rsq10,rinv10);
961 /* Compute parameters for interactions between i and j atoms */
962 qq10 = _mm_mul_pd(iq1,jq0);
964 /* EWALD ELECTROSTATICS */
966 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
967 ewrt = _mm_mul_pd(r10,ewtabscale);
968 ewitab = _mm_cvttpd_epi32(ewrt);
970 eweps = _mm_frcz_pd(ewrt);
972 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
974 twoeweps = _mm_add_pd(eweps,eweps);
975 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
977 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
978 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
980 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
984 fscal = _mm_and_pd(fscal,cutoff_mask);
986 /* Update vectorial force */
987 fix1 = _mm_macc_pd(dx10,fscal,fix1);
988 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
989 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
991 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
992 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
993 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
997 /**************************
998 * CALCULATE INTERACTIONS *
999 **************************/
1001 if (gmx_mm_any_lt(rsq20,rcutoff2))
1004 r20 = _mm_mul_pd(rsq20,rinv20);
1006 /* Compute parameters for interactions between i and j atoms */
1007 qq20 = _mm_mul_pd(iq2,jq0);
1009 /* EWALD ELECTROSTATICS */
1011 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1012 ewrt = _mm_mul_pd(r20,ewtabscale);
1013 ewitab = _mm_cvttpd_epi32(ewrt);
1015 eweps = _mm_frcz_pd(ewrt);
1017 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1019 twoeweps = _mm_add_pd(eweps,eweps);
1020 gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1022 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1023 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1025 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1029 fscal = _mm_and_pd(fscal,cutoff_mask);
1031 /* Update vectorial force */
1032 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1033 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1034 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1036 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1037 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1038 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1042 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1044 /* Inner loop uses 150 flops */
1047 if(jidx<j_index_end)
1051 j_coord_offsetA = DIM*jnrA;
1053 /* load j atom coordinates */
1054 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1057 /* Calculate displacement vector */
1058 dx00 = _mm_sub_pd(ix0,jx0);
1059 dy00 = _mm_sub_pd(iy0,jy0);
1060 dz00 = _mm_sub_pd(iz0,jz0);
1061 dx10 = _mm_sub_pd(ix1,jx0);
1062 dy10 = _mm_sub_pd(iy1,jy0);
1063 dz10 = _mm_sub_pd(iz1,jz0);
1064 dx20 = _mm_sub_pd(ix2,jx0);
1065 dy20 = _mm_sub_pd(iy2,jy0);
1066 dz20 = _mm_sub_pd(iz2,jz0);
1068 /* Calculate squared distance and things based on it */
1069 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1070 rsq10 = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1071 rsq20 = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1073 rinv00 = gmx_mm_invsqrt_pd(rsq00);
1074 rinv10 = gmx_mm_invsqrt_pd(rsq10);
1075 rinv20 = gmx_mm_invsqrt_pd(rsq20);
1077 rinvsq00 = _mm_mul_pd(rinv00,rinv00);
1078 rinvsq10 = _mm_mul_pd(rinv10,rinv10);
1079 rinvsq20 = _mm_mul_pd(rinv20,rinv20);
1081 /* Load parameters for j particles */
1082 jq0 = _mm_load_sd(charge+jnrA+0);
1083 vdwjidx0A = 2*vdwtype[jnrA+0];
1085 fjx0 = _mm_setzero_pd();
1086 fjy0 = _mm_setzero_pd();
1087 fjz0 = _mm_setzero_pd();
1089 /**************************
1090 * CALCULATE INTERACTIONS *
1091 **************************/
1093 if (gmx_mm_any_lt(rsq00,rcutoff2))
1096 r00 = _mm_mul_pd(rsq00,rinv00);
1098 /* Compute parameters for interactions between i and j atoms */
1099 qq00 = _mm_mul_pd(iq0,jq0);
1100 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1101 c6grid_00 = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1103 /* EWALD ELECTROSTATICS */
1105 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1106 ewrt = _mm_mul_pd(r00,ewtabscale);
1107 ewitab = _mm_cvttpd_epi32(ewrt);
1109 eweps = _mm_frcz_pd(ewrt);
1111 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1113 twoeweps = _mm_add_pd(eweps,eweps);
