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.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
46 #include "gromacs/math/vec.h"
49 #include "gromacs/simd/math_x86_sse2_single.h"
50 #include "kernelutil_x86_sse2_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse2_single
54 * Electrostatics interaction: Coulomb
55 * VdW interaction: CubicSplineTable
56 * Geometry: Particle-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse2_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 SSE, 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 tx,ty,tz,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;
90 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
93 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
96 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
97 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
99 __m128i ifour = _mm_set1_epi32(4);
100 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
102 __m128 dummy_mask,cutoff_mask;
103 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
104 __m128 one = _mm_set1_ps(1.0);
105 __m128 two = _mm_set1_ps(2.0);
111 jindex = nlist->jindex;
113 shiftidx = nlist->shift;
115 shiftvec = fr->shift_vec[0];
116 fshift = fr->fshift[0];
117 facel = _mm_set1_ps(fr->epsfac);
118 charge = mdatoms->chargeA;
119 nvdwtype = fr->ntype;
121 vdwtype = mdatoms->typeA;
123 vftab = kernel_data->table_vdw->data;
124 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
126 /* Avoid stupid compiler warnings */
127 jnrA = jnrB = jnrC = jnrD = 0;
136 for(iidx=0;iidx<4*DIM;iidx++)
141 /* Start outer loop over neighborlists */
142 for(iidx=0; iidx<nri; iidx++)
144 /* Load shift vector for this list */
145 i_shift_offset = DIM*shiftidx[iidx];
147 /* Load limits for loop over neighbors */
148 j_index_start = jindex[iidx];
149 j_index_end = jindex[iidx+1];
151 /* Get outer coordinate index */
153 i_coord_offset = DIM*inr;
155 /* Load i particle coords and add shift vector */
156 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
158 fix0 = _mm_setzero_ps();
159 fiy0 = _mm_setzero_ps();
160 fiz0 = _mm_setzero_ps();
162 /* Load parameters for i particles */
163 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
164 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
166 /* Reset potential sums */
167 velecsum = _mm_setzero_ps();
168 vvdwsum = _mm_setzero_ps();
170 /* Start inner kernel loop */
171 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
174 /* Get j neighbor index, and coordinate index */
179 j_coord_offsetA = DIM*jnrA;
180 j_coord_offsetB = DIM*jnrB;
181 j_coord_offsetC = DIM*jnrC;
182 j_coord_offsetD = DIM*jnrD;
184 /* load j atom coordinates */
185 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
186 x+j_coord_offsetC,x+j_coord_offsetD,
189 /* Calculate displacement vector */
190 dx00 = _mm_sub_ps(ix0,jx0);
191 dy00 = _mm_sub_ps(iy0,jy0);
192 dz00 = _mm_sub_ps(iz0,jz0);
194 /* Calculate squared distance and things based on it */
195 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
197 rinv00 = gmx_mm_invsqrt_ps(rsq00);
199 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
201 /* Load parameters for j particles */
202 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
203 charge+jnrC+0,charge+jnrD+0);
204 vdwjidx0A = 2*vdwtype[jnrA+0];
205 vdwjidx0B = 2*vdwtype[jnrB+0];
206 vdwjidx0C = 2*vdwtype[jnrC+0];
207 vdwjidx0D = 2*vdwtype[jnrD+0];
209 /**************************
210 * CALCULATE INTERACTIONS *
211 **************************/
213 r00 = _mm_mul_ps(rsq00,rinv00);
215 /* Compute parameters for interactions between i and j atoms */
216 qq00 = _mm_mul_ps(iq0,jq0);
217 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
218 vdwparam+vdwioffset0+vdwjidx0B,
219 vdwparam+vdwioffset0+vdwjidx0C,
220 vdwparam+vdwioffset0+vdwjidx0D,
223 /* Calculate table index by multiplying r with table scale and truncate to integer */
224 rt = _mm_mul_ps(r00,vftabscale);
225 vfitab = _mm_cvttps_epi32(rt);
226 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
227 vfitab = _mm_slli_epi32(vfitab,3);
