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.
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_256_double.h"
50 #include "kernelutil_x86_avx_256_double.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_avx_256_double
54 * Electrostatics interaction: None
55 * VdW interaction: LennardJones
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_avx_256_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,C,D refer to j loop unrolling done with AVX, e.g. for the four 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;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
79 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
80 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
82 real *shiftvec,*fshift,*x,*f;
83 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
85 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 real * vdwioffsetptr0;
87 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
89 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
92 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
95 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
96 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
97 __m256d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
98 real rswitch_scalar,d_scalar;
99 __m256d dummy_mask,cutoff_mask;
100 __m128 tmpmask0,tmpmask1;
101 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
102 __m256d one = _mm256_set1_pd(1.0);
103 __m256d two = _mm256_set1_pd(2.0);
109 jindex = nlist->jindex;
111 shiftidx = nlist->shift;
113 shiftvec = fr->shift_vec[0];
114 fshift = fr->fshift[0];
115 nvdwtype = fr->ntype;
117 vdwtype = mdatoms->typeA;
119 rcutoff_scalar = fr->rvdw;
120 rcutoff = _mm256_set1_pd(rcutoff_scalar);
121 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
123 rswitch_scalar = fr->rvdw_switch;
124 rswitch = _mm256_set1_pd(rswitch_scalar);
125 /* Setup switch parameters */
126 d_scalar = rcutoff_scalar-rswitch_scalar;
127 d = _mm256_set1_pd(d_scalar);
128 swV3 = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
129 swV4 = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
130 swV5 = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
131 swF2 = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
132 swF3 = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
133 swF4 = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
135 /* Avoid stupid compiler warnings */
136 jnrA = jnrB = jnrC = jnrD = 0;
145 for(iidx=0;iidx<4*DIM;iidx++)
150 /* Start outer loop over neighborlists */
151 for(iidx=0; iidx<nri; iidx++)
153 /* Load shift vector for this list */
154 i_shift_offset = DIM*shiftidx[iidx];
156 /* Load limits for loop over neighbors */
157 j_index_start = jindex[iidx];
158 j_index_end = jindex[iidx+1];
160 /* Get outer coordinate index */
162 i_coord_offset = DIM*inr;
164 /* Load i particle coords and add shift vector */
165 gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
167 fix0 = _mm256_setzero_pd();
168 fiy0 = _mm256_setzero_pd();
169 fiz0 = _mm256_setzero_pd();
171 /* Load parameters for i particles */
172 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
174 /* Reset potential sums */
175 vvdwsum = _mm256_setzero_pd();
177 /* Start inner kernel loop */
178 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
181 /* Get j neighbor index, and coordinate index */
186 j_coord_offsetA = DIM*jnrA;
187 j_coord_offsetB = DIM*jnrB;
188 j_coord_offsetC = DIM*jnrC;
189 j_coord_offsetD = DIM*jnrD;
191 /* load j atom coordinates */
192 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
193 x+j_coord_offsetC,x+j_coord_offsetD,
196 /* Calculate displacement vector */
197 dx00 = _mm256_sub_pd(ix0,jx0);
198 dy00 = _mm256_sub_pd(iy0,jy0);
199 dz00 = _mm256_sub_pd(iz0,jz0);
201 /* Calculate squared distance and things based on it */
202 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
204 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
206 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
208 /* Load parameters for j particles */
209 vdwjidx0A = 2*vdwtype[jnrA+0];
210 vdwjidx0B = 2*vdwtype[jnrB+0];
211 vdwjidx0C = 2*vdwtype[jnrC+0];
212 vdwjidx0D = 2*vdwtype[jnrD+0];
214 /**************************
215 * CALCULATE INTERACTIONS *
216 **************************/
218 if (gmx_mm256_any_lt(rsq00,rcutoff2))
221 r00 = _mm256_mul_pd(rsq00,rinv00);
223 /* Compute parameters for interactions between i and j atoms */
224 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
225 vdwioffsetptr0+vdwjidx0B,
226 vdwioffsetptr0+vdwjidx0C,
227 vdwioffsetptr0+vdwjidx0D,
230 /* LENNARD-JONES DISPERSION/REPULSION */
232 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
233 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
234 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
235 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
236 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
238 d = _mm256_sub_pd(r00,rswitch);
239 d = _mm256_max_pd(d,_mm256_setzero_pd());
240 d2 = _mm256_mul_pd(d,d);
241 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)))))));
243 dsw = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
245 /* Evaluate switch function */
246 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
247 fvdw = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
248 vvdw = _mm256_mul_pd(vvdw,sw);
249 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
251 /* Update potential sum for this i atom from the interaction with this j atom. */
252 vvdw = _mm256_and_pd(vvdw,cutoff_mask);
253 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
257 fscal = _mm256_and_pd(fscal,cutoff_mask);
259 /* Calculate temporary vectorial force */
260 tx = _mm256_mul_pd(fscal,dx00);
261 ty = _mm256_mul_pd(fscal,dy00);
262 tz = _mm256_mul_pd(fscal,dz00);
264 /* Update vectorial force */
265 fix0 = _mm256_add_pd(fix0,tx);
266 fiy0 = _mm256_add_pd(fiy0,ty);
267 fiz0 = _mm256_add_pd(fiz0,tz);
269 fjptrA = f+j_coord_offsetA;
270 fjptrB = f+j_coord_offsetB;
271 fjptrC = f+j_coord_offsetC;
272 fjptrD = f+j_coord_offsetD;
273 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
277 /* Inner loop uses 59 flops */
283 /* Get j neighbor index, and coordinate index */
284 jnrlistA = jjnr[jidx];
285 jnrlistB = jjnr[jidx+1];
286 jnrlistC = jjnr[jidx+2];
287 jnrlistD = jjnr[jidx+3];
288 /* Sign of each element will be negative for non-real atoms.
