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.
44 #include "../nb_kernel.h"
45 #include "types/simple.h"
46 #include "gromacs/math/vec.h"
49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
50 #include "kernelutil_x86_avx_128_fma_single.h"
53 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_128_fma_single
54 * Electrostatics interaction: CubicSplineTable
55 * VdW interaction: None
56 * Geometry: Water4-Particle
57 * Calculate force/pot: PotentialAndForce
60 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_VF_avx_128_fma_single
61 (t_nblist * gmx_restrict nlist,
62 rvec * gmx_restrict xx,
63 rvec * gmx_restrict ff,
64 t_forcerec * gmx_restrict fr,
65 t_mdatoms * gmx_restrict mdatoms,
66 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
67 t_nrnb * gmx_restrict nrnb)
69 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
70 * just 0 for non-waters.
71 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
72 * jnr indices corresponding to data put in the four positions in the SIMD register.
74 int i_shift_offset,i_coord_offset,outeriter,inneriter;
75 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
76 int jnrA,jnrB,jnrC,jnrD;
77 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
78 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
79 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
81 real *shiftvec,*fshift,*x,*f;
82 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
84 __m128 fscal,rcutoff,rcutoff2,jidxall;
86 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
88 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
90 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
91 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
92 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
93 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
94 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
95 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
96 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
99 __m128i ifour = _mm_set1_epi32(4);
100 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
102 __m128 dummy_mask,cutoff_mask;
103 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
104 __m128 one = _mm_set1_ps(1.0);
105 __m128 two = _mm_set1_ps(2.0);
111 jindex = nlist->jindex;
113 shiftidx = nlist->shift;
115 shiftvec = fr->shift_vec[0];
116 fshift = fr->fshift[0];
117 facel = _mm_set1_ps(fr->epsfac);
118 charge = mdatoms->chargeA;
120 vftab = kernel_data->table_elec->data;
121 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
123 /* Setup water-specific parameters */
124 inr = nlist->iinr[0];
125 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
126 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
127 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
129 /* Avoid stupid compiler warnings */
130 jnrA = jnrB = jnrC = jnrD = 0;
139 for(iidx=0;iidx<4*DIM;iidx++)
144 /* Start outer loop over neighborlists */
145 for(iidx=0; iidx<nri; iidx++)
147 /* Load shift vector for this list */
148 i_shift_offset = DIM*shiftidx[iidx];
150 /* Load limits for loop over neighbors */
151 j_index_start = jindex[iidx];
152 j_index_end = jindex[iidx+1];
154 /* Get outer coordinate index */
156 i_coord_offset = DIM*inr;
158 /* Load i particle coords and add shift vector */
159 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
160 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
162 fix1 = _mm_setzero_ps();
163 fiy1 = _mm_setzero_ps();
164 fiz1 = _mm_setzero_ps();
165 fix2 = _mm_setzero_ps();
166 fiy2 = _mm_setzero_ps();
167 fiz2 = _mm_setzero_ps();
168 fix3 = _mm_setzero_ps();
169 fiy3 = _mm_setzero_ps();
170 fiz3 = _mm_setzero_ps();
172 /* Reset potential sums */
173 velecsum = _mm_setzero_ps();
175 /* Start inner kernel loop */
176 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
179 /* Get j neighbor index, and coordinate index */
184 j_coord_offsetA = DIM*jnrA;
185 j_coord_offsetB = DIM*jnrB;
186 j_coord_offsetC = DIM*jnrC;
187 j_coord_offsetD = DIM*jnrD;
189 /* load j atom coordinates */
190 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
191 x+j_coord_offsetC,x+j_coord_offsetD,
194 /* Calculate displacement vector */
195 dx10 = _mm_sub_ps(ix1,jx0);
196 dy10 = _mm_sub_ps(iy1,jy0);
197 dz10 = _mm_sub_ps(iz1,jz0);
198 dx20 = _mm_sub_ps(ix2,jx0);
199 dy20 = _mm_sub_ps(iy2,jy0);
200 dz20 = _mm_sub_ps(iz2,jz0);
201 dx30 = _mm_sub_ps(ix3,jx0);
202 dy30 = _mm_sub_ps(iy3,jy0);
203 dz30 = _mm_sub_ps(iz3,jz0);
205 /* Calculate squared distance and things based on it */
206 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
207 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
