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_256_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_avx_256_double.h"
48 #include "kernelutil_x86_avx_256_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_avx_256_double
52 * Electrostatics interaction: Ewald
53 * VdW interaction: None
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_avx_256_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,C,D refer to j loop unrolling done with AVX, e.g. for the four 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;
74 int jnrA,jnrB,jnrC,jnrD;
75 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
80 real *shiftvec,*fshift,*x,*f;
81 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
83 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 real * vdwioffsetptr0;
85 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
87 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
88 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
89 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
92 __m256d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
93 __m256d beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
95 __m256d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
96 real rswitch_scalar,d_scalar;
97 __m256d dummy_mask,cutoff_mask;
98 __m128 tmpmask0,tmpmask1;
99 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
100 __m256d one = _mm256_set1_pd(1.0);
101 __m256d two = _mm256_set1_pd(2.0);
107 jindex = nlist->jindex;
109 shiftidx = nlist->shift;
111 shiftvec = fr->shift_vec[0];
112 fshift = fr->fshift[0];
113 facel = _mm256_set1_pd(fr->epsfac);
114 charge = mdatoms->chargeA;
116 sh_ewald = _mm256_set1_pd(fr->ic->sh_ewald);
117 beta = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
118 beta2 = _mm256_mul_pd(beta,beta);
119 beta3 = _mm256_mul_pd(beta,beta2);
121 ewtab = fr->ic->tabq_coul_FDV0;
122 ewtabscale = _mm256_set1_pd(fr->ic->tabq_scale);
123 ewtabhalfspace = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
125 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
126 rcutoff_scalar = fr->rcoulomb;
127 rcutoff = _mm256_set1_pd(rcutoff_scalar);
128 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
130 rswitch_scalar = fr->rcoulomb_switch;
131 rswitch = _mm256_set1_pd(rswitch_scalar);
132 /* Setup switch parameters */
133 d_scalar = rcutoff_scalar-rswitch_scalar;
134 d = _mm256_set1_pd(d_scalar);
135 swV3 = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
136 swV4 = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
137 swV5 = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
138 swF2 = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
139 swF3 = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
140 swF4 = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
142 /* Avoid stupid compiler warnings */
143 jnrA = jnrB = jnrC = jnrD = 0;
152 for(iidx=0;iidx<4*DIM;iidx++)
157 /* Start outer loop over neighborlists */
158 for(iidx=0; iidx<nri; iidx++)
160 /* Load shift vector for this list */
161 i_shift_offset = DIM*shiftidx[iidx];
163 /* Load limits for loop over neighbors */
164 j_index_start = jindex[iidx];
165 j_index_end = jindex[iidx+1];
167 /* Get outer coordinate index */
169 i_coord_offset = DIM*inr;
171 /* Load i particle coords and add shift vector */
172 gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
174 fix0 = _mm256_setzero_pd();
175 fiy0 = _mm256_setzero_pd();
176 fiz0 = _mm256_setzero_pd();
178 /* Load parameters for i particles */
179 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
181 /* Reset potential sums */
182 velecsum = _mm256_setzero_pd();
184 /* Start inner kernel loop */
185 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
188 /* Get j neighbor index, and coordinate index */
193 j_coord_offsetA = DIM*jnrA;
194 j_coord_offsetB = DIM*jnrB;
195 j_coord_offsetC = DIM*jnrC;
196 j_coord_offsetD = DIM*jnrD;
198 /* load j atom coordinates */
199 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
200 x+j_coord_offsetC,x+j_coord_offsetD,
203 /* Calculate displacement vector */
204 dx00 = _mm256_sub_pd(ix0,jx0);
205 dy00 = _mm256_sub_pd(iy0,jy0);
206 dz00 = _mm256_sub_pd(iz0,jz0);
208 /* Calculate squared distance and things based on it */
209 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
211 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
213 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
215 /* Load parameters for j particles */
216 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
217 charge+jnrC+0,charge+jnrD+0);
219 /**************************
220 * CALCULATE INTERACTIONS *
221 **************************/
223 if (gmx_mm256_any_lt(rsq00,rcutoff2))
226 r00 = _mm256_mul_pd(rsq00,rinv00);
228 /* Compute parameters for interactions between i and j atoms */
229 qq00 = _mm256_mul_pd(iq0,jq0);
231 /* EWALD ELECTROSTATICS */
233 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
234 ewrt = _mm256_mul_pd(r00,ewtabscale);
235 ewitab = _mm256_cvttpd_epi32(ewrt);
236 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
237 ewitab = _mm_slli_epi32(ewitab,2);
238 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
239 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
240 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
241 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
242 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
243 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
244 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
245 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
246 felec = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
248 d = _mm256_sub_pd(r00,rswitch);
