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 "gromacs/legacyheaders/types/simple.h"
44 #include "gromacs/math/vec.h"
45 #include "gromacs/legacyheaders/nrnb.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_ElecEw_VdwCSTab_GeomP1P1_VF_sse4_1_single
52 * Electrostatics interaction: Ewald
53 * VdW interaction: CubicSplineTable
54 * Geometry: Particle-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecEw_VdwCSTab_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;
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;
101 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
103 __m128 dummy_mask,cutoff_mask;
104 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
105 __m128 one = _mm_set1_ps(1.0);
106 __m128 two = _mm_set1_ps(2.0);
112 jindex = nlist->jindex;
114 shiftidx = nlist->shift;
116 shiftvec = fr->shift_vec[0];
117 fshift = fr->fshift[0];
118 facel = _mm_set1_ps(fr->epsfac);
119 charge = mdatoms->chargeA;
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 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
128 ewtab = fr->ic->tabq_coul_FDV0;
129 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
130 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
132 /* Avoid stupid compiler warnings */
133 jnrA = jnrB = jnrC = jnrD = 0;
142 for(iidx=0;iidx<4*DIM;iidx++)
147 /* Start outer loop over neighborlists */
148 for(iidx=0; iidx<nri; iidx++)
150 /* Load shift vector for this list */
151 i_shift_offset = DIM*shiftidx[iidx];
153 /* Load limits for loop over neighbors */
154 j_index_start = jindex[iidx];
155 j_index_end = jindex[iidx+1];
157 /* Get outer coordinate index */
159 i_coord_offset = DIM*inr;
161 /* Load i particle coords and add shift vector */
162 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
164 fix0 = _mm_setzero_ps();
165 fiy0 = _mm_setzero_ps();
166 fiz0 = _mm_setzero_ps();
168 /* Load parameters for i particles */
169 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
170 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
172 /* Reset potential sums */
173 velecsum = _mm_setzero_ps();
174 vvdwsum = _mm_setzero_ps();
176 /* Start inner kernel loop */
177 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
180 /* Get j neighbor index, and coordinate index */
185 j_coord_offsetA = DIM*jnrA;
186 j_coord_offsetB = DIM*jnrB;
187 j_coord_offsetC = DIM*jnrC;
188 j_coord_offsetD = DIM*jnrD;
190 /* load j atom coordinates */
191 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
192 x+j_coord_offsetC,x+j_coord_offsetD,
195 /* Calculate displacement vector */
196 dx00 = _mm_sub_ps(ix0,jx0);
197 dy00 = _mm_sub_ps(iy0,jy0);
198 dz00 = _mm_sub_ps(iz0,jz0);
200 /* Calculate squared distance and things based on it */
201 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
203 rinv00 = gmx_mm_invsqrt_ps(rsq00);
205 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
207 /* Load parameters for j particles */
208 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
209 charge+jnrC+0,charge+jnrD+0);
210 vdwjidx0A = 2*vdwtype[jnrA+0];
211 vdwjidx0B = 2*vdwtype[jnrB+0];
212 vdwjidx0C = 2*vdwtype[jnrC+0];
213 vdwjidx0D = 2*vdwtype[jnrD+0];
215 /**************************
216 * CALCULATE INTERACTIONS *
217 **************************/
219 r00 = _mm_mul_ps(rsq00,rinv00);
221 /* Compute parameters for interactions between i and j atoms */
222 qq00 = _mm_mul_ps(iq0,jq0);
223 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
224 vdwparam+vdwioffset0+vdwjidx0B,
225 vdwparam+vdwioffset0+vdwjidx0C,
226 vdwparam+vdwioffset0+vdwjidx0D,
229 /* Calculate table index by multiplying r with table scale and truncate to integer */
230 rt = _mm_mul_ps(r00,vftabscale);
231 vfitab = _mm_cvttps_epi32(rt);
232 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
233 vfitab = _mm_slli_epi32(vfitab,3);
235 /* EWALD ELECTROSTATICS */
237 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
238 ewrt = _mm_mul_ps(r00,ewtabscale);
239 ewitab = _mm_cvttps_epi32(ewrt);
240 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
241 ewitab = _mm_slli_epi32(ewitab,2);
242 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
243 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
244 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
245 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
246 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
247 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