208 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
210 rinv10 = gmx_mm_invsqrt_ps(rsq10);
211 rinv20 = gmx_mm_invsqrt_ps(rsq20);
212 rinv30 = gmx_mm_invsqrt_ps(rsq30);
214 /* Load parameters for j particles */
215 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
216 charge+jnrC+0,charge+jnrD+0);
218 fjx0 = _mm_setzero_ps();
219 fjy0 = _mm_setzero_ps();
220 fjz0 = _mm_setzero_ps();
222 /**************************
223 * CALCULATE INTERACTIONS *
224 **************************/
226 r10 = _mm_mul_ps(rsq10,rinv10);
228 /* Compute parameters for interactions between i and j atoms */
229 qq10 = _mm_mul_ps(iq1,jq0);
231 /* Calculate table index by multiplying r with table scale and truncate to integer */
232 rt = _mm_mul_ps(r10,vftabscale);
233 vfitab = _mm_cvttps_epi32(rt);
235 vfeps = _mm_frcz_ps(rt);
237 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
239 twovfeps = _mm_add_ps(vfeps,vfeps);
240 vfitab = _mm_slli_epi32(vfitab,2);
242 /* CUBIC SPLINE TABLE ELECTROSTATICS */
243 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
244 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
245 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
246 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
247 _MM_TRANSPOSE4_PS(Y,F,G,H);
248 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
249 VV = _mm_macc_ps(vfeps,Fp,Y);
250 velec = _mm_mul_ps(qq10,VV);
251 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
252 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
254 /* Update potential sum for this i atom from the interaction with this j atom. */
255 velecsum = _mm_add_ps(velecsum,velec);
259 /* Update vectorial force */
260 fix1 = _mm_macc_ps(dx10,fscal,fix1);
261 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
262 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
264 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
265 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
266 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
268 /**************************
269 * CALCULATE INTERACTIONS *
270 **************************/
272 r20 = _mm_mul_ps(rsq20,rinv20);
274 /* Compute parameters for interactions between i and j atoms */
275 qq20 = _mm_mul_ps(iq2,jq0);
277 /* Calculate table index by multiplying r with table scale and truncate to integer */
278 rt = _mm_mul_ps(r20,vftabscale);
279 vfitab = _mm_cvttps_epi32(rt);
281 vfeps = _mm_frcz_ps(rt);
283 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
285 twovfeps = _mm_add_ps(vfeps,vfeps);
286 vfitab = _mm_slli_epi32(vfitab,2);
288 /* CUBIC SPLINE TABLE ELECTROSTATICS */
289 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
290 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
291 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
292 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
293 _MM_TRANSPOSE4_PS(Y,F,G,H);
294 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
295 VV = _mm_macc_ps(vfeps,Fp,Y);
296 velec = _mm_mul_ps(qq20,VV);
297 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
298 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
300 /* Update potential sum for this i atom from the interaction with this j atom. */
301 velecsum = _mm_add_ps(velecsum,velec);
305 /* Update vectorial force */
306 fix2 = _mm_macc_ps(dx20,fscal,fix2);
307 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
308 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
310 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
311 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
312 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
314 /**************************
315 * CALCULATE INTERACTIONS *
316 **************************/
318 r30 = _mm_mul_ps(rsq30,rinv30);
320 /* Compute parameters for interactions between i and j atoms */
321 qq30 = _mm_mul_ps(iq3,jq0);
323 /* Calculate table index by multiplying r with table scale and truncate to integer */
324 rt = _mm_mul_ps(r30,vftabscale);
325 vfitab = _mm_cvttps_epi32(rt);
327 vfeps = _mm_frcz_ps(rt);
329 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
331 twovfeps = _mm_add_ps(vfeps,vfeps);
332 vfitab = _mm_slli_epi32(vfitab,2);
334 /* CUBIC SPLINE TABLE ELECTROSTATICS */
335 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
336 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
337 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
338 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
339 _MM_TRANSPOSE4_PS(Y,F,G,H);
340 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
341 VV = _mm_macc_ps(vfeps,Fp,Y);
342 velec = _mm_mul_ps(qq30,VV);