249 d = _mm256_max_pd(d,_mm256_setzero_pd());
250 d2 = _mm256_mul_pd(d,d);
251 sw = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
253 dsw = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
255 /* Evaluate switch function */
256 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
257 felec = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
258 velec = _mm256_mul_pd(velec,sw);
259 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
261 /* Update potential sum for this i atom from the interaction with this j atom. */
262 velec = _mm256_and_pd(velec,cutoff_mask);
263 velecsum = _mm256_add_pd(velecsum,velec);
267 fscal = _mm256_and_pd(fscal,cutoff_mask);
269 /* Calculate temporary vectorial force */
270 tx = _mm256_mul_pd(fscal,dx00);
271 ty = _mm256_mul_pd(fscal,dy00);
272 tz = _mm256_mul_pd(fscal,dz00);
274 /* Update vectorial force */
275 fix0 = _mm256_add_pd(fix0,tx);
276 fiy0 = _mm256_add_pd(fiy0,ty);
277 fiz0 = _mm256_add_pd(fiz0,tz);
279 fjptrA = f+j_coord_offsetA;
280 fjptrB = f+j_coord_offsetB;
281 fjptrC = f+j_coord_offsetC;
282 fjptrD = f+j_coord_offsetD;
283 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
287 /* Inner loop uses 65 flops */
293 /* Get j neighbor index, and coordinate index */
294 jnrlistA = jjnr[jidx];
295 jnrlistB = jjnr[jidx+1];
296 jnrlistC = jjnr[jidx+2];
297 jnrlistD = jjnr[jidx+3];
298 /* Sign of each element will be negative for non-real atoms.
299 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
300 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
302 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
304 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
305 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
306 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
308 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
309 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
310 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
311 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
312 j_coord_offsetA = DIM*jnrA;
313 j_coord_offsetB = DIM*jnrB;
314 j_coord_offsetC = DIM*jnrC;
315 j_coord_offsetD = DIM*jnrD;
317 /* load j atom coordinates */
318 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
319 x+j_coord_offsetC,x+j_coord_offsetD,
322 /* Calculate displacement vector */
323 dx00 = _mm256_sub_pd(ix0,jx0);
324 dy00 = _mm256_sub_pd(iy0,jy0);
325 dz00 = _mm256_sub_pd(iz0,jz0);
327 /* Calculate squared distance and things based on it */
328 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
330 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
332 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
334 /* Load parameters for j particles */
335 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
336 charge+jnrC+0,charge+jnrD+0);
338 /**************************
339 * CALCULATE INTERACTIONS *
340 **************************/
342 if (gmx_mm256_any_lt(rsq00,rcutoff2))
345 r00 = _mm256_mul_pd(rsq00,rinv00);
346 r00 = _mm256_andnot_pd(dummy_mask,r00);
348 /* Compute parameters for interactions between i and j atoms */
349 qq00 = _mm256_mul_pd(iq0,jq0);
351 /* EWALD ELECTROSTATICS */
353 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
354 ewrt = _mm256_mul_pd(r00,ewtabscale);
355 ewitab = _mm256_cvttpd_epi32(ewrt);
356 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
357 ewitab = _mm_slli_epi32(ewitab,2);
358 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
359 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
360 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
361 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
362 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
363 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
364 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
365 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
366 felec = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
368 d = _mm256_sub_pd(r00,rswitch);
369 d = _mm256_max_pd(d,_mm256_setzero_pd());
370 d2 = _mm256_mul_pd(d,d);
371 sw = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
373 dsw = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
375 /* Evaluate switch function */
376 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
377 felec = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
378 velec = _mm256_mul_pd(velec,sw);
379 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
381 /* Update potential sum for this i atom from the interaction with this j atom. */
382 velec = _mm256_and_pd(velec,cutoff_mask);
383 velec = _mm256_andnot_pd(dummy_mask,velec);
384 velecsum = _mm256_add_pd(velecsum,velec);
388 fscal = _mm256_and_pd(fscal,cutoff_mask);
390 fscal = _mm256_andnot_pd(dummy_mask,fscal);
392 /* Calculate temporary vectorial force */
393 tx = _mm256_mul_pd(fscal,dx00);
394 ty = _mm256_mul_pd(fscal,dy00);
395 tz = _mm256_mul_pd(fscal,dz00);
397 /* Update vectorial force */
398 fix0 = _mm256_add_pd(fix0,tx);
399 fiy0 = _mm256_add_pd(fiy0,ty);
400 fiz0 = _mm256_add_pd(fiz0,tz);
402 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
403 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
404 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
405 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
406 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
410 /* Inner loop uses 66 flops */
413 /* End of innermost loop */
415 gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
416 f+i_coord_offset,fshift+i_shift_offset);
419 /* Update potential energies */
420 gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
422 /* Increment number of inner iterations */
