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_single 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_256_single.h"
48 #include "kernelutil_x86_avx_256_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_avx_256_single
52 * Electrostatics interaction: None
53 * VdW interaction: LennardJones
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_avx_256_single
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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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 jnrE,jnrF,jnrG,jnrH;
76 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
77 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
80 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
82 real *shiftvec,*fshift,*x,*f;
83 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
85 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
86 real * vdwioffsetptr0;
87 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
88 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
89 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
90 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
92 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
95 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
96 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
97 __m256 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
98 real rswitch_scalar,d_scalar;
99 __m256 dummy_mask,cutoff_mask;
100 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
101 __m256 one = _mm256_set1_ps(1.0);
102 __m256 two = _mm256_set1_ps(2.0);
108 jindex = nlist->jindex;
110 shiftidx = nlist->shift;
112 shiftvec = fr->shift_vec[0];
113 fshift = fr->fshift[0];
114 nvdwtype = fr->ntype;
116 vdwtype = mdatoms->typeA;
118 rcutoff_scalar = fr->rvdw;
119 rcutoff = _mm256_set1_ps(rcutoff_scalar);
120 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
122 rswitch_scalar = fr->rvdw_switch;
123 rswitch = _mm256_set1_ps(rswitch_scalar);
124 /* Setup switch parameters */
125 d_scalar = rcutoff_scalar-rswitch_scalar;
126 d = _mm256_set1_ps(d_scalar);
127 swV3 = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
128 swV4 = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
129 swV5 = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
130 swF2 = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
131 swF3 = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
132 swF4 = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
134 /* Avoid stupid compiler warnings */
135 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
148 for(iidx=0;iidx<4*DIM;iidx++)
153 /* Start outer loop over neighborlists */
154 for(iidx=0; iidx<nri; iidx++)
156 /* Load shift vector for this list */
157 i_shift_offset = DIM*shiftidx[iidx];
159 /* Load limits for loop over neighbors */
160 j_index_start = jindex[iidx];
161 j_index_end = jindex[iidx+1];
163 /* Get outer coordinate index */
165 i_coord_offset = DIM*inr;
167 /* Load i particle coords and add shift vector */
168 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
170 fix0 = _mm256_setzero_ps();
171 fiy0 = _mm256_setzero_ps();
172 fiz0 = _mm256_setzero_ps();
174 /* Load parameters for i particles */
175 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
177 /* Reset potential sums */
178 vvdwsum = _mm256_setzero_ps();
180 /* Start inner kernel loop */
181 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
184 /* 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;
197 j_coord_offsetE = DIM*jnrE;
198 j_coord_offsetF = DIM*jnrF;