343 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
344 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
346 /* Update potential sum for this i atom from the interaction with this j atom. */
347 velecsum = _mm_add_ps(velecsum,velec);
351 /* Update vectorial force */
352 fix3 = _mm_macc_ps(dx30,fscal,fix3);
353 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
354 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
356 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
357 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
358 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
360 fjptrA = f+j_coord_offsetA;
361 fjptrB = f+j_coord_offsetB;
362 fjptrC = f+j_coord_offsetC;
363 fjptrD = f+j_coord_offsetD;
365 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
367 /* Inner loop uses 138 flops */
373 /* Get j neighbor index, and coordinate index */
374 jnrlistA = jjnr[jidx];
375 jnrlistB = jjnr[jidx+1];
376 jnrlistC = jjnr[jidx+2];
377 jnrlistD = jjnr[jidx+3];
378 /* Sign of each element will be negative for non-real atoms.
379 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
380 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
382 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
383 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
384 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
385 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
386 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
387 j_coord_offsetA = DIM*jnrA;
388 j_coord_offsetB = DIM*jnrB;
389 j_coord_offsetC = DIM*jnrC;
390 j_coord_offsetD = DIM*jnrD;
392 /* load j atom coordinates */
393 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
394 x+j_coord_offsetC,x+j_coord_offsetD,
397 /* Calculate displacement vector */
398 dx10 = _mm_sub_ps(ix1,jx0);
399 dy10 = _mm_sub_ps(iy1,jy0);
400 dz10 = _mm_sub_ps(iz1,jz0);
401 dx20 = _mm_sub_ps(ix2,jx0);
402 dy20 = _mm_sub_ps(iy2,jy0);
403 dz20 = _mm_sub_ps(iz2,jz0);
404 dx30 = _mm_sub_ps(ix3,jx0);
405 dy30 = _mm_sub_ps(iy3,jy0);
406 dz30 = _mm_sub_ps(iz3,jz0);
408 /* Calculate squared distance and things based on it */
409 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
410 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
411 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
413 rinv10 = gmx_mm_invsqrt_ps(rsq10);
414 rinv20 = gmx_mm_invsqrt_ps(rsq20);
415 rinv30 = gmx_mm_invsqrt_ps(rsq30);
417 /* Load parameters for j particles */
418 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
419 charge+jnrC+0,charge+jnrD+0);
421 fjx0 = _mm_setzero_ps();
422 fjy0 = _mm_setzero_ps();
423 fjz0 = _mm_setzero_ps();
425 /**************************
426 * CALCULATE INTERACTIONS *
427 **************************/
429 r10 = _mm_mul_ps(rsq10,rinv10);
430 r10 = _mm_andnot_ps(dummy_mask,r10);
432 /* Compute parameters for interactions between i and j atoms */
433 qq10 = _mm_mul_ps(iq1,jq0);
435 /* Calculate table index by multiplying r with table scale and truncate to integer */
436 rt = _mm_mul_ps(r10,vftabscale);
437 vfitab = _mm_cvttps_epi32(rt);
439 vfeps = _mm_frcz_ps(rt);
441 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
443 twovfeps = _mm_add_ps(vfeps,vfeps);
444 vfitab = _mm_slli_epi32(vfitab,2);
446 /* CUBIC SPLINE TABLE ELECTROSTATICS */
447 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
448 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
449 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
450 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
451 _MM_TRANSPOSE4_PS(Y,F,G,H);
452 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
453 VV = _mm_macc_ps(vfeps,Fp,Y);
454 velec = _mm_mul_ps(qq10,VV);
455 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
456 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
458 /* Update potential sum for this i atom from the interaction with this j atom. */
459 velec = _mm_andnot_ps(dummy_mask,velec);
460 velecsum = _mm_add_ps(velecsum,velec);
464 fscal = _mm_andnot_ps(dummy_mask,fscal);
466 /* Update vectorial force */
467 fix1 = _mm_macc_ps(dx10,fscal,fix1);
468 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
469 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
471 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
472 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
473 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
475 /**************************
476 * CALCULATE INTERACTIONS *
477 **************************/
479 r20 = _mm_mul_ps(rsq20,rinv20);
480 r20 = _mm_andnot_ps(dummy_mask,r20);
482 /* Compute parameters for interactions between i and j atoms */
483 qq20 = _mm_mul_ps(iq2,jq0);
