?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Removed include simple.h from nb_kernel_avx_128_fma_XX
[gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_avx_128_fma_single / nb_kernel_ElecNone_VdwLJSw_GeomP1P1_avx_128_fma_single.c
blob6c151e92d7154d2f2c999b6f377ecc2cb9e70d89
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 avx_128_fma_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_avx_128_fma_single.h"
49 #include "kernelutil_x86_avx_128_fma_single.h"
50
51 /*
52 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_avx_128_fma_single
53 * Electrostatics interaction: None
54 * VdW interaction: LennardJones
55 * Geometry: Particle-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_avx_128_fma_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 AVX_128, 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 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 int nvdwtype;
90 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
91 int *vdwtype;
92 real *vdwparam;
93 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
94 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
95 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
96 real rswitch_scalar,d_scalar;
97 __m128 dummy_mask,cutoff_mask;
98 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
99 __m128 one = _mm_set1_ps(1.0);
100 __m128 two = _mm_set1_ps(2.0);
101 x = xx[0];
102 f = ff[0];
103
104 nri = nlist->nri;
105 iinr = nlist->iinr;
106 jindex = nlist->jindex;
107 jjnr = nlist->jjnr;
108 shiftidx = nlist->shift;
109 gid = nlist->gid;
110 shiftvec = fr->shift_vec[0];
111 fshift = fr->fshift[0];
112 nvdwtype = fr->ntype;
113 vdwparam = fr->nbfp;
114 vdwtype = mdatoms->typeA;
115
116 rcutoff_scalar = fr->rvdw;
117 rcutoff = _mm_set1_ps(rcutoff_scalar);
118 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
119
120 rswitch_scalar = fr->rvdw_switch;
121 rswitch = _mm_set1_ps(rswitch_scalar);
122 /* Setup switch parameters */
123 d_scalar = rcutoff_scalar-rswitch_scalar;
124 d = _mm_set1_ps(d_scalar);
125 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
126 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
127 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
128 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
129 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
130 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
131
132 /* Avoid stupid compiler warnings */
133 jnrA = jnrB = jnrC = jnrD = 0;
134 j_coord_offsetA = 0;
135 j_coord_offsetB = 0;
136 j_coord_offsetC = 0;
137 j_coord_offsetD = 0;
138
139 outeriter = 0;
140 inneriter = 0;
141
142 for(iidx=0;iidx<4*DIM;iidx++)
143 {
144 scratch[iidx] = 0.0;
145 }
146
147 /* Start outer loop over neighborlists */
148 for(iidx=0; iidx<nri; iidx++)
149 {
150 /* Load shift vector for this list */
151 i_shift_offset = DIM*shiftidx[iidx];
152
153 /* Load limits for loop over neighbors */
154 j_index_start = jindex[iidx];
155 j_index_end = jindex[iidx+1];
156
157 /* Get outer coordinate index */
158 inr = iinr[iidx];
159 i_coord_offset = DIM*inr;
160
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);
163
164 fix0 = _mm_setzero_ps();
165 fiy0 = _mm_setzero_ps();
166 fiz0 = _mm_setzero_ps();
167
168 /* Load parameters for i particles */
169 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
170
171 /* Reset potential sums */
172 vvdwsum = _mm_setzero_ps();
173
174 /* Start inner kernel loop */
175 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
176 {
177
178 /* Get j neighbor index, and coordinate index */
179 jnrA = jjnr[jidx];
180 jnrB = jjnr[jidx+1];
181 jnrC = jjnr[jidx+2];
182 jnrD = jjnr[jidx+3];
183 j_coord_offsetA = DIM*jnrA;
184 j_coord_offsetB = DIM*jnrB;
185 j_coord_offsetC = DIM*jnrC;
186 j_coord_offsetD = DIM*jnrD;
187
188 /* load j atom coordinates */
189 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
190 x+j_coord_offsetC,x+j_coord_offsetD,
191 &jx0,&jy0,&jz0);
192
193 /* Calculate displacement vector */
194 dx00 = _mm_sub_ps(ix0,jx0);
195 dy00 = _mm_sub_ps(iy0,jy0);
196 dz00 = _mm_sub_ps(iz0,jz0);
197
198 /* Calculate squared distance and things based on it */
199 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
200
201 rinv00 = gmx_mm_invsqrt_ps(rsq00);
202
203 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
204
205 /* Load parameters for j particles */
206 vdwjidx0A = 2*vdwtype[jnrA+0];
207 vdwjidx0B = 2*vdwtype[jnrB+0];
208 vdwjidx0C = 2*vdwtype[jnrC+0];
209 vdwjidx0D = 2*vdwtype[jnrD+0];
210
211 /**************************
212 * CALCULATE INTERACTIONS *