199 j_coord_offsetG = DIM*jnrG;
200 j_coord_offsetH = DIM*jnrH;
202 /* load j atom coordinates */
203 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
204 x+j_coord_offsetC,x+j_coord_offsetD,
205 x+j_coord_offsetE,x+j_coord_offsetF,
206 x+j_coord_offsetG,x+j_coord_offsetH,
209 /* Calculate displacement vector */
210 dx00 = _mm256_sub_ps(ix0,jx0);
211 dy00 = _mm256_sub_ps(iy0,jy0);
212 dz00 = _mm256_sub_ps(iz0,jz0);
214 /* Calculate squared distance and things based on it */
215 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
217 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
219 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
221 /* Load parameters for j particles */
222 vdwjidx0A = 2*vdwtype[jnrA+0];
223 vdwjidx0B = 2*vdwtype[jnrB+0];
224 vdwjidx0C = 2*vdwtype[jnrC+0];
225 vdwjidx0D = 2*vdwtype[jnrD+0];
226 vdwjidx0E = 2*vdwtype[jnrE+0];
227 vdwjidx0F = 2*vdwtype[jnrF+0];
228 vdwjidx0G = 2*vdwtype[jnrG+0];
229 vdwjidx0H = 2*vdwtype[jnrH+0];
231 /**************************
232 * CALCULATE INTERACTIONS *
233 **************************/
235 if (gmx_mm256_any_lt(rsq00,rcutoff2))
238 r00 = _mm256_mul_ps(rsq00,rinv00);
240 /* Compute parameters for interactions between i and j atoms */
241 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
242 vdwioffsetptr0+vdwjidx0B,
243 vdwioffsetptr0+vdwjidx0C,
244 vdwioffsetptr0+vdwjidx0D,
245 vdwioffsetptr0+vdwjidx0E,
246 vdwioffsetptr0+vdwjidx0F,
247 vdwioffsetptr0+vdwjidx0G,
248 vdwioffsetptr0+vdwjidx0H,
251 /* LENNARD-JONES DISPERSION/REPULSION */
253 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
254 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
255 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
256 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
257 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
259 d = _mm256_sub_ps(r00,rswitch);
260 d = _mm256_max_ps(d,_mm256_setzero_ps());
261 d2 = _mm256_mul_ps(d,d);
262 sw = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
264 dsw = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
266 /* Evaluate switch function */
267 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
268 fvdw = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
269 vvdw = _mm256_mul_ps(vvdw,sw);
270 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
272 /* Update potential sum for this i atom from the interaction with this j atom. */
273 vvdw = _mm256_and_ps(vvdw,cutoff_mask);
274 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
278 fscal = _mm256_and_ps(fscal,cutoff_mask);
280 /* Calculate temporary vectorial force */
281 tx = _mm256_mul_ps(fscal,dx00);
282 ty = _mm256_mul_ps(fscal,dy00);
283 tz = _mm256_mul_ps(fscal,dz00);
285 /* Update vectorial force */
286 fix0 = _mm256_add_ps(fix0,tx);
287 fiy0 = _mm256_add_ps(fiy0,ty);
288 fiz0 = _mm256_add_ps(fiz0,tz);
290 fjptrA = f+j_coord_offsetA;
291 fjptrB = f+j_coord_offsetB;
292 fjptrC = f+j_coord_offsetC;
293 fjptrD = f+j_coord_offsetD;
294 fjptrE = f+j_coord_offsetE;
295 fjptrF = f+j_coord_offsetF;
296 fjptrG = f+j_coord_offsetG;
297 fjptrH = f+j_coord_offsetH;
298 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
302 /* Inner loop uses 59 flops */
308 /* Get j neighbor index, and coordinate index */
309 jnrlistA = jjnr[jidx];
310 jnrlistB = jjnr[jidx+1];
311 jnrlistC = jjnr[jidx+2];
312 jnrlistD = jjnr[jidx+3];
313 jnrlistE = jjnr[jidx+4];
314 jnrlistF = jjnr[jidx+5];
315 jnrlistG = jjnr[jidx+6];
316 jnrlistH = jjnr[jidx+7];
317 /* Sign of each element will be negative for non-real atoms.