485 /* Calculate table index by multiplying r with table scale and truncate to integer */
486 rt = _mm_mul_ps(r20,vftabscale);
487 vfitab = _mm_cvttps_epi32(rt);
489 vfeps = _mm_frcz_ps(rt);
491 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
493 twovfeps = _mm_add_ps(vfeps,vfeps);
494 vfitab = _mm_slli_epi32(vfitab,2);
496 /* CUBIC SPLINE TABLE ELECTROSTATICS */
497 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
498 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
499 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
500 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
501 _MM_TRANSPOSE4_PS(Y,F,G,H);
502 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
503 VV = _mm_macc_ps(vfeps,Fp,Y);
504 velec = _mm_mul_ps(qq20,VV);
505 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
506 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
508 /* Update potential sum for this i atom from the interaction with this j atom. */
509 velec = _mm_andnot_ps(dummy_mask,velec);
510 velecsum = _mm_add_ps(velecsum,velec);
514 fscal = _mm_andnot_ps(dummy_mask,fscal);
516 /* Update vectorial force */
517 fix2 = _mm_macc_ps(dx20,fscal,fix2);
518 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
519 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
521 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
522 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
523 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
525 /**************************
526 * CALCULATE INTERACTIONS *
527 **************************/
529 r30 = _mm_mul_ps(rsq30,rinv30);
530 r30 = _mm_andnot_ps(dummy_mask,r30);
532 /* Compute parameters for interactions between i and j atoms */
533 qq30 = _mm_mul_ps(iq3,jq0);
535 /* Calculate table index by multiplying r with table scale and truncate to integer */
536 rt = _mm_mul_ps(r30,vftabscale);
537 vfitab = _mm_cvttps_epi32(rt);
539 vfeps = _mm_frcz_ps(rt);
541 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
543 twovfeps = _mm_add_ps(vfeps,vfeps);
544 vfitab = _mm_slli_epi32(vfitab,2);
546 /* CUBIC SPLINE TABLE ELECTROSTATICS */
547 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
548 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
549 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
550 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
551 _MM_TRANSPOSE4_PS(Y,F,G,H);
552 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
553 VV = _mm_macc_ps(vfeps,Fp,Y);
554 velec = _mm_mul_ps(qq30,VV);
555 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
556 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
558 /* Update potential sum for this i atom from the interaction with this j atom. */
559 velec = _mm_andnot_ps(dummy_mask,velec);
560 velecsum = _mm_add_ps(velecsum,velec);
564 fscal = _mm_andnot_ps(dummy_mask,fscal);
566 /* Update vectorial force */
567 fix3 = _mm_macc_ps(dx30,fscal,fix3);
568 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
569 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
571 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
572 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
573 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
575 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
576 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
577 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
578 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
580 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
582 /* Inner loop uses 141 flops */
585 /* End of innermost loop */
587 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
588 f+i_coord_offset+DIM,fshift+i_shift_offset);
591 /* Update potential energies */
592 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
594 /* Increment number of inner iterations */
595 inneriter += j_index_end - j_index_start;
597 /* Outer loop uses 19 flops */
600 /* Increment number of outer iterations */
603 /* Update outer/inner flops */
605 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*141);
608 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_128_fma_single
609 * Electrostatics interaction: CubicSplineTable
610 * VdW interaction: None
611 * Geometry: Water4-Particle
612 * Calculate force/pot: Force
615 nb_kernel_ElecCSTab_VdwNone_GeomW4P1_F_avx_128_fma_single
616 (t_nblist * gmx_restrict nlist,
617 rvec * gmx_restrict xx,
618 rvec * gmx_restrict ff,
619 t_forcerec * gmx_restrict fr,
620 t_mdatoms * gmx_restrict mdatoms,
621 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
622 t_nrnb * gmx_restrict nrnb)
624 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
625 * just 0 for non-waters.
