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 avx_128_fma_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_avx_128_fma_single.h"
48 #include "kernelutil_x86_avx_128_fma_single.h"
51 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_128_fma_single
52 * Electrostatics interaction: CubicSplineTable
53 * VdW interaction: None
54 * Geometry: Water4-Particle
55 * Calculate force/pot: PotentialAndForce
58 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_128_fma_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 AVX_128, 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 fscal,rcutoff,rcutoff2,jidxall;
84 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
86 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
88 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
89 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
90 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
91 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
92 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
93 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
94 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
97 __m128i ifour = _mm_set1_epi32(4);
98 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
100 __m128 dummy_mask,cutoff_mask;
101 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
102 __m128 one = _mm_set1_ps(1.0);
103 __m128 two = _mm_set1_ps(2.0);
109 jindex = nlist->jindex;
111 shiftidx = nlist->shift;
113 shiftvec = fr->shift_vec[0];
114 fshift = fr->fshift[0];
115 facel = _mm_set1_ps(fr->epsfac);
116 charge = mdatoms->chargeA;
118 vftab = kernel_data->table_elec->data;
119 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
121 /* Setup water-specific parameters */
122 inr = nlist->iinr[0];
123 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
124 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
125 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
127 /* Avoid stupid compiler warnings */
128 jnrA = jnrB = jnrC = jnrD = 0;
137 for(iidx=0;iidx<4*DIM;iidx++)
142 /* Start outer loop over neighborlists */
143 for(iidx=0; iidx<nri; iidx++)
145 /* Load shift vector for this list */
146 i_shift_offset = DIM*shiftidx[iidx];
148 /* Load limits for loop over neighbors */
149 j_index_start = jindex[iidx];
150 j_index_end = jindex[iidx+1];
152 /* Get outer coordinate index */
154 i_coord_offset = DIM*inr;
156 /* Load i particle coords and add shift vector */
157 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
158 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
160 fix1 = _mm_setzero_ps();
161 fiy1 = _mm_setzero_ps();
162 fiz1 = _mm_setzero_ps();
163 fix2 = _mm_setzero_ps();
164 fiy2 = _mm_setzero_ps();
165 fiz2 = _mm_setzero_ps();
166 fix3 = _mm_setzero_ps();
167 fiy3 = _mm_setzero_ps();
168 fiz3 = _mm_setzero_ps();
170 /* Reset potential sums */
171 velecsum = _mm_setzero_ps();
173 /* Start inner kernel loop */
174 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
177 /* Get j neighbor index, and coordinate index */
182 j_coord_offsetA = DIM*jnrA;
183 j_coord_offsetB = DIM*jnrB;
184 j_coord_offsetC = DIM*jnrC;
185 j_coord_offsetD = DIM*jnrD;
187 /* load j atom coordinates */
188 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
189 x+j_coord_offsetC,x+j_coord_offsetD,
192 /* Calculate displacement vector */
193 dx10 = _mm_sub_ps(ix1,jx0);
194 dy10 = _mm_sub_ps(iy1,jy0);
195 dz10 = _mm_sub_ps(iz1,jz0);
196 dx20 = _mm_sub_ps(ix2,jx0);
197 dy20 = _mm_sub_ps(iy2,jy0);
198 dz20 = _mm_sub_ps(iz2,jz0);
199 dx30 = _mm_sub_ps(ix3,jx0);
200 dy30 = _mm_sub_ps(iy3,jy0);
201 dz30 = _mm_sub_ps(iz3,jz0);
203 /* Calculate squared distance and things based on it */
204 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
205 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
206 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
208 rinv10 = gmx_mm_invsqrt_ps(rsq10);
209 rinv20 = gmx_mm_invsqrt_ps(rsq20);
210 rinv30 = gmx_mm_invsqrt_ps(rsq30);
212 /* Load parameters for j particles */
213 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
214 charge+jnrC+0,charge+jnrD+0);
216 fjx0 = _mm_setzero_ps();
217 fjy0 = _mm_setzero_ps();
218 fjz0 = _mm_setzero_ps();
220 /**************************
221 * CALCULATE INTERACTIONS *
222 **************************/
224 r10 = _mm_mul_ps(rsq10,rinv10);
226 /* Compute parameters for interactions between i and j atoms */
227 qq10 = _mm_mul_ps(iq1,jq0);
229 /* Calculate table index by multiplying r with table scale and truncate to integer */
230 rt = _mm_mul_ps(r10,vftabscale);
231 vfitab = _mm_cvttps_epi32(rt);
233 vfeps = _mm_frcz_ps(rt);
235 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
237 twovfeps = _mm_add_ps(vfeps,vfeps);
