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_ElecRF_VdwCSTab_GeomP1P1_VF_sse2_single
52 * Electrostatics interaction: ReactionField
53 * VdW interaction: CubicSplineTable
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecRF_VdwCSTab_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;
88 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
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);
97 __m128i ifour = _mm_set1_epi32(4);
98 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
100 __m128 dummy_mask,cutoff_mask;
101 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
102 __m128 one = _mm_set1_ps(1.0);
103 __m128 two = _mm_set1_ps(2.0);
109 jindex = nlist->jindex;
111 shiftidx = nlist->shift;
113 shiftvec = fr->shift_vec[0];
114 fshift = fr->fshift[0];
115 facel = _mm_set1_ps(fr->epsfac);
116 charge = mdatoms->chargeA;
117 krf = _mm_set1_ps(fr->ic->k_rf);
118 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
119 crf = _mm_set1_ps(fr->ic->c_rf);
120 nvdwtype = fr->ntype;
122 vdwtype = mdatoms->typeA;
124 vftab = kernel_data->table_vdw->data;
125 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
127 /* Avoid stupid compiler warnings */
128 jnrA = jnrB = jnrC = jnrD = 0;
137 for(iidx=0;iidx<4*DIM;iidx++)
142 /* Start outer loop over neighborlists */
143 for(iidx=0; iidx<nri; iidx++)
145 /* Load shift vector for this list */
146 i_shift_offset = DIM*shiftidx[iidx];
148 /* Load limits for loop over neighbors */
149 j_index_start = jindex[iidx];
150 j_index_end = jindex[iidx+1];
152 /* Get outer coordinate index */
154 i_coord_offset = DIM*inr;
156 /* Load i particle coords and add shift vector */
157 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
159 fix0 = _mm_setzero_ps();
160 fiy0 = _mm_setzero_ps();
161 fiz0 = _mm_setzero_ps();
163 /* Load parameters for i particles */
164 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
165 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
167 /* Reset potential sums */
168 velecsum = _mm_setzero_ps();
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 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
204 charge+jnrC+0,charge+jnrD+0);
205 vdwjidx0A = 2*vdwtype[jnrA+0];
206 vdwjidx0B = 2*vdwtype[jnrB+0];
207 vdwjidx0C = 2*vdwtype[jnrC+0];
208 vdwjidx0D = 2*vdwtype[jnrD+0];
210 /**************************
211 * CALCULATE INTERACTIONS *
212 **************************/
214 r00 = _mm_mul_ps(rsq00,rinv00);
216 /* Compute parameters for interactions between i and j atoms */
217 qq00 = _mm_mul_ps(iq0,jq0);
218 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
219 vdwparam+vdwioffset0+vdwjidx0B,
220 vdwparam+vdwioffset0+vdwjidx0C,
221 vdwparam+vdwioffset0+vdwjidx0D,
224 /* Calculate table index by multiplying r with table scale and truncate to integer */
225 rt = _mm_mul_ps(r00,vftabscale);
226 vfitab = _mm_cvttps_epi32(rt);
227 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
228 vfitab = _mm_slli_epi32(vfitab,3);
230 /* REACTION-FIELD ELECTROSTATICS */
231 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
232 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
234 /* CUBIC SPLINE TABLE DISPERSION */
235 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
236 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
237 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
238 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
239 _MM_TRANSPOSE4_PS(Y,F,G,H);
240 Heps = _mm_mul_ps(vfeps,H);
241 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
242 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
243 vvdw6 = _mm_mul_ps(c6_00,VV);
244 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
245 fvdw6 = _mm_mul_ps(c6_00,FF);
247 /* CUBIC SPLINE TABLE REPULSION */
248 vfitab = _mm_add_epi32(vfitab,ifour);
249 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
250 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
251 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
252 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
253 _MM_TRANSPOSE4_PS(Y,F,G,H);
254 Heps = _mm_mul_ps(vfeps,H);
255 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
256 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
257 vvdw12 = _mm_mul_ps(c12_00,VV);
258 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