248 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
249 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
250 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
252 /* CUBIC SPLINE TABLE DISPERSION */
253 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
254 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
255 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
256 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
257 _MM_TRANSPOSE4_PS(Y,F,G,H);
258 Heps = _mm_mul_ps(vfeps,H);
259 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
260 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
261 vvdw6 = _mm_mul_ps(c6_00,VV);
262 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
263 fvdw6 = _mm_mul_ps(c6_00,FF);
265 /* CUBIC SPLINE TABLE REPULSION */
266 vfitab = _mm_add_epi32(vfitab,ifour);
267 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
268 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
269 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
270 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
271 _MM_TRANSPOSE4_PS(Y,F,G,H);
272 Heps = _mm_mul_ps(vfeps,H);
273 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
274 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
275 vvdw12 = _mm_mul_ps(c12_00,VV);
276 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
277 fvdw12 = _mm_mul_ps(c12_00,FF);
278 vvdw = _mm_add_ps(vvdw12,vvdw6);
279 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
281 /* Update potential sum for this i atom from the interaction with this j atom. */
282 velecsum = _mm_add_ps(velecsum,velec);
283 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
285 fscal = _mm_add_ps(felec,fvdw);
287 /* Calculate temporary vectorial force */
288 tx = _mm_mul_ps(fscal,dx00);
289 ty = _mm_mul_ps(fscal,dy00);
290 tz = _mm_mul_ps(fscal,dz00);
292 /* Update vectorial force */
293 fix0 = _mm_add_ps(fix0,tx);
294 fiy0 = _mm_add_ps(fiy0,ty);
295 fiz0 = _mm_add_ps(fiz0,tz);
297 fjptrA = f+j_coord_offsetA;
298 fjptrB = f+j_coord_offsetB;
299 fjptrC = f+j_coord_offsetC;
300 fjptrD = f+j_coord_offsetD;
301 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
303 /* Inner loop uses 75 flops */
309 /* Get j neighbor index, and coordinate index */
310 jnrlistA = jjnr[jidx];
311 jnrlistB = jjnr[jidx+1];
312 jnrlistC = jjnr[jidx+2];
313 jnrlistD = jjnr[jidx+3];
314 /* Sign of each element will be negative for non-real atoms.
315 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
316 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
318 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
319 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
320 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
321 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
322 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
323 j_coord_offsetA = DIM*jnrA;
324 j_coord_offsetB = DIM*jnrB;
325 j_coord_offsetC = DIM*jnrC;
326 j_coord_offsetD = DIM*jnrD;
328 /* load j atom coordinates */
329 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
330 x+j_coord_offsetC,x+j_coord_offsetD,
333 /* Calculate displacement vector */
334 dx00 = _mm_sub_ps(ix0,jx0);
335 dy00 = _mm_sub_ps(iy0,jy0);
336 dz00 = _mm_sub_ps(iz0,jz0);
338 /* Calculate squared distance and things based on it */
339 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
341 rinv00 = gmx_mm_invsqrt_ps(rsq00);
343 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
345 /* Load parameters for j particles */
346 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
347 charge+jnrC+0,charge+jnrD+0);
348 vdwjidx0A = 2*vdwtype[jnrA+0];
349 vdwjidx0B = 2*vdwtype[jnrB+0];
350 vdwjidx0C = 2*vdwtype[jnrC+0];
351 vdwjidx0D = 2*vdwtype[jnrD+0];
353 /**************************
354 * CALCULATE INTERACTIONS *
355 **************************/
357 r00 = _mm_mul_ps(rsq00,rinv00);
358 r00 = _mm_andnot_ps(dummy_mask,r00);
360 /* Compute parameters for interactions between i and j atoms */
361 qq00 = _mm_mul_ps(iq0,jq0);
362 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
363 vdwparam+vdwioffset0+vdwjidx0B,
364 vdwparam+vdwioffset0+vdwjidx0C,
365 vdwparam+vdwioffset0+vdwjidx0D,
368 /* Calculate table index by multiplying r with table scale and truncate to integer */
369 rt = _mm_mul_ps(r00,vftabscale);
370 vfitab = _mm_cvttps_epi32(rt);
371 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
372 vfitab = _mm_slli_epi32(vfitab,3);
