?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed simple.h from nb_kernel_sse2_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_sse2_single / nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_sse2_single.c
blob11f81543b0fa1187defba236f3edc0937a8be5df
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015, 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.
8 *
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.
13 *
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.
18 *
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.
23 *
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.
31 *
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.
34 */
35 /*
36 * Note: this file was generated by the GROMACS sse2_single kernel generator.
37 */
38 #include "gmxpre.h"
39
40 #include "config.h"
41
42 #include <math.h>
43
44 #include "../nb_kernel.h"
45 #include "gromacs/math/vec.h"
46 #include "gromacs/legacyheaders/nrnb.h"
47
48 #include "gromacs/simd/math_x86_sse2_single.h"
49 #include "kernelutil_x86_sse2_single.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse2_single
53 * Electrostatics interaction: Coulomb
54 * VdW interaction: CubicSplineTable
55 * Geometry: Particle-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse2_single
60 (t_nblist * gmx_restrict nlist,
61 rvec * gmx_restrict xx,
62 rvec * gmx_restrict ff,
63 t_forcerec * gmx_restrict fr,
64 t_mdatoms * gmx_restrict mdatoms,
65 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
66 t_nrnb * gmx_restrict nrnb)
67 {
68 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
69 * just 0 for non-waters.
70 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
71 * jnr indices corresponding to data put in the four positions in the SIMD register.
72 */
73 int i_shift_offset,i_coord_offset,outeriter,inneriter;
74 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
75 int jnrA,jnrB,jnrC,jnrD;
76 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
77 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
78 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
79 real rcutoff_scalar;
80 real *shiftvec,*fshift,*x,*f;
81 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
82 real scratch[4*DIM];
83 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
84 int vdwioffset0;
85 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
86 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
87 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
88 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
89 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
90 real *charge;
91 int nvdwtype;
92 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
93 int *vdwtype;
94 real *vdwparam;
95 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
96 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
97 __m128i vfitab;
98 __m128i ifour = _mm_set1_epi32(4);
99 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
100 real *vftab;
101 __m128 dummy_mask,cutoff_mask;
102 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
103 __m128 one = _mm_set1_ps(1.0);
104 __m128 two = _mm_set1_ps(2.0);
105 x = xx[0];
106 f = ff[0];
107
108 nri = nlist->nri;
109 iinr = nlist->iinr;
110 jindex = nlist->jindex;
111 jjnr = nlist->jjnr;
112 shiftidx = nlist->shift;
113 gid = nlist->gid;
114 shiftvec = fr->shift_vec[0];
115 fshift = fr->fshift[0];
116 facel = _mm_set1_ps(fr->epsfac);
117 charge = mdatoms->chargeA;
118 nvdwtype = fr->ntype;
119 vdwparam = fr->nbfp;
120 vdwtype = mdatoms->typeA;
121
122 vftab = kernel_data->table_vdw->data;
123 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
124
125 /* Avoid stupid compiler warnings */
126 jnrA = jnrB = jnrC = jnrD = 0;
127 j_coord_offsetA = 0;
128 j_coord_offsetB = 0;
129 j_coord_offsetC = 0;
130 j_coord_offsetD = 0;
131
132 outeriter = 0;
133 inneriter = 0;
134
135 for(iidx=0;iidx<4*DIM;iidx++)
136 {
137 scratch[iidx] = 0.0;
138 }
139
140 /* Start outer loop over neighborlists */
141 for(iidx=0; iidx<nri; iidx++)
142 {
143 /* Load shift vector for this list */
144 i_shift_offset = DIM*shiftidx[iidx];
145
146 /* Load limits for loop over neighbors */
147 j_index_start = jindex[iidx];
148 j_index_end = jindex[iidx+1];
149
150 /* Get outer coordinate index */
151 inr = iinr[iidx];
152 i_coord_offset = DIM*inr;
153
154 /* Load i particle coords and add shift vector */
155 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
156
157 fix0 = _mm_setzero_ps();
158 fiy0 = _mm_setzero_ps();
159 fiz0 = _mm_setzero_ps();
160
161 /* Load parameters for i particles */
