2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
36 * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
48 #include "kernelutil_x86_avx_128_fma_double.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJ_GeomP1P1_VF_avx_128_fma_double
52 * Electrostatics interaction: None
53 * VdW interaction: LennardJones
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecNone_VdwLJ_GeomP1P1_VF_avx_128_fma_double
59 (t_nblist * gmx_restrict nlist,
60 rvec * gmx_restrict xx,
61 rvec * gmx_restrict ff,
62 t_forcerec * gmx_restrict fr,
63 t_mdatoms * gmx_restrict mdatoms,
64 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
65 t_nrnb * gmx_restrict nrnb)
67 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
68 * just 0 for non-waters.
69 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
70 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 int i_shift_offset,i_coord_offset,outeriter,inneriter;
73 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int j_coord_offsetA,j_coord_offsetB;
76 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
78 real *shiftvec,*fshift,*x,*f;
79 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
81 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
82 int vdwjidx0A,vdwjidx0B;
83 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
84 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
86 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
89 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
90 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
91 __m128d dummy_mask,cutoff_mask;
92 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
93 __m128d one = _mm_set1_pd(1.0);
94 __m128d two = _mm_set1_pd(2.0);
100 jindex = nlist->jindex;
102 shiftidx = nlist->shift;
104 shiftvec = fr->shift_vec[0];
105 fshift = fr->fshift[0];
106 nvdwtype = fr->ntype;
108 vdwtype = mdatoms->typeA;
110 /* Avoid stupid compiler warnings */
118 /* Start outer loop over neighborlists */
119 for(iidx=0; iidx<nri; iidx++)
121 /* Load shift vector for this list */
122 i_shift_offset = DIM*shiftidx[iidx];
124 /* Load limits for loop over neighbors */
125 j_index_start = jindex[iidx];
126 j_index_end = jindex[iidx+1];
128 /* Get outer coordinate index */
130 i_coord_offset = DIM*inr;
132 /* Load i particle coords and add shift vector */
133 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
135 fix0 = _mm_setzero_pd();
136 fiy0 = _mm_setzero_pd();
137 fiz0 = _mm_setzero_pd();
139 /* Load parameters for i particles */
140 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
142 /* Reset potential sums */
143 vvdwsum = _mm_setzero_pd();
145 /* Start inner kernel loop */
146 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
149 /* Get j neighbor index, and coordinate index */
152 j_coord_offsetA = DIM*jnrA;
153 j_coord_offsetB = DIM*jnrB;
155 /* load j atom coordinates */
156 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
159 /* Calculate displacement vector */
160 dx00 = _mm_sub_pd(ix0,jx0);
161 dy00 = _mm_sub_pd(iy0,jy0);
162 dz00 = _mm_sub_pd(iz0,jz0);
164 /* Calculate squared distance and things based on it */
165 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
167 rinvsq00 = gmx_mm_inv_pd(rsq00);
169 /* Load parameters for j particles */
170 vdwjidx0A = 2*vdwtype[jnrA+0];
171 vdwjidx0B = 2*vdwtype[jnrB+0];
173 /**************************
174 * CALCULATE INTERACTIONS *
175 **************************/
177 /* Compute parameters for interactions between i and j atoms */
178 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
179 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
181 /* LENNARD-JONES DISPERSION/REPULSION */
183 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
184 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
185 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
186 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
187 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
189 /* Update potential sum for this i atom from the interaction with this j atom. */
190 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
194 /* Update vectorial force */
195 fix0 = _mm_macc_pd(dx00,fscal,fix0);
196 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
197 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
199 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
200 _mm_mul_pd(dx00,fscal),
201 _mm_mul_pd(dy00,fscal),
202 _mm_mul_pd(dz00,fscal));
204 /* Inner loop uses 35 flops */
211 j_coord_offsetA = DIM*jnrA;
213 /* load j atom coordinates */
214 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
217 /* Calculate displacement vector */
218 dx00 = _mm_sub_pd(ix0,jx0);
219 dy00 = _mm_sub_pd(iy0,jy0);
220 dz00 = _mm_sub_pd(iz0,jz0);
222 /* Calculate squared distance and things based on it */
223 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
225 rinvsq00 = gmx_mm_inv_pd(rsq00);
227 /* Load parameters for j particles */
228 vdwjidx0A = 2*vdwtype[jnrA+0];
230 /**************************
231 * CALCULATE INTERACTIONS *
232 **************************/
234 /* Compute parameters for interactions between i and j atoms */
235 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
237 /* LENNARD-JONES DISPERSION/REPULSION */
239 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
240 vvdw6 = _mm_mul_pd(c6_00,rinvsix);
241 vvdw12 = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
242 vvdw = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
243 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
245 /* Update potential sum for this i atom from the interaction with this j atom. */
246 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
247 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
251 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
253 /* Update vectorial force */
254 fix0 = _mm_macc_pd(dx00,fscal,fix0);
255 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
256 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
258 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
259 _mm_mul_pd(dx00,fscal),
260 _mm_mul_pd(dy00,fscal),
261 _mm_mul_pd(dz00,fscal));
263 /* Inner loop uses 35 flops */
266 /* End of innermost loop */
268 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
269 f+i_coord_offset,fshift+i_shift_offset);
272 /* Update potential energies */
273 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
275 /* Increment number of inner iterations */
276 inneriter += j_index_end - j_index_start;
278 /* Outer loop uses 7 flops */
281 /* Increment number of outer iterations */
284 /* Update outer/inner flops */
286 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*35);
289 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_128_fma_double
290 * Electrostatics interaction: None
291 * VdW interaction: LennardJones
292 * Geometry: Particle-Particle
293 * Calculate force/pot: Force
296 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_128_fma_double
297 (t_nblist * gmx_restrict nlist,
298 rvec * gmx_restrict xx,
299 rvec * gmx_restrict ff,
300 t_forcerec * gmx_restrict fr,
301 t_mdatoms * gmx_restrict mdatoms,
302 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
303 t_nrnb * gmx_restrict nrnb)
305 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
306 * just 0 for non-waters.
