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_VdwLJEw_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_VdwLJEw_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 /* Avoid stupid compiler warnings */
125 jnrA = jnrB = jnrC = jnrD = 0;
134 for(iidx=0;iidx<4*DIM;iidx++)
139 /* Start outer loop over neighborlists */
140 for(iidx=0; iidx<nri; iidx++)
142 /* Load shift vector for this list */
143 i_shift_offset = DIM*shiftidx[iidx];
145 /* Load limits for loop over neighbors */
146 j_index_start = jindex[iidx];
147 j_index_end = jindex[iidx+1];
149 /* Get outer coordinate index */
151 i_coord_offset = DIM*inr;
153 /* Load i particle coords and add shift vector */
154 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
156 fix0 = _mm_setzero_ps();
157 fiy0 = _mm_setzero_ps();
158 fiz0 = _mm_setzero_ps();
160 /* Load parameters for i particles */
161 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
163 /* Reset potential sums */
164 vvdwsum = _mm_setzero_ps();
166 /* Start inner kernel loop */
167 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
170 /* Get j neighbor index, and coordinate index */
175 j_coord_offsetA = DIM*jnrA;
176 j_coord_offsetB = DIM*jnrB;
177 j_coord_offsetC = DIM*jnrC;
178 j_coord_offsetD = DIM*jnrD;
180 /* load j atom coordinates */
181 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
182 x+j_coord_offsetC,x+j_coord_offsetD,
185 /* Calculate displacement vector */
186 dx00 = _mm_sub_ps(ix0,jx0);
187 dy00 = _mm_sub_ps(iy0,jy0);
188 dz00 = _mm_sub_ps(iz0,jz0);
190 /* Calculate squared distance and things based on it */
191 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
193 rinv00 = gmx_mm_invsqrt_ps(rsq00);
195 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
197 /* Load parameters for j particles */
198 vdwjidx0A = 2*vdwtype[jnrA+0];
199 vdwjidx0B = 2*vdwtype[jnrB+0];
200 vdwjidx0C = 2*vdwtype[jnrC+0];
201 vdwjidx0D = 2*vdwtype[jnrD+0];
203 /**************************
204 * CALCULATE INTERACTIONS *
205 **************************/
207 r00 = _mm_mul_ps(rsq00,rinv00);
209 /* Compute parameters for interactions between i and j atoms */
210 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
211 vdwparam+vdwioffset0+vdwjidx0B,
212 vdwparam+vdwioffset0+vdwjidx0C,
213 vdwparam+vdwioffset0+vdwjidx0D,
216 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
217 vdwgridparam+vdwioffset0+vdwjidx0B,
218 vdwgridparam+vdwioffset0+vdwjidx0C,
219 vdwgridparam+vdwioffset0+vdwjidx0D);
221 /* Analytical LJ-PME */
222 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
223 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
224 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
225 exponent = gmx_simd_exp_r(ewcljrsq);
226 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
227 poly = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
228 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
229 vvdw6 = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
230 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
231 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
232 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
233 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);
235 /* Update potential sum for this i atom from the interaction with this j atom. */
236 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
240 /* Update vectorial force */
241 fix0 = _mm_macc_ps(dx00,fscal,fix0);
242 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
243 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
245 fjptrA = f+j_coord_offsetA;
246 fjptrB = f+j_coord_offsetB;
247 fjptrC = f+j_coord_offsetC;
248 fjptrD = f+j_coord_offsetD;
249 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
250 _mm_mul_ps(dx00,fscal),
251 _mm_mul_ps(dy00,fscal),
252 _mm_mul_ps(dz00,fscal));
254 /* Inner loop uses 50 flops */
260 /* Get j neighbor index, and coordinate index */
261 jnrlistA = jjnr[jidx];
262 jnrlistB = jjnr[jidx+1];
263 jnrlistC = jjnr[jidx+2];
264 jnrlistD = jjnr[jidx+3];
265 /* Sign of each element will be negative for non-real atoms.