289 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
290 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
292 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
294 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
295 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
296 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
298 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
299 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
300 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
301 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
302 j_coord_offsetA = DIM*jnrA;
303 j_coord_offsetB = DIM*jnrB;
304 j_coord_offsetC = DIM*jnrC;
305 j_coord_offsetD = DIM*jnrD;
307 /* load j atom coordinates */
308 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
309 x+j_coord_offsetC,x+j_coord_offsetD,
312 /* Calculate displacement vector */
313 dx00 = _mm256_sub_pd(ix0,jx0);
314 dy00 = _mm256_sub_pd(iy0,jy0);
315 dz00 = _mm256_sub_pd(iz0,jz0);
317 /* Calculate squared distance and things based on it */
318 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
320 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
322 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
324 /* Load parameters for j particles */
325 vdwjidx0A = 2*vdwtype[jnrA+0];
326 vdwjidx0B = 2*vdwtype[jnrB+0];
327 vdwjidx0C = 2*vdwtype[jnrC+0];
328 vdwjidx0D = 2*vdwtype[jnrD+0];
330 /**************************
331 * CALCULATE INTERACTIONS *
332 **************************/
334 if (gmx_mm256_any_lt(rsq00,rcutoff2))
337 r00 = _mm256_mul_pd(rsq00,rinv00);
338 r00 = _mm256_andnot_pd(dummy_mask,r00);
340 /* Compute parameters for interactions between i and j atoms */
341 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
342 vdwioffsetptr0+vdwjidx0B,
343 vdwioffsetptr0+vdwjidx0C,
344 vdwioffsetptr0+vdwjidx0D,
347 /* LENNARD-JONES DISPERSION/REPULSION */
349 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
350 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
351 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
352 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
353 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
355 d = _mm256_sub_pd(r00,rswitch);
356 d = _mm256_max_pd(d,_mm256_setzero_pd());
357 d2 = _mm256_mul_pd(d,d);
358 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)))))));
360 dsw = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
362 /* Evaluate switch function */
363 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
364 fvdw = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
365 vvdw = _mm256_mul_pd(vvdw,sw);
366 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
368 /* Update potential sum for this i atom from the interaction with this j atom. */
369 vvdw = _mm256_and_pd(vvdw,cutoff_mask);
370 vvdw = _mm256_andnot_pd(dummy_mask,vvdw);
371 vvdwsum = _mm256_add_pd(vvdwsum,vvdw);
375 fscal = _mm256_and_pd(fscal,cutoff_mask);
377 fscal = _mm256_andnot_pd(dummy_mask,fscal);
379 /* Calculate temporary vectorial force */
380 tx = _mm256_mul_pd(fscal,dx00);
381 ty = _mm256_mul_pd(fscal,dy00);
382 tz = _mm256_mul_pd(fscal,dz00);
384 /* Update vectorial force */
385 fix0 = _mm256_add_pd(fix0,tx);
386 fiy0 = _mm256_add_pd(fiy0,ty);
387 fiz0 = _mm256_add_pd(fiz0,tz);
389 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
390 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
391 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
392 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
393 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
397 /* Inner loop uses 60 flops */
400 /* End of innermost loop */
402 gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
403 f+i_coord_offset,fshift+i_shift_offset);
406 /* Update potential energies */
407 gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
409 /* Increment number of inner iterations */
410 inneriter += j_index_end - j_index_start;
412 /* Outer loop uses 7 flops */
415 /* Increment number of outer iterations */
418 /* Update outer/inner flops */
420 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*60);
423 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_256_double
424 * Electrostatics interaction: None
425 * VdW interaction: LennardJones
426 * Geometry: Particle-Particle
427 * Calculate force/pot: Force
430 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_256_double
431 (t_nblist * gmx_restrict nlist,
432 rvec * gmx_restrict xx,
433 rvec * gmx_restrict ff,
434 t_forcerec * gmx_restrict fr,
435 t_mdatoms * gmx_restrict mdatoms,
436 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
437 t_nrnb * gmx_restrict nrnb)
439 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
440 * just 0 for non-waters.
