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 sse2_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_sse2_single.h"
48 #include "kernelutil_x86_sse2_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_sse2_single
52 * Electrostatics interaction: None
53 * VdW interaction: LJEwald
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_VF_sse2_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 rcutoff_scalar = fr->rvdw;
123 rcutoff = _mm_set1_ps(rcutoff_scalar);
124 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
126 sh_vdw_invrcut6 = _mm_set1_ps(fr->ic->sh_invrc6);
127 rvdw = _mm_set1_ps(fr->rvdw);
129 /* Avoid stupid compiler warnings */
130 jnrA = jnrB = jnrC = jnrD = 0;
139 for(iidx=0;iidx<4*DIM;iidx++)
144 /* Start outer loop over neighborlists */
145 for(iidx=0; iidx<nri; iidx++)
147 /* Load shift vector for this list */
148 i_shift_offset = DIM*shiftidx[iidx];
150 /* Load limits for loop over neighbors */
151 j_index_start = jindex[iidx];
152 j_index_end = jindex[iidx+1];
154 /* Get outer coordinate index */
156 i_coord_offset = DIM*inr;
158 /* Load i particle coords and add shift vector */
159 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
161 fix0 = _mm_setzero_ps();
162 fiy0 = _mm_setzero_ps();
163 fiz0 = _mm_setzero_ps();
165 /* Load parameters for i particles */
166 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
168 /* Reset potential sums */
169 vvdwsum = _mm_setzero_ps();
171 /* Start inner kernel loop */
172 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
175 /* Get j neighbor index, and coordinate index */
180 j_coord_offsetA = DIM*jnrA;
181 j_coord_offsetB = DIM*jnrB;
182 j_coord_offsetC = DIM*jnrC;
183 j_coord_offsetD = DIM*jnrD;
185 /* load j atom coordinates */
186 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
187 x+j_coord_offsetC,x+j_coord_offsetD,
190 /* Calculate displacement vector */
191 dx00 = _mm_sub_ps(ix0,jx0);
192 dy00 = _mm_sub_ps(iy0,jy0);
193 dz00 = _mm_sub_ps(iz0,jz0);
195 /* Calculate squared distance and things based on it */
196 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
198 rinv00 = gmx_mm_invsqrt_ps(rsq00);
200 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
202 /* Load parameters for j particles */
203 vdwjidx0A = 2*vdwtype[jnrA+0];
204 vdwjidx0B = 2*vdwtype[jnrB+0];
205 vdwjidx0C = 2*vdwtype[jnrC+0];
206 vdwjidx0D = 2*vdwtype[jnrD+0];
208 /**************************
209 * CALCULATE INTERACTIONS *
210 **************************/
212 if (gmx_mm_any_lt(rsq00,rcutoff2))
215 r00 = _mm_mul_ps(rsq00,rinv00);
217 /* Compute parameters for interactions between i and j atoms */
218 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
219 vdwparam+vdwioffset0+vdwjidx0B,
220 vdwparam+vdwioffset0+vdwjidx0C,
221 vdwparam+vdwioffset0+vdwjidx0D,
223 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
224 vdwgridparam+vdwioffset0+vdwjidx0B,
225 vdwgridparam+vdwioffset0+vdwjidx0C,
226 vdwgridparam+vdwioffset0+vdwjidx0D);
228 /* Analytical LJ-PME */
229 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
230 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
231 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
232 exponent = gmx_simd_exp_r(ewcljrsq);
233 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
234 poly = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
235 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
236 vvdw6 = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
237 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
238 vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
239 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
240 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
241 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);
243 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
245 /* Update potential sum for this i atom from the interaction with this j atom. */
246 vvdw = _mm_and_ps(vvdw,cutoff_mask);
247 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
251 fscal = _mm_and_ps(fscal,cutoff_mask);
253 /* Calculate temporary vectorial force */
254 tx = _mm_mul_ps(fscal,dx00);
255 ty = _mm_mul_ps(fscal,dy00);
256 tz = _mm_mul_ps(fscal,dz00);
258 /* Update vectorial force */
259 fix0 = _mm_add_ps(fix0,tx);
260 fiy0 = _mm_add_ps(fiy0,ty);
261 fiz0 = _mm_add_ps(fiz0,tz);
263 fjptrA = f+j_coord_offsetA;
264 fjptrB = f+j_coord_offsetB;
265 fjptrC = f+j_coord_offsetC;
266 fjptrD = f+j_coord_offsetD;
267 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
271 /* Inner loop uses 62 flops */
277 /* Get j neighbor index, and coordinate index */
278 jnrlistA = jjnr[jidx];
279 jnrlistB = jjnr[jidx+1];
280 jnrlistC = jjnr[jidx+2];
281 jnrlistD = jjnr[jidx+3];
282 /* Sign of each element will be negative for non-real atoms.