229 /* COULOMB ELECTROSTATICS */
230 velec = _mm_mul_ps(qq00,rinv00);
231 felec = _mm_mul_ps(velec,rinvsq00);
233 /* CUBIC SPLINE TABLE DISPERSION */
234 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
235 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
236 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
237 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
238 _MM_TRANSPOSE4_PS(Y,F,G,H);
239 Heps = _mm_mul_ps(vfeps,H);
240 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
241 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
242 vvdw6 = _mm_mul_ps(c6_00,VV);
243 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
244 fvdw6 = _mm_mul_ps(c6_00,FF);
246 /* CUBIC SPLINE TABLE REPULSION */
247 vfitab = _mm_add_epi32(vfitab,ifour);
248 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
249 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
250 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
251 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
252 _MM_TRANSPOSE4_PS(Y,F,G,H);
253 Heps = _mm_mul_ps(vfeps,H);
254 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
255 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
256 vvdw12 = _mm_mul_ps(c12_00,VV);
257 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
258 fvdw12 = _mm_mul_ps(c12_00,FF);
259 vvdw = _mm_add_ps(vvdw12,vvdw6);
260 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
262 /* Update potential sum for this i atom from the interaction with this j atom. */
263 velecsum = _mm_add_ps(velecsum,velec);
264 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
266 fscal = _mm_add_ps(felec,fvdw);
268 /* Calculate temporary vectorial force */
269 tx = _mm_mul_ps(fscal,dx00);
270 ty = _mm_mul_ps(fscal,dy00);
271 tz = _mm_mul_ps(fscal,dz00);
273 /* Update vectorial force */
274 fix0 = _mm_add_ps(fix0,tx);
275 fiy0 = _mm_add_ps(fiy0,ty);
276 fiz0 = _mm_add_ps(fiz0,tz);
278 fjptrA = f+j_coord_offsetA;
279 fjptrB = f+j_coord_offsetB;
280 fjptrC = f+j_coord_offsetC;
281 fjptrD = f+j_coord_offsetD;
282 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
284 /* Inner loop uses 63 flops */
290 /* Get j neighbor index, and coordinate index */
291 jnrlistA = jjnr[jidx];
292 jnrlistB = jjnr[jidx+1];
293 jnrlistC = jjnr[jidx+2];
294 jnrlistD = jjnr[jidx+3];
295 /* Sign of each element will be negative for non-real atoms.
296 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
297 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
299 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
300 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
301 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
302 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
303 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
304 j_coord_offsetA = DIM*jnrA;
305 j_coord_offsetB = DIM*jnrB;
306 j_coord_offsetC = DIM*jnrC;
307 j_coord_offsetD = DIM*jnrD;
309 /* load j atom coordinates */
310 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
311 x+j_coord_offsetC,x+j_coord_offsetD,
314 /* Calculate displacement vector */
315 dx00 = _mm_sub_ps(ix0,jx0);
316 dy00 = _mm_sub_ps(iy0,jy0);
317 dz00 = _mm_sub_ps(iz0,jz0);
319 /* Calculate squared distance and things based on it */
320 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
322 rinv00 = gmx_mm_invsqrt_ps(rsq00);
324 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
326 /* Load parameters for j particles */
327 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
328 charge+jnrC+0,charge+jnrD+0);
329 vdwjidx0A = 2*vdwtype[jnrA+0];
330 vdwjidx0B = 2*vdwtype[jnrB+0];
331 vdwjidx0C = 2*vdwtype[jnrC+0];
332 vdwjidx0D = 2*vdwtype[jnrD+0];
334 /**************************
335 * CALCULATE INTERACTIONS *
336 **************************/
338 r00 = _mm_mul_ps(rsq00,rinv00);
339 r00 = _mm_andnot_ps(dummy_mask,r00);
341 /* Compute parameters for interactions between i and j atoms */
342 qq00 = _mm_mul_ps(iq0,jq0);
343 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
344 vdwparam+vdwioffset0+vdwjidx0B,