238 vfitab = _mm_slli_epi32(vfitab,2);
240 /* CUBIC SPLINE TABLE ELECTROSTATICS */
241 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
242 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
243 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
244 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
245 _MM_TRANSPOSE4_PS(Y,F,G,H);
246 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
247 VV = _mm_macc_ps(vfeps,Fp,Y);
248 velec = _mm_mul_ps(qq10,VV);
249 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
250 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
252 /* Update potential sum for this i atom from the interaction with this j atom. */
253 velecsum = _mm_add_ps(velecsum,velec);
257 /* Update vectorial force */
258 fix1 = _mm_macc_ps(dx10,fscal,fix1);
259 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
260 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
262 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
263 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
264 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
266 /**************************
267 * CALCULATE INTERACTIONS *
268 **************************/
270 r20 = _mm_mul_ps(rsq20,rinv20);
272 /* Compute parameters for interactions between i and j atoms */
273 qq20 = _mm_mul_ps(iq2,jq0);
275 /* Calculate table index by multiplying r with table scale and truncate to integer */
276 rt = _mm_mul_ps(r20,vftabscale);
277 vfitab = _mm_cvttps_epi32(rt);
279 vfeps = _mm_frcz_ps(rt);
281 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
283 twovfeps = _mm_add_ps(vfeps,vfeps);
284 vfitab = _mm_slli_epi32(vfitab,2);
286 /* CUBIC SPLINE TABLE ELECTROSTATICS */
287 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
288 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
289 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
290 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
291 _MM_TRANSPOSE4_PS(Y,F,G,H);
292 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
293 VV = _mm_macc_ps(vfeps,Fp,Y);
294 velec = _mm_mul_ps(qq20,VV);
295 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
296 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
298 /* Update potential sum for this i atom from the interaction with this j atom. */
299 velecsum = _mm_add_ps(velecsum,velec);
303 /* Update vectorial force */
304 fix2 = _mm_macc_ps(dx20,fscal,fix2);
305 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
306 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
308 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
309 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
310 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
312 /**************************
313 * CALCULATE INTERACTIONS *
314 **************************/
316 r30 = _mm_mul_ps(rsq30,rinv30);
318 /* Compute parameters for interactions between i and j atoms */
319 qq30 = _mm_mul_ps(iq3,jq0);
321 /* Calculate table index by multiplying r with table scale and truncate to integer */
322 rt = _mm_mul_ps(r30,vftabscale);
323 vfitab = _mm_cvttps_epi32(rt);
325 vfeps = _mm_frcz_ps(rt);
327 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
329 twovfeps = _mm_add_ps(vfeps,vfeps);
330 vfitab = _mm_slli_epi32(vfitab,2);
332 /* CUBIC SPLINE TABLE ELECTROSTATICS */
333 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
334 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
335 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
336 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
337 _MM_TRANSPOSE4_PS(Y,F,G,H);
338 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
339 VV = _mm_macc_ps(vfeps,Fp,Y);
340 velec = _mm_mul_ps(qq30,VV);
341 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
342 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
344 /* Update potential sum for this i atom from the interaction with this j atom. */
345 velecsum = _mm_add_ps(velecsum,velec);
349 /* Update vectorial force */
350 fix3 = _mm_macc_ps(dx30,fscal,fix3);
351 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
352 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
354 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
355 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
356 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
358 fjptrA = f+j_coord_offsetA;
359 fjptrB = f+j_coord_offsetB;
360 fjptrC = f+j_coord_offsetC;
361 fjptrD = f+j_coord_offsetD;
363 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
365 /* Inner loop uses 138 flops */
371 /* Get j neighbor index, and coordinate index */
372 jnrlistA = jjnr[jidx];
373 jnrlistB = jjnr[jidx+1];
374 jnrlistC = jjnr[jidx+2];
375 jnrlistD = jjnr[jidx+3];
376 /* Sign of each element will be negative for non-real atoms.
