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 sse4_1_single kernel generator.
42 #include "../nb_kernel.h"
43 #include "types/simple.h"
44 #include "gromacs/math/vec.h"
47 #include "gromacs/simd/math_x86_sse4_1_single.h"
48 #include "kernelutil_x86_sse4_1_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse4_1_single
52 * Electrostatics interaction: None
53 * VdW interaction: LJEwald
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sse4_1_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 refer to j loop unrolling done with SSE, e.g. for the four 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 jnrlistA,jnrlistB,jnrlistC,jnrlistD;
76 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
77 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real *shiftvec,*fshift,*x,*f;
80 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
85 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
86 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
87 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
89 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
92 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
93 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
95 __m128 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
97 __m128 one_half = _mm_set1_ps(0.5);
98 __m128 minus_one = _mm_set1_ps(-1.0);
99 __m128 dummy_mask,cutoff_mask;
100 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
101 __m128 one = _mm_set1_ps(1.0);
102 __m128 two = _mm_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;
117 vdwgridparam = fr->ljpme_c6grid;
118 sh_lj_ewald = _mm_set1_ps(fr->ic->sh_lj_ewald);
119 ewclj = _mm_set1_ps(fr->ewaldcoeff_lj);
120 ewclj2 = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
122 /* Avoid stupid compiler warnings */
123 jnrA = jnrB = jnrC = jnrD = 0;
132 for(iidx=0;iidx<4*DIM;iidx++)
137 /* Start outer loop over neighborlists */
138 for(iidx=0; iidx<nri; iidx++)
140 /* Load shift vector for this list */
141 i_shift_offset = DIM*shiftidx[iidx];
143 /* Load limits for loop over neighbors */
144 j_index_start = jindex[iidx];
145 j_index_end = jindex[iidx+1];
147 /* Get outer coordinate index */
149 i_coord_offset = DIM*inr;
151 /* Load i particle coords and add shift vector */
152 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
154 fix0 = _mm_setzero_ps();
155 fiy0 = _mm_setzero_ps();
156 fiz0 = _mm_setzero_ps();
158 /* Load parameters for i particles */
159 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
161 /* Reset potential sums */
162 vvdwsum = _mm_setzero_ps();
164 /* Start inner kernel loop */
165 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
168 /* Get j neighbor index, and coordinate index */
173 j_coord_offsetA = DIM*jnrA;
174 j_coord_offsetB = DIM*jnrB;
175 j_coord_offsetC = DIM*jnrC;
176 j_coord_offsetD = DIM*jnrD;
178 /* load j atom coordinates */
179 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
180 x+j_coord_offsetC,x+j_coord_offsetD,
183 /* Calculate displacement vector */
184 dx00 = _mm_sub_ps(ix0,jx0);
185 dy00 = _mm_sub_ps(iy0,jy0);
186 dz00 = _mm_sub_ps(iz0,jz0);
188 /* Calculate squared distance and things based on it */
189 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
191 rinv00 = gmx_mm_invsqrt_ps(rsq00);
193 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
195 /* Load parameters for j particles */
196 vdwjidx0A = 2*vdwtype[jnrA+0];
197 vdwjidx0B = 2*vdwtype[jnrB+0];
198 vdwjidx0C = 2*vdwtype[jnrC+0];
199 vdwjidx0D = 2*vdwtype[jnrD+0];
201 /**************************
202 * CALCULATE INTERACTIONS *
203 **************************/
205 r00 = _mm_mul_ps(rsq00,rinv00);
207 /* Compute parameters for interactions between i and j atoms */
208 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
209 vdwparam+vdwioffset0+vdwjidx0B,
210 vdwparam+vdwioffset0+vdwjidx0C,
211 vdwparam+vdwioffset0+vdwjidx0D,
214 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
215 vdwgridparam+vdwioffset0+vdwjidx0B,
216 vdwgridparam+vdwioffset0+vdwjidx0C,
217 vdwgridparam+vdwioffset0+vdwjidx0D);
219 /* Analytical LJ-PME */
220 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
221 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
222 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
223 exponent = gmx_simd_exp_r(ewcljrsq);
224 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
225 poly = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
226 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
227 vvdw6 = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
228 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
229 vvdw = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
230 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
231 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
233 /* Update potential sum for this i atom from the interaction with this j atom. */
234 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
238 /* Calculate temporary vectorial force */
239 tx = _mm_mul_ps(fscal,dx00);
240 ty = _mm_mul_ps(fscal,dy00);
241 tz = _mm_mul_ps(fscal,dz00);
243 /* Update vectorial force */
244 fix0 = _mm_add_ps(fix0,tx);
245 fiy0 = _mm_add_ps(fiy0,ty);
246 fiz0 = _mm_add_ps(fiz0,tz);
248 fjptrA = f+j_coord_offsetA;
249 fjptrB = f+j_coord_offsetB;
250 fjptrC = f+j_coord_offsetC;
251 fjptrD = f+j_coord_offsetD;
252 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
254 /* Inner loop uses 51 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_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
