?  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_ElecCSTab_VdwNone_GeomP1P1_sse2_single.c
blob8219cd416617d3e5c0a8536b71eed76c9258cece
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_ElecCSTab_VdwNone_GeomP1P1_VF_sse2_single
53 * Electrostatics interaction: CubicSplineTable
54 * VdW interaction: None
55 * Geometry: Particle-Particle
56 * Calculate force/pot: PotentialAndForce
57 */
58 void
59 nb_kernel_ElecCSTab_VdwNone_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 __m128i vfitab;
92 __m128i ifour = _mm_set1_epi32(4);
93 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
94 real *vftab;
95 __m128 dummy_mask,cutoff_mask;
96 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
97 __m128 one = _mm_set1_ps(1.0);
98 __m128 two = _mm_set1_ps(2.0);
99 x = xx[0];
100 f = ff[0];
101
102 nri = nlist->nri;
103 iinr = nlist->iinr;
104 jindex = nlist->jindex;
105 jjnr = nlist->jjnr;
106 shiftidx = nlist->shift;
107 gid = nlist->gid;
108 shiftvec = fr->shift_vec[0];
109 fshift = fr->fshift[0];
110 facel = _mm_set1_ps(fr->epsfac);
111 charge = mdatoms->chargeA;
112
113 vftab = kernel_data->table_elec->data;
114 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
115
116 /* Avoid stupid compiler warnings */
117 jnrA = jnrB = jnrC = jnrD = 0;
118 j_coord_offsetA = 0;
119 j_coord_offsetB = 0;
120 j_coord_offsetC = 0;
121 j_coord_offsetD = 0;
122
123 outeriter = 0;
124 inneriter = 0;
125
126 for(iidx=0;iidx<4*DIM;iidx++)
127 {
128 scratch[iidx] = 0.0;
129 }
130
131 /* Start outer loop over neighborlists */
132 for(iidx=0; iidx<nri; iidx++)
133 {
134 /* Load shift vector for this list */
135 i_shift_offset = DIM*shiftidx[iidx];
136
137 /* Load limits for loop over neighbors */
138 j_index_start = jindex[iidx];
139 j_index_end = jindex[iidx+1];
140
141 /* Get outer coordinate index */
142 inr = iinr[iidx];
143 i_coord_offset = DIM*inr;
144
145 /* Load i particle coords and add shift vector */
146 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
147
148 fix0 = _mm_setzero_ps();
149 fiy0 = _mm_setzero_ps();
150 fiz0 = _mm_setzero_ps();
151
152 /* Load parameters for i particles */
153 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
154
155 /* Reset potential sums */
156 velecsum = _mm_setzero_ps();
157
158 /* Start inner kernel loop */
159 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
160 {
161
162 /* Get j neighbor index, and coordinate index */
163 jnrA = jjnr[jidx];
164 jnrB = jjnr[jidx+1];
165 jnrC = jjnr[jidx+2];
166 jnrD = jjnr[jidx+3];
167 j_coord_offsetA = DIM*jnrA;
168 j_coord_offsetB = DIM*jnrB;
169 j_coord_offsetC = DIM*jnrC;
170 j_coord_offsetD = DIM*jnrD;
171
172 /* load j atom coordinates */
173 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
174 x+j_coord_offsetC,x+j_coord_offsetD,
175 &jx0,&jy0,&jz0);
176
177 /* Calculate displacement vector */
178 dx00 = _mm_sub_ps(ix0,jx0);
179 dy00 = _mm_sub_ps(iy0,jy0);
180 dz00 = _mm_sub_ps(iz0,jz0);
181
182 /* Calculate squared distance and things based on it */
183 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
184
185 rinv00 = gmx_mm_invsqrt_ps(rsq00);
186
187 /* Load parameters for j particles */
188 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
189 charge+jnrC+0,charge+jnrD+0);
190
191 /**************************
192 * CALCULATE INTERACTIONS *
193 **************************/
194
195 r00 = _mm_mul_ps(rsq00,rinv00);
196
197 /* Compute parameters for interactions between i and j atoms */
198 qq00 = _mm_mul_ps(iq0,jq0);
199
200 /* Calculate table index by multiplying r with table scale and truncate to integer */
201 rt = _mm_mul_ps(r00,vftabscale);
202 vfitab = _mm_cvttps_epi32(rt);
203 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
204 vfitab = _mm_slli_epi32(vfitab,2);
205
206 /* CUBIC SPLINE TABLE ELECTROSTATICS */
207 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
208 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
209 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
210 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
211 _MM_TRANSPOSE4_PS(Y,F,G,H);
212 Heps = _mm_mul_ps(vfeps,H);
213 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
214 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
215 velec = _mm_mul_ps(qq00,VV);
216 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
217 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
218
219 /* Update potential sum for this i atom from the interaction with this j atom. */
220 velecsum = _mm_add_ps(velecsum,velec);
221
222 fscal = felec;
223
224 /* Calculate temporary vectorial force */
225 tx = _mm_mul_ps(fscal,dx00);
226 ty = _mm_mul_ps(fscal,dy00);
227 tz = _mm_mul_ps(fscal,dz00);
228
229 /* Update vectorial force */
230 fix0 = _mm_add_ps(fix0,tx);
231 fiy0 = _mm_add_ps(fiy0,ty);
232 fiz0 = _mm_add_ps(fiz0,tz);
233
234 fjptrA = f+j_coord_offsetA;
235 fjptrB = f+j_coord_offsetB;
236 fjptrC = f+j_coord_offsetC;
237 fjptrD = f+j_coord_offsetD;
238 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
239
240 /* Inner loop uses 43 flops */
241 }
242
243 if(jidx<j_index_end)
244 {
245
246 /* Get j neighbor index, and coordinate index */
247 jnrlistA = jjnr[jidx];
248 jnrlistB = jjnr[jidx+1];
249 jnrlistC = jjnr[jidx+2];
250 jnrlistD = jjnr[jidx+3];
251 /* Sign of each element will be negative for non-real atoms.