374 /* EWALD ELECTROSTATICS */
376 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
377 ewrt = _mm_mul_ps(r00,ewtabscale);
378 ewitab = _mm_cvttps_epi32(ewrt);
379 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
380 ewitab = _mm_slli_epi32(ewitab,2);
381 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
382 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
383 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
384 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
385 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
386 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
387 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
388 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
389 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
391 /* CUBIC SPLINE TABLE DISPERSION */
392 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
393 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
394 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
395 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
396 _MM_TRANSPOSE4_PS(Y,F,G,H);
397 Heps = _mm_mul_ps(vfeps,H);
398 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
399 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
400 vvdw6 = _mm_mul_ps(c6_00,VV);
401 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
402 fvdw6 = _mm_mul_ps(c6_00,FF);
404 /* CUBIC SPLINE TABLE REPULSION */
405 vfitab = _mm_add_epi32(vfitab,ifour);
406 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
407 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
408 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
409 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
410 _MM_TRANSPOSE4_PS(Y,F,G,H);
411 Heps = _mm_mul_ps(vfeps,H);
412 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
413 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
414 vvdw12 = _mm_mul_ps(c12_00,VV);
415 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
416 fvdw12 = _mm_mul_ps(c12_00,FF);
417 vvdw = _mm_add_ps(vvdw12,vvdw6);
418 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
420 /* Update potential sum for this i atom from the interaction with this j atom. */
421 velec = _mm_andnot_ps(dummy_mask,velec);
422 velecsum = _mm_add_ps(velecsum,velec);
423 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
424 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
426 fscal = _mm_add_ps(felec,fvdw);
428 fscal = _mm_andnot_ps(dummy_mask,fscal);
430 /* Calculate temporary vectorial force */
431 tx = _mm_mul_ps(fscal,dx00);
432 ty = _mm_mul_ps(fscal,dy00);
433 tz = _mm_mul_ps(fscal,dz00);
435 /* Update vectorial force */
436 fix0 = _mm_add_ps(fix0,tx);
437 fiy0 = _mm_add_ps(fiy0,ty);
438 fiz0 = _mm_add_ps(fiz0,tz);
440 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
441 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
442 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
443 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
444 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
446 /* Inner loop uses 76 flops */
449 /* End of innermost loop */
451 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
452 f+i_coord_offset,fshift+i_shift_offset);
455 /* Update potential energies */
456 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
457 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
459 /* Increment number of inner iterations */
460 inneriter += j_index_end - j_index_start;
462 /* Outer loop uses 9 flops */
465 /* Increment number of outer iterations */
468 /* Update outer/inner flops */
470 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*76);
473 * Gromacs nonbonded kernel: nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse4_1_single
474 * Electrostatics interaction: Ewald
475 * VdW interaction: CubicSplineTable
476 * Geometry: Particle-Particle
477 * Calculate force/pot: Force
480 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse4_1_single
481 (t_nblist * gmx_restrict nlist,
482 rvec * gmx_restrict xx,
483 rvec * gmx_restrict ff,
484 t_forcerec * gmx_restrict fr,
485 t_mdatoms * gmx_restrict mdatoms,
486 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
487 t_nrnb * gmx_restrict nrnb)
489 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
490 * just 0 for non-waters.
491 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
492 * jnr indices corresponding to data put in the four positions in the SIMD register.