266 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
267 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
269 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
270 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
271 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
272 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
273 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
274 j_coord_offsetA = DIM*jnrA;
275 j_coord_offsetB = DIM*jnrB;
276 j_coord_offsetC = DIM*jnrC;
277 j_coord_offsetD = DIM*jnrD;
279 /* load j atom coordinates */
280 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
281 x+j_coord_offsetC,x+j_coord_offsetD,
284 /* Calculate displacement vector */
285 dx00 = _mm_sub_ps(ix0,jx0);
286 dy00 = _mm_sub_ps(iy0,jy0);
287 dz00 = _mm_sub_ps(iz0,jz0);
289 /* Calculate squared distance and things based on it */
290 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
292 rinv00 = gmx_mm_invsqrt_ps(rsq00);
294 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
296 /* Load parameters for j particles */
297 vdwjidx0A = 2*vdwtype[jnrA+0];
298 vdwjidx0B = 2*vdwtype[jnrB+0];
299 vdwjidx0C = 2*vdwtype[jnrC+0];
300 vdwjidx0D = 2*vdwtype[jnrD+0];
302 /**************************
303 * CALCULATE INTERACTIONS *
304 **************************/
306 r00 = _mm_mul_ps(rsq00,rinv00);
307 r00 = _mm_andnot_ps(dummy_mask,r00);
309 /* Compute parameters for interactions between i and j atoms */
310 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
311 vdwparam+vdwioffset0+vdwjidx0B,
312 vdwparam+vdwioffset0+vdwjidx0C,
313 vdwparam+vdwioffset0+vdwjidx0D,
316 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
317 vdwgridparam+vdwioffset0+vdwjidx0B,
318 vdwgridparam+vdwioffset0+vdwjidx0C,
319 vdwgridparam+vdwioffset0+vdwjidx0D);
321 /* Analytical LJ-PME */
322 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
323 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
324 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
325 exponent = gmx_simd_exp_r(ewcljrsq);
326 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
327 poly = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
328 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
329 vvdw6 = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
330 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
331 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
332 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
333 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);
335 /* Update potential sum for this i atom from the interaction with this j atom. */
336 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
337 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
341 fscal = _mm_andnot_ps(dummy_mask,fscal);
343 /* Update vectorial force */
344 fix0 = _mm_macc_ps(dx00,fscal,fix0);
345 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
346 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
348 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
349 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
350 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
351 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
352 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
353 _mm_mul_ps(dx00,fscal),
354 _mm_mul_ps(dy00,fscal),
355 _mm_mul_ps(dz00,fscal));
357 /* Inner loop uses 51 flops */
360 /* End of innermost loop */
362 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
363 f+i_coord_offset,fshift+i_shift_offset);
366 /* Update potential energies */
367 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
369 /* Increment number of inner iterations */
370 inneriter += j_index_end - j_index_start;
372 /* Outer loop uses 7 flops */
375 /* Increment number of outer iterations */
378 /* Update outer/inner flops */
380 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*51);
383 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_avx_128_fma_single
384 * Electrostatics interaction: None
385 * VdW interaction: LJEwald
386 * Geometry: Particle-Particle
387 * Calculate force/pot: Force
390 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_avx_128_fma_single
391 (t_nblist * gmx_restrict nlist,
392 rvec * gmx_restrict xx,
393 rvec * gmx_restrict ff,
394 t_forcerec * gmx_restrict fr,
395 t_mdatoms * gmx_restrict mdatoms,
396 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
397 t_nrnb * gmx_restrict nrnb)
399 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
400 * just 0 for non-waters.
401 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
402 * jnr indices corresponding to data put in the four positions in the SIMD register.