345 vdwparam+vdwioffset0+vdwjidx0C,
346 vdwparam+vdwioffset0+vdwjidx0D,
349 /* Calculate table index by multiplying r with table scale and truncate to integer */
350 rt = _mm_mul_ps(r00,vftabscale);
351 vfitab = _mm_cvttps_epi32(rt);
352 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
353 vfitab = _mm_slli_epi32(vfitab,3);
355 /* COULOMB ELECTROSTATICS */
356 velec = _mm_mul_ps(qq00,rinv00);
357 felec = _mm_mul_ps(velec,rinvsq00);
359 /* CUBIC SPLINE TABLE DISPERSION */
360 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
361 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
362 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
363 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
364 _MM_TRANSPOSE4_PS(Y,F,G,H);
365 Heps = _mm_mul_ps(vfeps,H);
366 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
367 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
368 vvdw6 = _mm_mul_ps(c6_00,VV);
369 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
370 fvdw6 = _mm_mul_ps(c6_00,FF);
372 /* CUBIC SPLINE TABLE REPULSION */
373 vfitab = _mm_add_epi32(vfitab,ifour);
374 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
375 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
376 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
377 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
378 _MM_TRANSPOSE4_PS(Y,F,G,H);
379 Heps = _mm_mul_ps(vfeps,H);
380 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
381 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
382 vvdw12 = _mm_mul_ps(c12_00,VV);
383 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
384 fvdw12 = _mm_mul_ps(c12_00,FF);
385 vvdw = _mm_add_ps(vvdw12,vvdw6);
386 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
388 /* Update potential sum for this i atom from the interaction with this j atom. */
389 velec = _mm_andnot_ps(dummy_mask,velec);
390 velecsum = _mm_add_ps(velecsum,velec);
391 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
392 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
394 fscal = _mm_add_ps(felec,fvdw);
396 fscal = _mm_andnot_ps(dummy_mask,fscal);
398 /* Calculate temporary vectorial force */
399 tx = _mm_mul_ps(fscal,dx00);
400 ty = _mm_mul_ps(fscal,dy00);
401 tz = _mm_mul_ps(fscal,dz00);
403 /* Update vectorial force */
404 fix0 = _mm_add_ps(fix0,tx);
405 fiy0 = _mm_add_ps(fiy0,ty);
406 fiz0 = _mm_add_ps(fiz0,tz);
408 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
409 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
410 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
411 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
412 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
414 /* Inner loop uses 64 flops */
417 /* End of innermost loop */
419 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
420 f+i_coord_offset,fshift+i_shift_offset);
423 /* Update potential energies */
424 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
425 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
427 /* Increment number of inner iterations */
428 inneriter += j_index_end - j_index_start;
430 /* Outer loop uses 9 flops */
433 /* Increment number of outer iterations */
436 /* Update outer/inner flops */
438 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*64);
441 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse2_single
442 * Electrostatics interaction: Coulomb
443 * VdW interaction: CubicSplineTable
444 * Geometry: Particle-Particle
445 * Calculate force/pot: Force
448 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse2_single
449 (t_nblist * gmx_restrict nlist,
450 rvec * gmx_restrict xx,
451 rvec * gmx_restrict ff,
452 t_forcerec * gmx_restrict fr,
453 t_mdatoms * gmx_restrict mdatoms,
454 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
455 t_nrnb * gmx_restrict nrnb)
457 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
458 * just 0 for non-waters.
459 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
460 * jnr indices corresponding to data put in the four positions in the SIMD register.