377 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
378 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
380 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
381 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
382 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
383 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
384 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
385 j_coord_offsetA = DIM*jnrA;
386 j_coord_offsetB = DIM*jnrB;
387 j_coord_offsetC = DIM*jnrC;
388 j_coord_offsetD = DIM*jnrD;
390 /* load j atom coordinates */
391 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
392 x+j_coord_offsetC,x+j_coord_offsetD,
395 /* Calculate displacement vector */
396 dx10 = _mm_sub_ps(ix1,jx0);
397 dy10 = _mm_sub_ps(iy1,jy0);
398 dz10 = _mm_sub_ps(iz1,jz0);
399 dx20 = _mm_sub_ps(ix2,jx0);
400 dy20 = _mm_sub_ps(iy2,jy0);
401 dz20 = _mm_sub_ps(iz2,jz0);
402 dx30 = _mm_sub_ps(ix3,jx0);
403 dy30 = _mm_sub_ps(iy3,jy0);
404 dz30 = _mm_sub_ps(iz3,jz0);
406 /* Calculate squared distance and things based on it */
407 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
408 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
409 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
411 rinv10 = gmx_mm_invsqrt_ps(rsq10);
412 rinv20 = gmx_mm_invsqrt_ps(rsq20);
413 rinv30 = gmx_mm_invsqrt_ps(rsq30);
415 /* Load parameters for j particles */
416 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
417 charge+jnrC+0,charge+jnrD+0);
419 fjx0 = _mm_setzero_ps();
420 fjy0 = _mm_setzero_ps();
421 fjz0 = _mm_setzero_ps();
423 /**************************
424 * CALCULATE INTERACTIONS *
425 **************************/
427 r10 = _mm_mul_ps(rsq10,rinv10);
428 r10 = _mm_andnot_ps(dummy_mask,r10);
430 /* Compute parameters for interactions between i and j atoms */
431 qq10 = _mm_mul_ps(iq1,jq0);
433 /* Calculate table index by multiplying r with table scale and truncate to integer */
434 rt = _mm_mul_ps(r10,vftabscale);
435 vfitab = _mm_cvttps_epi32(rt);
437 vfeps = _mm_frcz_ps(rt);
439 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
441 twovfeps = _mm_add_ps(vfeps,vfeps);
442 vfitab = _mm_slli_epi32(vfitab,2);
444 /* CUBIC SPLINE TABLE ELECTROSTATICS */
445 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
446 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
447 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
448 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
449 _MM_TRANSPOSE4_PS(Y,F,G,H);
450 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
451 VV = _mm_macc_ps(vfeps,Fp,Y);
452 velec = _mm_mul_ps(qq10,VV);
453 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
454 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
456 /* Update potential sum for this i atom from the interaction with this j atom. */
457 velec = _mm_andnot_ps(dummy_mask,velec);
458 velecsum = _mm_add_ps(velecsum,velec);
462 fscal = _mm_andnot_ps(dummy_mask,fscal);
464 /* Update vectorial force */
465 fix1 = _mm_macc_ps(dx10,fscal,fix1);
466 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
467 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
469 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
470 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
471 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
473 /**************************
474 * CALCULATE INTERACTIONS *
475 **************************/
477 r20 = _mm_mul_ps(rsq20,rinv20);
478 r20 = _mm_andnot_ps(dummy_mask,r20);
480 /* Compute parameters for interactions between i and j atoms */
481 qq20 = _mm_mul_ps(iq2,jq0);
483 /* Calculate table index by multiplying r with table scale and truncate to integer */
484 rt = _mm_mul_ps(r20,vftabscale);
485 vfitab = _mm_cvttps_epi32(rt);
487 vfeps = _mm_frcz_ps(rt);
489 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
491 twovfeps = _mm_add_ps(vfeps,vfeps);
492 vfitab = _mm_slli_epi32(vfitab,2);
494 /* CUBIC SPLINE TABLE ELECTROSTATICS */
495 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
496 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
497 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
498 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