404 int i_shift_offset,i_coord_offset,outeriter,inneriter;
405 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
406 int jnrA,jnrB,jnrC,jnrD;
407 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
408 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
409 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
411 real *shiftvec,*fshift,*x,*f;
412 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
414 __m128 fscal,rcutoff,rcutoff2,jidxall;
416 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
417 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
418 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
419 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
421 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
424 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
425 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
428 __m128 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
429 __m128 one_half = _mm_set1_ps(0.5);
430 __m128 minus_one = _mm_set1_ps(-1.0);
431 __m128 dummy_mask,cutoff_mask;
432 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
433 __m128 one = _mm_set1_ps(1.0);
434 __m128 two = _mm_set1_ps(2.0);
440 jindex = nlist->jindex;
442 shiftidx = nlist->shift;
444 shiftvec = fr->shift_vec[0];
445 fshift = fr->fshift[0];
446 nvdwtype = fr->ntype;
448 vdwtype = mdatoms->typeA;
449 vdwgridparam = fr->ljpme_c6grid;
450 sh_lj_ewald = _mm_set1_ps(fr->ic->sh_lj_ewald);
451 ewclj = _mm_set1_ps(fr->ewaldcoeff_lj);
452 ewclj2 = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
454 /* Avoid stupid compiler warnings */
455 jnrA = jnrB = jnrC = jnrD = 0;
464 for(iidx=0;iidx<4*DIM;iidx++)
469 /* Start outer loop over neighborlists */
470 for(iidx=0; iidx<nri; iidx++)
472 /* Load shift vector for this list */
473 i_shift_offset = DIM*shiftidx[iidx];
475 /* Load limits for loop over neighbors */
476 j_index_start = jindex[iidx];
477 j_index_end = jindex[iidx+1];
479 /* Get outer coordinate index */
481 i_coord_offset = DIM*inr;
483 /* Load i particle coords and add shift vector */
484 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
486 fix0 = _mm_setzero_ps();
487 fiy0 = _mm_setzero_ps();
488 fiz0 = _mm_setzero_ps();
490 /* Load parameters for i particles */
491 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
493 /* Start inner kernel loop */
494 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
497 /* Get j neighbor index, and coordinate index */
502 j_coord_offsetA = DIM*jnrA;
503 j_coord_offsetB = DIM*jnrB;
504 j_coord_offsetC = DIM*jnrC;
505 j_coord_offsetD = DIM*jnrD;
507 /* load j atom coordinates */
508 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
509 x+j_coord_offsetC,x+j_coord_offsetD,
512 /* Calculate displacement vector */
513 dx00 = _mm_sub_ps(ix0,jx0);
514 dy00 = _mm_sub_ps(iy0,jy0);
515 dz00 = _mm_sub_ps(iz0,jz0);
517 /* Calculate squared distance and things based on it */
518 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
520 rinv00 = gmx_mm_invsqrt_ps(rsq00);
522 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
524 /* Load parameters for j particles */
525 vdwjidx0A = 2*vdwtype[jnrA+0];
526 vdwjidx0B = 2*vdwtype[jnrB+0];
527 vdwjidx0C = 2*vdwtype[jnrC+0];
528 vdwjidx0D = 2*vdwtype[jnrD+0];
530 /**************************
531 * CALCULATE INTERACTIONS *
532 **************************/
534 r00 = _mm_mul_ps(rsq00,rinv00);
536 /* Compute parameters for interactions between i and j atoms */
537 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
538 vdwparam+vdwioffset0+vdwjidx0B,
539 vdwparam+vdwioffset0+vdwjidx0C,
540 vdwparam+vdwioffset0+vdwjidx0D,
543 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
544 vdwgridparam+vdwioffset0+vdwjidx0B,
545 vdwgridparam+vdwioffset0+vdwjidx0C,
546 vdwgridparam+vdwioffset0+vdwjidx0D);
548 /* Analytical LJ-PME */
549 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
550 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
551 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
552 exponent = gmx_simd_exp_r(ewcljrsq);
553 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
554 poly = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
555 /* f6A = 6 * C6grid * (1 - poly) */
556 f6A = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
557 /* f6B = C6grid * exponent * beta^6 */
558 f6B = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
559 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
560 fvdw = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
564 /* Update vectorial force */
565 fix0 = _mm_macc_ps(dx00,fscal,fix0);
566 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
567 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
569 fjptrA = f+j_coord_offsetA;
570 fjptrB = f+j_coord_offsetB;
571 fjptrC = f+j_coord_offsetC;
572 fjptrD = f+j_coord_offsetD;
573 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
574 _mm_mul_ps(dx00,fscal),
575 _mm_mul_ps(dy00,fscal),
576 _mm_mul_ps(dz00,fscal));
578 /* Inner loop uses 47 flops */
584 /* Get j neighbor index, and coordinate index */
585 jnrlistA = jjnr[jidx];
586 jnrlistB = jjnr[jidx+1];
587 jnrlistC = jjnr[jidx+2];
588 jnrlistD = jjnr[jidx+3];
589 /* Sign of each element will be negative for non-real atoms.