462 int i_shift_offset,i_coord_offset,outeriter,inneriter;
463 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
464 int jnrA,jnrB,jnrC,jnrD;
465 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
466 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
467 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
469 real *shiftvec,*fshift,*x,*f;
470 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
472 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
474 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
475 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
476 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
477 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
478 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
481 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
484 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
485 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
487 __m128i ifour = _mm_set1_epi32(4);
488 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
490 __m128 dummy_mask,cutoff_mask;
491 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
492 __m128 one = _mm_set1_ps(1.0);
493 __m128 two = _mm_set1_ps(2.0);
499 jindex = nlist->jindex;
501 shiftidx = nlist->shift;
503 shiftvec = fr->shift_vec[0];
504 fshift = fr->fshift[0];
505 facel = _mm_set1_ps(fr->epsfac);
506 charge = mdatoms->chargeA;
507 nvdwtype = fr->ntype;
509 vdwtype = mdatoms->typeA;
511 vftab = kernel_data->table_vdw->data;
512 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
514 /* Avoid stupid compiler warnings */
515 jnrA = jnrB = jnrC = jnrD = 0;
524 for(iidx=0;iidx<4*DIM;iidx++)
529 /* Start outer loop over neighborlists */
530 for(iidx=0; iidx<nri; iidx++)
532 /* Load shift vector for this list */
533 i_shift_offset = DIM*shiftidx[iidx];
535 /* Load limits for loop over neighbors */
536 j_index_start = jindex[iidx];
537 j_index_end = jindex[iidx+1];
539 /* Get outer coordinate index */
541 i_coord_offset = DIM*inr;
543 /* Load i particle coords and add shift vector */
544 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
546 fix0 = _mm_setzero_ps();
547 fiy0 = _mm_setzero_ps();
548 fiz0 = _mm_setzero_ps();
550 /* Load parameters for i particles */
551 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
552 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
554 /* Start inner kernel loop */
555 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
558 /* Get j neighbor index, and coordinate index */
563 j_coord_offsetA = DIM*jnrA;
564 j_coord_offsetB = DIM*jnrB;
565 j_coord_offsetC = DIM*jnrC;
566 j_coord_offsetD = DIM*jnrD;
568 /* load j atom coordinates */
569 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
570 x+j_coord_offsetC,x+j_coord_offsetD,
573 /* Calculate displacement vector */
574 dx00 = _mm_sub_ps(ix0,jx0);
575 dy00 = _mm_sub_ps(iy0,jy0);
576 dz00 = _mm_sub_ps(iz0,jz0);
578 /* Calculate squared distance and things based on it */
579 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
581 rinv00 = gmx_mm_invsqrt_ps(rsq00);
583 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
585 /* Load parameters for j particles */
586 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
587 charge+jnrC+0,charge+jnrD+0);
588 vdwjidx0A = 2*vdwtype[jnrA+0];
589 vdwjidx0B = 2*vdwtype[jnrB+0];
590 vdwjidx0C = 2*vdwtype[jnrC+0];
591 vdwjidx0D = 2*vdwtype[jnrD+0];
593 /**************************
594 * CALCULATE INTERACTIONS *
595 **************************/
597 r00 = _mm_mul_ps(rsq00,rinv00);
599 /* Compute parameters for interactions between i and j atoms */
600 qq00 = _mm_mul_ps(iq0,jq0);
601 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
602 vdwparam+vdwioffset0+vdwjidx0B,
603 vdwparam+vdwioffset0+vdwjidx0C,
604 vdwparam+vdwioffset0+vdwjidx0D,
607 /* Calculate table index by multiplying r with table scale and truncate to integer */
608 rt = _mm_mul_ps(r00,vftabscale);
609 vfitab = _mm_cvttps_epi32(rt);
610 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
611 vfitab = _mm_slli_epi32(vfitab,3);
613 /* COULOMB ELECTROSTATICS */
614 velec = _mm_mul_ps(qq00,rinv00);
615 felec = _mm_mul_ps(velec,rinvsq00);
617 /* CUBIC SPLINE TABLE DISPERSION */
618 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
619 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
620 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
621 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
622 _MM_TRANSPOSE4_PS(Y,F,G,H);
623 Heps = _mm_mul_ps(vfeps,H);
624 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
625 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
626 fvdw6 = _mm_mul_ps(c6_00,FF);
628 /* CUBIC SPLINE TABLE REPULSION */
629 vfitab = _mm_add_epi32(vfitab,ifour);
630 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
631 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
632 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
633 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
634 _MM_TRANSPOSE4_PS(Y,F,G,H);
635 Heps = _mm_mul_ps(vfeps,H);
636 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
637 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
638 fvdw12 = _mm_mul_ps(c12_00,FF);
639 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
641 fscal = _mm_add_ps(felec,fvdw);
643 /* Calculate temporary vectorial force */
644 tx = _mm_mul_ps(fscal,dx00);
645 ty = _mm_mul_ps(fscal,dy00);
646 tz = _mm_mul_ps(fscal,dz00);
648 /* Update vectorial force */
649 fix0 = _mm_add_ps(fix0,tx);
650 fiy0 = _mm_add_ps(fiy0,ty);
651 fiz0 = _mm_add_ps(fiz0,tz);
653 fjptrA = f+j_coord_offsetA;
654 fjptrB = f+j_coord_offsetB;
655 fjptrC = f+j_coord_offsetC;
656 fjptrD = f+j_coord_offsetD;
657 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
659 /* Inner loop uses 54 flops */
665 /* Get j neighbor index, and coordinate index */
666 jnrlistA = jjnr[jidx];
667 jnrlistB = jjnr[jidx+1];
668 jnrlistC = jjnr[jidx+2];
669 jnrlistD = jjnr[jidx+3];
670 /* Sign of each element will be negative for non-real atoms.