162 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
163 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
164
165 /* Reset potential sums */
166 velecsum = _mm_setzero_ps();
167 vvdwsum = _mm_setzero_ps();
168
169 /* Start inner kernel loop */
170 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
171 {
172
173 /* Get j neighbor index, and coordinate index */
174 jnrA = jjnr[jidx];
175 jnrB = jjnr[jidx+1];
176 jnrC = jjnr[jidx+2];
177 jnrD = jjnr[jidx+3];
178 j_coord_offsetA = DIM*jnrA;
179 j_coord_offsetB = DIM*jnrB;
180 j_coord_offsetC = DIM*jnrC;
181 j_coord_offsetD = DIM*jnrD;
182
183 /* load j atom coordinates */
184 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
185 x+j_coord_offsetC,x+j_coord_offsetD,
186 &jx0,&jy0,&jz0);
187
188 /* Calculate displacement vector */
189 dx00 = _mm_sub_ps(ix0,jx0);
190 dy00 = _mm_sub_ps(iy0,jy0);
191 dz00 = _mm_sub_ps(iz0,jz0);
192
193 /* Calculate squared distance and things based on it */
194 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
195
196 rinv00 = gmx_mm_invsqrt_ps(rsq00);
197
198 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
199
200 /* Load parameters for j particles */
201 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
202 charge+jnrC+0,charge+jnrD+0);
203 vdwjidx0A = 2*vdwtype[jnrA+0];
204 vdwjidx0B = 2*vdwtype[jnrB+0];
205 vdwjidx0C = 2*vdwtype[jnrC+0];
206 vdwjidx0D = 2*vdwtype[jnrD+0];
207
208 /**************************
209 * CALCULATE INTERACTIONS *
210 **************************/
211
212 r00 = _mm_mul_ps(rsq00,rinv00);
213
214 /* Compute parameters for interactions between i and j atoms */
215 qq00 = _mm_mul_ps(iq0,jq0);
216 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
217 vdwparam+vdwioffset0+vdwjidx0B,
218 vdwparam+vdwioffset0+vdwjidx0C,
219 vdwparam+vdwioffset0+vdwjidx0D,
220 &c6_00,&c12_00);
221
222 /* Calculate table index by multiplying r with table scale and truncate to integer */
223 rt = _mm_mul_ps(r00,vftabscale);
224 vfitab = _mm_cvttps_epi32(rt);
225 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
226 vfitab = _mm_slli_epi32(vfitab,3);
227
228 /* COULOMB ELECTROSTATICS */
229 velec = _mm_mul_ps(qq00,rinv00);
230 felec = _mm_mul_ps(velec,rinvsq00);
231
232 /* CUBIC SPLINE TABLE DISPERSION */
233 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
234 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
235 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
236 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
237 _MM_TRANSPOSE4_PS(Y,F,G,H);
238 Heps = _mm_mul_ps(vfeps,H);
239 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
240 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
241 vvdw6 = _mm_mul_ps(c6_00,VV);
242 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
243 fvdw6 = _mm_mul_ps(c6_00,FF);
244
245 /* CUBIC SPLINE TABLE REPULSION */
246 vfitab = _mm_add_epi32(vfitab,ifour);
247 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
248 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
249 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
250 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
251 _MM_TRANSPOSE4_PS(Y,F,G,H);
252 Heps = _mm_mul_ps(vfeps,H);
253 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
254 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
255 vvdw12 = _mm_mul_ps(c12_00,VV);
256 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
257 fvdw12 = _mm_mul_ps(c12_00,FF);
258 vvdw = _mm_add_ps(vvdw12,vvdw6);
259 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
260
261 /* Update potential sum for this i atom from the interaction with this j atom. */
262 velecsum = _mm_add_ps(velecsum,velec);
263 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
264
265 fscal = _mm_add_ps(felec,fvdw);
266
267 /* Calculate temporary vectorial force */
268 tx = _mm_mul_ps(fscal,dx00);
269 ty = _mm_mul_ps(fscal,dy00);
270 tz = _mm_mul_ps(fscal,dz00);
271
272 /* Update vectorial force */
273 fix0 = _mm_add_ps(fix0,tx);
274 fiy0 = _mm_add_ps(fiy0,ty);
275 fiz0 = _mm_add_ps(fiz0,tz);
276
277 fjptrA = f+j_coord_offsetA;
278 fjptrB = f+j_coord_offsetB;
279 fjptrC = f+j_coord_offsetC;
280 fjptrD = f+j_coord_offsetD;
281 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
282
283 /* Inner loop uses 63 flops */
284 }
285
286 if(jidx<j_index_end)
287 {
288
289 /* Get j neighbor index, and coordinate index */
290 jnrlistA = jjnr[jidx];
291 jnrlistB = jjnr[jidx+1];
292 jnrlistC = jjnr[jidx+2];
293 jnrlistD = jjnr[jidx+3];
294 /* Sign of each element will be negative for non-real atoms.