318 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
319 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
321 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
322 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
324 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
325 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
326 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
327 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
328 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
329 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
330 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
331 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
332 j_coord_offsetA = DIM*jnrA;
333 j_coord_offsetB = DIM*jnrB;
334 j_coord_offsetC = DIM*jnrC;
335 j_coord_offsetD = DIM*jnrD;
336 j_coord_offsetE = DIM*jnrE;
337 j_coord_offsetF = DIM*jnrF;
338 j_coord_offsetG = DIM*jnrG;
339 j_coord_offsetH = DIM*jnrH;
341 /* load j atom coordinates */
342 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
343 x+j_coord_offsetC,x+j_coord_offsetD,
344 x+j_coord_offsetE,x+j_coord_offsetF,
345 x+j_coord_offsetG,x+j_coord_offsetH,
348 /* Calculate displacement vector */
349 dx00 = _mm256_sub_ps(ix0,jx0);
350 dy00 = _mm256_sub_ps(iy0,jy0);
351 dz00 = _mm256_sub_ps(iz0,jz0);
353 /* Calculate squared distance and things based on it */
354 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
356 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
358 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
360 /* Load parameters for j particles */
361 vdwjidx0A = 2*vdwtype[jnrA+0];
362 vdwjidx0B = 2*vdwtype[jnrB+0];
363 vdwjidx0C = 2*vdwtype[jnrC+0];
364 vdwjidx0D = 2*vdwtype[jnrD+0];
365 vdwjidx0E = 2*vdwtype[jnrE+0];
366 vdwjidx0F = 2*vdwtype[jnrF+0];
367 vdwjidx0G = 2*vdwtype[jnrG+0];
368 vdwjidx0H = 2*vdwtype[jnrH+0];
370 /**************************
371 * CALCULATE INTERACTIONS *
372 **************************/
374 if (gmx_mm256_any_lt(rsq00,rcutoff2))
377 r00 = _mm256_mul_ps(rsq00,rinv00);
378 r00 = _mm256_andnot_ps(dummy_mask,r00);
380 /* Compute parameters for interactions between i and j atoms */
381 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
382 vdwioffsetptr0+vdwjidx0B,
383 vdwioffsetptr0+vdwjidx0C,
384 vdwioffsetptr0+vdwjidx0D,
385 vdwioffsetptr0+vdwjidx0E,
386 vdwioffsetptr0+vdwjidx0F,
387 vdwioffsetptr0+vdwjidx0G,
388 vdwioffsetptr0+vdwjidx0H,
391 /* LENNARD-JONES DISPERSION/REPULSION */
393 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
394 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
395 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
396 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
397 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
399 d = _mm256_sub_ps(r00,rswitch);
400 d = _mm256_max_ps(d,_mm256_setzero_ps());
401 d2 = _mm256_mul_ps(d,d);
402 sw = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
404 dsw = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
406 /* Evaluate switch function */
407 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
408 fvdw = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
409 vvdw = _mm256_mul_ps(vvdw,sw);
410 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
412 /* Update potential sum for this i atom from the interaction with this j atom. */
413 vvdw = _mm256_and_ps(vvdw,cutoff_mask);
414 vvdw = _mm256_andnot_ps(dummy_mask,vvdw);
415 vvdwsum = _mm256_add_ps(vvdwsum,vvdw);
419 fscal = _mm256_and_ps(fscal,cutoff_mask);
421 fscal = _mm256_andnot_ps(dummy_mask,fscal);
423 /* Calculate temporary vectorial force */
424 tx = _mm256_mul_ps(fscal,dx00);
425 ty = _mm256_mul_ps(fscal,dy00);
426 tz = _mm256_mul_ps(fscal,dz00);
428 /* Update vectorial force */
429 fix0 = _mm256_add_ps(fix0,tx);
430 fiy0 = _mm256_add_ps(fiy0,ty);
431 fiz0 = _mm256_add_ps(fiz0,tz);
433 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
434 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
435 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
436 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
437 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
438 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
439 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
440 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
441 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
445 /* Inner loop uses 60 flops */
448 /* End of innermost loop */
450 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
451 f+i_coord_offset,fshift+i_shift_offset);
454 /* Update potential energies */
455 gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
457 /* Increment number of inner iterations */
458 inneriter += j_index_end - j_index_start;
460 /* Outer loop uses 7 flops */
463 /* Increment number of outer iterations */
466 /* Update outer/inner flops */
468 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*60);
471 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_256_single
472 * Electrostatics interaction: None
473 * VdW interaction: LennardJones
474 * Geometry: Particle-Particle
475 * Calculate force/pot: Force
478 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_256_single
479 (t_nblist * gmx_restrict nlist,
480 rvec * gmx_restrict xx,
481 rvec * gmx_restrict ff,
482 t_forcerec * gmx_restrict fr,
483 t_mdatoms * gmx_restrict mdatoms,
484 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
485 t_nrnb * gmx_restrict nrnb)
487 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
488 * just 0 for non-waters.
489 * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
490 * jnr indices corresponding to data put in the four positions in the SIMD register.
