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_single kernel generator.
44 #include "../nb_kernel.h"
45 #include "gromacs/legacyheaders/types/simple.h"
46 #include "gromacs/math/vec.h"
47 #include "gromacs/legacyheaders/nrnb.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
50 #include "kernelutil_x86_avx_128_fma_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_avx_128_fma_single
54 * Electrostatics interaction: None
55 * VdW interaction: LJEwald
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_avx_128_fma_single
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_128, 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 j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
84 __m128 fscal,rcutoff,rcutoff2,jidxall;
86 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
87 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
88 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
89 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
91 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
94 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
95 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
98 __m128 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
99 __m128 one_half = _mm_set1_ps(0.5);
100 __m128 minus_one = _mm_set1_ps(-1.0);
101 __m128 dummy_mask,cutoff_mask;
102 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
103 __m128 one = _mm_set1_ps(1.0);
104 __m128 two = _mm_set1_ps(2.0);
110 jindex = nlist->jindex;
112 shiftidx = nlist->shift;
114 shiftvec = fr->shift_vec[0];
115 fshift = fr->fshift[0];
116 nvdwtype = fr->ntype;
118 vdwtype = mdatoms->typeA;
119 vdwgridparam = fr->ljpme_c6grid;
120 sh_lj_ewald = _mm_set1_ps(fr->ic->sh_lj_ewald);
121 ewclj = _mm_set1_ps(fr->ewaldcoeff_lj);
122 ewclj2 = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
124 rcutoff_scalar = fr->rvdw;
125 rcutoff = _mm_set1_ps(rcutoff_scalar);
126 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
128 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
129 rvdw = _mm_set1_ps(fr->rvdw);
131 /* Avoid stupid compiler warnings */
132 jnrA = jnrB = jnrC = jnrD = 0;
141 for(iidx=0;iidx<4*DIM;iidx++)
146 /* Start outer loop over neighborlists */
147 for(iidx=0; iidx<nri; iidx++)
149 /* Load shift vector for this list */
150 i_shift_offset = DIM*shiftidx[iidx];
152 /* Load limits for loop over neighbors */
153 j_index_start = jindex[iidx];
154 j_index_end = jindex[iidx+1];
156 /* Get outer coordinate index */
158 i_coord_offset = DIM*inr;
160 /* Load i particle coords and add shift vector */
161 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
163 fix0 = _mm_setzero_ps();
164 fiy0 = _mm_setzero_ps();
165 fiz0 = _mm_setzero_ps();
167 /* Load parameters for i particles */
168 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
170 /* Reset potential sums */
171 vvdwsum = _mm_setzero_ps();
173 /* Start inner kernel loop */
174 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
177 /* Get j neighbor index, and coordinate index */
182 j_coord_offsetA = DIM*jnrA;
183 j_coord_offsetB = DIM*jnrB;
184 j_coord_offsetC = DIM*jnrC;
185 j_coord_offsetD = DIM*jnrD;
187 /* load j atom coordinates */
188 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
189 x+j_coord_offsetC,x+j_coord_offsetD,
192 /* Calculate displacement vector */
193 dx00 = _mm_sub_ps(ix0,jx0);
194 dy00 = _mm_sub_ps(iy0,jy0);
195 dz00 = _mm_sub_ps(iz0,jz0);
197 /* Calculate squared distance and things based on it */
198 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
200 rinv00 = gmx_mm_invsqrt_ps(rsq00);
202 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
204 /* Load parameters for j particles */
205 vdwjidx0A = 2*vdwtype[jnrA+0];
206 vdwjidx0B = 2*vdwtype[jnrB+0];
207 vdwjidx0C = 2*vdwtype[jnrC+0];
208 vdwjidx0D = 2*vdwtype[jnrD+0];
210 /**************************
211 * CALCULATE INTERACTIONS *
212 **************************/
214 if (gmx_mm_any_lt(rsq00,rcutoff2))
217 r00 = _mm_mul_ps(rsq00,rinv00);
219 /* Compute parameters for interactions between i and j atoms */
220 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
221 vdwparam+vdwioffset0+vdwjidx0B,
222 vdwparam+vdwioffset0+vdwjidx0C,
223 vdwparam+vdwioffset0+vdwjidx0D,
226 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
227 vdwgridparam+vdwioffset0+vdwjidx0B,
228 vdwgridparam+vdwioffset0+vdwjidx0C,
229 vdwgridparam+vdwioffset0+vdwjidx0D);
231 /* Analytical LJ-PME */
232 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
233 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
234 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
235 exponent = gmx_simd_exp_r(ewcljrsq);
236 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
237 poly = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
238 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
239 vvdw6 = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
240 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
241 vvdw = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
242 _mm_mul_ps(_mm_sub_ps(vvdw6,_mm_macc_ps(c6grid_00,sh_lj_ewald,_mm_mul_ps(c6_00,sh_vdw_invrcut6))),one_sixth));
243 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
244 fvdw = _mm_mul_ps(_mm_add_ps(vvdw12,_mm_msub_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6),vvdw6)),rinvsq00);
246 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
248 /* Update potential sum for this i atom from the interaction with this j atom. */
249 vvdw = _mm_and_ps(vvdw,cutoff_mask);
250 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
254 fscal = _mm_and_ps(fscal,cutoff_mask);
256 /* Update vectorial force */
257 fix0 = _mm_macc_ps(dx00,fscal,fix0);
258 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
259 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
261 fjptrA = f+j_coord_offsetA;
262 fjptrB = f+j_coord_offsetB;
263 fjptrC = f+j_coord_offsetC;
264 fjptrD = f+j_coord_offsetD;
265 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
266 _mm_mul_ps(dx00,fscal),
267 _mm_mul_ps(dy00,fscal),
268 _mm_mul_ps(dz00,fscal));
272 /* Inner loop uses 59 flops */
278 /* Get j neighbor index, and coordinate index */
279 jnrlistA = jjnr[jidx];
280 jnrlistB = jjnr[jidx+1];
281 jnrlistC = jjnr[jidx+2];
282 jnrlistD = jjnr[jidx+3];
283 /* Sign of each element will be negative for non-real atoms.