307 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
308 * jnr indices corresponding to data put in the four positions in the SIMD register.
310 int i_shift_offset,i_coord_offset,outeriter,inneriter;
311 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
313 int j_coord_offsetA,j_coord_offsetB;
314 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
316 real *shiftvec,*fshift,*x,*f;
317 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
319 __m128d ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
320 int vdwjidx0A,vdwjidx0B;
321 __m128d jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
322 __m128d dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
324 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
327 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
328 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
329 __m128d dummy_mask,cutoff_mask;
330 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
331 __m128d one = _mm_set1_pd(1.0);
332 __m128d two = _mm_set1_pd(2.0);
338 jindex = nlist->jindex;
340 shiftidx = nlist->shift;
342 shiftvec = fr->shift_vec[0];
343 fshift = fr->fshift[0];
344 nvdwtype = fr->ntype;
346 vdwtype = mdatoms->typeA;
348 /* Avoid stupid compiler warnings */
356 /* Start outer loop over neighborlists */
357 for(iidx=0; iidx<nri; iidx++)
359 /* Load shift vector for this list */
360 i_shift_offset = DIM*shiftidx[iidx];
362 /* Load limits for loop over neighbors */
363 j_index_start = jindex[iidx];
364 j_index_end = jindex[iidx+1];
366 /* Get outer coordinate index */
368 i_coord_offset = DIM*inr;
370 /* Load i particle coords and add shift vector */
371 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
373 fix0 = _mm_setzero_pd();
374 fiy0 = _mm_setzero_pd();
375 fiz0 = _mm_setzero_pd();
377 /* Load parameters for i particles */
378 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
380 /* Start inner kernel loop */
381 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
384 /* Get j neighbor index, and coordinate index */
387 j_coord_offsetA = DIM*jnrA;
388 j_coord_offsetB = DIM*jnrB;
390 /* load j atom coordinates */
391 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
394 /* Calculate displacement vector */
395 dx00 = _mm_sub_pd(ix0,jx0);
396 dy00 = _mm_sub_pd(iy0,jy0);
397 dz00 = _mm_sub_pd(iz0,jz0);
399 /* Calculate squared distance and things based on it */
400 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
402 rinvsq00 = gmx_mm_inv_pd(rsq00);
404 /* Load parameters for j particles */
405 vdwjidx0A = 2*vdwtype[jnrA+0];
406 vdwjidx0B = 2*vdwtype[jnrB+0];
408 /**************************
409 * CALCULATE INTERACTIONS *
410 **************************/
412 /* Compute parameters for interactions between i and j atoms */
413 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
414 vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
416 /* LENNARD-JONES DISPERSION/REPULSION */
418 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
419 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
423 /* Update vectorial force */
424 fix0 = _mm_macc_pd(dx00,fscal,fix0);
425 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
426 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
428 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
429 _mm_mul_pd(dx00,fscal),
430 _mm_mul_pd(dy00,fscal),
431 _mm_mul_pd(dz00,fscal));
433 /* Inner loop uses 30 flops */
440 j_coord_offsetA = DIM*jnrA;
442 /* load j atom coordinates */
443 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
446 /* Calculate displacement vector */
447 dx00 = _mm_sub_pd(ix0,jx0);
448 dy00 = _mm_sub_pd(iy0,jy0);
449 dz00 = _mm_sub_pd(iz0,jz0);
451 /* Calculate squared distance and things based on it */
452 rsq00 = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
454 rinvsq00 = gmx_mm_inv_pd(rsq00);
456 /* Load parameters for j particles */
457 vdwjidx0A = 2*vdwtype[jnrA+0];
459 /**************************
460 * CALCULATE INTERACTIONS *
461 **************************/
463 /* Compute parameters for interactions between i and j atoms */
464 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
466 /* LENNARD-JONES DISPERSION/REPULSION */
468 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
469 fvdw = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
473 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
475 /* Update vectorial force */
476 fix0 = _mm_macc_pd(dx00,fscal,fix0);
477 fiy0 = _mm_macc_pd(dy00,fscal,fiy0);
478 fiz0 = _mm_macc_pd(dz00,fscal,fiz0);
480 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
481 _mm_mul_pd(dx00,fscal),
482 _mm_mul_pd(dy00,fscal),
483 _mm_mul_pd(dz00,fscal));
485 /* Inner loop uses 30 flops */
488 /* End of innermost loop */
490 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
491 f+i_coord_offset,fshift+i_shift_offset);
493 /* Increment number of inner iterations */
494 inneriter += j_index_end - j_index_start;
496 /* Outer loop uses 6 flops */
499 /* Increment number of outer iterations */
502 /* Update outer/inner flops */
504 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);