492 int i_shift_offset,i_coord_offset,outeriter,inneriter;
493 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
494 int jnrA,jnrB,jnrC,jnrD;
495 int jnrE,jnrF,jnrG,jnrH;
496 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
497 int jnrlistE,jnrlistF,jnrlistG,jnrlistH;
498 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
499 int j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
500 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
502 real *shiftvec,*fshift,*x,*f;
503 real *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
505 __m256 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
506 real * vdwioffsetptr0;
507 __m256 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
508 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
509 __m256 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
510 __m256 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
512 __m256 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
515 __m256 one_sixth = _mm256_set1_ps(1.0/6.0);
516 __m256 one_twelfth = _mm256_set1_ps(1.0/12.0);
517 __m256 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
518 real rswitch_scalar,d_scalar;
519 __m256 dummy_mask,cutoff_mask;
520 __m256 signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
521 __m256 one = _mm256_set1_ps(1.0);
522 __m256 two = _mm256_set1_ps(2.0);
528 jindex = nlist->jindex;
530 shiftidx = nlist->shift;
532 shiftvec = fr->shift_vec[0];
533 fshift = fr->fshift[0];
534 nvdwtype = fr->ntype;
536 vdwtype = mdatoms->typeA;
538 rcutoff_scalar = fr->rvdw;
539 rcutoff = _mm256_set1_ps(rcutoff_scalar);
540 rcutoff2 = _mm256_mul_ps(rcutoff,rcutoff);
542 rswitch_scalar = fr->rvdw_switch;
543 rswitch = _mm256_set1_ps(rswitch_scalar);
544 /* Setup switch parameters */
545 d_scalar = rcutoff_scalar-rswitch_scalar;
546 d = _mm256_set1_ps(d_scalar);
547 swV3 = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
548 swV4 = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
549 swV5 = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
550 swF2 = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
551 swF3 = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
552 swF4 = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
554 /* Avoid stupid compiler warnings */
555 jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
568 for(iidx=0;iidx<4*DIM;iidx++)
573 /* Start outer loop over neighborlists */
574 for(iidx=0; iidx<nri; iidx++)
576 /* Load shift vector for this list */
577 i_shift_offset = DIM*shiftidx[iidx];
579 /* Load limits for loop over neighbors */
580 j_index_start = jindex[iidx];
581 j_index_end = jindex[iidx+1];
583 /* Get outer coordinate index */
585 i_coord_offset = DIM*inr;
587 /* Load i particle coords and add shift vector */
588 gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
590 fix0 = _mm256_setzero_ps();
591 fiy0 = _mm256_setzero_ps();
592 fiz0 = _mm256_setzero_ps();
594 /* Load parameters for i particles */
595 vdwioffsetptr0 = vdwparam+2*nvdwtype*vdwtype[inr+0];
597 /* Start inner kernel loop */
598 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
601 /* Get j neighbor index, and coordinate index */
610 j_coord_offsetA = DIM*jnrA;
611 j_coord_offsetB = DIM*jnrB;
612 j_coord_offsetC = DIM*jnrC;
613 j_coord_offsetD = DIM*jnrD;
614 j_coord_offsetE = DIM*jnrE;