213 **************************/
214
215 if (gmx_mm_any_lt(rsq00,rcutoff2))
216 {
217
218 r00 = _mm_mul_ps(rsq00,rinv00);
219
220 /* Compute parameters for interactions between i and j atoms */
221 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
222 vdwparam+vdwioffset0+vdwjidx0B,
223 vdwparam+vdwioffset0+vdwjidx0C,
224 vdwparam+vdwioffset0+vdwjidx0D,
225 &c6_00,&c12_00);
226
227 /* LENNARD-JONES DISPERSION/REPULSION */
228
229 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
230 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
231 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
232 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
233 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
234
235 d = _mm_sub_ps(r00,rswitch);
236 d = _mm_max_ps(d,_mm_setzero_ps());
237 d2 = _mm_mul_ps(d,d);
238 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
239
240 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
241
242 /* Evaluate switch function */
243 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
244 fvdw = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
245 vvdw = _mm_mul_ps(vvdw,sw);
246 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
247
248 /* Update potential sum for this i atom from the interaction with this j atom. */
249 vvdw = _mm_and_ps(vvdw,cutoff_mask);
250 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
251
252 fscal = fvdw;
253
254 fscal = _mm_and_ps(fscal,cutoff_mask);
255
256 /* Update vectorial force */
257 fix0 = _mm_macc_ps(dx00,fscal,fix0);
258 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
259 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
260
261 fjptrA = f+j_coord_offsetA;
262 fjptrB = f+j_coord_offsetB;
263 fjptrC = f+j_coord_offsetC;
264 fjptrD = f+j_coord_offsetD;
265 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
266 _mm_mul_ps(dx00,fscal),
267 _mm_mul_ps(dy00,fscal),
268 _mm_mul_ps(dz00,fscal));
269
270 }
271
272 /* Inner loop uses 62 flops */
273 }
274
275 if(jidx<j_index_end)
276 {
277
278 /* Get j neighbor index, and coordinate index */
279 jnrlistA = jjnr[jidx];
280 jnrlistB = jjnr[jidx+1];
281 jnrlistC = jjnr[jidx+2];
282 jnrlistD = jjnr[jidx+3];
283 /* Sign of each element will be negative for non-real atoms.
284 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
285 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
286 */
287 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
288 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
289 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
290 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
291 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
292 j_coord_offsetA = DIM*jnrA;
293 j_coord_offsetB = DIM*jnrB;
294 j_coord_offsetC = DIM*jnrC;
295 j_coord_offsetD = DIM*jnrD;
296
297 /* load j atom coordinates */
298 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
299 x+j_coord_offsetC,x+j_coord_offsetD,
300 &jx0,&jy0,&jz0);
301
302 /* Calculate displacement vector */
303 dx00 = _mm_sub_ps(ix0,jx0);
304 dy00 = _mm_sub_ps(iy0,jy0);
305 dz00 = _mm_sub_ps(iz0,jz0);
306
307 /* Calculate squared distance and things based on it */
308 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
309
310 rinv00 = gmx_mm_invsqrt_ps(rsq00);
311
312 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
313
314 /* Load parameters for j particles */
315 vdwjidx0A = 2*vdwtype[jnrA+0];
316 vdwjidx0B = 2*vdwtype[jnrB+0];
317 vdwjidx0C = 2*vdwtype[jnrC+0];
318 vdwjidx0D = 2*vdwtype[jnrD+0];
319
320 /**************************
321 * CALCULATE INTERACTIONS *
322 **************************/
323
324 if (gmx_mm_any_lt(rsq00,rcutoff2))
325 {
326
327 r00 = _mm_mul_ps(rsq00,rinv00);
328 r00 = _mm_andnot_ps(dummy_mask,r00);
329
330 /* Compute parameters for interactions between i and j atoms */
331 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
332 vdwparam+vdwioffset0+vdwjidx0B,
333 vdwparam+vdwioffset0+vdwjidx0C,
334 vdwparam+vdwioffset0+vdwjidx0D,
335 &c6_00,&c12_00);
336
337 /* LENNARD-JONES DISPERSION/REPULSION */
338
339 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
340 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
341 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
342 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
343 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
344
345 d = _mm_sub_ps(r00,rswitch);
346 d = _mm_max_ps(d,_mm_setzero_ps());
347 d2 = _mm_mul_ps(d,d);
348 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
349