328 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
329 vvdw6 = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
330 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
331 vvdw = _mm_sub_ps(_mm_mul_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_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),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 /* Calculate temporary vectorial force */
344 tx = _mm_mul_ps(fscal,dx00);
345 ty = _mm_mul_ps(fscal,dy00);
346 tz = _mm_mul_ps(fscal,dz00);
348 /* Update vectorial force */
349 fix0 = _mm_add_ps(fix0,tx);
350 fiy0 = _mm_add_ps(fiy0,ty);
351 fiz0 = _mm_add_ps(fiz0,tz);
353 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
354 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
355 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
356 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
357 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
359 /* Inner loop uses 52 flops */
362 /* End of innermost loop */
364 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
365 f+i_coord_offset,fshift+i_shift_offset);
368 /* Update potential energies */
369 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
371 /* Increment number of inner iterations */
372 inneriter += j_index_end - j_index_start;
374 /* Outer loop uses 7 flops */
377 /* Increment number of outer iterations */
380 /* Update outer/inner flops */
382 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*52);
385 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse4_1_single
386 * Electrostatics interaction: None
387 * VdW interaction: LJEwald
388 * Geometry: Particle-Particle
389 * Calculate force/pot: Force
392 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sse4_1_single
393 (t_nblist * gmx_restrict nlist,
394 rvec * gmx_restrict xx,
395 rvec * gmx_restrict ff,
396 t_forcerec * gmx_restrict fr,
397 t_mdatoms * gmx_restrict mdatoms,
398 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
399 t_nrnb * gmx_restrict nrnb)
401 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
402 * just 0 for non-waters.
403 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
404 * jnr indices corresponding to data put in the four positions in the SIMD register.
406 int i_shift_offset,i_coord_offset,outeriter,inneriter;
407 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
408 int jnrA,jnrB,jnrC,jnrD;
409 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
410 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
411 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
413 real *shiftvec,*fshift,*x,*f;
414 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
416 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
418 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
419 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
420 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
421 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
423 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
426 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
427 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
429 __m128 ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
431 __m128 one_half = _mm_set1_ps(0.5);
432 __m128 minus_one = _mm_set1_ps(-1.0);
433 __m128 dummy_mask,cutoff_mask;
434 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
435 __m128 one = _mm_set1_ps(1.0);
436 __m128 two = _mm_set1_ps(2.0);
442 jindex = nlist->jindex;
444 shiftidx = nlist->shift;
446 shiftvec = fr->shift_vec[0];
447 fshift = fr->fshift[0];
448 nvdwtype = fr->ntype;
450 vdwtype = mdatoms->typeA;
451 vdwgridparam = fr->ljpme_c6grid;
452 sh_lj_ewald = _mm_set1_ps(fr->ic->sh_lj_ewald);
453 ewclj = _mm_set1_ps(fr->ewaldcoeff_lj);
454 ewclj2 = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
456 /* Avoid stupid compiler warnings */
457 jnrA = jnrB = jnrC = jnrD = 0;
466 for(iidx=0;iidx<4*DIM;iidx++)
471 /* Start outer loop over neighborlists */
472 for(iidx=0; iidx<nri; iidx++)
474 /* Load shift vector for this list */
475 i_shift_offset = DIM*shiftidx[iidx];
477 /* Load limits for loop over neighbors */
478 j_index_start = jindex[iidx];
479 j_index_end = jindex[iidx+1];
481 /* Get outer coordinate index */
483 i_coord_offset = DIM*inr;
485 /* Load i particle coords and add shift vector */
486 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
488 fix0 = _mm_setzero_ps();
489 fiy0 = _mm_setzero_ps();
490 fiz0 = _mm_setzero_ps();
492 /* Load parameters for i particles */
493 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
495 /* Start inner kernel loop */
496 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
499 /* Get j neighbor index, and coordinate index */
504 j_coord_offsetA = DIM*jnrA;
505 j_coord_offsetB = DIM*jnrB;
506 j_coord_offsetC = DIM*jnrC;
507 j_coord_offsetD = DIM*jnrD;
509 /* load j atom coordinates */
510 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
511 x+j_coord_offsetC,x+j_coord_offsetD,
514 /* Calculate displacement vector */
515 dx00 = _mm_sub_ps(ix0,jx0);
516 dy00 = _mm_sub_ps(iy0,jy0);
517 dz00 = _mm_sub_ps(iz0,jz0);
519 /* Calculate squared distance and things based on it */
520 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
522 rinv00 = gmx_mm_invsqrt_ps(rsq00);
524 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
526 /* Load parameters for j particles */
527 vdwjidx0A = 2*vdwtype[jnrA+0];
528 vdwjidx0B = 2*vdwtype[jnrB+0];
529 vdwjidx0C = 2*vdwtype[jnrC+0];
530 vdwjidx0D = 2*vdwtype[jnrD+0];