284 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
285 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
287 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
288 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
289 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
290 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
291 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
292 j_coord_offsetA = DIM*jnrA;
293 j_coord_offsetB = DIM*jnrB;
294 j_coord_offsetC = DIM*jnrC;
295 j_coord_offsetD = DIM*jnrD;
297 /* load j atom coordinates */
298 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
299 x+j_coord_offsetC,x+j_coord_offsetD,
302 /* Calculate displacement vector */
303 dx00 = _mm_sub_ps(ix0,jx0);
304 dy00 = _mm_sub_ps(iy0,jy0);
305 dz00 = _mm_sub_ps(iz0,jz0);
307 /* Calculate squared distance and things based on it */
308 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
310 rinv00 = gmx_mm_invsqrt_ps(rsq00);
312 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
314 /* Load parameters for j particles */
315 vdwjidx0A = 2*vdwtype[jnrA+0];
316 vdwjidx0B = 2*vdwtype[jnrB+0];
317 vdwjidx0C = 2*vdwtype[jnrC+0];
318 vdwjidx0D = 2*vdwtype[jnrD+0];
320 /**************************
321 * CALCULATE INTERACTIONS *
322 **************************/
324 if (gmx_mm_any_lt(rsq00,rcutoff2))
327 r00 = _mm_mul_ps(rsq00,rinv00);
328 r00 = _mm_andnot_ps(dummy_mask,r00);
330 /* Compute parameters for interactions between i and j atoms */
331 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
332 vdwparam+vdwioffset0+vdwjidx0B,
333 vdwparam+vdwioffset0+vdwjidx0C,
334 vdwparam+vdwioffset0+vdwjidx0D,
337 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
338 vdwgridparam+vdwioffset0+vdwjidx0B,
339 vdwgridparam+vdwioffset0+vdwjidx0C,
340 vdwgridparam+vdwioffset0+vdwjidx0D);
342 /* Analytical LJ-PME */
343 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
344 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
345 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
346 exponent = gmx_simd_exp_r(ewcljrsq);
347 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
348 poly = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
349 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
350 vvdw6 = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
351 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
352 vvdw = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
353 _mm_mul_ps(_mm_sub_ps(vvdw6,_mm_macc_ps(c6grid_00,sh_lj_ewald,_mm_mul_ps(c6_00,sh_vdw_invrcut6))),one_sixth));
354 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
355 fvdw = _mm_mul_ps(_mm_add_ps(vvdw12,_mm_msub_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6),vvdw6)),rinvsq00);
357 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
359 /* Update potential sum for this i atom from the interaction with this j atom. */
360 vvdw = _mm_and_ps(vvdw,cutoff_mask);
361 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
362 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
366 fscal = _mm_and_ps(fscal,cutoff_mask);
368 fscal = _mm_andnot_ps(dummy_mask,fscal);
370 /* Update vectorial force */
371 fix0 = _mm_macc_ps(dx00,fscal,fix0);
372 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
373 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
375 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
376 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
377 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
378 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
379 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
380 _mm_mul_ps(dx00,fscal),
381 _mm_mul_ps(dy00,fscal),
382 _mm_mul_ps(dz00,fscal));
386 /* Inner loop uses 60 flops */
389 /* End of innermost loop */
391 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
392 f+i_coord_offset,fshift+i_shift_offset);
395 /* Update potential energies */
396 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
398 /* Increment number of inner iterations */
399 inneriter += j_index_end - j_index_start;
401 /* Outer loop uses 7 flops */
404 /* Increment number of outer iterations */
407 /* Update outer/inner flops */
409 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*60);
412 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_avx_128_fma_single
413 * Electrostatics interaction: None
414 * VdW interaction: LJEwald
415 * Geometry: Particle-Particle
416 * Calculate force/pot: Force
419 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_avx_128_fma_single
420 (t_nblist * gmx_restrict nlist,
421 rvec * gmx_restrict xx,
422 rvec * gmx_restrict ff,
423 t_forcerec * gmx_restrict fr,
424 t_mdatoms * gmx_restrict mdatoms,
425 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
426 t_nrnb * gmx_restrict nrnb)
428 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
429 * just 0 for non-waters.