252 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
253 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
254 */
255 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
256 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
257 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
258 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
259 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
260 j_coord_offsetA = DIM*jnrA;
261 j_coord_offsetB = DIM*jnrB;
262 j_coord_offsetC = DIM*jnrC;
263 j_coord_offsetD = DIM*jnrD;
264
265 /* load j atom coordinates */
266 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
267 x+j_coord_offsetC,x+j_coord_offsetD,
268 &jx0,&jy0,&jz0);
269
270 /* Calculate displacement vector */
271 dx00 = _mm_sub_ps(ix0,jx0);
272 dy00 = _mm_sub_ps(iy0,jy0);
273 dz00 = _mm_sub_ps(iz0,jz0);
274
275 /* Calculate squared distance and things based on it */
276 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
277
278 rinv00 = gmx_mm_invsqrt_ps(rsq00);
279
280 /* Load parameters for j particles */
281 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
282 charge+jnrC+0,charge+jnrD+0);
283
284 /**************************
285 * CALCULATE INTERACTIONS *
286 **************************/
287
288 r00 = _mm_mul_ps(rsq00,rinv00);
289 r00 = _mm_andnot_ps(dummy_mask,r00);
290
291 /* Compute parameters for interactions between i and j atoms */
292 qq00 = _mm_mul_ps(iq0,jq0);
293
294 /* Calculate table index by multiplying r with table scale and truncate to integer */
295 rt = _mm_mul_ps(r00,vftabscale);
296 vfitab = _mm_cvttps_epi32(rt);
297 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
298 vfitab = _mm_slli_epi32(vfitab,2);
299
300 /* CUBIC SPLINE TABLE ELECTROSTATICS */
301 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
302 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
303 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
304 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
305 _MM_TRANSPOSE4_PS(Y,F,G,H);
306 Heps = _mm_mul_ps(vfeps,H);
307 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
308 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
309 velec = _mm_mul_ps(qq00,VV);
310 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
311 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
312
313 /* Update potential sum for this i atom from the interaction with this j atom. */
314 velec = _mm_andnot_ps(dummy_mask,velec);
315 velecsum = _mm_add_ps(velecsum,velec);
316
317 fscal = felec;
318
319 fscal = _mm_andnot_ps(dummy_mask,fscal);
320
321 /* Calculate temporary vectorial force */
322 tx = _mm_mul_ps(fscal,dx00);
323 ty = _mm_mul_ps(fscal,dy00);
324 tz = _mm_mul_ps(fscal,dz00);
325
326 /* Update vectorial force */
327 fix0 = _mm_add_ps(fix0,tx);
328 fiy0 = _mm_add_ps(fiy0,ty);
329 fiz0 = _mm_add_ps(fiz0,tz);
330
331 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
332 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
333 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
334 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
335 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
336
337 /* Inner loop uses 44 flops */
338 }
339
340 /* End of innermost loop */
341
342 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
343 f+i_coord_offset,fshift+i_shift_offset);
344
345 ggid = gid[iidx];
346 /* Update potential energies */
347 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
348
349 /* Increment number of inner iterations */
350 inneriter += j_index_end - j_index_start;
351
352 /* Outer loop uses 8 flops */
353 }
354
355 /* Increment number of outer iterations */
356 outeriter += nri;
357
358 /* Update outer/inner flops */
359
360 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*44);
361 }
362 /*
363 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse2_single
364 * Electrostatics interaction: CubicSplineTable
365 * VdW interaction: None
366 * Geometry: Particle-Particle
367 * Calculate force/pot: Force
368 */
369 void
370 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse2_single
371 (t_nblist * gmx_restrict nlist,
372 rvec * gmx_restrict xx,
373 rvec * gmx_restrict ff,
374 t_forcerec * gmx_restrict fr,
375 t_mdatoms * gmx_restrict mdatoms,
376 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
377 t_nrnb * gmx_restrict nrnb)
378 {
379 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
380 * just 0 for non-waters.