283 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
284 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
286 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
287 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
288 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
289 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
290 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
291 j_coord_offsetA = DIM*jnrA;
292 j_coord_offsetB = DIM*jnrB;
293 j_coord_offsetC = DIM*jnrC;
294 j_coord_offsetD = DIM*jnrD;
296 /* load j atom coordinates */
297 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
298 x+j_coord_offsetC,x+j_coord_offsetD,
301 /* Calculate displacement vector */
302 dx00 = _mm_sub_ps(ix0,jx0);
303 dy00 = _mm_sub_ps(iy0,jy0);
304 dz00 = _mm_sub_ps(iz0,jz0);
306 /* Calculate squared distance and things based on it */
307 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
309 rinv00 = gmx_mm_invsqrt_ps(rsq00);
311 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
313 /* Load parameters for j particles */
314 vdwjidx0A = 2*vdwtype[jnrA+0];
315 vdwjidx0B = 2*vdwtype[jnrB+0];
316 vdwjidx0C = 2*vdwtype[jnrC+0];
317 vdwjidx0D = 2*vdwtype[jnrD+0];
319 /**************************
320 * CALCULATE INTERACTIONS *
321 **************************/
323 if (gmx_mm_any_lt(rsq00,rcutoff2))
326 r00 = _mm_mul_ps(rsq00,rinv00);
327 r00 = _mm_andnot_ps(dummy_mask,r00);
329 /* Compute parameters for interactions between i and j atoms */
330 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
331 vdwparam+vdwioffset0+vdwjidx0B,
332 vdwparam+vdwioffset0+vdwjidx0C,
333 vdwparam+vdwioffset0+vdwjidx0D,
335 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
336 vdwgridparam+vdwioffset0+vdwjidx0B,
337 vdwgridparam+vdwioffset0+vdwjidx0C,
338 vdwgridparam+vdwioffset0+vdwjidx0D);
340 /* Analytical LJ-PME */
341 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
342 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
343 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
344 exponent = gmx_simd_exp_r(ewcljrsq);
345 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
346 poly = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
347 /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
348 vvdw6 = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
349 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
350 vvdw = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
351 _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_00,sh_vdw_invrcut6),_mm_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
352 /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
353 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);
355 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
357 /* Update potential sum for this i atom from the interaction with this j atom. */
358 vvdw = _mm_and_ps(vvdw,cutoff_mask);
359 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
360 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
364 fscal = _mm_and_ps(fscal,cutoff_mask);
366 fscal = _mm_andnot_ps(dummy_mask,fscal);
368 /* Calculate temporary vectorial force */
369 tx = _mm_mul_ps(fscal,dx00);
370 ty = _mm_mul_ps(fscal,dy00);
371 tz = _mm_mul_ps(fscal,dz00);
373 /* Update vectorial force */
374 fix0 = _mm_add_ps(fix0,tx);
375 fiy0 = _mm_add_ps(fiy0,ty);
376 fiz0 = _mm_add_ps(fiz0,tz);
378 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
379 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
380 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
381 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
382 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
386 /* Inner loop uses 63 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*63);
412 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sse2_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_sse2_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 SSE, 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 tx,ty,tz,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);
456 __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,
581 c6grid_00 = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
582 vdwgridparam+vdwioffset0+vdwjidx0B,
583 vdwgridparam+vdwioffset0+vdwjidx0C,
584 vdwgridparam+vdwioffset0+vdwjidx0D);
586 /* Analytical LJ-PME */
587 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