423 inneriter += j_index_end - j_index_start;
425 /* Outer loop uses 8 flops */
428 /* Increment number of outer iterations */
431 /* Update outer/inner flops */
433 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*66);
436 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_256_double
437 * Electrostatics interaction: Ewald
438 * VdW interaction: None
439 * Geometry: Particle-Particle
440 * Calculate force/pot: Force
443 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_256_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,C,D refer to j loop unrolling done with AVX, e.g. for the four 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;
459 int jnrA,jnrB,jnrC,jnrD;
460 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
461 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
462 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
463 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
465 real *shiftvec,*fshift,*x,*f;
466 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
468 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
469 real * vdwioffsetptr0;
470 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
471 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
472 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
473 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
474 __m256d velec,felec,velecsum,facel,crf,krf,krf2;
477 __m256d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
478 __m256d beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
480 __m256d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
481 real rswitch_scalar,d_scalar;
482 __m256d dummy_mask,cutoff_mask;
483 __m128 tmpmask0,tmpmask1;
484 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
485 __m256d one = _mm256_set1_pd(1.0);
486 __m256d two = _mm256_set1_pd(2.0);
492 jindex = nlist->jindex;
494 shiftidx = nlist->shift;
496 shiftvec = fr->shift_vec[0];
497 fshift = fr->fshift[0];
498 facel = _mm256_set1_pd(fr->epsfac);
499 charge = mdatoms->chargeA;
501 sh_ewald = _mm256_set1_pd(fr->ic->sh_ewald);
502 beta = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
503 beta2 = _mm256_mul_pd(beta,beta);
504 beta3 = _mm256_mul_pd(beta,beta2);
506 ewtab = fr->ic->tabq_coul_FDV0;
507 ewtabscale = _mm256_set1_pd(fr->ic->tabq_scale);
508 ewtabhalfspace = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
510 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
511 rcutoff_scalar = fr->rcoulomb;
512 rcutoff = _mm256_set1_pd(rcutoff_scalar);
513 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
515 rswitch_scalar = fr->rcoulomb_switch;
516 rswitch = _mm256_set1_pd(rswitch_scalar);
517 /* Setup switch parameters */
518 d_scalar = rcutoff_scalar-rswitch_scalar;
519 d = _mm256_set1_pd(d_scalar);
520 swV3 = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
521 swV4 = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
522 swV5 = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
523 swF2 = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
524 swF3 = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
525 swF4 = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
527 /* Avoid stupid compiler warnings */
528 jnrA = jnrB = jnrC = jnrD = 0;
537 for(iidx=0;iidx<4*DIM;iidx++)
542 /* Start outer loop over neighborlists */
543 for(iidx=0; iidx<nri; iidx++)
545 /* Load shift vector for this list */
546 i_shift_offset = DIM*shiftidx[iidx];
548 /* Load limits for loop over neighbors */
549 j_index_start = jindex[iidx];
550 j_index_end = jindex[iidx+1];
552 /* Get outer coordinate index */
554 i_coord_offset = DIM*inr;
556 /* Load i particle coords and add shift vector */
557 gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
559 fix0 = _mm256_setzero_pd();
560 fiy0 = _mm256_setzero_pd();
561 fiz0 = _mm256_setzero_pd();
563 /* Load parameters for i particles */
564 iq0 = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
566 /* Start inner kernel loop */
567 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
570 /* Get j neighbor index, and coordinate index */
575 j_coord_offsetA = DIM*jnrA;
576 j_coord_offsetB = DIM*jnrB;
577 j_coord_offsetC = DIM*jnrC;
578 j_coord_offsetD = DIM*jnrD;
580 /* load j atom coordinates */
581 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
582 x+j_coord_offsetC,x+j_coord_offsetD,
585 /* Calculate displacement vector */
586 dx00 = _mm256_sub_pd(ix0,jx0);
587 dy00 = _mm256_sub_pd(iy0,jy0);
588 dz00 = _mm256_sub_pd(iz0,jz0);
590 /* Calculate squared distance and things based on it */
591 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
593 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
595 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
597 /* Load parameters for j particles */
598 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
599 charge+jnrC+0,charge+jnrD+0);
601 /**************************
602 * CALCULATE INTERACTIONS *
603 **************************/
605 if (gmx_mm256_any_lt(rsq00,rcutoff2))
608 r00 = _mm256_mul_pd(rsq00,rinv00);
610 /* Compute parameters for interactions between i and j atoms */
611 qq00 = _mm256_mul_pd(iq0,jq0);
613 /* EWALD ELECTROSTATICS */
615 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
616 ewrt = _mm256_mul_pd(r00,ewtabscale);
617 ewitab = _mm256_cvttpd_epi32(ewrt);
618 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
619 ewitab = _mm_slli_epi32(ewitab,2);
620 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
621 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
622 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
623 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
624 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