381 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
382 * jnr indices corresponding to data put in the four positions in the SIMD register.
383 */
384 int i_shift_offset,i_coord_offset,outeriter,inneriter;
385 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
386 int jnrA,jnrB,jnrC,jnrD;
387 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
388 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
389 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
390 real rcutoff_scalar;
391 real *shiftvec,*fshift,*x,*f;
392 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
393 real scratch[4*DIM];
394 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
395 int vdwioffset0;
396 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
397 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
398 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
399 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
400 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
401 real *charge;
402 __m128i vfitab;
403 __m128i ifour = _mm_set1_epi32(4);
404 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
405 real *vftab;
406 __m128 dummy_mask,cutoff_mask;
407 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
408 __m128 one = _mm_set1_ps(1.0);
409 __m128 two = _mm_set1_ps(2.0);
410 x = xx[0];
411 f = ff[0];
412
413 nri = nlist->nri;
414 iinr = nlist->iinr;
415 jindex = nlist->jindex;
416 jjnr = nlist->jjnr;
417 shiftidx = nlist->shift;
418 gid = nlist->gid;
419 shiftvec = fr->shift_vec[0];
420 fshift = fr->fshift[0];
421 facel = _mm_set1_ps(fr->epsfac);
422 charge = mdatoms->chargeA;
423
424 vftab = kernel_data->table_elec->data;
425 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
426
427 /* Avoid stupid compiler warnings */
428 jnrA = jnrB = jnrC = jnrD = 0;
429 j_coord_offsetA = 0;
430 j_coord_offsetB = 0;
431 j_coord_offsetC = 0;
432 j_coord_offsetD = 0;
433
434 outeriter = 0;
435 inneriter = 0;
436
437 for(iidx=0;iidx<4*DIM;iidx++)
438 {
439 scratch[iidx] = 0.0;
440 }
441
442 /* Start outer loop over neighborlists */
443 for(iidx=0; iidx<nri; iidx++)
444 {
445 /* Load shift vector for this list */
446 i_shift_offset = DIM*shiftidx[iidx];
447
448 /* Load limits for loop over neighbors */
449 j_index_start = jindex[iidx];
450 j_index_end = jindex[iidx+1];
451
452 /* Get outer coordinate index */
453 inr = iinr[iidx];
454 i_coord_offset = DIM*inr;
455
456 /* Load i particle coords and add shift vector */
457 gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
458
459 fix0 = _mm_setzero_ps();
460 fiy0 = _mm_setzero_ps();
461 fiz0 = _mm_setzero_ps();
462
463 /* Load parameters for i particles */
464 iq0 = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
465
466 /* Start inner kernel loop */
467 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
468 {
469
470 /* Get j neighbor index, and coordinate index */
471 jnrA = jjnr[jidx];
472 jnrB = jjnr[jidx+1];
473 jnrC = jjnr[jidx+2];
474 jnrD = jjnr[jidx+3];
475 j_coord_offsetA = DIM*jnrA;
476 j_coord_offsetB = DIM*jnrB;
477 j_coord_offsetC = DIM*jnrC;
478 j_coord_offsetD = DIM*jnrD;
479
480 /* load j atom coordinates */
481 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
482 x+j_coord_offsetC,x+j_coord_offsetD,
483 &jx0,&jy0,&jz0);
484
485 /* Calculate displacement vector */
486 dx00 = _mm_sub_ps(ix0,jx0);
487 dy00 = _mm_sub_ps(iy0,jy0);
488 dz00 = _mm_sub_ps(iz0,jz0);
489
490 /* Calculate squared distance and things based on it */
491 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
492
493 rinv00 = gmx_mm_invsqrt_ps(rsq00);
494
495 /* Load parameters for j particles */
496 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
497 charge+jnrC+0,charge+jnrD+0);
498
499 /**************************
500 * CALCULATE INTERACTIONS *
501 **************************/
502
503 r00 = _mm_mul_ps(rsq00,rinv00);
504
505 /* Compute parameters for interactions between i and j atoms */
506 qq00 = _mm_mul_ps(iq0,jq0);
507
508 /* Calculate table index by multiplying r with table scale and truncate to integer */
509 rt = _mm_mul_ps(r00,vftabscale);
510 vfitab = _mm_cvttps_epi32(rt);
511 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
512 vfitab = _mm_slli_epi32(vfitab,2);
513
514 /* CUBIC SPLINE TABLE ELECTROSTATICS */
515 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
516 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
517 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
518 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