441 * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
442 * jnr indices corresponding to data put in the four positions in the SIMD register.
444 int i_shift_offset,i_coord_offset,outeriter,inneriter;
445 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
446 int jnrA,jnrB,jnrC,jnrD;
447 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
448 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
449 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
450 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
452 real *shiftvec,*fshift,*x,*f;
453 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
455 __m256d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
456 real * vdwioffsetptr0;
457 __m256d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
458 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
459 __m256d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
460 __m256d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
462 __m256d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
465 __m256d one_sixth = _mm256_set1_pd(1.0/6.0);
466 __m256d one_twelfth = _mm256_set1_pd(1.0/12.0);
467 __m256d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
468 real rswitch_scalar,d_scalar;
469 __m256d dummy_mask,cutoff_mask;
470 __m128 tmpmask0,tmpmask1;
471 __m256d signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
472 __m256d one = _mm256_set1_pd(1.0);
473 __m256d two = _mm256_set1_pd(2.0);
479 jindex = nlist->jindex;
481 shiftidx = nlist->shift;
483 shiftvec = fr->shift_vec[0];
484 fshift = fr->fshift[0];
485 nvdwtype = fr->ntype;
487 vdwtype = mdatoms->typeA;
489 rcutoff_scalar = fr->rvdw;
490 rcutoff = _mm256_set1_pd(rcutoff_scalar);
491 rcutoff2 = _mm256_mul_pd(rcutoff,rcutoff);
493 rswitch_scalar = fr->rvdw_switch;
494 rswitch = _mm256_set1_pd(rswitch_scalar);
495 /* Setup switch parameters */
496 d_scalar = rcutoff_scalar-rswitch_scalar;
497 d = _mm256_set1_pd(d_scalar);
498 swV3 = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
499 swV4 = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
500 swV5 = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
501 swF2 = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
502 swF3 = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
503 swF4 = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
505 /* Avoid stupid compiler warnings */
506 jnrA = jnrB = jnrC = jnrD = 0;
515 for(iidx=0;iidx<4*DIM;iidx++)
520 /* Start outer loop over neighborlists */
521 for(iidx=0; iidx<nri; iidx++)
523 /* Load shift vector for this list */
524 i_shift_offset = DIM*shiftidx[iidx];
526 /* Load limits for loop over neighbors */
527 j_index_start = jindex[iidx];
528 j_index_end = jindex[iidx+1];
530 /* Get outer coordinate index */
532 i_coord_offset = DIM*inr;
534 /* Load i particle coords and add shift vector */
535 gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
537 fix0 = _mm256_setzero_pd();
538 fiy0 = _mm256_setzero_pd();
539 fiz0 = _mm256_setzero_pd();
541 /* Load parameters for i particles */
542 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
544 /* Start inner kernel loop */
545 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
548 /* Get j neighbor index, and coordinate index */
553 j_coord_offsetA = DIM*jnrA;
554 j_coord_offsetB = DIM*jnrB;
555 j_coord_offsetC = DIM*jnrC;
556 j_coord_offsetD = DIM*jnrD;
558 /* load j atom coordinates */
559 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
560 x+j_coord_offsetC,x+j_coord_offsetD,
563 /* Calculate displacement vector */
564 dx00 = _mm256_sub_pd(ix0,jx0);
565 dy00 = _mm256_sub_pd(iy0,jy0);
566 dz00 = _mm256_sub_pd(iz0,jz0);
568 /* Calculate squared distance and things based on it */
569 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
571 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
573 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
575 /* Load parameters for j particles */
576 vdwjidx0A = 2*vdwtype[jnrA+0];
577 vdwjidx0B = 2*vdwtype[jnrB+0];
578 vdwjidx0C = 2*vdwtype[jnrC+0];
579 vdwjidx0D = 2*vdwtype[jnrD+0];
581 /**************************
582 * CALCULATE INTERACTIONS *
583 **************************/
585 if (gmx_mm256_any_lt(rsq00,rcutoff2))
588 r00 = _mm256_mul_pd(rsq00,rinv00);
590 /* Compute parameters for interactions between i and j atoms */
591 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
592 vdwioffsetptr0+vdwjidx0B,
593 vdwioffsetptr0+vdwjidx0C,
594 vdwioffsetptr0+vdwjidx0D,
597 /* LENNARD-JONES DISPERSION/REPULSION */
599 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
600 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
601 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
602 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
603 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
605 d = _mm256_sub_pd(r00,rswitch);
606 d = _mm256_max_pd(d,_mm256_setzero_pd());
607 d2 = _mm256_mul_pd(d,d);
608 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)))))));
610 dsw = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
612 /* Evaluate switch function */
613 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
614 fvdw = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
615 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
619 fscal = _mm256_and_pd(fscal,cutoff_mask);
621 /* Calculate temporary vectorial force */
622 tx = _mm256_mul_pd(fscal,dx00);
623 ty = _mm256_mul_pd(fscal,dy00);
624 tz = _mm256_mul_pd(fscal,dz00);
626 /* Update vectorial force */
627 fix0 = _mm256_add_pd(fix0,tx);
628 fiy0 = _mm256_add_pd(fiy0,ty);
629 fiz0 = _mm256_add_pd(fiz0,tz);
631 fjptrA = f+j_coord_offsetA;
632 fjptrB = f+j_coord_offsetB;
633 fjptrC = f+j_coord_offsetC;
634 fjptrD = f+j_coord_offsetD;
635 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
639 /* Inner loop uses 56 flops */
645 /* Get j neighbor index, and coordinate index */
646 jnrlistA = jjnr[jidx];
647 jnrlistB = jjnr[jidx+1];
648 jnrlistC = jjnr[jidx+2];
649 jnrlistD = jjnr[jidx+3];
650 /* Sign of each element will be negative for non-real atoms.