1114 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1115 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1116 felec = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1118 /* Analytical LJ-PME */
1119 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1120 ewcljrsq = _mm_mul_pd(ewclj2,rsq00);
1121 ewclj6 = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1122 exponent = gmx_simd_exp_d(ewcljrsq);
1123 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1124 poly = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1125 /* f6A = 6 * C6grid * (1 - poly) */
1126 f6A = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1127 /* f6B = C6grid * exponent * beta^6 */
1128 f6B = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1129 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1130 fvdw = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1132 cutoff_mask = _mm_cmplt_pd(rsq00,rcutoff2);
1134 fscal = _mm_add_pd(felec,fvdw);
1136 fscal = _mm_and_pd(fscal,cutoff_mask);
1138 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1140 /* Update vectorial force */
1141 fix0 = _mm_macc_pd(dx00,fscal,fix0);
1142 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
1143 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
1145 fjx0 = _mm_macc_pd(dx00,fscal,fjx0);
1146 fjy0 = _mm_macc_pd(dy00,fscal,fjy0);
1147 fjz0 = _mm_macc_pd(dz00,fscal,fjz0);
1151 /**************************
1152 * CALCULATE INTERACTIONS *
1153 **************************/
1155 if (gmx_mm_any_lt(rsq10,rcutoff2))
1158 r10 = _mm_mul_pd(rsq10,rinv10);
1160 /* Compute parameters for interactions between i and j atoms */
1161 qq10 = _mm_mul_pd(iq1,jq0);
1163 /* EWALD ELECTROSTATICS */
1165 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1166 ewrt = _mm_mul_pd(r10,ewtabscale);
1167 ewitab = _mm_cvttpd_epi32(ewrt);
1169 eweps = _mm_frcz_pd(ewrt);
1171 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1173 twoeweps = _mm_add_pd(eweps,eweps);
1174 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1175 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1176 felec = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1178 cutoff_mask = _mm_cmplt_pd(rsq10,rcutoff2);
1182 fscal = _mm_and_pd(fscal,cutoff_mask);
1184 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1186 /* Update vectorial force */
1187 fix1 = _mm_macc_pd(dx10,fscal,fix1);
1188 fiy1 = _mm_macc_pd(dy10,fscal,fiy1);
1189 fiz1 = _mm_macc_pd(dz10,fscal,fiz1);
1191 fjx0 = _mm_macc_pd(dx10,fscal,fjx0);
1192 fjy0 = _mm_macc_pd(dy10,fscal,fjy0);
1193 fjz0 = _mm_macc_pd(dz10,fscal,fjz0);
1197 /**************************
1198 * CALCULATE INTERACTIONS *
1199 **************************/
1201 if (gmx_mm_any_lt(rsq20,rcutoff2))
1204 r20 = _mm_mul_pd(rsq20,rinv20);
1206 /* Compute parameters for interactions between i and j atoms */
1207 qq20 = _mm_mul_pd(iq2,jq0);
1209 /* EWALD ELECTROSTATICS */
1211 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1212 ewrt = _mm_mul_pd(r20,ewtabscale);
1213 ewitab = _mm_cvttpd_epi32(ewrt);
1215 eweps = _mm_frcz_pd(ewrt);
1217 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1219 twoeweps = _mm_add_pd(eweps,eweps);
1220 gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1221 felec = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1222 felec = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1224 cutoff_mask = _mm_cmplt_pd(rsq20,rcutoff2);
1228 fscal = _mm_and_pd(fscal,cutoff_mask);
1230 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1232 /* Update vectorial force */
1233 fix2 = _mm_macc_pd(dx20,fscal,fix2);
1234 fiy2 = _mm_macc_pd(dy20,fscal,fiy2);
1235 fiz2 = _mm_macc_pd(dz20,fscal,fiz2);
1237 fjx0 = _mm_macc_pd(dx20,fscal,fjx0);
1238 fjy0 = _mm_macc_pd(dy20,fscal,fjy0);
1239 fjz0 = _mm_macc_pd(dz20,fscal,fjz0);
1243 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1245 /* Inner loop uses 150 flops */
1248 /* End of innermost loop */
1250 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1251 f+i_coord_offset,fshift+i_shift_offset);
1253 /* Increment number of inner iterations */
1254 inneriter += j_index_end - j_index_start;
1256 /* Outer loop uses 18 flops */
1259 /* Increment number of outer iterations */
1262 /* Update outer/inner flops */
1264 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*150);