499 _MM_TRANSPOSE4_PS(Y,F,G,H);
500 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
501 VV = _mm_macc_ps(vfeps,Fp,Y);
502 velec = _mm_mul_ps(qq20,VV);
503 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
504 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
506 /* Update potential sum for this i atom from the interaction with this j atom. */
507 velec = _mm_andnot_ps(dummy_mask,velec);
508 velecsum = _mm_add_ps(velecsum,velec);
512 fscal = _mm_andnot_ps(dummy_mask,fscal);
514 /* Update vectorial force */
515 fix2 = _mm_macc_ps(dx20,fscal,fix2);
516 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
517 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
519 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
520 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
521 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
523 /**************************
524 * CALCULATE INTERACTIONS *
525 **************************/
527 r30 = _mm_mul_ps(rsq30,rinv30);
528 r30 = _mm_andnot_ps(dummy_mask,r30);
530 /* Compute parameters for interactions between i and j atoms */
531 qq30 = _mm_mul_ps(iq3,jq0);
533 /* Calculate table index by multiplying r with table scale and truncate to integer */
534 rt = _mm_mul_ps(r30,vftabscale);
535 vfitab = _mm_cvttps_epi32(rt);
537 vfeps = _mm_frcz_ps(rt);
539 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
541 twovfeps = _mm_add_ps(vfeps,vfeps);
542 vfitab = _mm_slli_epi32(vfitab,2);
544 /* CUBIC SPLINE TABLE ELECTROSTATICS */
545 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
546 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
547 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
548 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
549 _MM_TRANSPOSE4_PS(Y,F,G,H);
550 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
551 VV = _mm_macc_ps(vfeps,Fp,Y);
552 velec = _mm_mul_ps(qq30,VV);
553 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
554 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
556 /* Update potential sum for this i atom from the interaction with this j atom. */
557 velec = _mm_andnot_ps(dummy_mask,velec);
558 velecsum = _mm_add_ps(velecsum,velec);
562 fscal = _mm_andnot_ps(dummy_mask,fscal);
564 /* Update vectorial force */
565 fix3 = _mm_macc_ps(dx30,fscal,fix3);
566 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
567 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
569 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
570 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
571 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
573 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
574 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
575 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
576 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
578 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
580 /* Inner loop uses 141 flops */
583 /* End of innermost loop */
585 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
586 f+i_coord_offset+DIM,fshift+i_shift_offset);
589 /* Update potential energies */
590 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
592 /* Increment number of inner iterations */
593 inneriter += j_index_end - j_index_start;
595 /* Outer loop uses 19 flops */
598 /* Increment number of outer iterations */
601 /* Update outer/inner flops */
603 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*141);
606 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_128_fma_single
607 * Electrostatics interaction: CubicSplineTable
608 * VdW interaction: None
609 * Geometry: Water4-Particle
610 * Calculate force/pot: Force
613 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_128_fma_single
614 (t_nblist * gmx_restrict nlist,
615 rvec * gmx_restrict xx,
616 rvec * gmx_restrict ff,
617 t_forcerec * gmx_restrict fr,
618 t_mdatoms * gmx_restrict mdatoms,
619 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
620 t_nrnb * gmx_restrict nrnb)
622 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
623 * just 0 for non-waters.
624 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
625 * jnr indices corresponding to data put in the four positions in the SIMD register.