626 * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
627 * jnr indices corresponding to data put in the four positions in the SIMD register.
629 int i_shift_offset,i_coord_offset,outeriter,inneriter;
630 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
631 int jnrA,jnrB,jnrC,jnrD;
632 int jnrlistA,jnrlistB,jnrlistC,jnrlistD;
633 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
634 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
636 real *shiftvec,*fshift,*x,*f;
637 real *fjptrA,*fjptrB,*fjptrC,*fjptrD;
639 __m128 fscal,rcutoff,rcutoff2,jidxall;
641 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
643 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
645 __m128 ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
646 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
647 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
648 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
649 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
650 __m128 dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
651 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
654 __m128i ifour = _mm_set1_epi32(4);
655 __m128 rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
657 __m128 dummy_mask,cutoff_mask;
658 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
659 __m128 one = _mm_set1_ps(1.0);
660 __m128 two = _mm_set1_ps(2.0);
666 jindex = nlist->jindex;
668 shiftidx = nlist->shift;
670 shiftvec = fr->shift_vec[0];
671 fshift = fr->fshift[0];
672 facel = _mm_set1_ps(fr->epsfac);
673 charge = mdatoms->chargeA;
675 vftab = kernel_data->table_elec->data;
676 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
678 /* Setup water-specific parameters */
679 inr = nlist->iinr[0];
680 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
681 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
682 iq3 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
684 /* Avoid stupid compiler warnings */
685 jnrA = jnrB = jnrC = jnrD = 0;
694 for(iidx=0;iidx<4*DIM;iidx++)
699 /* Start outer loop over neighborlists */
700 for(iidx=0; iidx<nri; iidx++)
702 /* Load shift vector for this list */
703 i_shift_offset = DIM*shiftidx[iidx];
705 /* Load limits for loop over neighbors */
706 j_index_start = jindex[iidx];
707 j_index_end = jindex[iidx+1];
709 /* Get outer coordinate index */
711 i_coord_offset = DIM*inr;
713 /* Load i particle coords and add shift vector */
714 gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
715 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
717 fix1 = _mm_setzero_ps();
718 fiy1 = _mm_setzero_ps();
719 fiz1 = _mm_setzero_ps();
720 fix2 = _mm_setzero_ps();
721 fiy2 = _mm_setzero_ps();
722 fiz2 = _mm_setzero_ps();
723 fix3 = _mm_setzero_ps();
724 fiy3 = _mm_setzero_ps();
725 fiz3 = _mm_setzero_ps();
727 /* Start inner kernel loop */
728 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
731 /* Get j neighbor index, and coordinate index */
736 j_coord_offsetA = DIM*jnrA;
737 j_coord_offsetB = DIM*jnrB;
738 j_coord_offsetC = DIM*jnrC;
739 j_coord_offsetD = DIM*jnrD;
741 /* load j atom coordinates */
742 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
743 x+j_coord_offsetC,x+j_coord_offsetD,
746 /* Calculate displacement vector */
747 dx10 = _mm_sub_ps(ix1,jx0);
748 dy10 = _mm_sub_ps(iy1,jy0);
749 dz10 = _mm_sub_ps(iz1,jz0);
750 dx20 = _mm_sub_ps(ix2,jx0);
751 dy20 = _mm_sub_ps(iy2,jy0);
752 dz20 = _mm_sub_ps(iz2,jz0);
753 dx30 = _mm_sub_ps(ix3,jx0);
754 dy30 = _mm_sub_ps(iy3,jy0);
755 dz30 = _mm_sub_ps(iz3,jz0);
757 /* Calculate squared distance and things based on it */
758 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
759 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
760 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
762 rinv10 = gmx_mm_invsqrt_ps(rsq10);
763 rinv20 = gmx_mm_invsqrt_ps(rsq20);
764 rinv30 = gmx_mm_invsqrt_ps(rsq30);
766 /* Load parameters for j particles */
767 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
768 charge+jnrC+0,charge+jnrD+0);
770 fjx0 = _mm_setzero_ps();
771 fjy0 = _mm_setzero_ps();
772 fjz0 = _mm_setzero_ps();
774 /**************************
775 * CALCULATE INTERACTIONS *
776 **************************/
778 r10 = _mm_mul_ps(rsq10,rinv10);
780 /* Compute parameters for interactions between i and j atoms */
781 qq10 = _mm_mul_ps(iq1,jq0);
783 /* Calculate table index by multiplying r with table scale and truncate to integer */