259 fvdw12 = _mm_mul_ps(c12_00,FF);
260 vvdw = _mm_add_ps(vvdw12,vvdw6);
261 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
263 /* Update potential sum for this i atom from the interaction with this j atom. */
264 velecsum = _mm_add_ps(velecsum,velec);
265 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
267 fscal = _mm_add_ps(felec,fvdw);
269 /* Calculate temporary vectorial force */
270 tx = _mm_mul_ps(fscal,dx00);
271 ty = _mm_mul_ps(fscal,dy00);
272 tz = _mm_mul_ps(fscal,dz00);
274 /* Update vectorial force */
275 fix0 = _mm_add_ps(fix0,tx);
276 fiy0 = _mm_add_ps(fiy0,ty);
277 fiz0 = _mm_add_ps(fiz0,tz);
279 fjptrA = f+j_coord_offsetA;
280 fjptrB = f+j_coord_offsetB;
281 fjptrC = f+j_coord_offsetC;
282 fjptrD = f+j_coord_offsetD;
283 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
285 /* Inner loop uses 67 flops */
291 /* Get j neighbor index, and coordinate index */
292 jnrlistA = jjnr[jidx];
293 jnrlistB = jjnr[jidx+1];
294 jnrlistC = jjnr[jidx+2];
295 jnrlistD = jjnr[jidx+3];
296 /* Sign of each element will be negative for non-real atoms.
297 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
298 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
300 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
301 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
302 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
303 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
304 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
305 j_coord_offsetA = DIM*jnrA;
306 j_coord_offsetB = DIM*jnrB;
307 j_coord_offsetC = DIM*jnrC;
308 j_coord_offsetD = DIM*jnrD;
310 /* load j atom coordinates */
311 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
312 x+j_coord_offsetC,x+j_coord_offsetD,
315 /* Calculate displacement vector */
316 dx00 = _mm_sub_ps(ix0,jx0);
317 dy00 = _mm_sub_ps(iy0,jy0);
318 dz00 = _mm_sub_ps(iz0,jz0);
320 /* Calculate squared distance and things based on it */
321 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
323 rinv00 = gmx_mm_invsqrt_ps(rsq00);
325 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
327 /* Load parameters for j particles */
328 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
329 charge+jnrC+0,charge+jnrD+0);
330 vdwjidx0A = 2*vdwtype[jnrA+0];
331 vdwjidx0B = 2*vdwtype[jnrB+0];
332 vdwjidx0C = 2*vdwtype[jnrC+0];
333 vdwjidx0D = 2*vdwtype[jnrD+0];
335 /**************************
336 * CALCULATE INTERACTIONS *
337 **************************/
339 r00 = _mm_mul_ps(rsq00,rinv00);
340 r00 = _mm_andnot_ps(dummy_mask,r00);
342 /* Compute parameters for interactions between i and j atoms */
343 qq00 = _mm_mul_ps(iq0,jq0);
344 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
345 vdwparam+vdwioffset0+vdwjidx0B,
346 vdwparam+vdwioffset0+vdwjidx0C,
347 vdwparam+vdwioffset0+vdwjidx0D,
350 /* Calculate table index by multiplying r with table scale and truncate to integer */
351 rt = _mm_mul_ps(r00,vftabscale);
352 vfitab = _mm_cvttps_epi32(rt);
353 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
354 vfitab = _mm_slli_epi32(vfitab,3);
356 /* REACTION-FIELD ELECTROSTATICS */
357 velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
358 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
360 /* CUBIC SPLINE TABLE DISPERSION */
361 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
362 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
363 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
364 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
365 _MM_TRANSPOSE4_PS(Y,F,G,H);
366 Heps = _mm_mul_ps(vfeps,H);
367 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
368 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
369 vvdw6 = _mm_mul_ps(c6_00,VV);
370 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
371 fvdw6 = _mm_mul_ps(c6_00,FF);
373 /* CUBIC SPLINE TABLE REPULSION */
374 vfitab = _mm_add_epi32(vfitab,ifour);
375 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
376 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
377 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
378 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
379 _MM_TRANSPOSE4_PS(Y,F,G,H);
380 Heps = _mm_mul_ps(vfeps,H);
381 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
382 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