532 /**************************
533 * CALCULATE INTERACTIONS *
534 **************************/
536 r00 = _mm_mul_ps(rsq00,rinv00);
538 /* Compute parameters for interactions between i and j atoms */
539 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
540 vdwparam+vdwioffset0+vdwjidx0B,
541 vdwparam+vdwioffset0+vdwjidx0C,
542 vdwparam+vdwioffset0+vdwjidx0D,
545 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
546 vdwgridparam+vdwioffset0+vdwjidx0B,
547 vdwgridparam+vdwioffset0+vdwjidx0C,
548 vdwgridparam+vdwioffset0+vdwjidx0D);
550 /* Analytical LJ-PME */
551 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
552 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
553 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
554 exponent = gmx_simd_exp_r(ewcljrsq);
555 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
556 poly = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
557 /* f6A = 6 * C6grid * (1 - poly) */
558 f6A = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
559 /* f6B = C6grid * exponent * beta^6 */
560 f6B = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
561 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
562 fvdw = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
566 /* Calculate temporary vectorial force */
567 tx = _mm_mul_ps(fscal,dx00);
568 ty = _mm_mul_ps(fscal,dy00);
569 tz = _mm_mul_ps(fscal,dz00);
571 /* Update vectorial force */
572 fix0 = _mm_add_ps(fix0,tx);
573 fiy0 = _mm_add_ps(fiy0,ty);
574 fiz0 = _mm_add_ps(fiz0,tz);
576 fjptrA = f+j_coord_offsetA;
577 fjptrB = f+j_coord_offsetB;
578 fjptrC = f+j_coord_offsetC;
579 fjptrD = f+j_coord_offsetD;
580 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
582 /* Inner loop uses 46 flops */
588 /* Get j neighbor index, and coordinate index */
589 jnrlistA = jjnr[jidx];
590 jnrlistB = jjnr[jidx+1];
591 jnrlistC = jjnr[jidx+2];
592 jnrlistD = jjnr[jidx+3];
593 /* Sign of each element will be negative for non-real atoms.
594 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
595 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
597 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
598 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
599 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
600 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
601 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
602 j_coord_offsetA = DIM*jnrA;
603 j_coord_offsetB = DIM*jnrB;
604 j_coord_offsetC = DIM*jnrC;
605 j_coord_offsetD = DIM*jnrD;
607 /* load j atom coordinates */
608 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
609 x+j_coord_offsetC,x+j_coord_offsetD,
612 /* Calculate displacement vector */
613 dx00 = _mm_sub_ps(ix0,jx0);
614 dy00 = _mm_sub_ps(iy0,jy0);
615 dz00 = _mm_sub_ps(iz0,jz0);
617 /* Calculate squared distance and things based on it */
618 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
620 rinv00 = gmx_mm_invsqrt_ps(rsq00);
622 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
624 /* Load parameters for j particles */
625 vdwjidx0A = 2*vdwtype[jnrA+0];
626 vdwjidx0B = 2*vdwtype[jnrB+0];
627 vdwjidx0C = 2*vdwtype[jnrC+0];
628 vdwjidx0D = 2*vdwtype[jnrD+0];
630 /**************************
631 * CALCULATE INTERACTIONS *
632 **************************/
634 r00 = _mm_mul_ps(rsq00,rinv00);
635 r00 = _mm_andnot_ps(dummy_mask,r00);
637 /* Compute parameters for interactions between i and j atoms */
638 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
639 vdwparam+vdwioffset0+vdwjidx0B,
640 vdwparam+vdwioffset0+vdwjidx0C,
641 vdwparam+vdwioffset0+vdwjidx0D,
644 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
645 vdwgridparam+vdwioffset0+vdwjidx0B,
646 vdwgridparam+vdwioffset0+vdwjidx0C,
647 vdwgridparam+vdwioffset0+vdwjidx0D);
649 /* Analytical LJ-PME */
650 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
651 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
652 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
653 exponent = gmx_simd_exp_r(ewcljrsq);
654 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
655 poly = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
656 /* f6A = 6 * C6grid * (1 - poly) */
657 f6A = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
658 /* f6B = C6grid * exponent * beta^6 */
659 f6B = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
660 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
661 fvdw = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
665 fscal = _mm_andnot_ps(dummy_mask,fscal);
667 /* Calculate temporary vectorial force */
668 tx = _mm_mul_ps(fscal,dx00);
669 ty = _mm_mul_ps(fscal,dy00);
670 tz = _mm_mul_ps(fscal,dz00);
672 /* Update vectorial force */
673 fix0 = _mm_add_ps(fix0,tx);
674 fiy0 = _mm_add_ps(fiy0,ty);
675 fiz0 = _mm_add_ps(fiz0,tz);
677 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
678 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
679 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
680 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
681 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
683 /* Inner loop uses 47 flops */
686 /* End of innermost loop */
688 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
689 f+i_coord_offset,fshift+i_shift_offset);
691 /* Increment number of inner iterations */
692 inneriter += j_index_end - j_index_start;
694 /* Outer loop uses 6 flops */
697 /* Increment number of outer iterations */
700 /* Update outer/inner flops */
702 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*47);