615 j_coord_offsetF = DIM*jnrF;
616 j_coord_offsetG = DIM*jnrG;
617 j_coord_offsetH = DIM*jnrH;
619 /* load j atom coordinates */
620 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
621 x+j_coord_offsetC,x+j_coord_offsetD,
622 x+j_coord_offsetE,x+j_coord_offsetF,
623 x+j_coord_offsetG,x+j_coord_offsetH,
626 /* Calculate displacement vector */
627 dx00 = _mm256_sub_ps(ix0,jx0);
628 dy00 = _mm256_sub_ps(iy0,jy0);
629 dz00 = _mm256_sub_ps(iz0,jz0);
631 /* Calculate squared distance and things based on it */
632 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
634 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
636 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
638 /* Load parameters for j particles */
639 vdwjidx0A = 2*vdwtype[jnrA+0];
640 vdwjidx0B = 2*vdwtype[jnrB+0];
641 vdwjidx0C = 2*vdwtype[jnrC+0];
642 vdwjidx0D = 2*vdwtype[jnrD+0];
643 vdwjidx0E = 2*vdwtype[jnrE+0];
644 vdwjidx0F = 2*vdwtype[jnrF+0];
645 vdwjidx0G = 2*vdwtype[jnrG+0];
646 vdwjidx0H = 2*vdwtype[jnrH+0];
648 /**************************
649 * CALCULATE INTERACTIONS *
650 **************************/
652 if (gmx_mm256_any_lt(rsq00,rcutoff2))
655 r00 = _mm256_mul_ps(rsq00,rinv00);
657 /* Compute parameters for interactions between i and j atoms */
658 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
659 vdwioffsetptr0+vdwjidx0B,
660 vdwioffsetptr0+vdwjidx0C,
661 vdwioffsetptr0+vdwjidx0D,
662 vdwioffsetptr0+vdwjidx0E,
663 vdwioffsetptr0+vdwjidx0F,
664 vdwioffsetptr0+vdwjidx0G,
665 vdwioffsetptr0+vdwjidx0H,
668 /* LENNARD-JONES DISPERSION/REPULSION */
670 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
671 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
672 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
673 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
674 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
676 d = _mm256_sub_ps(r00,rswitch);
677 d = _mm256_max_ps(d,_mm256_setzero_ps());
678 d2 = _mm256_mul_ps(d,d);
679 sw = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
681 dsw = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
683 /* Evaluate switch function */
684 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
685 fvdw = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
686 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
690 fscal = _mm256_and_ps(fscal,cutoff_mask);
692 /* Calculate temporary vectorial force */
693 tx = _mm256_mul_ps(fscal,dx00);
694 ty = _mm256_mul_ps(fscal,dy00);
695 tz = _mm256_mul_ps(fscal,dz00);
697 /* Update vectorial force */
698 fix0 = _mm256_add_ps(fix0,tx);
699 fiy0 = _mm256_add_ps(fiy0,ty);
700 fiz0 = _mm256_add_ps(fiz0,tz);
702 fjptrA = f+j_coord_offsetA;
703 fjptrB = f+j_coord_offsetB;
704 fjptrC = f+j_coord_offsetC;
705 fjptrD = f+j_coord_offsetD;
706 fjptrE = f+j_coord_offsetE;
707 fjptrF = f+j_coord_offsetF;
708 fjptrG = f+j_coord_offsetG;
709 fjptrH = f+j_coord_offsetH;
710 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
714 /* Inner loop uses 56 flops */
720 /* Get j neighbor index, and coordinate index */
721 jnrlistA = jjnr[jidx];
722 jnrlistB = jjnr[jidx+1];
723 jnrlistC = jjnr[jidx+2];
724 jnrlistD = jjnr[jidx+3];
725 jnrlistE = jjnr[jidx+4];
726 jnrlistF = jjnr[jidx+5];
727 jnrlistG = jjnr[jidx+6];
728 jnrlistH = jjnr[jidx+7];
729 /* Sign of each element will be negative for non-real atoms.