383 vvdw12 = _mm_mul_ps(c12_00,VV);
384 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
385 fvdw12 = _mm_mul_ps(c12_00,FF);
386 vvdw = _mm_add_ps(vvdw12,vvdw6);
387 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
389 /* Update potential sum for this i atom from the interaction with this j atom. */
390 velec = _mm_andnot_ps(dummy_mask,velec);
391 velecsum = _mm_add_ps(velecsum,velec);
392 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
393 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
395 fscal = _mm_add_ps(felec,fvdw);
397 fscal = _mm_andnot_ps(dummy_mask,fscal);
399 /* Calculate temporary vectorial force */
400 tx = _mm_mul_ps(fscal,dx00);
401 ty = _mm_mul_ps(fscal,dy00);
402 tz = _mm_mul_ps(fscal,dz00);
404 /* Update vectorial force */
405 fix0 = _mm_add_ps(fix0,tx);
406 fiy0 = _mm_add_ps(fiy0,ty);
407 fiz0 = _mm_add_ps(fiz0,tz);
409 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
410 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
411 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
412 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
413 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
415 /* Inner loop uses 68 flops */
418 /* End of innermost loop */
420 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
421 f+i_coord_offset,fshift+i_shift_offset);
424 /* Update potential energies */
425 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
426 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
428 /* Increment number of inner iterations */
429 inneriter += j_index_end - j_index_start;
431 /* Outer loop uses 9 flops */
434 /* Increment number of outer iterations */
437 /* Update outer/inner flops */
439 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
442 * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse2_single
443 * Electrostatics interaction: ReactionField
444 * VdW interaction: CubicSplineTable
445 * Geometry: Particle-Particle
446 * Calculate force/pot: Force
449 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse2_single
450 (t_nblist * gmx_restrict nlist,
451 rvec * gmx_restrict xx,
452 rvec * gmx_restrict ff,
453 t_forcerec * gmx_restrict fr,
454 t_mdatoms * gmx_restrict mdatoms,
455 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
456 t_nrnb * gmx_restrict nrnb)
458 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
459 * just 0 for non-waters.
460 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
461 * jnr indices corresponding to data put in the four positions in the SIMD register.
463 int i_shift_offset,i_coord_offset,outeriter,inneriter;
464 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
465 int jnrA,jnrB,jnrC,jnrD;
466 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
467 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
468 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
470 real *shiftvec,*fshift,*x,*f;
471 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
473 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
475 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
476 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
477 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
478 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
479 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
482 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
485 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
486 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
488 __m128i ifour = _mm_set1_epi32(4);
489 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
491 __m128 dummy_mask,cutoff_mask;
492 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
493 __m128 one = _mm_set1_ps(1.0);
494 __m128 two = _mm_set1_ps(2.0);
500 jindex = nlist->jindex;
502 shiftidx = nlist->shift;
504 shiftvec = fr->shift_vec[0];
505 fshift = fr->fshift[0];
506 facel = _mm_set1_ps(fr->epsfac);
507 charge = mdatoms->chargeA;
508 krf = _mm_set1_ps(fr->ic->k_rf);
509 krf2 = _mm_set1_ps(fr->ic->k_rf*2.0);
510 crf = _mm_set1_ps(fr->ic->c_rf);
511 nvdwtype = fr->ntype;
513 vdwtype = mdatoms->typeA;
515 vftab = kernel_data->table_vdw->data;
516 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
518 /* Avoid stupid compiler warnings */
519 jnrA = jnrB = jnrC = jnrD = 0;
528 for(iidx=0;iidx<4*DIM;iidx++)
533 /* Start outer loop over neighborlists */