627 int i_shift_offset,i_coord_offset,outeriter,inneriter;
628 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
629 int jnrA,jnrB,jnrC,jnrD;
630 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
631 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
632 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
634 real *shiftvec,*fshift,*x,*f;
635 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
637 __m128 fscal,rcutoff,rcutoff2,jidxall;
639 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
641 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
643 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
644 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
645 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
646 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
647 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
648 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
649 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
652 __m128i ifour = _mm_set1_epi32(4);
653 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
655 __m128 dummy_mask,cutoff_mask;
656 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
657 __m128 one = _mm_set1_ps(1.0);
658 __m128 two = _mm_set1_ps(2.0);
664 jindex = nlist->jindex;
666 shiftidx = nlist->shift;
668 shiftvec = fr->shift_vec[0];
669 fshift = fr->fshift[0];
670 facel = _mm_set1_ps(fr->epsfac);
671 charge = mdatoms->chargeA;
673 vftab = kernel_data->table_elec->data;
674 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
676 /* Setup water-specific parameters */
677 inr = nlist->iinr[0];
678 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
679 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
680 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
682 /* Avoid stupid compiler warnings */
683 jnrA = jnrB = jnrC = jnrD = 0;
692 for(iidx=0;iidx<4*DIM;iidx++)
697 /* Start outer loop over neighborlists */
698 for(iidx=0; iidx<nri; iidx++)
700 /* Load shift vector for this list */
701 i_shift_offset = DIM*shiftidx[iidx];
703 /* Load limits for loop over neighbors */
704 j_index_start = jindex[iidx];
705 j_index_end = jindex[iidx+1];
707 /* Get outer coordinate index */
709 i_coord_offset = DIM*inr;
711 /* Load i particle coords and add shift vector */
712 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
713 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
715 fix1 = _mm_setzero_ps();
716 fiy1 = _mm_setzero_ps();
717 fiz1 = _mm_setzero_ps();
718 fix2 = _mm_setzero_ps();
719 fiy2 = _mm_setzero_ps();
720 fiz2 = _mm_setzero_ps();
721 fix3 = _mm_setzero_ps();
722 fiy3 = _mm_setzero_ps();
723 fiz3 = _mm_setzero_ps();
725 /* Start inner kernel loop */
726 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
729 /* Get j neighbor index, and coordinate index */
734 j_coord_offsetA = DIM*jnrA;
735 j_coord_offsetB = DIM*jnrB;
736 j_coord_offsetC = DIM*jnrC;
737 j_coord_offsetD = DIM*jnrD;
739 /* load j atom coordinates */
740 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
741 x+j_coord_offsetC,x+j_coord_offsetD,
744 /* Calculate displacement vector */
745 dx10 = _mm_sub_ps(ix1,jx0);
746 dy10 = _mm_sub_ps(iy1,jy0);
747 dz10 = _mm_sub_ps(iz1,jz0);
748 dx20 = _mm_sub_ps(ix2,jx0);
749 dy20 = _mm_sub_ps(iy2,jy0);
750 dz20 = _mm_sub_ps(iz2,jz0);
751 dx30 = _mm_sub_ps(ix3,jx0);
752 dy30 = _mm_sub_ps(iy3,jy0);
753 dz30 = _mm_sub_ps(iz3,jz0);
755 /* Calculate squared distance and things based on it */
756 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
757 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
758 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
760 rinv10 = gmx_mm_invsqrt_ps(rsq10);
761 rinv20 = gmx_mm_invsqrt_ps(rsq20);
762 rinv30 = gmx_mm_invsqrt_ps(rsq30);
764 /* Load parameters for j particles */
765 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
766 charge+jnrC+0,charge+jnrD+0);
768 fjx0 = _mm_setzero_ps();
769 fjy0 = _mm_setzero_ps();
770 fjz0 = _mm_setzero_ps();
772 /**************************
773 * CALCULATE INTERACTIONS *
774 **************************/
776 r10 = _mm_mul_ps(rsq10,rinv10);
778 /* Compute parameters for interactions between i and j atoms */
779 qq10 = _mm_mul_ps(iq1,jq0);
781 /* Calculate table index by multiplying r with table scale and truncate to integer */
782 rt = _mm_mul_ps(r10,vftabscale);
783 vfitab = _mm_cvttps_epi32(rt);
785 vfeps = _mm_frcz_ps(rt);
787 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
789 twovfeps = _mm_add_ps(vfeps,vfeps);
790 vfitab = _mm_slli_epi32(vfitab,2);
792 /* CUBIC SPLINE TABLE ELECTROSTATICS */
793 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
794 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
795 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
796 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
797 _MM_TRANSPOSE4_PS(Y,F,G,H);
798 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
799 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
800 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
804 /* Update vectorial force */
805 fix1 = _mm_macc_ps(dx10,fscal,fix1);
806 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
807 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