784 rt = _mm_mul_ps(r10,vftabscale);
785 vfitab = _mm_cvttps_epi32(rt);
787 vfeps = _mm_frcz_ps(rt);
789 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
791 twovfeps = _mm_add_ps(vfeps,vfeps);
792 vfitab = _mm_slli_epi32(vfitab,2);
794 /* CUBIC SPLINE TABLE ELECTROSTATICS */
795 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
796 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
797 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
798 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
799 _MM_TRANSPOSE4_PS(Y,F,G,H);
800 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
801 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
802 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
806 /* Update vectorial force */
807 fix1 = _mm_macc_ps(dx10,fscal,fix1);
808 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
809 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
811 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
812 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
813 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
815 /**************************
816 * CALCULATE INTERACTIONS *
817 **************************/
819 r20 = _mm_mul_ps(rsq20,rinv20);
821 /* Compute parameters for interactions between i and j atoms */
822 qq20 = _mm_mul_ps(iq2,jq0);
824 /* Calculate table index by multiplying r with table scale and truncate to integer */
825 rt = _mm_mul_ps(r20,vftabscale);
826 vfitab = _mm_cvttps_epi32(rt);
828 vfeps = _mm_frcz_ps(rt);
830 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
832 twovfeps = _mm_add_ps(vfeps,vfeps);
833 vfitab = _mm_slli_epi32(vfitab,2);
835 /* CUBIC SPLINE TABLE ELECTROSTATICS */
836 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
837 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
838 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
839 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
840 _MM_TRANSPOSE4_PS(Y,F,G,H);
841 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
842 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
843 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
847 /* Update vectorial force */
848 fix2 = _mm_macc_ps(dx20,fscal,fix2);
849 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
850 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
852 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
853 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
854 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
856 /**************************
857 * CALCULATE INTERACTIONS *
858 **************************/
860 r30 = _mm_mul_ps(rsq30,rinv30);
862 /* Compute parameters for interactions between i and j atoms */
863 qq30 = _mm_mul_ps(iq3,jq0);
865 /* Calculate table index by multiplying r with table scale and truncate to integer */
866 rt = _mm_mul_ps(r30,vftabscale);
867 vfitab = _mm_cvttps_epi32(rt);
869 vfeps = _mm_frcz_ps(rt);
871 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
873 twovfeps = _mm_add_ps(vfeps,vfeps);
874 vfitab = _mm_slli_epi32(vfitab,2);
876 /* CUBIC SPLINE TABLE ELECTROSTATICS */
877 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
878 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
879 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
880 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
881 _MM_TRANSPOSE4_PS(Y,F,G,H);
882 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
883 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
884 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
888 /* Update vectorial force */
889 fix3 = _mm_macc_ps(dx30,fscal,fix3);
890 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
891 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
893 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
894 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
895 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
897 fjptrA = f+j_coord_offsetA;
898 fjptrB = f+j_coord_offsetB;
899 fjptrC = f+j_coord_offsetC;
900 fjptrD = f+j_coord_offsetD;
902 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
904 /* Inner loop uses 126 flops */
910 /* Get j neighbor index, and coordinate index */
911 jnrlistA = jjnr[jidx];
912 jnrlistB = jjnr[jidx+1];
913 jnrlistC = jjnr[jidx+2];
914 jnrlistD = jjnr[jidx+3];
915 /* Sign of each element will be negative for non-real atoms.