534 for(iidx=0; iidx<nri; iidx++)
536 /* Load shift vector for this list */
537 i_shift_offset = DIM*shiftidx[iidx];
539 /* Load limits for loop over neighbors */
540 j_index_start = jindex[iidx];
541 j_index_end = jindex[iidx+1];
543 /* Get outer coordinate index */
545 i_coord_offset = DIM*inr;
547 /* Load i particle coords and add shift vector */
548 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
550 fix0 = _mm_setzero_ps();
551 fiy0 = _mm_setzero_ps();
552 fiz0 = _mm_setzero_ps();
554 /* Load parameters for i particles */
555 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
556 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
558 /* Start inner kernel loop */
559 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
562 /* Get j neighbor index, and coordinate index */
567 j_coord_offsetA = DIM*jnrA;
568 j_coord_offsetB = DIM*jnrB;
569 j_coord_offsetC = DIM*jnrC;
570 j_coord_offsetD = DIM*jnrD;
572 /* load j atom coordinates */
573 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
574 x+j_coord_offsetC,x+j_coord_offsetD,
577 /* Calculate displacement vector */
578 dx00 = _mm_sub_ps(ix0,jx0);
579 dy00 = _mm_sub_ps(iy0,jy0);
580 dz00 = _mm_sub_ps(iz0,jz0);
582 /* Calculate squared distance and things based on it */
583 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
585 rinv00 = gmx_mm_invsqrt_ps(rsq00);
587 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
589 /* Load parameters for j particles */
590 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
591 charge+jnrC+0,charge+jnrD+0);
592 vdwjidx0A = 2*vdwtype[jnrA+0];
593 vdwjidx0B = 2*vdwtype[jnrB+0];
594 vdwjidx0C = 2*vdwtype[jnrC+0];
595 vdwjidx0D = 2*vdwtype[jnrD+0];
597 /**************************
598 * CALCULATE INTERACTIONS *
599 **************************/
601 r00 = _mm_mul_ps(rsq00,rinv00);
603 /* Compute parameters for interactions between i and j atoms */
604 qq00 = _mm_mul_ps(iq0,jq0);
605 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
606 vdwparam+vdwioffset0+vdwjidx0B,
607 vdwparam+vdwioffset0+vdwjidx0C,
608 vdwparam+vdwioffset0+vdwjidx0D,
611 /* Calculate table index by multiplying r with table scale and truncate to integer */
612 rt = _mm_mul_ps(r00,vftabscale);
613 vfitab = _mm_cvttps_epi32(rt);
614 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
615 vfitab = _mm_slli_epi32(vfitab,3);
617 /* REACTION-FIELD ELECTROSTATICS */
618 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
620 /* CUBIC SPLINE TABLE DISPERSION */
621 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
622 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
623 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
624 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
625 _MM_TRANSPOSE4_PS(Y,F,G,H);
626 Heps = _mm_mul_ps(vfeps,H);
627 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
628 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
629 fvdw6 = _mm_mul_ps(c6_00,FF);
631 /* CUBIC SPLINE TABLE REPULSION */
632 vfitab = _mm_add_epi32(vfitab,ifour);
633 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
634 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
635 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
636 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
637 _MM_TRANSPOSE4_PS(Y,F,G,H);
638 Heps = _mm_mul_ps(vfeps,H);
639 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
640 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
641 fvdw12 = _mm_mul_ps(c12_00,FF);
642 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
644 fscal = _mm_add_ps(felec,fvdw);
646 /* Calculate temporary vectorial force */
647 tx = _mm_mul_ps(fscal,dx00);
648 ty = _mm_mul_ps(fscal,dy00);
649 tz = _mm_mul_ps(fscal,dz00);
651 /* Update vectorial force */
652 fix0 = _mm_add_ps(fix0,tx);
653 fiy0 = _mm_add_ps(fiy0,ty);
654 fiz0 = _mm_add_ps(fiz0,tz);
656 fjptrA = f+j_coord_offsetA;
657 fjptrB = f+j_coord_offsetB;
658 fjptrC = f+j_coord_offsetC;
659 fjptrD = f+j_coord_offsetD;
660 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
662 /* Inner loop uses 54 flops */
668 /* Get j neighbor index, and coordinate index */
669 jnrlistA = jjnr[jidx];
670 jnrlistB = jjnr[jidx+1];
671 jnrlistC = jjnr[jidx+2];
672 jnrlistD = jjnr[jidx+3];
673 /* Sign of each element will be negative for non-real atoms.