295 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
296 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
297 */
298 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
299 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
300 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
301 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
302 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
303 j_coord_offsetA = DIM*jnrA;
304 j_coord_offsetB = DIM*jnrB;
305 j_coord_offsetC = DIM*jnrC;
306 j_coord_offsetD = DIM*jnrD;
307
308 /* load j atom coordinates */
309 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
310 x+j_coord_offsetC,x+j_coord_offsetD,
311 &jx0,&jy0,&jz0);
312
313 /* Calculate displacement vector */
314 dx00 = _mm_sub_ps(ix0,jx0);
315 dy00 = _mm_sub_ps(iy0,jy0);
316 dz00 = _mm_sub_ps(iz0,jz0);
317
318 /* Calculate squared distance and things based on it */
319 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
320
321 rinv00 = gmx_mm_invsqrt_ps(rsq00);
322
323 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
324
325 /* Load parameters for j particles */
326 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
327 charge+jnrC+0,charge+jnrD+0);
328 vdwjidx0A = 2*vdwtype[jnrA+0];
329 vdwjidx0B = 2*vdwtype[jnrB+0];
330 vdwjidx0C = 2*vdwtype[jnrC+0];
331 vdwjidx0D = 2*vdwtype[jnrD+0];
332
333 /**************************
334 * CALCULATE INTERACTIONS *
335 **************************/
336
337 r00 = _mm_mul_ps(rsq00,rinv00);
338 r00 = _mm_andnot_ps(dummy_mask,r00);
339
340 /* Compute parameters for interactions between i and j atoms */
341 qq00 = _mm_mul_ps(iq0,jq0);
342 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
343 vdwparam+vdwioffset0+vdwjidx0B,
344 vdwparam+vdwioffset0+vdwjidx0C,
345 vdwparam+vdwioffset0+vdwjidx0D,
346 &c6_00,&c12_00);
347
348 /* Calculate table index by multiplying r with table scale and truncate to integer */
349 rt = _mm_mul_ps(r00,vftabscale);
350 vfitab = _mm_cvttps_epi32(rt);
351 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
352 vfitab = _mm_slli_epi32(vfitab,3);
353
354 /* COULOMB ELECTROSTATICS */
355 velec = _mm_mul_ps(qq00,rinv00);
356 felec = _mm_mul_ps(velec,rinvsq00);
357
358 /* CUBIC SPLINE TABLE DISPERSION */
359 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
360 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
361 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
362 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
363 _MM_TRANSPOSE4_PS(Y,F,G,H);
364 Heps = _mm_mul_ps(vfeps,H);
365 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
366 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
367 vvdw6 = _mm_mul_ps(c6_00,VV);
368 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
369 fvdw6 = _mm_mul_ps(c6_00,FF);
370
371 /* CUBIC SPLINE TABLE REPULSION */
372 vfitab = _mm_add_epi32(vfitab,ifour);
373 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
374 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
375 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
376 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
377 _MM_TRANSPOSE4_PS(Y,F,G,H);
378 Heps = _mm_mul_ps(vfeps,H);
379 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
380 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
381 vvdw12 = _mm_mul_ps(c12_00,VV);
382 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
383 fvdw12 = _mm_mul_ps(c12_00,FF);
384 vvdw = _mm_add_ps(vvdw12,vvdw6);
385 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
386
387 /* Update potential sum for this i atom from the interaction with this j atom. */
388 velec = _mm_andnot_ps(dummy_mask,velec);
389 velecsum = _mm_add_ps(velecsum,velec);
390 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
391 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
392
393 fscal = _mm_add_ps(felec,fvdw);
394
395 fscal = _mm_andnot_ps(dummy_mask,fscal);
396
397 /* Calculate temporary vectorial force */
398 tx = _mm_mul_ps(fscal,dx00);
399 ty = _mm_mul_ps(fscal,dy00);
400 tz = _mm_mul_ps(fscal,dz00);
401
402 /* Update vectorial force */
403 fix0 = _mm_add_ps(fix0,tx);
404 fiy0 = _mm_add_ps(fiy0,ty);
405 fiz0 = _mm_add_ps(fiz0,tz);
406
407 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
408 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