350 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
351
352 /* Evaluate switch function */
353 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
354 fvdw = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
355 vvdw = _mm_mul_ps(vvdw,sw);
356 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
357
358 /* Update potential sum for this i atom from the interaction with this j atom. */
359 vvdw = _mm_and_ps(vvdw,cutoff_mask);
360 vvdw = _mm_andnot_ps(dummy_mask,vvdw);
361 vvdwsum = _mm_add_ps(vvdwsum,vvdw);
362
363 fscal = fvdw;
364
365 fscal = _mm_and_ps(fscal,cutoff_mask);
366
367 fscal = _mm_andnot_ps(dummy_mask,fscal);
368
369 /* Update vectorial force */
370 fix0 = _mm_macc_ps(dx00,fscal,fix0);
371 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
372 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
373
374 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
375 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
376 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
377 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
378 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
379 _mm_mul_ps(dx00,fscal),
380 _mm_mul_ps(dy00,fscal),
381 _mm_mul_ps(dz00,fscal));
382
383 }
384
385 /* Inner loop uses 63 flops */
386 }
387
388 /* End of innermost loop */
389
390 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
391 f+i_coord_offset,fshift+i_shift_offset);
392
393 ggid = gid[iidx];
394 /* Update potential energies */
395 gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
396
397 /* Increment number of inner iterations */
398 inneriter += j_index_end - j_index_start;
399
400 /* Outer loop uses 7 flops */
401 }
402
403 /* Increment number of outer iterations */
404 outeriter += nri;
405
406 /* Update outer/inner flops */
407
408 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*63);
409 }
410 /*
411 * Gromacs nonbonded kernel: nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_128_fma_single
412 * Electrostatics interaction: None
413 * VdW interaction: LennardJones
414 * Geometry: Particle-Particle
415 * Calculate force/pot: Force
416 */
417 void
418 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_128_fma_single
419 (t_nblist * gmx_restrict nlist,
420 rvec * gmx_restrict xx,
421 rvec * gmx_restrict ff,
422 t_forcerec * gmx_restrict fr,
423 t_mdatoms * gmx_restrict mdatoms,
424 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
425 t_nrnb * gmx_restrict nrnb)
426 {
427 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
428 * just 0 for non-waters.
429 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
430 * jnr indices corresponding to data put in the four positions in the SIMD register.
431 */
432 int i_shift_offset,i_coord_offset,outeriter,inneriter;
433 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
434 int jnrA,jnrB,jnrC,jnrD;
435 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
436 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
437 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
438 real rcutoff_scalar;
439 real *shiftvec,*fshift,*x,*f;
440 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
441 real scratch[4*DIM];
442 __m128 fscal,rcutoff,rcutoff2,jidxall;
443 int vdwioffset0;
444 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
445 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
446 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
447 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
448 int nvdwtype;
449 __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
450 int *vdwtype;
451 real *vdwparam;
452 __m128 one_sixth = _mm_set1_ps(1.0/6.0);
453 __m128 one_twelfth = _mm_set1_ps(1.0/12.0);
454 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
455 real rswitch_scalar,d_scalar;
456 __m128 dummy_mask,cutoff_mask;
457 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
458 __m128 one = _mm_set1_ps(1.0);
459 __m128 two = _mm_set1_ps(2.0);
460 x = xx[0];
461 f = ff[0];
462
463 nri = nlist->nri;
464 iinr = nlist->iinr;
465 jindex = nlist->jindex;
466 jjnr = nlist->jjnr;
467 shiftidx = nlist->shift;
468 gid = nlist->gid;
469 shiftvec = fr->shift_vec[0];
470 fshift = fr->fshift[0];
471 nvdwtype = fr->ntype;
472 vdwparam = fr->nbfp;
473 vdwtype = mdatoms->typeA;
474
475 rcutoff_scalar = fr->rvdw;
476 rcutoff = _mm_set1_ps(rcutoff_scalar);
477 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
478
479 rswitch_scalar = fr->rvdw_switch;
480 rswitch = _mm_set1_ps(rswitch_scalar);
481 /* Setup switch parameters */
482 d_scalar = rcutoff_scalar-rswitch_scalar;
483 d = _mm_set1_ps(d_scalar);