809 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
810 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
811 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
813 /**************************
814 * CALCULATE INTERACTIONS *
815 **************************/
817 r20 = _mm_mul_ps(rsq20,rinv20);
819 /* Compute parameters for interactions between i and j atoms */
820 qq20 = _mm_mul_ps(iq2,jq0);
822 /* Calculate table index by multiplying r with table scale and truncate to integer */
823 rt = _mm_mul_ps(r20,vftabscale);
824 vfitab = _mm_cvttps_epi32(rt);
826 vfeps = _mm_frcz_ps(rt);
828 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
830 twovfeps = _mm_add_ps(vfeps,vfeps);
831 vfitab = _mm_slli_epi32(vfitab,2);
833 /* CUBIC SPLINE TABLE ELECTROSTATICS */
834 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
835 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
836 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
837 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
838 _MM_TRANSPOSE4_PS(Y,F,G,H);
839 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
840 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
841 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
845 /* Update vectorial force */
846 fix2 = _mm_macc_ps(dx20,fscal,fix2);
847 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
848 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
850 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
851 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
852 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
854 /**************************
855 * CALCULATE INTERACTIONS *
856 **************************/
858 r30 = _mm_mul_ps(rsq30,rinv30);
860 /* Compute parameters for interactions between i and j atoms */
861 qq30 = _mm_mul_ps(iq3,jq0);
863 /* Calculate table index by multiplying r with table scale and truncate to integer */
864 rt = _mm_mul_ps(r30,vftabscale);
865 vfitab = _mm_cvttps_epi32(rt);
867 vfeps = _mm_frcz_ps(rt);
869 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
871 twovfeps = _mm_add_ps(vfeps,vfeps);
872 vfitab = _mm_slli_epi32(vfitab,2);
874 /* CUBIC SPLINE TABLE ELECTROSTATICS */
875 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
876 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
877 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
878 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
879 _MM_TRANSPOSE4_PS(Y,F,G,H);
880 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
881 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
882 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
886 /* Update vectorial force */
887 fix3 = _mm_macc_ps(dx30,fscal,fix3);
888 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
889 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
891 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
892 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
893 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
895 fjptrA = f+j_coord_offsetA;
896 fjptrB = f+j_coord_offsetB;
897 fjptrC = f+j_coord_offsetC;
898 fjptrD = f+j_coord_offsetD;
900 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
902 /* Inner loop uses 126 flops */
908 /* Get j neighbor index, and coordinate index */
909 jnrlistA = jjnr[jidx];
910 jnrlistB = jjnr[jidx+1];
911 jnrlistC = jjnr[jidx+2];
912 jnrlistD = jjnr[jidx+3];
913 /* Sign of each element will be negative for non-real atoms.
914 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
915 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
917 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
918 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
919 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
920 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
921 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
922 j_coord_offsetA = DIM*jnrA;
923 j_coord_offsetB = DIM*jnrB;
924 j_coord_offsetC = DIM*jnrC;
925 j_coord_offsetD = DIM*jnrD;
927 /* load j atom coordinates */
928 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
929 x+j_coord_offsetC,x+j_coord_offsetD,
932 /* Calculate displacement vector */
933 dx10 = _mm_sub_ps(ix1,jx0);
934 dy10 = _mm_sub_ps(iy1,jy0);
935 dz10 = _mm_sub_ps(iz1,jz0);
936 dx20 = _mm_sub_ps(ix2,jx0);
937 dy20 = _mm_sub_ps(iy2,jy0);
938 dz20 = _mm_sub_ps(iz2,jz0);
939 dx30 = _mm_sub_ps(ix3,jx0);
940 dy30 = _mm_sub_ps(iy3,jy0);
941 dz30 = _mm_sub_ps(iz3,jz0);
943 /* Calculate squared distance and things based on it */
944 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
945 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
946 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
948 rinv10 = gmx_mm_invsqrt_ps(rsq10);
949 rinv20 = gmx_mm_invsqrt_ps(rsq20);
950 rinv30 = gmx_mm_invsqrt_ps(rsq30);
952 /* Load parameters for j particles */
953 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
954 charge+jnrC+0,charge+jnrD+0);
956 fjx0 = _mm_setzero_ps();
957 fjy0 = _mm_setzero_ps();
958 fjz0 = _mm_setzero_ps();
960 /**************************
961 * CALCULATE INTERACTIONS *
962 **************************/
964 r10 = _mm_mul_ps(rsq10,rinv10);