409 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
410 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
411 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
412
413 /* Inner loop uses 64 flops */
414 }
415
416 /* End of innermost loop */
417
418 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
419 f+i_coord_offset,fshift+i_shift_offset);
420
421 ggid = gid[iidx];
422 /* Update potential energies */
423 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
424 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
425
426 /* Increment number of inner iterations */
427 inneriter += j_index_end - j_index_start;
428
429 /* Outer loop uses 9 flops */
430 }
431
432 /* Increment number of outer iterations */
433 outeriter += nri;
434
435 /* Update outer/inner flops */
436
437 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*64);
438 }
439 /*
440 * Gromacs nonbonded kernel: nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse2_single
441 * Electrostatics interaction: Coulomb
442 * VdW interaction: CubicSplineTable
443 * Geometry: Particle-Particle
444 * Calculate force/pot: Force
445 */
446 void
447 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse2_single
448 (t_nblist * gmx_restrict nlist,
449 rvec * gmx_restrict xx,
450 rvec * gmx_restrict ff,
451 t_forcerec * gmx_restrict fr,
452 t_mdatoms * gmx_restrict mdatoms,
453 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
454 t_nrnb * gmx_restrict nrnb)
455 {
456 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
457 * just 0 for non-waters.
458 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
459 * jnr indices corresponding to data put in the four positions in the SIMD register.
460 */
461 int i_shift_offset,i_coord_offset,outeriter,inneriter;
462 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
463 int jnrA,jnrB,jnrC,jnrD;
464 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
465 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
466 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
467 real rcutoff_scalar;
468 real *shiftvec,*fshift,*x,*f;
469 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
470 real scratch[4*DIM];
471 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
472 int vdwioffset0;
473 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
474 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
475 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
476 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
477 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
478 real *charge;
479 int nvdwtype;
480 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
481 int *vdwtype;
482 real *vdwparam;
483 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
484 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
485 __m128i vfitab;
486 __m128i ifour = _mm_set1_epi32(4);
487 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
488 real *vftab;
489 __m128 dummy_mask,cutoff_mask;
490 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
491 __m128 one = _mm_set1_ps(1.0);
492 __m128 two = _mm_set1_ps(2.0);
493 x = xx[0];
494 f = ff[0];
495
496 nri = nlist->nri;
497 iinr = nlist->iinr;
498 jindex = nlist->jindex;
499 jjnr = nlist->jjnr;
500 shiftidx = nlist->shift;
501 gid = nlist->gid;
502 shiftvec = fr->shift_vec[0];
503 fshift = fr->fshift[0];
504 facel = _mm_set1_ps(fr->epsfac);
505 charge = mdatoms->chargeA;
506 nvdwtype = fr->ntype;
507 vdwparam = fr->nbfp;
508 vdwtype = mdatoms->typeA;
509
510 vftab = kernel_data->table_vdw->data;
511 vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale);
512
513 /* Avoid stupid compiler warnings */
514 jnrA = jnrB = jnrC = jnrD = 0;
515 j_coord_offsetA = 0;
516 j_coord_offsetB = 0;
517 j_coord_offsetC = 0;
518 j_coord_offsetD = 0;
519
520 outeriter = 0;
521 inneriter = 0;
522
523 for(iidx=0;iidx<4*DIM;iidx++)
524 {
525 scratch[iidx] = 0.0;
526 }
527
528 /* Start outer loop over neighborlists */
529 for(iidx=0; iidx<nri; iidx++)
530 {
531 /* Load shift vector for this list */
532 i_shift_offset = DIM*shiftidx[iidx];
533
534 /* Load limits for loop over neighbors */
535 j_index_start = jindex[iidx];
536 j_index_end = jindex[iidx+1];
537
538 /* Get outer coordinate index */
539 inr = iinr[iidx];
540 i_coord_offset = DIM*inr;