484 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
485 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
486 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
487 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
488 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
489 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
490
491 /* Avoid stupid compiler warnings */
492 jnrA = jnrB = jnrC = jnrD = 0;
493 j_coord_offsetA = 0;
494 j_coord_offsetB = 0;
495 j_coord_offsetC = 0;
496 j_coord_offsetD = 0;
497
498 outeriter = 0;
499 inneriter = 0;
500
501 for(iidx=0;iidx<4*DIM;iidx++)
502 {
503 scratch[iidx] = 0.0;
504 }
505
506 /* Start outer loop over neighborlists */
507 for(iidx=0; iidx<nri; iidx++)
508 {
509 /* Load shift vector for this list */
510 i_shift_offset = DIM*shiftidx[iidx];
511
512 /* Load limits for loop over neighbors */
513 j_index_start = jindex[iidx];
514 j_index_end = jindex[iidx+1];
515
516 /* Get outer coordinate index */
517 inr = iinr[iidx];
518 i_coord_offset = DIM*inr;
519
520 /* Load i particle coords and add shift vector */
521 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
522
523 fix0 = _mm_setzero_ps();
524 fiy0 = _mm_setzero_ps();
525 fiz0 = _mm_setzero_ps();
526
527 /* Load parameters for i particles */
528 vdwioffset0 = 2*nvdwtype*vdwtype[inr+0];
529
530 /* Start inner kernel loop */
531 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
532 {
533
534 /* Get j neighbor index, and coordinate index */
535 jnrA = jjnr[jidx];
536 jnrB = jjnr[jidx+1];
537 jnrC = jjnr[jidx+2];
538 jnrD = jjnr[jidx+3];
539 j_coord_offsetA = DIM*jnrA;
540 j_coord_offsetB = DIM*jnrB;
541 j_coord_offsetC = DIM*jnrC;
542 j_coord_offsetD = DIM*jnrD;
543
544 /* load j atom coordinates */
545 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
546 x+j_coord_offsetC,x+j_coord_offsetD,
547 &jx0,&jy0,&jz0);
548
549 /* Calculate displacement vector */
550 dx00 = _mm_sub_ps(ix0,jx0);
551 dy00 = _mm_sub_ps(iy0,jy0);
552 dz00 = _mm_sub_ps(iz0,jz0);
553
554 /* Calculate squared distance and things based on it */
555 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
556
557 rinv00 = gmx_mm_invsqrt_ps(rsq00);
558
559 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
560
561 /* Load parameters for j particles */
562 vdwjidx0A = 2*vdwtype[jnrA+0];
563 vdwjidx0B = 2*vdwtype[jnrB+0];
564 vdwjidx0C = 2*vdwtype[jnrC+0];
565 vdwjidx0D = 2*vdwtype[jnrD+0];
566
567 /**************************
568 * CALCULATE INTERACTIONS *
569 **************************/
570
571 if (gmx_mm_any_lt(rsq00,rcutoff2))
572 {
573
574 r00 = _mm_mul_ps(rsq00,rinv00);
575
576 /* Compute parameters for interactions between i and j atoms */
577 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
578 vdwparam+vdwioffset0+vdwjidx0B,
579 vdwparam+vdwioffset0+vdwjidx0C,
580 vdwparam+vdwioffset0+vdwjidx0D,
581 &c6_00,&c12_00);
582
583 /* LENNARD-JONES DISPERSION/REPULSION */
584
585 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
586 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
587 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
588 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
589 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
590
591 d = _mm_sub_ps(r00,rswitch);
592 d = _mm_max_ps(d,_mm_setzero_ps());
593 d2 = _mm_mul_ps(d,d);
594 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
595
596 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
597
598 /* Evaluate switch function */
599 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
600 fvdw = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
601 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
602
603 fscal = fvdw;
604
605 fscal = _mm_and_ps(fscal,cutoff_mask);
606
607 /* Update vectorial force */
608 fix0 = _mm_macc_ps(dx00,fscal,fix0);
609 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
610 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
611
612 fjptrA = f+j_coord_offsetA;
613 fjptrB = f+j_coord_offsetB;
614 fjptrC = f+j_coord_offsetC;
615 fjptrD = f+j_coord_offsetD;
616 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
617 _mm_mul_ps(dx00,fscal),
618 _mm_mul_ps(dy00,fscal),
619 _mm_mul_ps(dz00,fscal));
620
621 }
622
623 /* Inner loop uses 59 flops */
624 }
625
626 if(jidx<j_index_end)
627 {
628
629 /* Get j neighbor index, and coordinate index */
630 jnrlistA = jjnr[jidx];
631 jnrlistB = jjnr[jidx+1];
632 jnrlistC = jjnr[jidx+2];
633 jnrlistD = jjnr[jidx+3];
634 /* Sign of each element will be negative for non-real atoms.