674 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
675 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
677 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
678 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
679 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
680 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
681 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
682 j_coord_offsetA = DIM*jnrA;
683 j_coord_offsetB = DIM*jnrB;
684 j_coord_offsetC = DIM*jnrC;
685 j_coord_offsetD = DIM*jnrD;
687 /* load j atom coordinates */
688 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
689 x+j_coord_offsetC,x+j_coord_offsetD,
692 /* Calculate displacement vector */
693 dx00 = _mm_sub_ps(ix0,jx0);
694 dy00 = _mm_sub_ps(iy0,jy0);
695 dz00 = _mm_sub_ps(iz0,jz0);
697 /* Calculate squared distance and things based on it */
698 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
700 rinv00 = gmx_mm_invsqrt_ps(rsq00);
702 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
704 /* Load parameters for j particles */
705 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
706 charge+jnrC+0,charge+jnrD+0);
707 vdwjidx0A = 2*vdwtype[jnrA+0];
708 vdwjidx0B = 2*vdwtype[jnrB+0];
709 vdwjidx0C = 2*vdwtype[jnrC+0];
710 vdwjidx0D = 2*vdwtype[jnrD+0];
712 /**************************
713 * CALCULATE INTERACTIONS *
714 **************************/
716 r00 = _mm_mul_ps(rsq00,rinv00);
717 r00 = _mm_andnot_ps(dummy_mask,r00);
719 /* Compute parameters for interactions between i and j atoms */
720 qq00 = _mm_mul_ps(iq0,jq0);
721 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
722 vdwparam+vdwioffset0+vdwjidx0B,
723 vdwparam+vdwioffset0+vdwjidx0C,
724 vdwparam+vdwioffset0+vdwjidx0D,
727 /* Calculate table index by multiplying r with table scale and truncate to integer */
728 rt = _mm_mul_ps(r00,vftabscale);
729 vfitab = _mm_cvttps_epi32(rt);
730 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
731 vfitab = _mm_slli_epi32(vfitab,3);
733 /* REACTION-FIELD ELECTROSTATICS */
734 felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
736 /* CUBIC SPLINE TABLE DISPERSION */
737 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
738 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
739 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
740 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
741 _MM_TRANSPOSE4_PS(Y,F,G,H);
742 Heps = _mm_mul_ps(vfeps,H);
743 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
744 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
745 fvdw6 = _mm_mul_ps(c6_00,FF);
747 /* CUBIC SPLINE TABLE REPULSION */
748 vfitab = _mm_add_epi32(vfitab,ifour);
749 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
750 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
751 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
752 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
753 _MM_TRANSPOSE4_PS(Y,F,G,H);
754 Heps = _mm_mul_ps(vfeps,H);
755 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
756 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
757 fvdw12 = _mm_mul_ps(c12_00,FF);
758 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
760 fscal = _mm_add_ps(felec,fvdw);
762 fscal = _mm_andnot_ps(dummy_mask,fscal);
764 /* Calculate temporary vectorial force */
765 tx = _mm_mul_ps(fscal,dx00);
766 ty = _mm_mul_ps(fscal,dy00);
767 tz = _mm_mul_ps(fscal,dz00);
769 /* Update vectorial force */
770 fix0 = _mm_add_ps(fix0,tx);
771 fiy0 = _mm_add_ps(fiy0,ty);
772 fiz0 = _mm_add_ps(fiz0,tz);
774 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
775 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
776 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
777 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
778 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
780 /* Inner loop uses 55 flops */
783 /* End of innermost loop */
785 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
786 f+i_coord_offset,fshift+i_shift_offset);
788 /* Increment number of inner iterations */
789 inneriter += j_index_end - j_index_start;
791 /* Outer loop uses 7 flops */
794 /* Increment number of outer iterations */
797 /* Update outer/inner flops */
799 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*55);