730 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
731 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
733 dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
734 gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
736 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
737 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
738 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
739 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
740 jnrE = (jnrlistE>=0) ? jnrlistE : 0;
741 jnrF = (jnrlistF>=0) ? jnrlistF : 0;
742 jnrG = (jnrlistG>=0) ? jnrlistG : 0;
743 jnrH = (jnrlistH>=0) ? jnrlistH : 0;
744 j_coord_offsetA = DIM*jnrA;
745 j_coord_offsetB = DIM*jnrB;
746 j_coord_offsetC = DIM*jnrC;
747 j_coord_offsetD = DIM*jnrD;
748 j_coord_offsetE = DIM*jnrE;
749 j_coord_offsetF = DIM*jnrF;
750 j_coord_offsetG = DIM*jnrG;
751 j_coord_offsetH = DIM*jnrH;
753 /* load j atom coordinates */
754 gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
755 x+j_coord_offsetC,x+j_coord_offsetD,
756 x+j_coord_offsetE,x+j_coord_offsetF,
757 x+j_coord_offsetG,x+j_coord_offsetH,
760 /* Calculate displacement vector */
761 dx00 = _mm256_sub_ps(ix0,jx0);
762 dy00 = _mm256_sub_ps(iy0,jy0);
763 dz00 = _mm256_sub_ps(iz0,jz0);
765 /* Calculate squared distance and things based on it */
766 rsq00 = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
768 rinv00 = gmx_mm256_invsqrt_ps(rsq00);
770 rinvsq00 = _mm256_mul_ps(rinv00,rinv00);
772 /* Load parameters for j particles */
773 vdwjidx0A = 2*vdwtype[jnrA+0];
774 vdwjidx0B = 2*vdwtype[jnrB+0];
775 vdwjidx0C = 2*vdwtype[jnrC+0];
776 vdwjidx0D = 2*vdwtype[jnrD+0];
777 vdwjidx0E = 2*vdwtype[jnrE+0];
778 vdwjidx0F = 2*vdwtype[jnrF+0];
779 vdwjidx0G = 2*vdwtype[jnrG+0];
780 vdwjidx0H = 2*vdwtype[jnrH+0];
782 /**************************
783 * CALCULATE INTERACTIONS *
784 **************************/
786 if (gmx_mm256_any_lt(rsq00,rcutoff2))
789 r00 = _mm256_mul_ps(rsq00,rinv00);
790 r00 = _mm256_andnot_ps(dummy_mask,r00);
792 /* Compute parameters for interactions between i and j atoms */
793 gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
794 vdwioffsetptr0+vdwjidx0B,
795 vdwioffsetptr0+vdwjidx0C,
796 vdwioffsetptr0+vdwjidx0D,
797 vdwioffsetptr0+vdwjidx0E,
798 vdwioffsetptr0+vdwjidx0F,
799 vdwioffsetptr0+vdwjidx0G,
800 vdwioffsetptr0+vdwjidx0H,
803 /* LENNARD-JONES DISPERSION/REPULSION */
805 rinvsix = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
806 vvdw6 = _mm256_mul_ps(c6_00,rinvsix);
807 vvdw12 = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
808 vvdw = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
809 fvdw = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
811 d = _mm256_sub_ps(r00,rswitch);
812 d = _mm256_max_ps(d,_mm256_setzero_ps());
813 d2 = _mm256_mul_ps(d,d);
814 sw = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
816 dsw = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
818 /* Evaluate switch function */
819 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
820 fvdw = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
821 cutoff_mask = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
825 fscal = _mm256_and_ps(fscal,cutoff_mask);
827 fscal = _mm256_andnot_ps(dummy_mask,fscal);
829 /* Calculate temporary vectorial force */
830 tx = _mm256_mul_ps(fscal,dx00);
831 ty = _mm256_mul_ps(fscal,dy00);
832 tz = _mm256_mul_ps(fscal,dz00);
834 /* Update vectorial force */
835 fix0 = _mm256_add_ps(fix0,tx);
836 fiy0 = _mm256_add_ps(fiy0,ty);
837 fiz0 = _mm256_add_ps(fiz0,tz);
839 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
840 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
841 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
842 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
843 fjptrE = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
844 fjptrF = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
845 fjptrG = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
846 fjptrH = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
847 gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
851 /* Inner loop uses 57 flops */
854 /* End of innermost loop */
856 gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
857 f+i_coord_offset,fshift+i_shift_offset);
859 /* Increment number of inner iterations */
860 inneriter += j_index_end - j_index_start;
862 /* Outer loop uses 6 flops */
865 /* Increment number of outer iterations */
868 /* Update outer/inner flops */
870 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*57);