635 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
636 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
637 */
638 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
639 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
640 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
641 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
642 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
643 j_coord_offsetA = DIM*jnrA;
644 j_coord_offsetB = DIM*jnrB;
645 j_coord_offsetC = DIM*jnrC;
646 j_coord_offsetD = DIM*jnrD;
647
648 /* load j atom coordinates */
649 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
650 x+j_coord_offsetC,x+j_coord_offsetD,
651 &jx0,&jy0,&jz0);
652
653 /* Calculate displacement vector */
654 dx00 = _mm_sub_ps(ix0,jx0);
655 dy00 = _mm_sub_ps(iy0,jy0);
656 dz00 = _mm_sub_ps(iz0,jz0);
657
658 /* Calculate squared distance and things based on it */
659 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
660
661 rinv00 = gmx_mm_invsqrt_ps(rsq00);
662
663 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
664
665 /* Load parameters for j particles */
666 vdwjidx0A = 2*vdwtype[jnrA+0];
667 vdwjidx0B = 2*vdwtype[jnrB+0];
668 vdwjidx0C = 2*vdwtype[jnrC+0];
669 vdwjidx0D = 2*vdwtype[jnrD+0];
670
671 /**************************
672 * CALCULATE INTERACTIONS *
673 **************************/
674
675 if (gmx_mm_any_lt(rsq00,rcutoff2))
676 {
677
678 r00 = _mm_mul_ps(rsq00,rinv00);
679 r00 = _mm_andnot_ps(dummy_mask,r00);
680
681 /* Compute parameters for interactions between i and j atoms */
682 gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
683 vdwparam+vdwioffset0+vdwjidx0B,
684 vdwparam+vdwioffset0+vdwjidx0C,
685 vdwparam+vdwioffset0+vdwjidx0D,
686 &c6_00,&c12_00);
687
688 /* LENNARD-JONES DISPERSION/REPULSION */
689
690 rinvsix = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
691 vvdw6 = _mm_mul_ps(c6_00,rinvsix);
692 vvdw12 = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
693 vvdw = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
694 fvdw = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
695
696 d = _mm_sub_ps(r00,rswitch);
697 d = _mm_max_ps(d,_mm_setzero_ps());
698 d2 = _mm_mul_ps(d,d);
699 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
700
701 dsw = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
702
703 /* Evaluate switch function */
704 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
705 fvdw = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
706 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
707
708 fscal = fvdw;
709
710 fscal = _mm_and_ps(fscal,cutoff_mask);
711
712 fscal = _mm_andnot_ps(dummy_mask,fscal);
713
714 /* Update vectorial force */
715 fix0 = _mm_macc_ps(dx00,fscal,fix0);
716 fiy0 = _mm_macc_ps(dy00,fscal,fiy0);
717 fiz0 = _mm_macc_ps(dz00,fscal,fiz0);
718
719 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
720 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
721 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
722 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
723 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
724 _mm_mul_ps(dx00,fscal),
725 _mm_mul_ps(dy00,fscal),
726 _mm_mul_ps(dz00,fscal));
727
728 }
729
730 /* Inner loop uses 60 flops */
731 }
732
733 /* End of innermost loop */
734
735 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
736 f+i_coord_offset,fshift+i_shift_offset);
737
738 /* Increment number of inner iterations */
739 inneriter += j_index_end - j_index_start;
740
741 /* Outer loop uses 6 flops */
742 }
743
744 /* Increment number of outer iterations */
745 outeriter += nri;
746
747 /* Update outer/inner flops */
748
749 inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*60);
750 }
The ultimate molecular dynamics simulation package