916 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
917 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
919 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
920 jnrA = (jnrlistA>=0) ? jnrlistA : 0;
921 jnrB = (jnrlistB>=0) ? jnrlistB : 0;
922 jnrC = (jnrlistC>=0) ? jnrlistC : 0;
923 jnrD = (jnrlistD>=0) ? jnrlistD : 0;
924 j_coord_offsetA = DIM*jnrA;
925 j_coord_offsetB = DIM*jnrB;
926 j_coord_offsetC = DIM*jnrC;
927 j_coord_offsetD = DIM*jnrD;
929 /* load j atom coordinates */
930 gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
931 x+j_coord_offsetC,x+j_coord_offsetD,
934 /* Calculate displacement vector */
935 dx10 = _mm_sub_ps(ix1,jx0);
936 dy10 = _mm_sub_ps(iy1,jy0);
937 dz10 = _mm_sub_ps(iz1,jz0);
938 dx20 = _mm_sub_ps(ix2,jx0);
939 dy20 = _mm_sub_ps(iy2,jy0);
940 dz20 = _mm_sub_ps(iz2,jz0);
941 dx30 = _mm_sub_ps(ix3,jx0);
942 dy30 = _mm_sub_ps(iy3,jy0);
943 dz30 = _mm_sub_ps(iz3,jz0);
945 /* Calculate squared distance and things based on it */
946 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
947 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
948 rsq30 = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
950 rinv10 = gmx_mm_invsqrt_ps(rsq10);
951 rinv20 = gmx_mm_invsqrt_ps(rsq20);
952 rinv30 = gmx_mm_invsqrt_ps(rsq30);
954 /* Load parameters for j particles */
955 jq0 = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
956 charge+jnrC+0,charge+jnrD+0);
958 fjx0 = _mm_setzero_ps();
959 fjy0 = _mm_setzero_ps();
960 fjz0 = _mm_setzero_ps();
962 /**************************
963 * CALCULATE INTERACTIONS *
964 **************************/
966 r10 = _mm_mul_ps(rsq10,rinv10);
967 r10 = _mm_andnot_ps(dummy_mask,r10);
969 /* Compute parameters for interactions between i and j atoms */
970 qq10 = _mm_mul_ps(iq1,jq0);
972 /* Calculate table index by multiplying r with table scale and truncate to integer */
973 rt = _mm_mul_ps(r10,vftabscale);
974 vfitab = _mm_cvttps_epi32(rt);
976 vfeps = _mm_frcz_ps(rt);
978 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
980 twovfeps = _mm_add_ps(vfeps,vfeps);
981 vfitab = _mm_slli_epi32(vfitab,2);
983 /* CUBIC SPLINE TABLE ELECTROSTATICS */
984 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
985 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
986 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
987 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
988 _MM_TRANSPOSE4_PS(Y,F,G,H);
989 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
990 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
991 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
995 fscal = _mm_andnot_ps(dummy_mask,fscal);
997 /* Update vectorial force */
998 fix1 = _mm_macc_ps(dx10,fscal,fix1);
999 fiy1 = _mm_macc_ps(dy10,fscal,fiy1);
1000 fiz1 = _mm_macc_ps(dz10,fscal,fiz1);
1002 fjx0 = _mm_macc_ps(dx10,fscal,fjx0);
1003 fjy0 = _mm_macc_ps(dy10,fscal,fjy0);
1004 fjz0 = _mm_macc_ps(dz10,fscal,fjz0);
1006 /**************************
1007 * CALCULATE INTERACTIONS *
1008 **************************/
1010 r20 = _mm_mul_ps(rsq20,rinv20);
1011 r20 = _mm_andnot_ps(dummy_mask,r20);
1013 /* Compute parameters for interactions between i and j atoms */
1014 qq20 = _mm_mul_ps(iq2,jq0);
1016 /* Calculate table index by multiplying r with table scale and truncate to integer */