671 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
672 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
674 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
675 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
676 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
677 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
678 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
679 j_coord_offsetA = DIM*jnrA;
680 j_coord_offsetB = DIM*jnrB;
681 j_coord_offsetC = DIM*jnrC;
682 j_coord_offsetD = DIM*jnrD;
684 /* load j atom coordinates */
685 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
686 x+j_coord_offsetC,x+j_coord_offsetD,
689 /* Calculate displacement vector */
690 dx00 = _mm_sub_ps(ix0,jx0);
691 dy00 = _mm_sub_ps(iy0,jy0);
692 dz00 = _mm_sub_ps(iz0,jz0);
694 /* Calculate squared distance and things based on it */
695 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
697 rinv00 = gmx_mm_invsqrt_ps(rsq00);
699 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
701 /* Load parameters for j particles */
702 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
703 charge+jnrC+0,charge+jnrD+0);
704 vdwjidx0A = 2*vdwtype[jnrA+0];
705 vdwjidx0B = 2*vdwtype[jnrB+0];
706 vdwjidx0C = 2*vdwtype[jnrC+0];
707 vdwjidx0D = 2*vdwtype[jnrD+0];
709 /**************************
710 * CALCULATE INTERACTIONS *
711 **************************/
713 r00 = _mm_mul_ps(rsq00,rinv00);
714 r00 = _mm_andnot_ps(dummy_mask,r00);
716 /* Compute parameters for interactions between i and j atoms */
717 qq00 = _mm_mul_ps(iq0,jq0);
718 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
719 vdwparam+vdwioffset0+vdwjidx0B,
720 vdwparam+vdwioffset0+vdwjidx0C,
721 vdwparam+vdwioffset0+vdwjidx0D,
724 /* Calculate table index by multiplying r with table scale and truncate to integer */
725 rt = _mm_mul_ps(r00,vftabscale);
726 vfitab = _mm_cvttps_epi32(rt);
727 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
728 vfitab = _mm_slli_epi32(vfitab,3);
730 /* COULOMB ELECTROSTATICS */
731 velec = _mm_mul_ps(qq00,rinv00);
732 felec = _mm_mul_ps(velec,rinvsq00);
734 /* CUBIC SPLINE TABLE DISPERSION */
735 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
736 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
737 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
738 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
739 _MM_TRANSPOSE4_PS(Y,F,G,H);
740 Heps = _mm_mul_ps(vfeps,H);
741 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
742 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
743 fvdw6 = _mm_mul_ps(c6_00,FF);
745 /* CUBIC SPLINE TABLE REPULSION */
746 vfitab = _mm_add_epi32(vfitab,ifour);
747 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
748 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
749 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
750 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
751 _MM_TRANSPOSE4_PS(Y,F,G,H);
752 Heps = _mm_mul_ps(vfeps,H);
753 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
754 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
755 fvdw12 = _mm_mul_ps(c12_00,FF);
756 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
758 fscal = _mm_add_ps(felec,fvdw);
760 fscal = _mm_andnot_ps(dummy_mask,fscal);
762 /* Calculate temporary vectorial force */
763 tx = _mm_mul_ps(fscal,dx00);
764 ty = _mm_mul_ps(fscal,dy00);
765 tz = _mm_mul_ps(fscal,dz00);
767 /* Update vectorial force */
768 fix0 = _mm_add_ps(fix0,tx);
769 fiy0 = _mm_add_ps(fiy0,ty);
770 fiz0 = _mm_add_ps(fiz0,tz);
772 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
773 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
774 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
775 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
776 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
778 /* Inner loop uses 55 flops */
781 /* End of innermost loop */
783 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
784 f+i_coord_offset,fshift+i_shift_offset);
786 /* Increment number of inner iterations */
787 inneriter += j_index_end - j_index_start;
789 /* Outer loop uses 7 flops */
792 /* Increment number of outer iterations */
795 /* Update outer/inner flops */
797 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*55);