590 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
591 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
593 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
594 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
595 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
596 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
597 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
598 j_coord_offsetA = DIM*jnrA;
599 j_coord_offsetB = DIM*jnrB;
600 j_coord_offsetC = DIM*jnrC;
601 j_coord_offsetD = DIM*jnrD;
603 /* load j atom coordinates */
604 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
605 x+j_coord_offsetC,x+j_coord_offsetD,
608 /* Calculate displacement vector */
609 dx00 = _mm_sub_ps(ix0,jx0);
610 dy00 = _mm_sub_ps(iy0,jy0);
611 dz00 = _mm_sub_ps(iz0,jz0);
613 /* Calculate squared distance and things based on it */
614 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
616 rinv00 = gmx_mm_invsqrt_ps(rsq00);
618 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
620 /* Load parameters for j particles */
621 vdwjidx0A = 2*vdwtype[jnrA+0];
622 vdwjidx0B = 2*vdwtype[jnrB+0];
623 vdwjidx0C = 2*vdwtype[jnrC+0];
624 vdwjidx0D = 2*vdwtype[jnrD+0];
626 /**************************
627 * CALCULATE INTERACTIONS *
628 **************************/
630 r00 = _mm_mul_ps(rsq00,rinv00);
631 r00 = _mm_andnot_ps(dummy_mask,r00);
633 /* Compute parameters for interactions between i and j atoms */
634 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
635 vdwparam+vdwioffset0+vdwjidx0B,
636 vdwparam+vdwioffset0+vdwjidx0C,
637 vdwparam+vdwioffset0+vdwjidx0D,
640 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
641 vdwgridparam+vdwioffset0+vdwjidx0B,
642 vdwgridparam+vdwioffset0+vdwjidx0C,
643 vdwgridparam+vdwioffset0+vdwjidx0D);
645 /* Analytical LJ-PME */
646 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
647 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
648 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
649 exponent = gmx_simd_exp_r(ewcljrsq);
650 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
651 poly = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
652 /* f6A = 6 * C6grid * (1 - poly) */
653 f6A = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
654 /* f6B = C6grid * exponent * beta^6 */
655 f6B = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
656 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
657 fvdw = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
661 fscal = _mm_andnot_ps(dummy_mask,fscal);
663 /* Update vectorial force */
664 fix0 = _mm_macc_ps(dx00,fscal,fix0);
665 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
666 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
668 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
669 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
670 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
671 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
672 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
673 _mm_mul_ps(dx00,fscal),
674 _mm_mul_ps(dy00,fscal),
675 _mm_mul_ps(dz00,fscal));
677 /* Inner loop uses 48 flops */
680 /* End of innermost loop */
682 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
683 f+i_coord_offset,fshift+i_shift_offset);
685 /* Increment number of inner iterations */
686 inneriter += j_index_end - j_index_start;
688 /* Outer loop uses 6 flops */
691 /* Increment number of outer iterations */
694 /* Update outer/inner flops */
696 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*48);