965 r10 = _mm_andnot_ps(dummy_mask,r10);
967 /* Compute parameters for interactions between i and j atoms */
968 qq10 = _mm_mul_ps(iq1,jq0);
970 /* Calculate table index by multiplying r with table scale and truncate to integer */
971 rt = _mm_mul_ps(r10,vftabscale);
972 vfitab = _mm_cvttps_epi32(rt);
974 vfeps = _mm_frcz_ps(rt);
976 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
978 twovfeps = _mm_add_ps(vfeps,vfeps);
979 vfitab = _mm_slli_epi32(vfitab,2);
981 /* CUBIC SPLINE TABLE ELECTROSTATICS */
982 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
983 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
984 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
985 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
986 _MM_TRANSPOSE4_PS(Y,F,G,H);
987 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
988 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
989 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
993 fscal = _mm_andnot_ps(dummy_mask,fscal);
995 /* Update vectorial force */
996 fix1 = _mm_macc_ps(dx10,fscal,fix1);
997 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
998 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1000 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1001 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1002 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1004 /**************************
1005 * CALCULATE INTERACTIONS *
1006 **************************/
1008 r20 = _mm_mul_ps(rsq20,rinv20);
1009 r20 = _mm_andnot_ps(dummy_mask,r20);
1011 /* Compute parameters for interactions between i and j atoms */
1012 qq20 = _mm_mul_ps(iq2,jq0);
1014 /* Calculate table index by multiplying r with table scale and truncate to integer */
1015 rt = _mm_mul_ps(r20,vftabscale);
1016 vfitab = _mm_cvttps_epi32(rt);
1018 vfeps = _mm_frcz_ps(rt);
1020 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1022 twovfeps = _mm_add_ps(vfeps,vfeps);
1023 vfitab = _mm_slli_epi32(vfitab,2);
1025 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1026 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1027 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1028 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1029 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1030 _MM_TRANSPOSE4_PS(Y,F,G,H);
1031 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1032 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1033 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1037 fscal = _mm_andnot_ps(dummy_mask,fscal);
1039 /* Update vectorial force */
1040 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1041 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1042 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1044 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1045 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1046 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1048 /**************************
1049 * CALCULATE INTERACTIONS *
1050 **************************/
1052 r30 = _mm_mul_ps(rsq30,rinv30);
1053 r30 = _mm_andnot_ps(dummy_mask,r30);
1055 /* Compute parameters for interactions between i and j atoms */
1056 qq30 = _mm_mul_ps(iq3,jq0);
1058 /* Calculate table index by multiplying r with table scale and truncate to integer */
1059 rt = _mm_mul_ps(r30,vftabscale);
1060 vfitab = _mm_cvttps_epi32(rt);
1062 vfeps = _mm_frcz_ps(rt);
1064 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1066 twovfeps = _mm_add_ps(vfeps,vfeps);
1067 vfitab = _mm_slli_epi32(vfitab,2);
1069 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1070 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1071 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1072 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1073 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1074 _MM_TRANSPOSE4_PS(Y,F,G,H);
1075 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1076 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1077 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1081 fscal = _mm_andnot_ps(dummy_mask,fscal);
1083 /* Update vectorial force */
1084 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1085 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1086 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1088 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1089 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1090 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1092 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1093 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1094 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1095 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1097 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1099 /* Inner loop uses 129 flops */
1102 /* End of innermost loop */
1104 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1105 f+i_coord_offset+DIM,fshift+i_shift_offset);
1107 /* Increment number of inner iterations */
1108 inneriter += j_index_end - j_index_start;
1110 /* Outer loop uses 18 flops */
1113 /* Increment number of outer iterations */
1116 /* Update outer/inner flops */
1118 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*129);