430 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
431 * jnr indices corresponding to data put in the four positions in the SIMD register.
433 int i_shift_offset,i_coord_offset,outeriter,inneriter;
434 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
435 int jnrA,jnrB,jnrC,jnrD;
436 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
437 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
438 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
440 real *shiftvec,*fshift,*x,*f;
441 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
443 __m128 fscal,rcutoff,rcutoff2,jidxall;
445 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
446 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
447 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
448 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
450 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
453 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
454 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
457 __m128 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
458 __m128 one_half = _mm_set1_ps(0.5);
459 __m128 minus_one = _mm_set1_ps(-1.0);
460 __m128 dummy_mask,cutoff_mask;
461 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
462 __m128 one = _mm_set1_ps(1.0);
463 __m128 two = _mm_set1_ps(2.0);
469 jindex = nlist->jindex;
471 shiftidx = nlist->shift;
473 shiftvec = fr->shift_vec[0];
474 fshift = fr->fshift[0];
475 nvdwtype = fr->ntype;
477 vdwtype = mdatoms->typeA;
478 vdwgridparam = fr->ljpme_c6grid;
479 sh_lj_ewald = _mm_set1_ps(fr->ic->sh_lj_ewald);
480 ewclj = _mm_set1_ps(fr->ewaldcoeff_lj);
481 ewclj2 = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
483 rcutoff_scalar = fr->rvdw;
484 rcutoff = _mm_set1_ps(rcutoff_scalar);
485 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
487 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
488 rvdw = _mm_set1_ps(fr->rvdw);
490 /* Avoid stupid compiler warnings */
491 jnrA = jnrB = jnrC = jnrD = 0;
500 for(iidx=0;iidx<4*DIM;iidx++)
505 /* Start outer loop over neighborlists */
506 for(iidx=0; iidx<nri; iidx++)
508 /* Load shift vector for this list */
509 i_shift_offset = DIM*shiftidx[iidx];
511 /* Load limits for loop over neighbors */
512 j_index_start = jindex[iidx];
513 j_index_end = jindex[iidx+1];
515 /* Get outer coordinate index */
517 i_coord_offset = DIM*inr;
519 /* Load i particle coords and add shift vector */
520 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
522 fix0 = _mm_setzero_ps();
523 fiy0 = _mm_setzero_ps();
524 fiz0 = _mm_setzero_ps();
526 /* Load parameters for i particles */
527 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
529 /* Start inner kernel loop */
530 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
533 /* Get j neighbor index, and coordinate index */
538 j_coord_offsetA = DIM*jnrA;
539 j_coord_offsetB = DIM*jnrB;
540 j_coord_offsetC = DIM*jnrC;
541 j_coord_offsetD = DIM*jnrD;
543 /* load j atom coordinates */
544 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
545 x+j_coord_offsetC,x+j_coord_offsetD,
548 /* Calculate displacement vector */
549 dx00 = _mm_sub_ps(ix0,jx0);
550 dy00 = _mm_sub_ps(iy0,jy0);
551 dz00 = _mm_sub_ps(iz0,jz0);
553 /* Calculate squared distance and things based on it */
554 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
556 rinv00 = gmx_mm_invsqrt_ps(rsq00);
558 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
560 /* Load parameters for j particles */
561 vdwjidx0A = 2*vdwtype[jnrA+0];
562 vdwjidx0B = 2*vdwtype[jnrB+0];
563 vdwjidx0C = 2*vdwtype[jnrC+0];
564 vdwjidx0D = 2*vdwtype[jnrD+0];
566 /**************************
567 * CALCULATE INTERACTIONS *
568 **************************/
570 if (gmx_mm_any_lt(rsq00,rcutoff2))
573 r00 = _mm_mul_ps(rsq00,rinv00);
575 /* Compute parameters for interactions between i and j atoms */
576 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
577 vdwparam+vdwioffset0+vdwjidx0B,
578 vdwparam+vdwioffset0+vdwjidx0C,
579 vdwparam+vdwioffset0+vdwjidx0D,
582 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
583 vdwgridparam+vdwioffset0+vdwjidx0B,
584 vdwgridparam+vdwioffset0+vdwjidx0C,
585 vdwgridparam+vdwioffset0+vdwjidx0D);
587 /* Analytical LJ-PME */
588 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
589 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
590 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
591 exponent = gmx_simd_exp_r(ewcljrsq);
592 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
593 poly = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
594 /* f6A = 6 * C6grid * (1 - poly) */
595 f6A = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
596 /* f6B = C6grid * exponent * beta^6 */