1017 rt = _mm_mul_ps(r20,vftabscale);
1018 vfitab = _mm_cvttps_epi32(rt);
1020 vfeps = _mm_frcz_ps(rt);
1022 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1024 twovfeps = _mm_add_ps(vfeps,vfeps);
1025 vfitab = _mm_slli_epi32(vfitab,2);
1027 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1028 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1029 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1030 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1031 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1032 _MM_TRANSPOSE4_PS(Y,F,G,H);
1033 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1034 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1035 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1039 fscal = _mm_andnot_ps(dummy_mask,fscal);
1041 /* Update vectorial force */
1042 fix2 = _mm_macc_ps(dx20,fscal,fix2);
1043 fiy2 = _mm_macc_ps(dy20,fscal,fiy2);
1044 fiz2 = _mm_macc_ps(dz20,fscal,fiz2);
1046 fjx0 = _mm_macc_ps(dx20,fscal,fjx0);
1047 fjy0 = _mm_macc_ps(dy20,fscal,fjy0);
1048 fjz0 = _mm_macc_ps(dz20,fscal,fjz0);
1050 /**************************
1051 * CALCULATE INTERACTIONS *
1052 **************************/
1054 r30 = _mm_mul_ps(rsq30,rinv30);
1055 r30 = _mm_andnot_ps(dummy_mask,r30);
1057 /* Compute parameters for interactions between i and j atoms */
1058 qq30 = _mm_mul_ps(iq3,jq0);
1060 /* Calculate table index by multiplying r with table scale and truncate to integer */
1061 rt = _mm_mul_ps(r30,vftabscale);
1062 vfitab = _mm_cvttps_epi32(rt);
1064 vfeps = _mm_frcz_ps(rt);
1066 vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1068 twovfeps = _mm_add_ps(vfeps,vfeps);
1069 vfitab = _mm_slli_epi32(vfitab,2);
1071 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1072 Y = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1073 F = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1074 G = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1075 H = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1076 _MM_TRANSPOSE4_PS(Y,F,G,H);
1077 Fp = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1078 FF = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1079 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1083 fscal = _mm_andnot_ps(dummy_mask,fscal);
1085 /* Update vectorial force */
1086 fix3 = _mm_macc_ps(dx30,fscal,fix3);
1087 fiy3 = _mm_macc_ps(dy30,fscal,fiy3);
1088 fiz3 = _mm_macc_ps(dz30,fscal,fiz3);
1090 fjx0 = _mm_macc_ps(dx30,fscal,fjx0);
1091 fjy0 = _mm_macc_ps(dy30,fscal,fjy0);
1092 fjz0 = _mm_macc_ps(dz30,fscal,fjz0);
1094 fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1095 fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1096 fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1097 fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1099 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1101 /* Inner loop uses 129 flops */
1104 /* End of innermost loop */
1106 gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1107 f+i_coord_offset+DIM,fshift+i_shift_offset);
1109 /* Increment number of inner iterations */
1110 inneriter += j_index_end - j_index_start;
1112 /* Outer loop uses 18 flops */
1115 /* Increment number of outer iterations */
1118 /* Update outer/inner flops */
1120 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*129);