651 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
652 * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
654 tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
656 tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
657 tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
658 dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
660 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
661 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
662 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
663 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
664 j_coord_offsetA = DIM*jnrA;
665 j_coord_offsetB = DIM*jnrB;
666 j_coord_offsetC = DIM*jnrC;
667 j_coord_offsetD = DIM*jnrD;
669 /* load j atom coordinates */
670 gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
671 x+j_coord_offsetC,x+j_coord_offsetD,
674 /* Calculate displacement vector */
675 dx00 = _mm256_sub_pd(ix0,jx0);
676 dy00 = _mm256_sub_pd(iy0,jy0);
677 dz00 = _mm256_sub_pd(iz0,jz0);
679 /* Calculate squared distance and things based on it */
680 rsq00 = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
682 rinv00 = gmx_mm256_invsqrt_pd(rsq00);
684 rinvsq00 = _mm256_mul_pd(rinv00,rinv00);
686 /* Load parameters for j particles */
687 vdwjidx0A = 2*vdwtype[jnrA+0];
688 vdwjidx0B = 2*vdwtype[jnrB+0];
689 vdwjidx0C = 2*vdwtype[jnrC+0];
690 vdwjidx0D = 2*vdwtype[jnrD+0];
692 /**************************
693 * CALCULATE INTERACTIONS *
694 **************************/
696 if (gmx_mm256_any_lt(rsq00,rcutoff2))
699 r00 = _mm256_mul_pd(rsq00,rinv00);
700 r00 = _mm256_andnot_pd(dummy_mask,r00);
702 /* Compute parameters for interactions between i and j atoms */
703 gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
704 vdwioffsetptr0+vdwjidx0B,
705 vdwioffsetptr0+vdwjidx0C,
706 vdwioffsetptr0+vdwjidx0D,
709 /* LENNARD-JONES DISPERSION/REPULSION */
711 rinvsix = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
712 vvdw6 = _mm256_mul_pd(c6_00,rinvsix);
713 vvdw12 = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
714 vvdw = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
715 fvdw = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
717 d = _mm256_sub_pd(r00,rswitch);
718 d = _mm256_max_pd(d,_mm256_setzero_pd());
719 d2 = _mm256_mul_pd(d,d);
720 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)))))));
722 dsw = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
724 /* Evaluate switch function */
725 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
726 fvdw = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
727 cutoff_mask = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
731 fscal = _mm256_and_pd(fscal,cutoff_mask);
733 fscal = _mm256_andnot_pd(dummy_mask,fscal);
735 /* Calculate temporary vectorial force */
736 tx = _mm256_mul_pd(fscal,dx00);
737 ty = _mm256_mul_pd(fscal,dy00);
738 tz = _mm256_mul_pd(fscal,dz00);
740 /* Update vectorial force */
741 fix0 = _mm256_add_pd(fix0,tx);
742 fiy0 = _mm256_add_pd(fiy0,ty);
743 fiz0 = _mm256_add_pd(fiz0,tz);
745 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
746 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
747 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
748 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
749 gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
753 /* Inner loop uses 57 flops */
756 /* End of innermost loop */
758 gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
759 f+i_coord_offset,fshift+i_shift_offset);
761 /* Increment number of inner iterations */
762 inneriter += j_index_end - j_index_start;
764 /* Outer loop uses 6 flops */
767 /* Increment number of outer iterations */
770 /* Update outer/inner flops */
772 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*57);