625 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
626 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
627 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
628 felec = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
630 d = _mm256_sub_pd(r00,rswitch);
631 d = _mm256_max_pd(d,_mm256_setzero_pd());
632 d2 = _mm256_mul_pd(d,d);
633 sw = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
635 dsw = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
637 /* Evaluate switch function */
638 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
639 felec = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
640 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
644 fscal = _mm256_and_pd(fscal,cutoff_mask);
646 /* Calculate temporary vectorial force */
647 tx = _mm256_mul_pd(fscal,dx00);
648 ty = _mm256_mul_pd(fscal,dy00);
649 tz = _mm256_mul_pd(fscal,dz00);
651 /* Update vectorial force */
652 fix0 = _mm256_add_pd(fix0,tx);
653 fiy0 = _mm256_add_pd(fiy0,ty);
654 fiz0 = _mm256_add_pd(fiz0,tz);
656 fjptrA = f+j_coord_offsetA;
657 fjptrB = f+j_coord_offsetB;
658 fjptrC = f+j_coord_offsetC;
659 fjptrD = f+j_coord_offsetD;
660 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
664 /* Inner loop uses 62 flops */
670 /* Get j neighbor index, and coordinate index */
671 jnrlistA = jjnr[jidx];
672 jnrlistB = jjnr[jidx+1];
673 jnrlistC = jjnr[jidx+2];
674 jnrlistD = jjnr[jidx+3];
675 /* Sign of each element will be negative for non-real atoms.
676 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
677 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
679 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
681 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
682 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
683 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
685 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
686 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
687 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
688 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
689 j_coord_offsetA = DIM*jnrA;
690 j_coord_offsetB = DIM*jnrB;
691 j_coord_offsetC = DIM*jnrC;
692 j_coord_offsetD = DIM*jnrD;
694 /* load j atom coordinates */
695 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
696 x+j_coord_offsetC,x+j_coord_offsetD,
699 /* Calculate displacement vector */
700 dx00 = _mm256_sub_pd(ix0,jx0);
701 dy00 = _mm256_sub_pd(iy0,jy0);
702 dz00 = _mm256_sub_pd(iz0,jz0);
704 /* Calculate squared distance and things based on it */
705 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
707 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
709 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
711 /* Load parameters for j particles */
712 jq0 = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
713 charge+jnrC+0,charge+jnrD+0);
715 /**************************
716 * CALCULATE INTERACTIONS *
717 **************************/
719 if (gmx_mm256_any_lt(rsq00,rcutoff2))
722 r00 = _mm256_mul_pd(rsq00,rinv00);
723 r00 = _mm256_andnot_pd(dummy_mask,r00);
725 /* Compute parameters for interactions between i and j atoms */
726 qq00 = _mm256_mul_pd(iq0,jq0);
728 /* EWALD ELECTROSTATICS */
730 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
731 ewrt = _mm256_mul_pd(r00,ewtabscale);
732 ewitab = _mm256_cvttpd_epi32(ewrt);
733 eweps = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
734 ewitab = _mm_slli_epi32(ewitab,2);
735 ewtabF = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
736 ewtabD = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
737 ewtabV = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
738 ewtabFn = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
739 GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
740 felec = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
741 velec = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
742 velec = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
743 felec = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
745 d = _mm256_sub_pd(r00,rswitch);
746 d = _mm256_max_pd(d,_mm256_setzero_pd());
747 d2 = _mm256_mul_pd(d,d);
748 sw = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
750 dsw = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
752 /* Evaluate switch function */
753 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
754 felec = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(velec,dsw)) );
755 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
759 fscal = _mm256_and_pd(fscal,cutoff_mask);
761 fscal = _mm256_andnot_pd(dummy_mask,fscal);
763 /* Calculate temporary vectorial force */
764 tx = _mm256_mul_pd(fscal,dx00);
765 ty = _mm256_mul_pd(fscal,dy00);
766 tz = _mm256_mul_pd(fscal,dz00);
768 /* Update vectorial force */
769 fix0 = _mm256_add_pd(fix0,tx);
770 fiy0 = _mm256_add_pd(fiy0,ty);
771 fiz0 = _mm256_add_pd(fiz0,tz);
773 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
774 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
775 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
776 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
777 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
781 /* Inner loop uses 63 flops */
784 /* End of innermost loop */
786 gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
787 f+i_coord_offset,fshift+i_shift_offset);
789 /* Increment number of inner iterations */
790 inneriter += j_index_end - j_index_start;
792 /* Outer loop uses 7 flops */
795 /* Increment number of outer iterations */
798 /* Update outer/inner flops */
800 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*63);