541
542 /* Load i particle coords and add shift vector */
543 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
544
545 fix0 = _mm_setzero_ps();
546 fiy0 = _mm_setzero_ps();
547 fiz0 = _mm_setzero_ps();
548
549 /* Load parameters for i particles */
550 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
551 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
552
553 /* Start inner kernel loop */
554 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
555 {
556
557 /* Get j neighbor index, and coordinate index */
558 jnrA = jjnr[jidx];
559 jnrB = jjnr[jidx+1];
560 jnrC = jjnr[jidx+2];
561 jnrD = jjnr[jidx+3];
562 j_coord_offsetA = DIM*jnrA;
563 j_coord_offsetB = DIM*jnrB;
564 j_coord_offsetC = DIM*jnrC;
565 j_coord_offsetD = DIM*jnrD;
566
567 /* load j atom coordinates */
568 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
569 x+j_coord_offsetC,x+j_coord_offsetD,
570 &jx0,&jy0,&jz0);
571
572 /* Calculate displacement vector */
573 dx00 = _mm_sub_ps(ix0,jx0);
574 dy00 = _mm_sub_ps(iy0,jy0);
575 dz00 = _mm_sub_ps(iz0,jz0);
576
577 /* Calculate squared distance and things based on it */
578 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
579
580 rinv00 = gmx_mm_invsqrt_ps(rsq00);
581
582 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
583
584 /* Load parameters for j particles */
585 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
586 charge+jnrC+0,charge+jnrD+0);
587 vdwjidx0A = 2*vdwtype[jnrA+0];
588 vdwjidx0B = 2*vdwtype[jnrB+0];
589 vdwjidx0C = 2*vdwtype[jnrC+0];
590 vdwjidx0D = 2*vdwtype[jnrD+0];
591
592 /**************************
593 * CALCULATE INTERACTIONS *
594 **************************/
595
596 r00 = _mm_mul_ps(rsq00,rinv00);
597
598 /* Compute parameters for interactions between i and j atoms */
599 qq00 = _mm_mul_ps(iq0,jq0);
600 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
601 vdwparam+vdwioffset0+vdwjidx0B,
602 vdwparam+vdwioffset0+vdwjidx0C,
603 vdwparam+vdwioffset0+vdwjidx0D,
604 &c6_00,&c12_00);
605
606 /* Calculate table index by multiplying r with table scale and truncate to integer */
607 rt = _mm_mul_ps(r00,vftabscale);
608 vfitab = _mm_cvttps_epi32(rt);
609 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
610 vfitab = _mm_slli_epi32(vfitab,3);
611
612 /* COULOMB ELECTROSTATICS */
613 velec = _mm_mul_ps(qq00,rinv00);
614 felec = _mm_mul_ps(velec,rinvsq00);
615
616 /* CUBIC SPLINE TABLE DISPERSION */
617 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
618 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
619 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
620 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
621 _MM_TRANSPOSE4_PS(Y,F,G,H);
622 Heps = _mm_mul_ps(vfeps,H);
623 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
624 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
625 fvdw6 = _mm_mul_ps(c6_00,FF);
626
627 /* CUBIC SPLINE TABLE REPULSION */
628 vfitab = _mm_add_epi32(vfitab,ifour);
629 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
630 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
631 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
632 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
633 _MM_TRANSPOSE4_PS(Y,F,G,H);
634 Heps = _mm_mul_ps(vfeps,H);
635 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
636 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
637 fvdw12 = _mm_mul_ps(c12_00,FF);
638 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
639
640 fscal = _mm_add_ps(felec,fvdw);
641
642 /* Calculate temporary vectorial force */
643 tx = _mm_mul_ps(fscal,dx00);
644 ty = _mm_mul_ps(fscal,dy00);
645 tz = _mm_mul_ps(fscal,dz00);
646
647 /* Update vectorial force */
648 fix0 = _mm_add_ps(fix0,tx);
649 fiy0 = _mm_add_ps(fiy0,ty);
650 fiz0 = _mm_add_ps(fiz0,tz);
651
652 fjptrA = f+j_coord_offsetA;
653 fjptrB = f+j_coord_offsetB;
654 fjptrC = f+j_coord_offsetC;
655 fjptrD = f+j_coord_offsetD;
656 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
657
658 /* Inner loop uses 54 flops */
659 }
660
661 if(jidx<j_index_end)
662 {
663
664 /* Get j neighbor index, and coordinate index */
665 jnrlistA = jjnr[jidx];
666 jnrlistB = jjnr[jidx+1];
667 jnrlistC = jjnr[jidx+2];
668 jnrlistD = jjnr[jidx+3];
669 /* Sign of each element will be negative for non-real atoms.