494 int i_shift_offset,i_coord_offset,outeriter,inneriter;
495 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
496 int jnrA,jnrB,jnrC,jnrD;
497 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
498 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
499 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
501 real *shiftvec,*fshift,*x,*f;
502 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
504 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
506 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
507 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
508 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
509 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
510 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
513 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
516 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
517 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
519 __m128i ifour = _mm_set1_epi32(4);
520 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
523 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
525 __m128 dummy_mask,cutoff_mask;
526 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
527 __m128 one = _mm_set1_ps(1.0);
528 __m128 two = _mm_set1_ps(2.0);
534 jindex = nlist->jindex;
536 shiftidx = nlist->shift;
538 shiftvec = fr->shift_vec[0];
539 fshift = fr->fshift[0];
540 facel = _mm_set1_ps(fr->epsfac);
541 charge = mdatoms->chargeA;
542 nvdwtype = fr->ntype;
544 vdwtype = mdatoms->typeA;
546 vftab = kernel_data->table_vdw->data;
547 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
549 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
550 ewtab = fr->ic->tabq_coul_F;
551 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
552 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
554 /* Avoid stupid compiler warnings */
555 jnrA = jnrB = jnrC = jnrD = 0;
564 for(iidx=0;iidx<4*DIM;iidx++)
569 /* Start outer loop over neighborlists */
570 for(iidx=0; iidx<nri; iidx++)
572 /* Load shift vector for this list */
573 i_shift_offset = DIM*shiftidx[iidx];
575 /* Load limits for loop over neighbors */
576 j_index_start = jindex[iidx];
577 j_index_end = jindex[iidx+1];
579 /* Get outer coordinate index */
581 i_coord_offset = DIM*inr;
583 /* Load i particle coords and add shift vector */
584 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
586 fix0 = _mm_setzero_ps();
587 fiy0 = _mm_setzero_ps();
588 fiz0 = _mm_setzero_ps();
590 /* Load parameters for i particles */
591 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
592 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
594 /* Start inner kernel loop */
595 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
598 /* Get j neighbor index, and coordinate index */
603 j_coord_offsetA = DIM*jnrA;
604 j_coord_offsetB = DIM*jnrB;
605 j_coord_offsetC = DIM*jnrC;
606 j_coord_offsetD = DIM*jnrD;
608 /* load j atom coordinates */
609 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
610 x+j_coord_offsetC,x+j_coord_offsetD,
613 /* Calculate displacement vector */
614 dx00 = _mm_sub_ps(ix0,jx0);
615 dy00 = _mm_sub_ps(iy0,jy0);
616 dz00 = _mm_sub_ps(iz0,jz0);
618 /* Calculate squared distance and things based on it */
619 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
621 rinv00 = gmx_mm_invsqrt_ps(rsq00);
623 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
625 /* Load parameters for j particles */
626 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
627 charge+jnrC+0,charge+jnrD+0);
628 vdwjidx0A = 2*vdwtype[jnrA+0];
629 vdwjidx0B = 2*vdwtype[jnrB+0];
630 vdwjidx0C = 2*vdwtype[jnrC+0];
631 vdwjidx0D = 2*vdwtype[jnrD+0];
633 /**************************
634 * CALCULATE INTERACTIONS *
635 **************************/
637 r00 = _mm_mul_ps(rsq00,rinv00);
639 /* Compute parameters for interactions between i and j atoms */
640 qq00 = _mm_mul_ps(iq0,jq0);
641 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
642 vdwparam+vdwioffset0+vdwjidx0B,
643 vdwparam+vdwioffset0+vdwjidx0C,
644 vdwparam+vdwioffset0+vdwjidx0D,
647 /* Calculate table index by multiplying r with table scale and truncate to integer */
648 rt = _mm_mul_ps(r00,vftabscale);
649 vfitab = _mm_cvttps_epi32(rt);
650 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
651 vfitab = _mm_slli_epi32(vfitab,3);
653 /* EWALD ELECTROSTATICS */
655 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
656 ewrt = _mm_mul_ps(r00,ewtabscale);
657 ewitab = _mm_cvttps_epi32(ewrt);
658 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
659 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
660 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
662 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
663 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
665 /* CUBIC SPLINE TABLE DISPERSION */
666 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
667 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
668 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
669 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
670 _MM_TRANSPOSE4_PS(Y,F,G,H);
671 Heps = _mm_mul_ps(vfeps,H);
672 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
673 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
674 fvdw6 = _mm_mul_ps(c6_00,FF);
676 /* CUBIC SPLINE TABLE REPULSION */
677 vfitab = _mm_add_epi32(vfitab,ifour);
678 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
679 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
680 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
681 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
682 _MM_TRANSPOSE4_PS(Y,F,G,H);
683 Heps = _mm_mul_ps(vfeps,H);
684 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
685 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
686 fvdw12 = _mm_mul_ps(c12_00,FF);
687 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
689 fscal = _mm_add_ps(felec,fvdw);
691 /* Calculate temporary vectorial force */
692 tx = _mm_mul_ps(fscal,dx00);
693 ty = _mm_mul_ps(fscal,dy00);
694 tz = _mm_mul_ps(fscal,dz00);
696 /* Update vectorial force */
697 fix0 = _mm_add_ps(fix0,tx);
698 fiy0 = _mm_add_ps(fiy0,ty);
699 fiz0 = _mm_add_ps(fiz0,tz);
701 fjptrA = f+j_coord_offsetA;
702 fjptrB = f+j_coord_offsetB;
703 fjptrC = f+j_coord_offsetC;
704 fjptrD = f+j_coord_offsetD;
705 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
707 /* Inner loop uses 62 flops */
713 /* Get j neighbor index, and coordinate index */
714 jnrlistA = jjnr[jidx];
715 jnrlistB = jjnr[jidx+1];
716 jnrlistC = jjnr[jidx+2];
717 jnrlistD = jjnr[jidx+3];
718 /* Sign of each element will be negative for non-real atoms.