519 _MM_TRANSPOSE4_PS(Y,F,G,H);
520 Heps = _mm_mul_ps(vfeps,H);
521 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
522 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
523 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
524
525 fscal = felec;
526
527 /* Calculate temporary vectorial force */
528 tx = _mm_mul_ps(fscal,dx00);
529 ty = _mm_mul_ps(fscal,dy00);
530 tz = _mm_mul_ps(fscal,dz00);
531
532 /* Update vectorial force */
533 fix0 = _mm_add_ps(fix0,tx);
534 fiy0 = _mm_add_ps(fiy0,ty);
535 fiz0 = _mm_add_ps(fiz0,tz);
536
537 fjptrA = f+j_coord_offsetA;
538 fjptrB = f+j_coord_offsetB;
539 fjptrC = f+j_coord_offsetC;
540 fjptrD = f+j_coord_offsetD;
541 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
542
543 /* Inner loop uses 39 flops */
544 }
545
546 if(jidx<j_index_end)
547 {
548
549 /* Get j neighbor index, and coordinate index */
550 jnrlistA = jjnr[jidx];
551 jnrlistB = jjnr[jidx+1];
552 jnrlistC = jjnr[jidx+2];
553 jnrlistD = jjnr[jidx+3];
554 /* Sign of each element will be negative for non-real atoms.
555 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
556 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
557 */
558 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
559 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
560 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
561 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
562 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
563 j_coord_offsetA = DIM*jnrA;
564 j_coord_offsetB = DIM*jnrB;
565 j_coord_offsetC = DIM*jnrC;
566 j_coord_offsetD = DIM*jnrD;
567
568 /* load j atom coordinates */
569 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
570 x+j_coord_offsetC,x+j_coord_offsetD,
571 &jx0,&jy0,&jz0);
572
573 /* Calculate displacement vector */
574 dx00 = _mm_sub_ps(ix0,jx0);
575 dy00 = _mm_sub_ps(iy0,jy0);
576 dz00 = _mm_sub_ps(iz0,jz0);
577
578 /* Calculate squared distance and things based on it */
579 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
580
581 rinv00 = gmx_mm_invsqrt_ps(rsq00);
582
583 /* Load parameters for j particles */
584 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
585 charge+jnrC+0,charge+jnrD+0);
586
587 /**************************
588 * CALCULATE INTERACTIONS *
589 **************************/
590
591 r00 = _mm_mul_ps(rsq00,rinv00);
592 r00 = _mm_andnot_ps(dummy_mask,r00);
593
594 /* Compute parameters for interactions between i and j atoms */
595 qq00 = _mm_mul_ps(iq0,jq0);
596
597 /* Calculate table index by multiplying r with table scale and truncate to integer */
598 rt = _mm_mul_ps(r00,vftabscale);
599 vfitab = _mm_cvttps_epi32(rt);
600 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
601 vfitab = _mm_slli_epi32(vfitab,2);
602
603 /* CUBIC SPLINE TABLE ELECTROSTATICS */
604 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
605 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
606 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
607 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
608 _MM_TRANSPOSE4_PS(Y,F,G,H);
609 Heps = _mm_mul_ps(vfeps,H);
610 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
611 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
612 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
613
614 fscal = felec;
615
616 fscal = _mm_andnot_ps(dummy_mask,fscal);
617
618 /* Calculate temporary vectorial force */
619 tx = _mm_mul_ps(fscal,dx00);
620 ty = _mm_mul_ps(fscal,dy00);
621 tz = _mm_mul_ps(fscal,dz00);
622
623 /* Update vectorial force */
624 fix0 = _mm_add_ps(fix0,tx);
625 fiy0 = _mm_add_ps(fiy0,ty);
626 fiz0 = _mm_add_ps(fiz0,tz);
627
628 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
629 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
630 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
631 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
632 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
633
634 /* Inner loop uses 40 flops */
635 }
636
637 /* End of innermost loop */
638
639 gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
640 f+i_coord_offset,fshift+i_shift_offset);
641
642 /* Increment number of inner iterations */
643 inneriter += j_index_end - j_index_start;
644
645 /* Outer loop uses 7 flops */
646 }
647
648 /* Increment number of outer iterations */
649 outeriter += nri;
650
651 /* Update outer/inner flops */
652
653 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*40);
654 }
The ultimate molecular dynamics simulation package