670 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
671 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
672 */
673 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
674 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
675 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
676 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
677 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
678 j_coord_offsetA = DIM*jnrA;
679 j_coord_offsetB = DIM*jnrB;
680 j_coord_offsetC = DIM*jnrC;
681 j_coord_offsetD = DIM*jnrD;
682
683 /* load j atom coordinates */
684 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
685 x+j_coord_offsetC,x+j_coord_offsetD,
686 &jx0,&jy0,&jz0);
687
688 /* Calculate displacement vector */
689 dx00 = _mm_sub_ps(ix0,jx0);
690 dy00 = _mm_sub_ps(iy0,jy0);
691 dz00 = _mm_sub_ps(iz0,jz0);
692
693 /* Calculate squared distance and things based on it */
694 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
695
696 rinv00 = gmx_mm_invsqrt_ps(rsq00);
697
698 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
699
700 /* Load parameters for j particles */
701 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
702 charge+jnrC+0,charge+jnrD+0);
703 vdwjidx0A = 2*vdwtype[jnrA+0];
704 vdwjidx0B = 2*vdwtype[jnrB+0];
705 vdwjidx0C = 2*vdwtype[jnrC+0];
706 vdwjidx0D = 2*vdwtype[jnrD+0];
707
708 /**************************
709 * CALCULATE INTERACTIONS *
710 **************************/
711
712 r00 = _mm_mul_ps(rsq00,rinv00);
713 r00 = _mm_andnot_ps(dummy_mask,r00);
714
715 /* Compute parameters for interactions between i and j atoms */
716 qq00 = _mm_mul_ps(iq0,jq0);
717 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
718 vdwparam+vdwioffset0+vdwjidx0B,
719 vdwparam+vdwioffset0+vdwjidx0C,
720 vdwparam+vdwioffset0+vdwjidx0D,
721 &c6_00,&c12_00);
722
723 /* Calculate table index by multiplying r with table scale and truncate to integer */
724 rt = _mm_mul_ps(r00,vftabscale);
725 vfitab = _mm_cvttps_epi32(rt);
726 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
727 vfitab = _mm_slli_epi32(vfitab,3);
728
729 /* COULOMB ELECTROSTATICS */
730 velec = _mm_mul_ps(qq00,rinv00);
731 felec = _mm_mul_ps(velec,rinvsq00);
732
733 /* CUBIC SPLINE TABLE DISPERSION */
734 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
735 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
736 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
737 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
738 _MM_TRANSPOSE4_PS(Y,F,G,H);
739 Heps = _mm_mul_ps(vfeps,H);
740 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
741 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
742 fvdw6 = _mm_mul_ps(c6_00,FF);
743
744 /* CUBIC SPLINE TABLE REPULSION */
745 vfitab = _mm_add_epi32(vfitab,ifour);
746 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
747 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
748 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
749 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
750 _MM_TRANSPOSE4_PS(Y,F,G,H);
751 Heps = _mm_mul_ps(vfeps,H);
752 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
753 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
754 fvdw12 = _mm_mul_ps(c12_00,FF);
755 fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
756
757 fscal = _mm_add_ps(felec,fvdw);
758
759 fscal = _mm_andnot_ps(dummy_mask,fscal);
760
761 /* Calculate temporary vectorial force */
762 tx = _mm_mul_ps(fscal,dx00);
763 ty = _mm_mul_ps(fscal,dy00);
764 tz = _mm_mul_ps(fscal,dz00);
765
766 /* Update vectorial force */
767 fix0 = _mm_add_ps(fix0,tx);
768 fiy0 = _mm_add_ps(fiy0,ty);
769 fiz0 = _mm_add_ps(fiz0,tz);
770
771 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
772 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
773 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
774 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
775 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
776
777 /* Inner loop uses 55 flops */
778 }
779
780 /* End of innermost loop */
781
782 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
783 f+i_coord_offset,fshift+i_shift_offset);
784
785 /* Increment number of inner iterations */
786 inneriter += j_index_end - j_index_start;
787
788 /* Outer loop uses 7 flops */
789 }
790
791 /* Increment number of outer iterations */
792 outeriter += nri;
793
794 /* Update outer/inner flops */
795
796 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*55);
797 }
The ultimate molecular dynamics simulation package