597 f6B = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
598 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
599 fvdw = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
601 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
605 fscal = _mm_and_ps(fscal,cutoff_mask);
607 /* Update vectorial force */
608 fix0 = _mm_macc_ps(dx00,fscal,fix0);
609 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
610 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
612 fjptrA = f+j_coord_offsetA;
613 fjptrB = f+j_coord_offsetB;
614 fjptrC = f+j_coord_offsetC;
615 fjptrD = f+j_coord_offsetD;
616 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
617 _mm_mul_ps(dx00,fscal),
618 _mm_mul_ps(dy00,fscal),
619 _mm_mul_ps(dz00,fscal));
623 /* Inner loop uses 50 flops */
629 /* Get j neighbor index, and coordinate index */
630 jnrlistA = jjnr[jidx];
631 jnrlistB = jjnr[jidx+1];
632 jnrlistC = jjnr[jidx+2];
633 jnrlistD = jjnr[jidx+3];
634 /* Sign of each element will be negative for non-real atoms.
635 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
636 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
638 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
639 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
640 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
641 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
642 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
643 j_coord_offsetA = DIM*jnrA;
644 j_coord_offsetB = DIM*jnrB;
645 j_coord_offsetC = DIM*jnrC;
646 j_coord_offsetD = DIM*jnrD;
648 /* load j atom coordinates */
649 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
650 x+j_coord_offsetC,x+j_coord_offsetD,
653 /* Calculate displacement vector */
654 dx00 = _mm_sub_ps(ix0,jx0);
655 dy00 = _mm_sub_ps(iy0,jy0);
656 dz00 = _mm_sub_ps(iz0,jz0);
658 /* Calculate squared distance and things based on it */
659 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
661 rinv00 = gmx_mm_invsqrt_ps(rsq00);
663 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
665 /* Load parameters for j particles */
666 vdwjidx0A = 2*vdwtype[jnrA+0];
667 vdwjidx0B = 2*vdwtype[jnrB+0];
668 vdwjidx0C = 2*vdwtype[jnrC+0];
669 vdwjidx0D = 2*vdwtype[jnrD+0];
671 /**************************
672 * CALCULATE INTERACTIONS *
673 **************************/
675 if (gmx_mm_any_lt(rsq00,rcutoff2))
678 r00 = _mm_mul_ps(rsq00,rinv00);
679 r00 = _mm_andnot_ps(dummy_mask,r00);
681 /* Compute parameters for interactions between i and j atoms */
682 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
683 vdwparam+vdwioffset0+vdwjidx0B,
684 vdwparam+vdwioffset0+vdwjidx0C,
685 vdwparam+vdwioffset0+vdwjidx0D,
688 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
689 vdwgridparam+vdwioffset0+vdwjidx0B,
690 vdwgridparam+vdwioffset0+vdwjidx0C,
691 vdwgridparam+vdwioffset0+vdwjidx0D);
693 /* Analytical LJ-PME */
694 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
695 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
696 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
697 exponent = gmx_simd_exp_r(ewcljrsq);
698 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
699 poly = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
700 /* f6A = 6 * C6grid * (1 - poly) */
701 f6A = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
702 /* f6B = C6grid * exponent * beta^6 */
703 f6B = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
704 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
705 fvdw = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
707 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
711 fscal = _mm_and_ps(fscal,cutoff_mask);
713 fscal = _mm_andnot_ps(dummy_mask,fscal);
715 /* Update vectorial force */
716 fix0 = _mm_macc_ps(dx00,fscal,fix0);
717 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
718 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
720 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
721 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
722 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
723 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
724 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
725 _mm_mul_ps(dx00,fscal),
726 _mm_mul_ps(dy00,fscal),
727 _mm_mul_ps(dz00,fscal));
731 /* Inner loop uses 51 flops */
734 /* End of innermost loop */
736 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
737 f+i_coord_offset,fshift+i_shift_offset);
739 /* Increment number of inner iterations */
740 inneriter += j_index_end - j_index_start;
742 /* Outer loop uses 6 flops */
745 /* Increment number of outer iterations */
748 /* Update outer/inner flops */
750 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*51);