719 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
720 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
722 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
723 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
724 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
725 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
726 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
727 j_coord_offsetA = DIM*jnrA;
728 j_coord_offsetB = DIM*jnrB;
729 j_coord_offsetC = DIM*jnrC;
730 j_coord_offsetD = DIM*jnrD;
732 /* load j atom coordinates */
733 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
734 x+j_coord_offsetC,x+j_coord_offsetD,
737 /* Calculate displacement vector */
738 dx00 = _mm_sub_ps(ix0,jx0);
739 dy00 = _mm_sub_ps(iy0,jy0);
740 dz00 = _mm_sub_ps(iz0,jz0);
742 /* Calculate squared distance and things based on it */
743 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
745 rinv00 = gmx_mm_invsqrt_ps(rsq00);
747 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
749 /* Load parameters for j particles */
750 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
751 charge+jnrC+0,charge+jnrD+0);
752 vdwjidx0A = 2*vdwtype[jnrA+0];
753 vdwjidx0B = 2*vdwtype[jnrB+0];
754 vdwjidx0C = 2*vdwtype[jnrC+0];
755 vdwjidx0D = 2*vdwtype[jnrD+0];
757 /**************************
758 * CALCULATE INTERACTIONS *
759 **************************/
761 r00 = _mm_mul_ps(rsq00,rinv00);
762 r00 = _mm_andnot_ps(dummy_mask,r00);
764 /* Compute parameters for interactions between i and j atoms */
765 qq00 = _mm_mul_ps(iq0,jq0);
766 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
767 vdwparam+vdwioffset0+vdwjidx0B,
768 vdwparam+vdwioffset0+vdwjidx0C,
769 vdwparam+vdwioffset0+vdwjidx0D,
772 /* Calculate table index by multiplying r with table scale and truncate to integer */
773 rt = _mm_mul_ps(r00,vftabscale);
774 vfitab = _mm_cvttps_epi32(rt);
775 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
776 vfitab = _mm_slli_epi32(vfitab,3);
778 /* EWALD ELECTROSTATICS */
780 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
781 ewrt = _mm_mul_ps(r00,ewtabscale);
782 ewitab = _mm_cvttps_epi32(ewrt);
783 eweps = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
784 gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
785 ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
787 felec = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
788 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
790 /* CUBIC SPLINE TABLE DISPERSION */
791 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
792 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
793 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
794 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
795 _MM_TRANSPOSE4_PS(Y,F,G,H);
796 Heps = _mm_mul_ps(vfeps,H);
797 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
798 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
799 fvdw6 = _mm_mul_ps(c6_00,FF);
801 /* CUBIC SPLINE TABLE REPULSION */
802 vfitab = _mm_add_epi32(vfitab,ifour);
803 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
804 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
805 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
806 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
807 _MM_TRANSPOSE4_PS(Y,F,G,H);
808 Heps = _mm_mul_ps(vfeps,H);
809 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
810 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
811 fvdw12 = _mm_mul_ps(c12_00,FF);
812 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
814 fscal = _mm_add_ps(felec,fvdw);
816 fscal = _mm_andnot_ps(dummy_mask,fscal);
818 /* Calculate temporary vectorial force */
819 tx = _mm_mul_ps(fscal,dx00);
820 ty = _mm_mul_ps(fscal,dy00);
821 tz = _mm_mul_ps(fscal,dz00);
823 /* Update vectorial force */
824 fix0 = _mm_add_ps(fix0,tx);
825 fiy0 = _mm_add_ps(fiy0,ty);
826 fiz0 = _mm_add_ps(fiz0,tz);
828 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
829 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
830 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
831 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
832 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
834 /* Inner loop uses 63 flops */
837 /* End of innermost loop */
839 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
840 f+i_coord_offset,fshift+i_shift_offset);
842 /* Increment number of inner iterations */
843 inneriter += j_index_end - j_index_start;
845 /* Outer loop uses 7 flops */
848 /* Increment number of outer iterations */
851 /* Update outer/inner flops */
853 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*63);