588 ewcljrsq = _mm_mul_ps(ewclj2,rsq00);
589 ewclj6 = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
590 exponent = gmx_simd_exp_r(ewcljrsq);
591 /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
592 poly = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
593 /* f6A = 6 * C6grid * (1 - poly) */
594 f6A = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
595 /* f6B = C6grid * exponent * beta^6 */
596 f6B = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
597 /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
598 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);
600 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
604 fscal = _mm_and_ps(fscal,cutoff_mask);
606 /* Calculate temporary vectorial force */
607 tx = _mm_mul_ps(fscal,dx00);
608 ty = _mm_mul_ps(fscal,dy00);
609 tz = _mm_mul_ps(fscal,dz00);
611 /* Update vectorial force */
612 fix0 = _mm_add_ps(fix0,tx);
613 fiy0 = _mm_add_ps(fiy0,ty);
614 fiz0 = _mm_add_ps(fiz0,tz);
616 fjptrA = f+j_coord_offsetA;
617 fjptrB = f+j_coord_offsetB;
618 fjptrC = f+j_coord_offsetC;
619 fjptrD = f+j_coord_offsetD;
620 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
624 /* Inner loop uses 49 flops */
630 /* Get j neighbor index, and coordinate index */
631 jnrlistA = jjnr[jidx];
632 jnrlistB = jjnr[jidx+1];
633 jnrlistC = jjnr[jidx+2];
634 jnrlistD = jjnr[jidx+3];
635 /* Sign of each element will be negative for non-real atoms.
636 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
637 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
639 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
640 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
641 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
642 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
643 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
644 j_coord_offsetA = DIM*jnrA;
645 j_coord_offsetB = DIM*jnrB;
646 j_coord_offsetC = DIM*jnrC;
647 j_coord_offsetD = DIM*jnrD;
649 /* load j atom coordinates */
650 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
651 x+j_coord_offsetC,x+j_coord_offsetD,
654 /* Calculate displacement vector */
655 dx00 = _mm_sub_ps(ix0,jx0);
656 dy00 = _mm_sub_ps(iy0,jy0);
657 dz00 = _mm_sub_ps(iz0,jz0);
659 /* Calculate squared distance and things based on it */
660 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
662 rinv00 = gmx_mm_invsqrt_ps(rsq00);
664 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
666 /* Load parameters for j particles */
667 vdwjidx0A = 2*vdwtype[jnrA+0];
668 vdwjidx0B = 2*vdwtype[jnrB+0];
669 vdwjidx0C = 2*vdwtype[jnrC+0];
670 vdwjidx0D = 2*vdwtype[jnrD+0];
672 /**************************
673 * CALCULATE INTERACTIONS *
674 **************************/
676 if (gmx_mm_any_lt(rsq00,rcutoff2))
679 r00 = _mm_mul_ps(rsq00,rinv00);
680 r00 = _mm_andnot_ps(dummy_mask,r00);
682 /* Compute parameters for interactions between i and j atoms */
683 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
684 vdwparam+vdwioffset0+vdwjidx0B,
685 vdwparam+vdwioffset0+vdwjidx0C,
686 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_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
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_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_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 /* Calculate temporary vectorial force */
716 tx = _mm_mul_ps(fscal,dx00);
717 ty = _mm_mul_ps(fscal,dy00);
718 tz = _mm_mul_ps(fscal,dz00);
720 /* Update vectorial force */
721 fix0 = _mm_add_ps(fix0,tx);
722 fiy0 = _mm_add_ps(fiy0,ty);
723 fiz0 = _mm_add_ps(fiz0,tz);
725 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
726 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
727 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
728 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
729 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
733 /* Inner loop uses 50 flops */
736 /* End of innermost loop */
738 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
739 f+i_coord_offset,fshift+i_shift_offset);
741 /* Increment number of inner iterations */
742 inneriter += j_index_end - j_index_start;
744 /* Outer loop uses 6 flops */
747 /* Increment number of outer iterations */
750 /* Update outer/inner flops */
752 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*50);